A Structured Review and Critical Synthesis of Multi-Criteria Decision- Making Models Integrated with Machine Learning for Water Resource Management

A Structured Review and Critical Synthesis of Multi-Criteria Decision- Making Models Integrated with Machine Learning for Water Resource Management | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Systematic Review A Structured Review and Critical Synthesis of Multi-Criteria Decision- Making Models Integrated with Machine Learning for Water Resource Management Murphy Bonkogia Lomboli, Opeyeolu Timothy Laseinde This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9145184/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 17 You are reading this latest preprint version Abstract Water resource management (WRM) is increasingly shaped by interconnected environmental, technical, social, and institutional pressures that make single-method analysis insufficient for robust decision support. The study follows a structured review methodology with systematic search and screening, combined with critical thematic synthesis to identify and synthesise studies on integrated MCDM-ML applications in WRM, with emphasis on conceptual integration patterns, domain-specific applications, methodological trade-offs, reporting gaps, and future implementation directions. Critically examining how multi-criteria decision-making (MCDM) models and machine learning (ML) techniques are being integrated to support more effective, transparent and context-appropriate WRM. The review identifies three broad integration patterns, namely sequential, parallel, and fully coupled frameworks. Across groundwater potential and recharge assessment, water demand forecasting and supply planning, flood risk and hydrological hazard management, water quality assessment and pollution control, sediment and catchment management, and urban water loss management, a consistent pattern emerges. ML contributes most strongly to prediction, classification, and pattern detection, while MCDM strengthens criteria weighting, prioritisation, and final decision structuring. The review also shows that although integrated systems often report strong case-specific results, cross-study comparison remains limited by inconsistent performance metrics, uneven validation procedures, weak transparency in weighting structures, limited uncertainty treatment, and poor reproducibility. In response, the paper proposes a minimum reporting framework and highlights key future directions, including explainable and trustworthy AI, real-time and IoT-enabled decision support, cloud and edge deployment, and policy-aligned stakeholder-centred systems. Overall, integrated MCDM-ML approaches show strong potential for WRM, but their future value will depend on clearer reporting, stronger validation, and closer alignment with real decision contexts. Water resource management multi-criteria decision-making machine learning decision support systems systematic literature review explainable artificial intelligence Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1. Introduction 1.1 Water resource management as a multi-dimensional decision problem Water resource management (WRM) has increasingly become complex due to the combined effects of population growth, urbanisation, industrial expansion, climate variability, land-use change as well as rising pressure on finite freshwater systems (He et al. 2020 ; Li et al. 2022 ). These pressures do not affect water systems in isolation, they fundamentally influence other major factors such as water availability, quality, allocation, infrastructure planning, ecosystem protection and long-term sustainability at the same time. As a result, WRM is no longer a matter of managing supply alone, but of balancing environmental, technical, social, economic and institutional considerations under conditions of uncertainty. Decision-makers are often therefore required to assess multiple and often conflicting criteria while responding to dynamic hydrological conditions as well as context-specific management priorities (Miller and Belton 2014 ). In practice, WRM decisions often involve trade-offs between short-term operational demands and long-term sustainability goals. For example, water allocation strategies may improve immediate service delivery while increasing ecological stress, and infrastructure interventions may strengthen supply reliability while introducing substantial cost and governance challenges (Zhu et al. 2019 ; Poli et al. 2024 ). These trade-offs are further complicated by incomplete data, spatial and temporal variability as well as the need to incorporate both quantitative evidence and expert judgment into decision processes. Because of this, WRM is best understood as a multi-dimensional decision environment in which robust planning depends on methods that can capture complexity, uncertainty and competing priorities in a structured and defensible way. 1.2 Why MCDM and ML are increasingly combined Multi-Criteria Decision-Making (MCDM) and Machine Learning (ML) have emerged as two important methodological approaches for addressing the complexity of WRM, but they contribute in different ways. MCDM methods are well suited to decision contexts where multiple criteria must be evaluated simultaneously, especially when factors such as stakeholder preferences, weighting of alternatives and transparent prioritisation are central to the problem (Kumar 2025 ; Kocaman and Asan 2025 ). In contrast, ML methods are valuable for and in identifying hidden patterns within large datasets, forecasting future conditions, classifying system states and supporting adaptive decision-making in data-rich and uncertain environments. Their growing use in WRM reflects the need for methods that can move beyond conventional deterministic analysis and respond to increasingly complex water-related challenges (Drogkoula et al. 2023 ). The increasing integration of MCDM and ML is driven by their complementary strengths. ML can improve predictive capability by modelling non-linear relationships in hydrological, environmental, and demand-related datasets, while MCDM can translate those predictive outputs into structured decision support that reflects practical priorities and competing objectives. In this sense, ML helps answer what is likely to happen, while MCDM helps in evaluating what should be prioritised once that information is available (Schuwirth et al. 2018 ; Mohammadifar et al. 2023 ). This combination is particularly relevant in WRM applications such as groundwater potential mapping, flood risk assessment, water demand forecasting, water quality evaluation and infrastructure planning, where technical performance alone is not enough and decisions must also remain interpretable, actionable, and context sensitive, a summary of these is provided in Fig. 1 . 1.3 Gap in the existing review literature Although the application of MCDM and ML in WRM has grown significantly over time, the review literature has not always kept pace with the way these methods are now being used together. Existing discussions often focus on MCDM methods or ML models separately, in other cases they summarise applications descriptively without critically examining how integrated approaches are structured, where they are most effective, what limitations they share and how comparable their results really are across studies (Hajkowicz and Collins 2007 ; Calizaya et al. 2010 ). As a result, the literature still lacks a sufficiently focused synthesis of how MCDM-ML combinations are being applied across WRM domains and what methodological patterns can be identified from that body of work. A further gap lies in the limited critical evaluation of performance reporting, interpretability, transferability and reproducibility in existing studies. Many published applications report strong results, but they often do so by using different datasets, validation strategies, criteria structures and performance metrics, which makes direct comparison difficult (Ghobadi and Kang 2023 ; Ahmed et al. 2024 ). In addition, there is limited consistency in how studies explain the interaction between predictive modelling and decision-ranking processes. This weakens the ability of researchers and practitioners to determine which integrated approaches are genuinely robust, which are context-dependent and which are difficult to generalise beyond individual case studies (Ali et al. 2023 ; Costa et al. 2024 ; Drogkoula et al. 2025 ). These limitations create a clear need for a review that goes beyond method description and instead offers structured thematic synthesis, critical comparison and clearer guidance for future research and application 1.4 Aim and review questions This review aims to critically examine how MCDM and ML have been integrated in WRM and to evaluate the extent to which these integrated approaches support more effective, transparent and context-appropriate decision-making. Rather than treating MCDM and ML as separate methodological streams, the review considers their joint use across major WRM application areas and assesses the strengths, limitations and practical implications of different combinations. The article is therefore positioned not only as a summary of existing studies, but as a synthesis of the conceptual patterns, application trends, methodological trade-offs as well as reporting gaps that currently shape the field. To guide this review, four main questions are addressed: How are MCDM and ML being integrated in water resource management? Which combinations of MCDM and ML are most commonly applied across different WRM domains, and for what reasons? What trade-offs emerge among predictive performance, interpretability, data requirements, scalability, and contextual suitability? What methodological and reporting gaps continue to limit comparability, reproducibility and broader implementation of integrated MCDM-ML approaches in WRM? These questions provide the foundation for the sections that follow as summarized in Table 1 , including the review design, conceptual integration framework, thematic synthesis, critical comparison, and the proposed directions for future research and implementation. Table 1 Review questions summary Review question Focus Section(s) addressing it Type of synthesis How are MCDM and ML being integrated in water resource management? Integration logic, roles of MCDM and ML, and major integration patterns Section 3 Conceptual synthesis Which combinations of MCDM and ML are most commonly applied across WRM domains, and for what reasons? Domain-specific application patterns and commonly used method combinations Section 4 Thematic synthesis What trade-offs emerge among predictive performance, interpretability, data requirements, scalability, and contextual suitability? Strengths, limitations, and comparative performance of integrated approaches Section 5 Critical comparative analysis What methodological and reporting gaps continue to limit comparability, reproducibility, and broader implementation? Metric inconsistency, validation gaps, reproducibility, and framework development Sections 6 and 7 Gap analysis and future-oriented synthesis 2. Review design and literature selection 2.1 Review approach This study adopts a structured and systematic review approach combined with critical synthesis to examine the integration of Multi-Criteria Decision-Making (MCDM) and Machine Learning (ML) in water resource management. While systematic search and screening procedures were applied to ensure transparency and relevance, the review is positioned as a critical thematic synthesis rather than a fully PRISMA-compliant systematic review. This was important because the objective of the study was not only to summarise existing work, but also to critically compare how integrated MCDM-ML approaches have been applied across different WRM domains, how they have been evaluated, and where important methodological gaps remain (Snyder 2019 ; Page et al. 2021a ). Rather than treating the review as a purely narrative survey, the study followed a structured evidence-synthesis logic in which the review questions, eligibility criteria, screening steps, appraisal approach and synthesis categories were defined in advance. This approach made it possible to organise the literature around explicit themes such as integration patterns, WRM application domains, interpretability, scalability, and reporting quality, which is more consistent with a critical review than a descriptive listing of studies. The review process follows key principles of systematic literature reviews, including defined search strategies, eligibility criteria, and structured synthesis; however, a formal PRISMA flow diagram and quantitative study selection reporting were not included, as the focus of this work is on conceptual integration, thematic synthesis and critical comparison across studies. 2.2 Search sources and search strategy The literature search was conducted across five major academic databases namely: Scopus, Web of Science, Google Scholar, IEEE Xplore and SpringerLink. These sources were selected because together they provide broad coverage of peer-reviewed work in water resource management, environmental modelling, artificial intelligence, machine learning as well as decision-support research. The search strategy was designed to retrieve studies that explicitly addressed the integration of MCDM and ML rather than studies focused on either method family in isolation. A combination of keywords and Boolean operators was used to identify relevant studies. The main search terms included “Multi-Criteria Decision-Making” or “MCDM”, “Machine Learning” or “Artificial Intelligence”, “Water Resource Management” or “Water Availability” and “Integration” or “Combined Approach”. These terms were combined using AND/OR operators to improve search sensitivity while retaining topical relevance. Searches were limited to studies published in English, and the formal study selection for the reviewed articles covered literature up to 2024, while additional supporting references published up to 2025 were used where relevant to strengthen the conceptual framing, emerging directions, and discussion. A representative search string used across databases was: (‘Multi-Criteria Decision-Making’ OR MCDM) AND (‘Machine Learning’ OR ‘Artificial Intelligence’) AND (‘Water Resource Management’ OR ‘Water Availability’) AND (‘Integration’ OR ‘Combined Approach’). 2.3 Eligibility criteria To maintain relevance and comparability, the review applied explicit inclusion and exclusion criteria during study selection. Studies were included if they addressed water resource management or a closely related water-sector application, explicitly integrated at least one MCDM method with at least one ML technique, reported a clear methodological workflow and presented sufficient information on model application, outputs or performance evaluation. In contrast, studies were excluded if they used MCDM or ML as standalone approaches without integration, fell outside the WRM context, lacked sufficient methodological detail, or were not peer-reviewed academic contributions. The eligibility process was also guided by the need to preserve the analytical focus of the review. Since the manuscript is concerned with the comparative value of integrated decision-support systems, studies were screened not only for topical relevance but also for their ability to contribute to cross-study synthesis in areas such as integration logic, data requirements, interpretability, and reported strengths or limitations. This helped avoid an overly broad review corpus and ensured that the final set of studies remained aligned with the core review questions. Table 2 Inclusion and exclusion criteria applied in study selection Criterion Inclusion Exclusion Study focus Explicit integration of at least one MCDM method with at least one ML technique Use of MCDM only or ML only without integrated application Application area Water resource management or a directly related water-sector decision/problem Studies outside WRM or only loosely related to water issues Publication type Peer-reviewed journal articles and conference papers with sufficient methodological detail Editorials, opinion pieces, theses without accessible review status, and non-scholarly material Methodological clarity Clear description of workflow, inputs, methods, and outputs Insufficient methodological detail to support comparison Performance or decision output Reports model outputs, ranking outcomes, validation results, or comparable analytical findings No usable results, no evaluation basis, or purely conceptual discussion without application Language and time frame English-language studies published up to 2024 Non-English studies and material outside the defined search period 2.4 Screening and study selection The screening and study selection process followed the PRISMA logic of identification, screening, eligibility assessment and final inclusion. The records retrieved from the selected databases were first screened by title and abstract to remove clearly irrelevant studies and duplicates. Full-text screening was then conducted on the remaining records using the predefined inclusion and exclusion criteria (Page et al. 2021b ; Sohrabi et al. 2021 ). This staged approach improved consistency in study selection and provided a clear basis for documenting how the final review dataset was assembled. At the full-text stage, particular attention was given to whether a study genuinely presented an integrated MCDM-ML workflow, rather than simply discussing both methods in the same paper. This distinction was particularly important because several water-related studies apply predictive ML models and decision tools in parallel without formally linking them in a single analytical framework. Only those studies that contributed meaningfully to the review’s comparative objective were retained for the final synthesis. 2.5 Data extraction and synthesis framework A structured extraction template was used to record comparable information from each included study. The extracted items included the WRM application area, study objective, MCDM technique used, ML technique used, mode of integration, type of data employed, performance metrics reported, validation approach, interpretability features and the main strengths as well as limitations identified in the study. Recording these categories consistently was necessary for the later thematic and comparative synthesis developed in Sections 4 to 6. The synthesis framework was designed to go beyond summary description. Instead of reviewing each article in isolation, extracted studies were grouped according to WRM application domain and then compared using recurring analytical dimensions such as predictive performance, transparency, weighting sensitivity, computational demand, data intensity and transferability (Snyder 2019 ). This structure supports the critical orientation of the manuscript and directly addresses the need for thematic rather than purely procedural synthesis. 2.6 Quality appraisal and risk-of-bias considerations To strengthen methodological consistency, the included studies were appraised using an adapted Critical Appraisal Skills Programme (CASP) logic (Schabbauer 2012 ; Kolaski et al. 2023 ). The appraisal focused on whether each study had a clear objective, an appropriate design, transparent data and modelling procedures, a justifiable integration workflow, sufficient reporting of outputs or performance, and a discussion of limitations relevant to interpretation and use. Although CASP was originally developed for broader evidence appraisal, its structured question-based approach provided a practical basis for judging methodological soundness and risk of bias across heterogeneous WRM studies. Risk of bias was considered in relation to common weaknesses observed in the reviewed literature, including incomplete reporting of data sources, unclear validation procedures, insufficient explanation of criteria weighting, weak justification for method selection and limited treatment of uncertainty (Long et al. 2020 ; Shaheen et al. 2023 ). These issues do not necessarily invalidate the studies, but they do affect how confidently their findings can be compared and generalised. For that reason, appraisal in this review was used not to exclude studies mechanically, but to support a more careful interpretation of the evidence base. 2.7 Limitations of the review process Although the review was designed to be systematic and transparent, several limitations should be acknowledged. First, the search was restricted to selected academic databases and English-language publications, which may have excluded relevant studies published in other languages or in less visible outlets. Secondly, the emphasis on peer-reviewed literature improved quality control but may have reduced visibility of technical reports, policy documents and grey literature that could contain useful implementation insights, especially in developing-country WRM contexts. Thirdly, despite the use of structured screening and appraisal procedures, some degree of interpretive judgement remained necessary during eligibility assessment and comparative synthesis. A further limitation arises from the heterogeneity of the reviewed studies themselves. Differences in datasets, performance metrics, validation strategies, decision criteria and reporting depth constrained the extent to which studies could be compared on a fully standardised basis. This is not only a limitation of the present review, but also an indication of a broader reporting problem in the MCDM-ML literature for WRM, which is addressed later in Section 6. 3. Conceptual framework for MCDM-ML integration in water resource management 3.1 Roles of MCDM in integrated systems In integrated water resource management systems, MCDM plays an important role in structuring decisions that involve multiple, and often conflicting, objectives. Water management decisions are rarely guided by a single consideration. They often involve several factors such as economic, environmental, technical and social factors that must be assessed together. In this context, MCDM provides a practical and systematic way of identifying relevant criteria, assigning their relative importance, and evaluating alternatives in a transparent manner. This makes it particularly useful in WRM, where the quality of a decision depends not only on technical evidence, but also on how competing priorities are balanced (Mendoza and Martins 2006 ). Another important role of MCDM in integrated systems is that it helps convert technical outputs into decision-ready results. In many WRM applications, modelling and forecasting tools generate large amounts of information, but those outputs still need to be interpreted in a way that supports real planning choices. MCDM therefore helps organise that process by comparing alternatives against selected criteria and producing a structured basis for ranking or prioritisation. This is especially useful in water resource planning, where decision-makers often need to choose between competing interventions, management scenarios, or infrastructure options under uncertain conditions (Lai et al. 2008 ). In integrated MCDM-ML systems, MCDM often functions as the decision layer that receives and interprets outputs from predictive or simulation-based models. In other words, ML may be used to estimate future conditions, classify risks, or predict performance, while MCDM is used to determine which alternative should be preferred once those results are available. This role is valuable because predictive performance on its own does not resolve the wider decision problem. Water resource decisions still require trade-offs, judgment, and preference-based evaluation. In basin-scale water planning, for example, MCDM has been used to combine environmental, social and economic criteria in order to support more balanced and defensible decisions (Calizaya et al. 2010 ). MCDM also strengthens integrated systems by improving transparency and making the final decision process easier to explain. Compared with purely data-driven approaches, MCDM makes it possible to show how criteria were selected, how alternatives were assessed, and why one option performed better than another. This is important in WRM because many decisions affect public systems, long-term planning, and policy implementation. For this reason, the role of MCDM in integrated systems as summarised in Fig. 2 , is not only analytical, but also interpretive and practical. It helps ensure that technically informed decisions remain understandable, defensible, and aligned with broader management objectives (Xi and Poh 2015 ). Across the reviewed literature, MCDM is most frequently used as the final decision-structuring layer rather than as a predictive tool. However, its effectiveness depends heavily on how criteria are selected and weighted, which introduces subjectivity into otherwise data-driven workflows. This creates a recurring trade-off in integrated systems: while MCDM improves transparency and interpretability, it can also introduce variability in outcomes when weighting structures are not robustly justified or validated. 3.2 Roles of ML in integrated systems Machine Learning (ML) primarily contributes to integrated systems for water resource management by identifying patterns in complicated datasets and producing predictions that are challenging to acquire through traditional analytical techniques alone. Several interrelated factors such as rainfall, temperature, streamflow, land use, groundwater behaviour, water demand and water quality conditions, have an impact on water systems. ML is helpful for finding patterns, predicting future conditions, and strengthening the analytical foundation upon which management decisions are made because these interactions are frequently nonlinear and dynamic. By transforming massive and diverse water-related datasets into data that may aid in planning, risk assessment, and operational response, machine learning (ML) thereby enhances integrated systems (Ahmed et al. 2024 ). One of the roles of ML in integrated systems is its ability to support classification, clustering, anomaly detection as well as other forms of data-driven interpretation that go beyond simple prediction. In WRM, this is important because decision support often depends on recognising system states, identifying high-risk areas or distinguishing between management conditions that are not immediately visible through descriptive statistics alone (Ghobadi and Kang 2023 ). Reviews of ML in WRM have shown that its applications commonly extend across three broad functions, namely prediction, clustering and reinforcement-based optimisation, which makes it a flexible component within integrated decision-support systems. ML also plays an important role in improving responsiveness and adaptability in integrated systems. Because many water-related problems evolve over time including drought conditions, water demand shifts and water quality deterioration, decision support cannot rely only on static analysis. ML models can be trained on historical and real-time data to update forecasts, detect changes early, and improve situational awareness under changing environmental conditions (Liu et al. 2024 ). This makes them particularly useful in integrated systems where the purpose is not only to understand present conditions, but also to support forward-looking and adaptive management. At the same time, the role of ML in integrated systems is not limited to its predictive strength. It also affects how useful the system becomes in practice. Recent work in hydroclimatic and water-related modelling has shown that interpretability and explainability are becoming increasingly important, especially where model outputs are expected to inform public decisions or policy-facing applications (Başağaoğlu et al. 2022 ). This means that, within integrated MCDM-ML systems, the value of ML lies not only in its ability to improve estimation accuracy, but also in its ability to produce outputs that can be meaningfully incorporated into transparent and defensible decision processes, a standard procedure in shown in Fig. 3 . However, across the reviewed WRM studies, the strength of ML in prediction does not automatically translate into improved decision-making. In many cases, high-performing models are used without sufficient explanation of how their outputs influence final decisions. This highlights a key limitation: predictive accuracy alone is insufficient unless it is explicitly linked to structured decision frameworks such as MCDM. 3.3 Main integration patterns The integration of MCDM and ML in water resource management does not follow a single fixed structure. Across the literature, the two methods are combined in different ways depending on the decision problem, the type of data available, and the role expected from each method in the overall workflow. In most cases, the integration is not simply about combining two techniques for novelty (Gebre et al. 2021 ; Ali et al. 2023 ). It is done because ML and MCDM solve different parts of the problem. ML is generally used to learn from data, detect patterns, or generate predictions, while MCDM is used to evaluate alternatives, weigh criteria, and support final prioritisation. For that reason, the structure of integration usually reflects the order in which prediction and decision support are needed in practice. One of the most common patterns is sequential integration. In this arrangement, ML is applied first to produce outputs such as forecasts, classifications, susceptibility estimates, or scenario results, and these outputs are then passed into an MCDM framework for ranking or decision-making. This pattern is common because it separates analytical prediction from final prioritisation in a clear and practical way. In water-related applications, this is especially useful where the first task is to estimate future conditions or identify suitable zones, and the second task is to choose among alternatives using environmental, economic, or policy-related criteria. A recent systematic review of integrated multicriteria and data-driven methods also shows that many hybrid systems are built around this staged logic, where one method produces the informational base and the other supports the final decision layer (Costa et al. 2024 ; Nasiri Khiavi et al. 2025 ). A second pattern is parallel integration, where MCDM and ML are applied alongside each other rather than in a strict sequence. In this case, the two methods may be used to analyse the same problem from different angles, after which their outputs are compared, combined, or jointly interpreted. This pattern is often used where researchers want to evaluate whether data-driven and criteria-based methods lead to similar conclusions, or where each method captures a different dimension of the problem. In environmental mapping studies, for example, MCDM may be used to represent expert-based weighting while ML is used to test predictive performance from data, allowing both perspectives to inform the final assessment (He et al. 2020 ; Khalil et al. 2022 ). A third pattern is fully coupled or hybrid integration, where MCDM and ML are embedded within a single decision-support framework rather than treated as separate stages. In such systems, the interaction between the two methods is tighter. MCDM outputs may be used to support feature weighting, rule development, or optimisation, while ML outputs may directly shape the criteria structure, scenario evaluation, or final recommendation process. This type of integration is usually more complex, but it can be more useful in situations where prediction, optimisation, and decision support must interact continuously rather than sequentially. Water demand planning studies that combine intelligent forecasting with multi-criteria optimisation illustrate this more tightly linked form of integration (Sharma 2022 ; Poursaeid 2025 ). Taken together, these patterns show that integration is best understood as a design choice rather than a standard formula. Sequential integration tends to be clearer and easier to interpret. Parallel integration is useful where comparison and triangulation are important. Fully coupled integration is more ambitious and can be more powerful, but it usually demands stronger methodological justification and clearer reporting. For this reason, understanding the main integration patterns is important not only for describing the literature, but also for evaluating which integrated systems are actually suitable for different WRM contexts. Figure 4 summarises the main ways in which MCDM and ML are combined in WRM, namely sequential, parallel, and fully coupled integration patterns. Despite these distinctions, the literature does not consistently demonstrate that more complex integration leads to better decision outcomes, as fully coupled systems often lack transparency and reproducibility compared to simpler sequential approaches. 4. Thematic synthesis by WRM application domain 4.1 Groundwater potential and recharge assessment Groundwater potential and recharge assessment is one of the strongest application areas for integrated geospatial, MCDM, and ML approaches in WRM. A study conducted by (Kanji and Das 2025a ) in the Kangsabati Upper Catchment showed how this integration can move beyond simple potential mapping toward recharge feasibility assessment. Their study combined XGBoost with AHP and considered both hydrogeological conditions and water quality constituents. The results showed that XGBoost outperformed SVM, RF, and ANN, achieving 81% accuracy, while the integrated workflow also identified suitable artificial recharge zones and proposed recharge structures for different lithological groups. This is important because it shows that integrated systems can support both resource identification and management action, rather than stopping at spatial classification alone. Another study conducted by (Ghosh et al. 2022 ) in the Kangsabati River basin (shown in Fig. 5 ) illustrates the continued relevance of MCDM-based groundwater zoning where structured weighting of conditioning factors is required. Using MCDA and AHP, the authors integrated geology, geomorphology, elevation, slope, drainage, lineament, curvature, topographic wetness, land use/land cover, and soil to delineate groundwater potential zones. Their results showed that 14.62% of the basin fell within high potential zones and 6.11% within very high potential zones. More importantly, the mapped zones were checked against pre-monsoon and post-monsoon groundwater depth data and ROC-based validation, which strengthened confidence in the applicability of the method. This study shows that MCDM remains particularly useful where the main goal is transparent spatial prioritisation grounded in hydrogeological reasoning. Adapted from (Ghosh et al. 2022 ). Assessment of groundwater potential zone using MCDA and AHP techniques: case study from a tropical river basin of India. Applied Water Science, 12:37, under the Creative Commons Attribution (CC BY 4.0) license. (Anand et al. 2025 ) shows a more recent shift toward ensemble ML models for groundwater potential assessment, especially in data-rich and urban settings. Their study using the methodology in Fig. 6 , compared six ML models and found that Random Forest was the most reliable, with an AUC of 0.91, while AdaBoost achieved the highest MCC. The sensitivity analysis further showed that geomorphology, elevation, and rainfall were among the most influential predictors. A key observation is that Which suggests that the strongest direction Which suggests that the strongest directionthe study did not just compare models by accuracy but also examined how changes in specific input layers affected prediction outcomes. This gives the assessment more interpretive value and suggests that recent ML-based groundwater studies are increasingly moving toward stronger model diagnostics instead of relying only on final performance scores. Adapted from (Anand et al. 2025 ). Evaluating groundwater potential with the synergistic use of geospatial methods and advanced machine learning approaches. Discover Cities, 2:56, under the Creative Commons Attribution (CC BY 4.0) license. (Akbari et al. 2021 ) adds another useful dimension by comparing different MCDM weighting frameworks for groundwater recharge potential mapping. Using AHP, BWM, and FUCOM, the authors showed that groundwater recharge zones can be highly sensitive to how weights are assigned to conditioning factors. Across their models, land use and lithology emerged as the most important variables, while slope angle had the lowest influence. Their findings suggest that recharge assessment is not only a mapping problem, but also a weighting problem, because different decision structures can influence the final spatial pattern even when the same environmental layers are used. This is important for the present review because it highlights that, in groundwater studies, methodological transparency is just as important as predictive performance. Taken together, these studies show a clear pattern in the groundwater literature. MCDM-based approaches remain strong where transparent weighting, hydrogeological reasoning, and practical prioritisation are central. ML-based approaches are increasingly preferred where the data structure is more complex and higher predictive discrimination is required. The most useful direction, however, appears to be integration. Studies that combine predictive ML with structured MCDM are better placed to support groundwater management decisions that need both analytical strength and practical interpretability. This is especially important for recharge assessment, where the objective is not only to identify where groundwater potential exists, but also where intervention is feasible and justifiable. Table 3 summarises the main groundwater potential and recharge assessment studies discussed in this subsection, with emphasis on their methodological approach, key inputs, and main contribution to the field. Table 3 Comparative overview of groundwater potential and recharge assessment studies reviewed in this subsection Study Study area Approach Key conditioning factors Main result Main contribution (Kanji and Das 2025a ) Kangsabati Upper Catchment, India XGBoost, SVM, RF, and ANN integrated with AHP Hydrogeological conditions and water quality constituents XGBoost achieved the highest performance with 81% accuracy, and the framework identified suitable artificial recharge zones Demonstrated that integrated ML-MCDM systems can support both groundwater potential mapping and recharge feasibility planning (Ghosh et al. 2022 ) Kangsabati River basin, India MCDA integrated with AHP Geology, geomorphology, elevation, slope, drainage density, lineament density, curvature, topographic wetness index, land use/land cover, and soil 14.62% of the basin was classified as high groundwater potential and 6.11% as very high potential, with validation using groundwater depth data and ROC analysis Showed the continued value of transparent factor weighting for groundwater potential zoning and practical spatial prioritisation (Anand et al. 2025 ) Urban groundwater assessment context Comparative geospatial ML modelling using six ML algorithms Geomorphology, elevation, rainfall, and other groundwater conditioning factors Random Forest achieved the best predictive performance with an AUC of 0.91, while AdaBoost produced the highest MCC Highlighted the strength of ensemble ML models and the importance of sensitivity analysis in groundwater potential assessment (Akbari et al. 2021 ) Groundwater recharge assessment context Comparative MCDM weighting using AHP, BWM, and FUCOM Land use, lithology, slope angle, and other recharge-related environmental layers Land use and lithology were the most influential variables, while slope angle had the least influence across the models Showed that recharge mapping outcomes are sensitive to the choice of weighting framework and that methodological transparency is critical 4.2 Water demand forecasting and supply planning In a review conducted by (Niknam et al. 2022 ) it was shown that water demand forecasting has become a core planning tool rather than only an operational exercise. Their review of more than 100 studies found that traditional time-series models and artificial neural networks were among the most widely used methods, but they also made it clear that no single forecasting method performs best in every system. Instead, model choice depends on the time horizon, the type of data available, and the management objective, whether that is pumping efficiency, leakage control, pressure management, or long-term supply planning. This is an important point for this review because it shows that forecasting models only become useful in practice when they are aligned with a clear planning purpose. (Sharma 2022 ) moved this discussion further by explicitly linking water demand forecasting to supply-side decision-making in Saudi Arabia, as shown in their workflow. The study proposed a combined framework based on Multi-Criteria Optimization and Intelligent Water Demand Forecasting, where monthly demand data were used to compare alternative water management responses, including interbasin transfer, rainwater harvesting, greywater recycling, water recycling, and irrigation-related options. The study reported a prediction accuracy of 98.96% and an optimization ratio of 97.87% and argued that forecasting should not end with demand estimation alone but should feed directly into the evaluation of practical supply and reuse strategies. Even though the paper is more application-driven than critical, it is useful here because it shows how hybrid ML-MCDM frameworks can connect demand prediction to actual planning choices. (Khalilzadeh et al. 2025 ) further adds a different but equally relevant planning dimension by showing that supply planning is not only about matching demand and source options, but also about managing project risk. Their framework combined Fuzzy DEMATEL, Fuzzy ANP, and ANN to assess environmental risks in water supply projects and showed that technical risks received the highest weight in both the ANP and ANN assessments, while supplier-related and communication-related risks also emerged as important decision variables. Although the study is not a demand forecasting paper in the narrow sense, it strengthens this subsection because it shows that supply planning becomes more robust when prediction and prioritisation are combined with formal risk assessment. In other words, planning future water supply requires not only knowing how much water may be needed, but also understanding which project and implementation risks are most likely to undermine delivery. Essentially, these studies show that water demand forecasting and supply planning are most effective when treated as linked parts of the same decision problem. Forecasting improves the timing and direction of planning, while MCDM-based optimisation and risk assessment help decision-makers compare options, account for uncertainty, and justify final choices. This suggests that the strongest direction in this domain is not forecasting alone, and not planning alone, but integrated systems that connect predictive outputs to structured, defensible supply decisions. Taken together, these studies show that water demand forecasting becomes significantly more valuable when it is directly linked to decision-making frameworks. Forecasting alone provides limited practical benefit unless it is embedded within optimisation or risk-based evaluation structures. This distinguishes the water demand domain from others, where prediction is often treated as an end rather than a means to structured planning. 4.3 Flood risk and hydrological hazard management Flood risk and hydrological hazard management are some of the clearest areas where integrated MCDM-ML systems have shown practical value. A study conducted by Pham et al. in Quang Nam Province, Vietnam, developed a flood risk framework that combined hybrid artificial intelligence models with MCDA by linking a flood susceptibility map to a flood consequences map. Their results showed that the Bagging-DT model performed best, with an AUC of 0.96, outperforming AdaBoost-DT and the single DT model. What makes this study especially useful is that it did not stop at susceptibility mapping (Pham et al. 2021 ). It moved further into risk assessment by incorporating human health and financial consequences, which makes the final output more relevant for flood management than hazard mapping alone. (Gholami et al. 2025 ) then showed how this application area is now moving beyond conventional machine learning into deep learning. In southern Iran, they combined a bidirectional LSTM (bLSTM) model with the COPRAS MCDM method to produce a flood risk map that integrated hazard and vulnerability. Their results showed that 15.8% of the study area fell within the high-risk class and 14.6% within the very high-risk class. The study is important because it shows that deep learning can improve the hazard side of flood analysis, while MCDM still remains necessary for vulnerability weighting and final risk interpretation. It also makes clear that topographic variables such as TWI, river density, TPI, SPI, slope, elevation, and distance to river remain dominant even in more advanced modelling workflows. (Rashidi Shikhteymour et al. 2023 ) adds an important social dimension to this domain. Their work in Abarkuh County, Iran, integrated SVM-based flood hazard mapping with ANP-DEMATEL-based vulnerability assessment. The SVM model performed best among the tested ML models, and the final integrated results showed that 6% of the study area was classified as high or very high flood risk. A key strength of this study is that it treated flood risk as the interaction between physical hazard and social vulnerability rather than as a purely hydrological problem. This makes the approach more useful for local authorities because it helps identify where flood exposure coincides with weak adaptive capacity and socio-economic sensitivity. Furthermore, (Asiri et al. 2024 ) extended this integration logic into the coastal flood context by combining AHP with SVM and Decision Tree models under RCP 2.6 and RCP 8.5 scenarios. Their results showed that the AHP-DT model achieved the best performance, with an AUC of 0.95. This study is useful because it brings climate-change-induced sea-level rise and coastal flood drivers into the flood risk workflow. It also shows that integrated MCDM-ML approaches are not limited to inland flood susceptibility studies but can be adapted to exposure-prone coastal environments where scenario-based planning is essential. A different perspective is provided by (Akay 2021 ), who compared statistical, fuzzy logic, and MCDM methods for flood hazard susceptibility mapping in Turkey. Although the study did not build a full integrated ML-MCDM risk model in the same way as the other papers, it remains useful for this subsection because it showed that model choice strongly affects mapping outcomes. The IoE model performed best under ROC-based evaluation, while PCA and AHP performed better under SCAI-based assessment. This is an important reminder that, in flood and hazard management, strong predictive performance does not always align perfectly with spatial reliability or interpretive usefulness. That point matters for this review because it reinforces the need for careful model selection and transparent validation when integrated systems are used for planning and intervention. These studies show that the flood literature has moved well beyond single-model hazard mapping. A clear pattern has emerged in which ML or DL models are increasingly used to estimate hazard or susceptibility, while MCDM methods are used to handle consequences, vulnerability, exposure, weighting, and final prioritisation. The strongest studies are those that treat flood risk as a combined physical and socio-economic problem rather than a prediction problem alone. This is particularly important in hydrological hazard management, where decisions about intervention, infrastructure, and resource allocation depend on more than identifying where flooding is likely to occur. Figure 7 summarises how the reviewed studies integrate machine learning and multi-criteria decision-making approaches for flood risk and hydrological hazard management. The diagram highlights how predictive models are used to estimate flood susceptibility or hazard, while MCDM frameworks support vulnerability weighting, risk classification, and final decision-oriented flood risk mapping. 4.4 Water quality assessment and pollution control (Das 2025a ) conducted a study on the Mahanadi River and showed how water quality assessment can be strengthened when GIS, MCDM, and machine learning are used together in a single framework. The study combined DEMATEL-based WQI with Random Forest, Decision Tree, Multilayer Perceptron, Naïve Bayes, and a Borda Scoring Algorithm, and used IDW interpolation to map spatial variation in pollution. The results showed that 31.58% of the tested locations were classified as poor water, 15.79% as very poor, and 5.26% as unsuitable, while the Random Forest model produced the best overall classification performance, with 90.50% accuracy, 99.87% sensitivity, and 74.56% specificity. The study is particularly useful because it did not only classify water quality status, but also moved into pollution source apportionment, identifying domestic organic debris and expanding human activity as major contributors to degradation. This makes the framework more useful for pollution control than WQI reporting alone. A second study conducted by (Das 2025b ) in the same catchment approached the problem through a different MCDM structure by integrating CILOS-based WQI, EDAS, ANN, and GIS for river water quality management. The study found that 31.58% of sites fell under poor water quality and 5.26% under very poor water quality, while the ANN model achieved an R² of 0.96 across training, cross-validation, and testing. It also identified total coliform, total Kjeldahl nitrogen, electrical conductivity, total dissolved solids, and chloride as major contributors to water quality decline. What makes this study important for the present review is that it shows how integrated systems can move beyond broad quality classification toward ranking polluted sites, identifying influential pollutants, and supporting more targeted water quality control strategies.(Schuwirth et al. 2018 ) adds an important management perspective to this domain by showing that integrated water quality assessment should not only identify current degradation but also evaluate future management alternatives. Their work combined multi-criteria decision analysis, integrated water quality modelling, and scenario planning to compare the future cost-effectiveness of management options under uncertainty. This is valuable for the present subsection because it shows that water quality assessment becomes more useful for pollution control when model outputs are linked to structured decision support rather than treated as stand-alone environmental indicators. In practical terms, this means integrated ML-MCDM type systems are most effective when they help decision-makers move from diagnosis to intervention. Taken together, these studies show that water quality applications are strongest when ML supports classification, pollutant identification, and spatial pattern recognition, while MCDM structures site ranking, intervention prioritisation, and management trade-offs. The most useful integrated systems are those that move beyond water quality diagnosis and support pollution control decisions under uncertainty. 4.5 Sediment, watershed, and catchment management In a work conducted by (Ghosh and Mukhopadhyay 2021 ), it was shown that sediment and watershed management can be strengthened when sub-watershed prioritisation is treated as a multi-model decision problem rather than a single-method exercise. In the Dwarkeswar River basin, they used an MCDM-based ensemble approach that combined SAW, COPRAS, ARAS, TOPSIS, and MOORA, while AHP was used to assign weights to the morphometric parameters. Their results showed that the COPRAS model produced the highest prediction accuracy, but the final prioritisation was based on an adjusted ensemble ranking rather than a single model output. This is important because it addresses a recurring problem in catchment management, namely that different prioritisation models can produce different rankings for the same watershed. By averaging statistically meaningful model outputs, the study moved toward a more defensible basis for identifying erosion-prone sub-watersheds and allocating management effort. (Kanji and Das 2025b ) expanded this line of work by integrating MCDM and ensemble machine learning to prioritise sub-watersheds based on surface runoff potentialities in the upper Kangsabati catchment. Their study considered morpho-tectonic, hydrological, and physical characteristics together, and then linked TOPSIS and VIKOR with Extra Trees, XGBoost, Bagging, and Voting models. The results showed that TOPCOM-ETR, VIKHYP-ETR, and TOPMPT-ETR were the most effective combinations, with low RMSE values and strong discrimination of runoff-generating sub-watersheds. What makes this study especially useful for this subsection is that it shows watershed and catchment management is no longer limited to erosion ranking based on morphometry alone. It is now moving toward more integrated frameworks that combine terrain structure, hydrological behaviour, and machine learning-based prioritisation to guide runoff control and water resource management at sub-watershed scale. Taken together, these studies show that sediment, watershed, and catchment management is increasingly shifting from single-factor prioritisation toward integrated decision-support systems. The older strength of MCDM in this domain lies in transparent ranking and management prioritisation, especially where morphometric and erosion-related indicators dominate. The newer contribution of ML is that it improves the ability to handle non-linear watershed behaviour and more diverse input structures. The strongest direction therefore appears to be integration, where MCDM supports prioritisation and interpretability, while ML strengthens predictive discrimination across sub-watersheds. 4.6 Urban water systems and loss management (Ayad et al. 2021 ) showed that urban water loss management becomes more effective when field activities, hydraulic modelling, optimisation, and GIS are treated as part of one workflow rather than as separate utility functions. Their approach combined EPANET, genetic algorithms, SCE-UA, field leak detection, and GIS-based data handling to calibrate the network, estimate physical losses, identify faulty meters, and improve pipe roughness estimation. What makes this study important for this subsection is that it framed non-revenue water reduction as an integrated operational problem, where leak quantification, network calibration, and spatial decision support must work together if utilities are to move beyond reactive repair practices. In a review by (Romero-Ben et al. 2023 ) it was shown that leak detection and localization in urban water distribution networks have shifted from purely model-based approaches toward broader combinations of model-based, data-driven, and mixed strategies. Their review explained that utilities are increasingly relying on software-based methods linked to SCADA, flow and pressure sensing, and district-metered structures to improve how quickly leaks are detected and located. This is important because it shows that urban loss management is no longer only about identifying a leak after major failure. It is now tied to continuous monitoring, network partitioning, and automated decision support, all of which make integrated MCDM-ML type systems more relevant for urban utilities. A review conducted by (Serafeim et al. 2024 ) reinforced this by showing that leakage control in urban systems depends not only on detection technology, but also on how utilities estimate losses, understand influencing factors, and choose mitigation strategies. Their review covered the water balance, BABE, and minimum night flow approaches, while also showing that leakage severity is shaped by infrastructure condition, maintenance quality, monitoring capability, and broader water stress conditions. For this review, that is important because it shows that loss management in urban systems is both a technical and managerial decision problem. In other words, reducing leakage is not simply a matter of better sensors, but of combining estimation, prioritisation, and intervention in a structured way. (Xu et al. 2014 ) adds a useful environmental perspective by showing that leakage control in urban distribution networks produces benefits that extend beyond direct water savings. Their review concluded that leakage can be further reduced through better detection capability, improved pipe maintenance strategies, and pressure regulation, while also noting that these actions reduce energy use and greenhouse gas emissions. This makes the urban water loss problem especially relevant in WRM because it connects infrastructure efficiency with broader environmental sustainability. It also supports the argument that urban loss management should be treated as a strategic part of integrated water system planning rather than only as a maintenance task. Compared to other WRM domains, urban water loss management shows a stronger emphasis on operational integration rather than methodological innovation. The focus is less on developing new hybrid models and more on combining monitoring, modelling, and optimisation into actionable workflows. This suggests that, in practice, integration effectiveness is often determined more by implementation design than by algorithmic complexity. To summarize on this, section 4 shows that the integration of MCDM and ML has been applied across a wide range of water resource management domains, including groundwater potential and recharge assessment, water demand forecasting and supply planning, flood risk and hydrological hazard management, water quality assessment and pollution control, sediment and catchment prioritisation, and urban water loss management. Across these domains, a clear pattern emerges: ML strengthens prediction, classification, and pattern detection, while MCDM improves weighting, prioritisation, and the translation of technical outputs into practical decisions, Table 4 provides a cross-domain quantitative comparison across the studies. The strongest studies are those that do not treat prediction and decision support as separate tasks but integrate them into coherent decision-support frameworks. However, the depth of this integration varies significantly across domains, with many studies still relying on loosely connected workflows where predictive outputs are not fully embedded into decision structures. This limits the practical applicability of otherwise high-performing models and highlights a key gap in the literature: the need for more tightly integrated, transparent, and reproducible MCDM-ML frameworks that explicitly link prediction to actionable decision-making. Figure 8 summarises the thematic pattern across all WRM domains discussed in Section 4. Table 4 Cross-domain quantitative comparison of integrated MCDM–ML studies reviewed Study Domain MCDM ML Integration pattern Metric(s) reported Main result Validation Comparative note (Ghosh et al. 2022 ) Groundwater potential MCDA, AHP No ML (MCDM-only approach) MCDM-led spatial prioritisation ROC, spatial validation High potential 14.62%, very high 6.11% Pre-/post-monsoon depth + ROC Strong transparency, limited direct comparison with ML-heavy studies (Anand et al. 2025 ) Groundwater potential Not explicitly used RF, AdaBoost, etc. Parallel / comparative AUC, MCC, CA, F1, Precision, Recall RF AUC 0.91; AdaBoost highest MCC Held-out validation Strong predictive comparison, weaker decision-layer transparency (Kanji and Das 2025a ) Groundwater recharge AHP XGBoost, SVM, RF, ANN Sequential Accuracy XGBoost 81% Model comparison Useful integrated recharge planning, but limited metric breadth (Sharma 2022 ) Demand forecasting and supply planning Multi-criteria optimisation Intelligent forecasting Fully coupled Prediction accuracy, optimisation ratio 98.96% accuracy; 97.87% optimisation ratio Simulation-based Strong application claim, but metric not directly comparable across domains (Pham et al. 2021 ) Flood risk MCDA Bagging-DT, AdaBoost-DT, DT Sequential AUC Bagging-DT AUC 0.96 ROC/AUC Strong hazard discrimination, but risk depends on consequence integration (Gholami et al. 2025 ) Flood risk COPRAS bLSTM Fully coupled Risk-class distribution 15.8% high risk, 14.6% very high risk Spatial risk mapping Good hybrid logic, limited cross-study comparability (Das 2025a ) Water quality DEMATEL, WQI, Borda RF, DT, MLP, NB Fully coupled Accuracy, sensitivity, specificity RF 90.50% accuracy, 99.87% sensitivity Classification validation Strong diagnostic framework, but metrics differ from other domains (Das 2025b ) Water quality CILOS, EDAS ANN Sequential R² ANN R² 0.96 Train/cross-val/test Good predictive fit, weak comparability with AUC-based studies Although the reviewed studies frequently report strong case-specific performance, direct cross-study comparison remains limited because different WRM domains rely on different evaluation logics and reporting conventions. Groundwater and flood studies most commonly report discrimination-based metrics such as AUC, ROC, and MCC, whereas water quality and demand forecasting studies more often report accuracy, sensitivity, specificity, R² or optimisation ratios. This makes it difficult to identify a single best integrated MCDM–ML configuration across domains. Instead, the quantitative evidence suggests a domain-dependent pattern: ensemble ML models often provide strong predictive discrimination in groundwater and flood applications, while MCDM components add the greatest value in weighting, prioritisation, and intervention structuring. The comparison also shows that studies using integrated frameworks often report stronger practical relevance than studies relying on prediction alone, but this practical advantage is rarely evaluated using standardised metrics. Table 5 Most commonly reported quantitative metrics across WRM domains Domain Most common metrics Typical output type Groundwater potential/recharge AUC, ROC, Accuracy, MCC Spatial suitability / classification Flood risk/hazard AUC, ROC, risk-class distribution Hazard/risk discrimination Water demand forecasting RMSE, MAE, R², optimisation ratio Forecasting / planning Water quality Accuracy, sensitivity, specificity, R² Classification / pollutant prioritisation Catchment/watershed prioritisation Ranking outputs, RMSE in hybrid studies Prioritisation / runoff discrimination Urban water loss Operational indicators, calibration outputs System optimisation / intervention support 5. Critical comparison of integrated approaches 5.1 Predictive performance versus transparency One of the clearest trade-offs across the studies reviewed in Section 4 is between predictive performance and decision transparency. In several domains, ML-driven models produced stronger predictive or classificatory performance than more traditional rule-based or purely weighted approaches. This was especially visible in groundwater assessment, where XGBoost and Random Forest showed stronger discrimination than several competing models, and in flood risk studies where hybrid or ensemble models such as Bagging-DT, bLSTM, and AHP-DT achieved high predictive performance. These results suggest that integrated approaches benefit when ML is used to capture non-linear relationships in complex environmental datasets. However, the gain in predictive strength does not automatically make the overall system easier to interpret, especially when the decision pathway between input variables, intermediate outputs, and final recommendations is not clearly explained. By contrast, MCDM-led approaches were generally more transparent in showing how criteria were selected, weighted, and translated into final rankings or priority zones. This was particularly evident in groundwater recharge assessment and watershed prioritisation studies, where methods such as AHP, FUCOM, BWM, TOPSIS, and VIKOR made the logic of prioritisation easier to follow. In these cases, the main strength was not necessarily superior predictive accuracy, but rather the ability to justify why one area, option, or intervention was ranked above another. This is an important advantage in WRM because many planning decisions need to be explained to stakeholders, utilities, or public agencies, not just technically optimised. The most effective integrated approaches were therefore not the ones that maximised predictive performance alone, but the ones that balanced predictive strength with a decision structure that remained understandable and practically useful. This was evident in studies on water demand planning, water quality management, and urban water systems, where forecasting or classification became more valuable once linked to optimisation, ranking, source identification, or operational prioritisation. In such cases, ML improved the analytical depth of the system, while MCDM or related decision structures improved how the results could be acted on. In other words, predictive accuracy became more meaningful when it was embedded in a framework that supported intervention rather than only diagnosis. Taken together, the comparison suggests that predictive performance and transparency should not be treated as competing goals in absolute terms, but as design variables that need to be balanced according to the WRM context. Where the main need is strong spatial discrimination or forecasting accuracy, more advanced ML components may be justified. Where the main need is defensible prioritisation, stakeholder communication, or policy-facing justification, transparent MCDM structures remain especially valuable. The strongest integrated approaches are those that combine both, allowing technically strong results to remain interpretable enough for real planning and management use. 5.2 Sensitivity to weighting and expert judgment Another important issue across the integrated approaches reviewed in Section 4 is the extent to which final outputs remain sensitive to weighting structures and expert judgment. This is especially visible in groundwater and recharge assessment, where the same environmental layers can produce different priority zones depending on how criteria are weighted. A study conducted by Akbari et al. showed that groundwater recharge maps changed across AHP, BWM, and FUCOM, even though the underlying conditioning factors remained broadly the same. Their results make it clear that the final pattern is not determined only by environmental data, but also by how decision importance is assigned to factors such as land use, lithology, and slope. This matters because it shows that integrated systems are not purely objective, even when they include strong predictive or spatial components. The decision structure itself can shape the output in meaningful ways. A similar issue appears in watershed and catchment management. The study conducted by Ghosh and Mukhopadhyay showed that different MCDM models can generate different prioritisation results for the same sub-watersheds, even when the same morphometric parameters are used. In their study, SAW, COPRAS, ARAS, TOPSIS, and MOORA did not produce identical rankings, which is why the authors moved toward an ensemble-adjusted result instead of relying on a single model. This is important because it suggests that weighting and ranking sensitivity is not only a groundwater issue. It also affects sediment and catchment management, where management decisions may shift depending on which prioritisation structure is selected. In such cases, expert judgment remains influential because it affects both the assignment of weights and the choice of ranking logic. In flood risk and hazard management, the role of expert judgment becomes even more significant because vulnerability, consequences, and exposure are often harder to quantify than physical hazard variables. Rashidi Shikhteymour et al. showed that flood risk results depended not only on the hazard map produced by SVM, but also on the ANP-DEMATEL structure used to assess social vulnerability. This means that even where the predictive model performs well, the final risk outcome still depends on how social and environmental criteria are framed and weighted. The same issue is visible in coastal flood assessment, where AHP was used alongside SVM and Decision Tree models to incorporate scenario-driven flood risk factors. These examples show that integrated approaches often remain partly judgment-driven, especially when they move from hazard estimation into final risk interpretation. In practical terms, this comparison suggests that sensitivity to weighting and expert judgment should be treated as a normal feature of integrated MCDM-ML systems rather than a weakness to be ignored. The issue is not that expert input exists, but whether it is applied transparently, justified clearly, and tested for robustness. Where weighting choices are left unexplained, the credibility of the final output becomes weaker, even if predictive performance appears strong. By contrast, studies that make their weighting logic explicit, compare alternative structures, or use ensemble ranking approaches provide more defensible results. For WRM, this is important because many interventions are not chosen only on technical grounds. They are also shaped by planning priorities, institutional preferences, and local decision contexts. Table 6 Sensitivity of integrated MCDM-ML approaches to weighting structures and expert judgment across WRM domains WRM domain Integrated approach Where weighting or judgment enters Main sensitivity observed Practical implication Groundwater recharge assessment AHP, BWM, FUCOM with spatial evaluation Factor weighting of recharge-related variables Different weighting frameworks produced different recharge zone patterns Weighting structure should be reported clearly and, where possible, compared across methods Watershed and catchment management AHP with SAW, COPRAS, ARAS, TOPSIS, MOORA Parameter weighting and ranking logic Different MCDM models produced different prioritisation outputs Ensemble or comparative ranking can improve robustness Flood risk management SVM with ANP-DEMATEL; AHP with SVM/DT Vulnerability weighting, consequence assessment, and scenario framing Final risk classes depended on social and environmental weighting choices Hazard prediction should be complemented by transparent vulnerability weighting Water demand and supply planning Forecasting with multi-criteria optimisation Ranking of supply and reuse options Final planning choices depended on prioritisation rules beyond forecast output Forecast accuracy alone is not enough for defensible supply planning Water quality and pollution control WQI with DEMATEL, CILOS, EDAS, ML models Parameter weighting, site ranking, and pollutant importance Priority sites and pollutant rankings changed with decision structure Pollution control frameworks should explain how criteria importance is assigned 5.3 Data requirements, computational cost, and scalability The studies reviewed in Section 4 show that integrated MCDM-ML approaches differ significantly in their data requirements, computational burden, and scalability. In general, MCDM-led approaches tend to be easier to apply where datasets are limited but expert knowledge and spatial layers are available. This was particularly clear in groundwater zoning, recharge assessment, and watershed prioritisation studies, where methods such as AHP, FUCOM, BWM, TOPSIS, and related ranking models worked effectively with mapped conditioning factors, expert-derived weights, and GIS-based overlays. These approaches are usually less computationally demanding than ensemble ML or deep learning models, and they are often easier to implement in settings where technical infrastructure or long-term historical datasets are limited. Their main advantage is that they remain workable under relatively modest data conditions, although that often comes at the cost of lower adaptive or predictive capacity. By contrast, the ML-heavy and ensemble-based approaches reviewed in Section 4 generally required richer datasets and greater computational effort, but they also offered stronger predictive discrimination in more complex environments. This was evident in groundwater studies using Random Forest, XGBoost, AdaBoost, and other ensemble models, as well as in flood risk studies that used Bagging-DT and bLSTM. These methods were better able to capture non-linear interactions and variable importance across multiple predictors, but their performance depended more strongly on data volume, input diversity, and validation design. In practical terms, this means that high-performing integrated models are often more suitable in data-rich environments where computing capability, preprocessing capacity, and technical expertise are available. Their scalability is potentially strong, but only where the supporting data ecosystem is sufficiently developed. A similar contrast appears in application domains linked to urban systems, water quality management, and demand forecasting. In these areas, the integration of prediction, optimisation, and spatial or operational decision support can become computationally demanding because the workflow extends beyond one analytical step. For example, water demand forecasting linked to multi-criteria optimisation, urban non-revenue water reduction linked to hydraulic calibration, and water quality classification linked to pollutant ranking all require more than simple model execution. They involve repeated data handling, scenario evaluation, or system-level integration across forecasting, optimisation, and prioritisation layers. This increases their practical value, but it also raises the demand for reliable databases, technical calibration, and implementation capacity. In this sense, scalability is not only about whether a model can process more data, but also whether the integrated workflow can be maintained under real operational conditions. Overall, the comparison suggests that there is no single best level of complexity for integrated MCDM-ML systems. Simpler MCDM-oriented frameworks may be more scalable in low-resource contexts because they are easier to apply, explain, and reproduce. More advanced ML-integrated systems may offer stronger analytical performance, but they usually require more extensive data, greater computational support, and stronger implementation capacity. For WRM practice, this means the choice of approach should be guided not only by expected accuracy, but also by whether the available data, computational infrastructure, and institutional capacity can support the model at the scale required. 5.4 Suitability under data scarcity and uncertainty Suitability under data scarcity and uncertainty is one of the most important points of comparison across the integrated approaches reviewed in Section 4. In general, MCDM-oriented frameworks appear more workable where measured data are limited, because they can still operate using mapped conditioning factors, expert judgment, and GIS-derived variables. This was especially visible in groundwater zoning and recharge assessment studies, where prioritisation could still be performed even when the system depended more on weighted environmental layers than on long continuous monitoring records. In such contexts, MCDM provides a practical way of supporting decisions where uncertainty is high and observational data are incomplete. By contrast, ML-dominant approaches are usually more sensitive to data scarcity. Their predictive advantage depends on the availability of enough representative input data for training, calibration, and validation. This was clear in groundwater and flood studies where ensemble models and deep learning methods performed strongly, but only because the workflows were supported by multiple input layers and relatively rich datasets. Where such data are sparse, the robustness of the model becomes harder to defend, even if the final outputs appear accurate. A similar issue appears in operational domains such as water demand forecasting, urban water loss management, and water quality control. These systems are useful when continuous or regularly updated datasets are available, but their practical value weakens when monitoring is inconsistent, field measurements are sparse, or uncertainty in implementation conditions is high. In such cases, integrated frameworks still remain valuable, but simpler and more transparent structures may be more suitable than highly data-hungry predictive systems. Overall, the comparison suggests that integrated MCDM-ML systems are most suitable under data scarcity when the decision framework can tolerate uncertainty explicitly. Approaches that combine moderate predictive support with transparent ranking or prioritisation are often more defensible than highly complex models built on weak data foundations. For WRM practice, this means that model suitability should be judged not only by accuracy, but also by whether the available data can support stable, credible, and repeatable decisions under uncertain conditions. 5.5 Transferability across contexts Transferability across contexts remains limited in many of the integrated approaches reviewed above. Most studies were developed for specific basins, catchments, cities, or hazard-prone regions, and their performance was closely tied to local data structure, environmental conditions, and decision priorities. This means that a model or framework that performs well in one area cannot automatically be assumed to perform equally well in another. This issue is especially clear in groundwater, flood, and watershed studies. In these domains, the importance of factors such as lithology, slope, rainfall, drainage density, land use, and terrain structure changes from one location to another. As a result, both the predictive model and the weighting structure may need to be recalibrated when transferred to a new setting. The challenge is not only technical. It also affects whether the final rankings or risk classes remain meaningful under different physical conditions. A similar limitation appears in urban and operational domains. Water demand forecasting, leakage control, and water quality assessment depend heavily on local infrastructure, monitoring systems, consumption behaviour, and management practice. This reduces direct transferability because even if the integration logic remains useful, the final model still depends on context-specific operational conditions. In such cases, what transfers best is often the framework rather than the exact calibrated model. Overall, this suggests that transferability is stronger at the level of integration design than at the level of exact model output. Sequential, parallel, or hybrid MCDM-ML structures can often be adapted to new settings, but their criteria weights, predictor importance, validation results, and final priorities usually require local adjustment. For WRM practice, this means that integrated approaches should be viewed as adaptable frameworks rather than universally transferable solutions. 6. Reporting gaps and a comparative framework for future studies 6.1 Metric inconsistency and benchmarking problems One of the clearest problems across the studies reviewed in Section 4 is the lack of consistency in how performance is measured and reported. Different studies used different validation metrics, and in some cases the reported outputs were not directly comparable. This makes it difficult to judge whether one integrated approach is genuinely stronger than another, or whether it only appears better because it was evaluated differently. This inconsistency was especially visible in groundwater and flood studies. Some studies reported AUC, others used accuracy, MCC, or ROC-based validation, while some MCDM-led studies relied more on spatial agreement or ranking consistency than on predictive metrics. As a result, studies addressing similar WRM problems often cannot be compared on a common performance basis, even when they appear to target the same type of output. A similar issue appears in water quality, urban water, and supply-planning studies. In these domains, some studies reported classification accuracy, others reported R², while others emphasised optimisation ratios, ranking outputs, or operational usefulness. This broad variation weakens cross-study benchmarking because predictive quality, decision quality, and practical utility are often mixed together without a shared reporting structure. Benchmarking is also weakened by the absence of common datasets, standard validation protocols, and shared comparison baselines. Many studies are designed around local case data, which is valuable for practical relevance, but it means that performance is tested under different data conditions, scales, and assumptions. In that setting, even strong results are hard to generalise because the benchmark itself is unstable. Overall, the problem is not only that metrics differs, but that the literature still lacks a common benchmarking logic for integrated MCDM-ML systems in WRM. 6.2 Validation and reproducibility gaps Another issue across the reviewed studies is the limited consistency in validation procedures. Many studies reported strong model performance, but the validation approaches differed widely. Some studies relied on ROC curves or AUC values, while others used accuracy, R², or confusion matrix measures. Because validation procedures vary, it becomes difficult to determine whether results reflect genuine methodological strength or simply differences in evaluation design. Validation gaps are also visible in spatial applications such as groundwater potential mapping and flood risk assessment. In several cases, models were validated using historical data or a limited number of field observations. While this approach can demonstrate internal consistency, it does not always confirm whether the model will perform reliably under different environmental conditions or future scenarios. This is particularly important in WRM, where models are often used to support long-term planning decisions rather than only retrospective analysis. Reproducibility also remains a challenge. Many studies provide a description of their workflow but do not include enough detail about data preprocessing, parameter settings, or weighting procedures for the results to be easily replicated. This is especially noticeable in integrated systems where multiple techniques are combined, such as GIS analysis, machine learning models, and MCDM ranking frameworks. Without clear documentation of each step, reproducing the results in another study area becomes difficult. The issue becomes even more complex in operational domains such as water demand forecasting, water quality management, and urban water loss reduction. These applications often depend on local datasets, monitoring systems, and institutional practices that are not always accessible to other researchers. As a result, even when the methodological framework is well described, the data required to reproduce the analysis may not be available. Overall, the literature shows that stronger validation protocols and clearer methodological reporting are needed for integrated MCDM-ML systems in WRM. Transparent validation procedures, accessible datasets, and well-documented workflows would improve the credibility of future studies and make it easier to test whether integrated approaches perform consistently across different contexts. 6.3 Proposed minimum reporting framework Based on the issues identified in Sections 6.1 and 6.2, a minimum reporting framework is needed for future integrated MCDM-ML studies in WRM. The purpose of such a framework is not to force all studies into one methodological design. Rather, it is to ensure that studies report enough information for readers to understand how the system was built, how it was validated, and how the final decision outputs were generated. Without this, cross-study comparison will remain weak, and reproducibility will continue to be limited. At a minimum, future studies should clearly report the WRM application domain, study objective, data sources, input variables, temporal and spatial scale, and the exact MCDM and ML methods used. This is necessary because many integrated studies appear similar at the title level, but differ substantially in their data structure, modelling depth, and decision logic. Clear reporting at this level would make it easier to compare studies across domains such as groundwater, flood risk, water quality, and urban water management. Studies should also report how integration actually occurs. This includes whether the approach is sequential, parallel, or fully coupled, what outputs are passed from one method to another, and how the final recommendation or prioritisation is produced. This is especially important because integrated systems are often presented as hybrid frameworks, but the interaction between the ML and MCDM components is not always explained in enough detail to show how the final decision support output was constructed. A further requirement is transparent reporting of validation, weighting, uncertainty, and sensitivity procedures. Future studies should state the validation metrics used, the reason for selecting them, the source of weights or preferences, and whether sensitivity analysis was performed. Where expert judgment is used, the basis for that judgment should be made explicit. Where uncertainty is present, it should be acknowledged and, where possible, tested rather than left implicit. Finally, studies should report enough practical information to support reuse and adaptation. This includes software or platform details, preprocessing steps, calibration choices, and any constraints that may affect implementation in other settings. For WRM, this matters because the usefulness of an integrated system depends not only on academic performance, but also on whether the framework can be interpreted, replicated, and adapted for real planning and management use. This framework can serve as a baseline checklist for future integrated WRM studies and as a basis for improving cross-study comparison, transparency, and reproducibility. Table 7 Proposed minimum reporting framework for integrated MCDM-ML studies in water resource management Reporting element Minimum information that should be reported Why it matters Study context WRM domain, study objective, and decision problem Clarifies what the integrated system is intended to solve Data description Data sources, variables, sample size, spatial scale, temporal scale Supports interpretation of model suitability and comparability MCDM component Method used, criteria, weighting approach, source of weights Makes the decision structure transparent ML component Model used, input features, preprocessing steps, calibration or training details Improves reproducibility and technical clarity Integration structure Sequential, parallel, or fully coupled design; how outputs move between methods Shows how the hybrid system actually works Validation approach Metrics used, validation procedure, benchmark or comparison basis Allows fairer performance assessment across studies Sensitivity and uncertainty Sensitivity testing, uncertainty treatment, robustness checks Strengthens confidence in the reported results Output format Type of final output, such as map, ranking, forecast, classification, or intervention priority Clarifies how the system supports decision-making Implementation relevance Software/tools used, data availability, operational constraints, transferability limits Helps others assess practical applicability Limitations Main methodological, data, or contextual limitations Prevents overstatement and improves interpretive caution 7. Emerging research and implementation directions in MCDM-ML for WRM 7.1 Explainable and trustworthy AI Explainable and trustworthy AI is becoming an important direction in WRM because strong predictive performance alone is no longer enough for practical decision support. The review conducted by (Başağaoğlu et al. 2022 ) showed that interpretable and explainable AI methods are increasingly being applied across hydroclimatic domains such as groundwater, streamflow, water quality, floods, and droughts. Their review makes it clear that explainability is valuable because it helps reveal how predictors influence model outputs, which is especially important where AI results may affect planning, regulation, or public-facing decisions. For this review, that is important because integrated MCDM-ML systems depend not only on accurate outputs, but also on decision pathways that users can understand and justify. (Infant et al. 2025 ) further showed that explainable AI is becoming increasingly relevant in urban water systems, especially in hydrological modelling, demand prediction, and leak detection. Their review discussed tools such as SHAP, LIME, and counterfactual analysis, and argued that explainability improves both model understanding and practical confidence in deployment. This matters for the present manuscript because future MCDM-ML frameworks in WRM will need to do more than generate outputs. They will also need to show why certain variables, risks, or options are driving the final recommendation. A much broader review conducted by (Schiller et al. 2025 ) on AI in environmental and Earth system sciences added an important caution. In their review it was found that although XAI methods are often presented as a way to increase trust, very few studies explicitly examine trustworthiness itself. In other words, explainability and trust are related, but they are not the same. This is an important distinction for WRM because a model may be partially interpretable while still raising concerns about robustness, fairness, uncertainty, or governance suitability. For that reason, future integrated systems will need to treat trustworthiness as a design requirement rather than assuming that explanation alone is enough. A practical example is provided by (Maußner et al. 2025 ) who developed explainable AI for water demand forecasting and linked it to the idea of reliable water supply decision-making. Their work is useful because it moves the discussion from principle to application. It shows that explainability can help utilities understand model behaviour in operational settings rather than treating AI as a black box. For the future of integrated MCDM-ML systems, this suggests that explainable AI is most valuable when it supports better human judgment, clearer prioritisation, and more defensible implementation decisions. The issue is no longer only whether AI can improve prediction. It is whether its outputs can be interpreted, questioned, trusted, and integrated into transparent decision-support structures. In this sense, explainability strengthens the ML side of integrated systems, while trustworthiness determines whether those systems can be responsibly used in real water management settings. 7.2 Real-time and IoT-enabled decision support Real-time monitoring has also shown to improve the ability of water utilities to manage infrastructure and reduce operational losses. (Singh and Ahmed 2021 ; Roostaei et al. 2023 ) demonstrated that IoT-enabled sensing systems can support continuous monitoring of distribution networks and improve leak detection, pressure management, and demand tracking. By combining sensor data with predictive analytics, utilities can detect anomalies and system failures earlier than would be possible with manual inspection alone. This development is particularly relevant for integrated MCDM-ML frameworks because real-time data streams provide the dynamic inputs needed for predictive models while decision-support layers help prioritise operational responses. IoT technologies also play an important role in environmental monitoring and early warning systems. (Kumar et al. 2025 ) showed that IoT-enabled hydrological monitoring can support flood forecasting and water quality monitoring through continuous data acquisition from distributed sensors. These systems improve situational awareness by allowing models to update predictions as new observations become available. In this context, real-time data streams strengthen the predictive capacity of ML models while enabling decision-support frameworks to react to changing conditions more quickly than static planning systems. Continuous monitoring improves the availability and timeliness of data, while integrated analytical frameworks allow this information to be translated into actionable decisions. For integrated MCDM-ML systems, this means that future implementations will likely operate in more dynamic environments where models are updated regularly and decisions are supported by live system data rather than static datasets. 7.3 Cloud and edge deployment for scalable WRM As integrated WRM systems move from offline analysis toward real-time monitoring and operational control, cloud and edge deployment is becoming increasingly important. Large-scale sensor networks in water distribution and monitoring systems now generate data volumes that are difficult to manage through fully centralised architectures alone. For this reason, cloud platforms remain useful for storage, historical analysis, coordination, and large-model training, while edge computing is increasingly valuable for local preprocessing, near-real-time inference, and faster operational response (Pagano et al. 2025 ). In practical terms, this means future MCDM-ML systems are likely to operate across layered architectures in which cloud services support system-wide intelligence and edge nodes support immediate field-level decisions. The value of edge deployment becomes even clearer in time-sensitive WRM applications such as leak detection, anomaly detection, water quality alerts, and distributed environmental monitoring. By processing part of the data closer to the source, edge architectures can reduce latency, lower data transmission load, improve energy efficiency, and reduce operating cost while still supporting machine learning functions at local nodes. This is especially relevant for water systems that depend on rapid detection and response, because delayed transmission to remote servers can weaken the practical value of otherwise strong analytical models (Roostaei et al. 2023 ). For integrated MCDM-ML systems, edge deployment therefore improves not only technical speed, but also the timeliness of prioritisation and intervention. At the same time, scalable deployment is not only a matter of adding sensors or moving models to the cloud. It also depends on communication reliability, data quality, infrastructure change, privacy protection, and the ability to keep models updated as network conditions evolve. Recent reviews of smart water distribution systems show that real-time monitoring, digital twins, uncertainty-aware forecasting, and explainable AI are likely to become more important as utilities move toward more connected and adaptive systems (Taloma et al. 2025 ). This suggests that future MCDM-ML frameworks in WRM should be designed not only for analytical performance, but also for operational scalability, maintainability, and secure long-term deployment. 7.4 Policy-aligned and stakeholder-centred systems The future usefulness of integrated MCDM-ML systems in WRM will depend not only on methodological sophistication, but also on whether they align with governance realities and stakeholder needs. Water decisions affect multiple users whose priorities, risks, and values are not always the same, and this makes technically strong models insufficient on their own. Evidence from participatory water governance shows that community engagement improves the legitimacy and durability of water interventions, especially when local actors are involved in planning, monitoring, and implementation rather than being treated only as end users of expert-generated outputs (Ahmadi et al. 2020 ; Moreira et al. 2024 ). For integrated WRM systems, this means decision-support tools should be designed to support participation, communication, and shared understanding, not just optimisation. This is especially important where water systems are characterised by conflicting interests, institutional fragmentation, or unequal influence among users. Stakeholder-based decision support research has shown that management scenarios become more realistic and more sustainable when the characteristics, interests, and influence of different actors are explicitly built into the analytical process. In practice, this means future MCDM-ML systems should increasingly move toward structures that do not only predict outcomes or rank options, but also make room for negotiated priorities, decentralised decisions, and trade-off awareness across environmental, social, and economic dimensions. Policy alignment is equally important. Water management tools are more likely to be adopted when they fit local legal frameworks, institutional mandates, allocation rules, and practitioner realities. Studies on water allocation planning and decision support have shown that even well-developed tools are often underused when they do not match operational needs or when they are introduced without sufficient stakeholder participation during development (Pearson et al. 2010 ; Nel et al. 2022 ). This means the next generation of integrated WRM systems should be designed with implementation in mind from the beginning, including how they fit existing governance structures, regulatory requirements, and the decision habits of practitioners. A further shift is needed from technically correct systems toward socially workable systems. Sustainable decision support in water management increasingly depends on adaptive processes, social learning, and the ability to evaluate options in ways that remain understandable across professional and community boundaries. In that sense, policy-aligned and stakeholder-centred systems are not a soft addition to MCDM-ML integration. They are part of what makes these systems usable in practice. The strongest future frameworks will therefore be those that connect predictive intelligence with transparent prioritisation, institutional fit, and meaningful stakeholder engagement. 8. Conclusion This review examined how MCDM and ML have been integrated across major WRM domains and showed that the combination is now being used far beyond isolated experimental studies. Across groundwater assessment, demand forecasting, flood risk management, water quality control, watershed prioritisation, and urban water systems, a consistent pattern emerged. ML contributes most strongly where prediction, classification, and pattern detection are required, while MCDM contributes most strongly where weighting, prioritisation, and final decision structuring are necessary. The literature therefore suggests that the value of integration lies not in combining methods for novelty, but in linking analytical prediction with decision-ready interpretation in ways that better reflect the complexity of WRM. At the same time, the review also shows that the field has not yet reached methodological maturity. Strong case-specific results are now common, but cross-study comparison remains weakened by inconsistent metrics, varied validation procedures, limited transparency in weighting structures, uneven treatment of uncertainty, and weak reproducibility. These limitations reduce the extent to which current findings can be generalised across settings and make it difficult to identify which integrated approaches are most robust under different WRM conditions. For that reason, future work should move beyond reporting good outcomes and focus more deliberately on validation design, sensitivity analysis, reporting consistency, and practical transferability. Overall, the strongest direction for future research and implementation is clear. Integrated MCDM-ML systems in WRM should become more explainable, more adaptive, more operationally scalable, and more closely aligned with stakeholder and policy contexts. In practice, this means combining predictive strength with transparent decision logic, real-time or near-real-time data support, clearer reporting frameworks, and implementation pathways that are credible to both technical experts and decision-makers. When these conditions are met, integrated systems are better positioned to support water management decisions that are not only accurate, but also defensible, context-appropriate, and useful in practice. A key immediate priority for the field is the adoption of clearer reporting and validation standards so that integrated WRM studies become easier to compare, reproduce, and implement. Declarations Funding: This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Competing interests: The authors declare that they have no competing interests. Clinical trial number: not applicable. Ethics approval: not applicable. Consent to participate: not applicable. Consent for publication: not applicable. Data Availability Statement: Not applicable. Author Contribution Murphy Bonkogia Lomboli - conceptualization, methodology, literature search, screening, analysis, synthesis, visualisation, and writing of the original draft.Opeyeolu Timothy Laseinde - supervision, methodological guidance, critical review, validation, and editing of the manuscript. Both authors read and approved the final manuscript. References Ahmadi A, Kerachian R, Skardi MJE, Abdolhay A. A stakeholder-based decision support system to manage water resources. J Hydrol (Amst). 2020;589. https://doi.org/10.1016/j.jhydrol.2020.125138 . Ahmed AA, Sayed S, Abdoulhalik A, et al. Applications of machine learning to water resources management: A review of present status and future opportunities. J Clean Prod. 2024;441:140715. https://doi.org/10.1016/j.jclepro.2024.140715 . Akay H. Flood hazards susceptibility mapping using statistical, fuzzy logic, and MCDM methods. Soft comput. 2021;25:9325–46. https://doi.org/10.1007/s00500-021-05903-1 . Akbari M, Meshram SG, Krishna RS, et al. Identification of the Groundwater Potential Recharge Zones Using MCDM Models: Full Consistency Method (FUCOM), Best Worst Method (BWM) and Analytic Hierarchy Process (AHP). Water Resour Manage. 2021;35:4727–45. https://doi.org/10.1007/s11269-021-02924-1 . Ali R, Hussain A, Nazir S, et al. Intelligent Decision Support Systems—An Analysis of Machine Learning and Multicriteria Decision-Making Methods. Appl Sci. 2023;13:12426. https://doi.org/10.3390/app132212426 . Anand V, Rajput VD, Minkina T et al. (2025) Evaluating groundwater potential with the synergistic use of geospatial methods and advanced machine learning approaches. Discover Cities 2:. https://doi.org/10.1007/s44327-025-00095-x Asiri MM, Aldehim G, Alruwais N, et al. Coastal Flood risk assessment using ensemble multi-criteria decision-making with machine learning approaches. Environ Res. 2024;245. https://doi.org/10.1016/j.envres.2023.118042 . Ayad A, Khalifa A, Fawy MEL, Moawad A. An integrated approach for non-revenue water reduction in water distribution networks based on field activities, optimisation, and GIS applications. Ain Shams Eng J. 2021;12:3509–20. https://doi.org/10.1016/j.asej.2021.04.007 . Başağaoğlu H, Chakraborty D, Lago C, Do, et al. A Review on Interpretable and Explainable Artificial Intelligence in Hydroclimatic Applications. Water (Basel). 2022;14:1230. https://doi.org/10.3390/w14081230 . Calizaya A, Meixner O, Bengtsson L, Berndtsson R. Multi-criteria decision analysis (MCDA) for integrated water resources management (IWRM) in the Lake Poopo basin, Bolivia. Water Resour Manage. 2010;24:2267–89. https://doi.org/10.1007/s11269-009-9551-x . Costa AP, de Choren A, de Pereira R M, et al. Integrating multicriteria decision making and principal component analysis: a systematic literature review. Cogent Eng. 2024;11. https://doi.org/10.1080/23311916.2024.2374944 . Das A. (2025a) An optimization based framework for water quality assessment and pollution source apportionment employing GIS and machine learning techniques for smart surface water governance. Discover Environment 3:. https://doi.org/10.1007/s44274-025-00327-2 Das A. (2025b) Water pollution and water quality assessment and application of criterion impact loss (CILOS), geographical information system (GIS), artificial neural network (ANN) and decision-learning technique in river water quality management: An experiment on the Mahanadi catchment, Odisha, India. Desalination Water Treat 321:. https://doi.org/10.1016/j.dwt.2024.100969 Drogkoula M, Kokkinos K, Samaras N. A Comprehensive Survey of Machine Learning Methodologies with Emphasis in Water Resources Management. Appl Sci. 2023;13:12147. https://doi.org/10.3390/app132212147 . Drogkoula M, Samaras N, Iatrellis O, et al. Systematic review and topic classification of soft computing and machine learning in water resources management. Discover Sustain. 2025;6:860. https://doi.org/10.1007/s43621-025-01832-3 . Gebre SL, Cattrysse D, Van Orshoven J. Multi-Criteria Decision-Making Methods to Address Water Allocation Problems: A Systematic Review. Water (Basel). 2021;13:125. https://doi.org/10.3390/w13020125 . Ghobadi F, Kang D. Application of Machine Learning in Water Resources Management: A Systematic Literature Review. Water (Basel). 2023;15:620. https://doi.org/10.3390/w15040620 . Gholami H, Mohammadifar A, Golzari S, et al. Mapping flood risk using a workflow including deep learning and MCDM– Application to southern Iran. Urban Clim. 2025;59. https://doi.org/10.1016/j.uclim.2024.102272 . Ghosh A, Adhikary PP, Bera B, et al. Assessment of groundwater potential zone using MCDA and AHP techniques: case study from a tropical river basin of India. Appl Water Sci. 2022;12. https://doi.org/10.1007/s13201-021-01548-5 . Ghosh B, Mukhopadhyay S. Erosion susceptibility mapping of sub-watersheds for management prioritization using MCDM-based ensemble approach. Arab J Geosci. 2021;14:36. https://doi.org/10.1007/s12517-020-06297-4 . Hajkowicz S, Collins K. A Review of Multiple Criteria Analysis for Water Resource Planning and Management. Water Resour Manage. 2007;21:1553–66. https://doi.org/10.1007/s11269-006-9112-5 . He C, Huang G, Liu L, et al. Multi-dimensional diagnosis model for the sustainable development of regions facing water scarcity problem: A case study for Guangdong, China. Sci Total Environ. 2020. https://doi.org/10.1016/j.scitotenv.2020.139394 . 734:. Infant SS, Vickram S, Saravanan A, et al. Explainable artificial intelligence for sustainable urban water systems engineering. Results Eng. 2025;25:104349. https://doi.org/10.1016/j.rineng.2025.104349 . Kanji S, Das S. Assessing groundwater potentialities and replenishment feasibility using machine learning and MCDM models considering hydro-geological aspects and water quality constituents. Environ Earth Sci. 2025a;84. https://doi.org/10.1007/s12665-024-11996-2 . Kanji S, Das S. Exploring the morpho-tectonic nature, hydrological and physical characteristics of a watershed and prioritizing sub-watersheds surface runoff potentialities by integrating MCDM and ensemble machine learning models. J Environ Manage. 2025b;386:125772. https://doi.org/10.1016/j.jenvman.2025.125772 . Khalil U, Imtiaz I, Aslam B, et al. Comparative analysis of machine learning and multi-criteria decision making techniques for landslide susceptibility mapping of Muzaffarabad district. Front Environ Sci. 2022;10. https://doi.org/10.3389/fenvs.2022.1028373 . Khalilzadeh M, Banihashemi SA, Heidari A, et al. Risk Analysis and Assessment of Water Supply Projects Using the Fuzzy DEMATEL-ANP and Artificial Neural Network Methods. Water (Switzerland). 2025;17. https://doi.org/10.3390/w17131995 . Kocaman H, Asan U. Integration Modes Between MCDM Methods and Machine Learning Algorithms: A Structured Approach for Framework Development. Mathematics. 2025;14:33. https://doi.org/10.3390/math14010033 . Kolaski K, Logan LR, Ioannidis JPA. Guidance to best tools and practices for systematic reviews. Syst Rev. 2023;12. https://doi.org/10.1186/s13643-023-02255-9 . Kumar N, Kumar S, Yadav S. Smart Water Resource Management using IoT and Data Analytics. J Water Eng Manage. 2025;6:45–9. https://doi.org/10.47884/jweam.v6i3pp45 . Kumar R. A Comprehensive Review of MCDM Methods, Applications, and Emerging Trends. Decis Mak Adv. 2025;3:185–99. https://doi.org/10.31181/dma31202569 . Lai E, Lundie S, Ashbolt NJ. Review of multi-criteria decision aid for integrated sustainability assessment of urban water systems. Urban Water J. 2008;5:315–27. https://doi.org/10.1080/15730620802041038 . Li M, Cao X, Liu D, et al. Sustainable management of agricultural water and land resources under changing climate and socio-economic conditions: A multi-dimensional optimization approach. Agric Water Manag. 2022;259. https://doi.org/10.1016/j.agwat.2021.107235 . Liu C, Xu J, Li X, et al. Water resource forecasting with machine learning and deep learning: A scientometric analysis. Artif Intell Geosci. 2024;5:100084. https://doi.org/10.1016/j.aiig.2024.100084 . Long HA, French DP, Brooks JM. Optimising the value of the critical appraisal skills programme (CASP) tool for quality appraisal in qualitative evidence synthesis. Res Methods Med Health Sci. 2020;1:31–42. https://doi.org/10.1177/2632084320947559 . Maußner C, Oberascher M, Autengruber A, et al. Explainable artificial intelligence for reliable water demand forecasting to increase trust in predictions. Water Res. 2025;268. https://doi.org/10.1016/j.watres.2024.122779 . Mendoza GA, Martins H. Multi-criteria decision analysis in natural resource management: A critical review of methods and new modelling paradigms. Ecol Manage. 2006;230:1–22. https://doi.org/10.1016/j.foreco.2006.03.023 . Miller KA, Belton V. Water resource management and climate change adaptation: a holistic and multiple criteria perspective. Mitig Adapt Strateg Glob Chang. 2014;19:289–308. https://doi.org/10.1007/s11027-013-9537-0 . Mohammadifar A, Gholami H, Golzari S. Novel integrated modelling based on multiplicative long short-term memory (mLSTM) deep learning model and ensemble multi-criteria decision making (MCDM) models for mapping flood risk. J Environ Manage. 2023;345. https://doi.org/10.1016/j.jenvman.2023.118838 . Moreira FD, Fonseca PRS, Miranda RM, et al. Stakeholder engagement for inclusive water governance in a rural community in Brazil. Front Water. 2024;6. https://doi.org/10.3389/frwa.2024.1378514 . Nasiri Khiavi A, Vafakhah M, Kim D, et al. Integrating Remote Sensing, Machine Learning, and Local Knowledge for Innovative Flood Susceptibility and Vulnerability Mapping. J Flood Risk Manag. 2025;18. https://doi.org/10.1111/jfr3.70149 . Nel JB, Mativenga PT, Marnewick AL. A Framework to Support the Selection of an Appropriate Water Allocation Planning and Decision Support Scheme. Water (Switzerland). 2022;14. https://doi.org/10.3390/w14121854 . Niknam A, Zare HK, Hosseininasab H, et al. A Critical Review of Short-Term Water Demand Forecasting Tools—What Method Should I Use? Sustainability. 2022;14:5412. https://doi.org/10.3390/su14095412 . Pagano A, Garlisi D, Tinnirello I, et al. A survey on massive IoT for water distribution systems: Challenges, simulation tools, and guidelines for large-scale deployment. Ad Hoc Netw. 2025;168:103714. https://doi.org/10.1016/j.adhoc.2024.103714 . Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021a;372:n71. https://doi.org/10.1136/bmj.n71 . Page MJ, Moher D, Bossuyt PM, et al. PRISMA 2020 explanation and elaboration: updated guidance and exemplars for reporting systematic reviews. BMJ. 2021b;372:n160. https://doi.org/10.1136/bmj.n160 . Pearson LJ, Coggan A, Proctor W, Smith TF. A sustainable decision support framework for Urban water management. Water Resour Manage. 2010;24:363–76. https://doi.org/10.1007/s11269-009-9450-1 . Pham BT, Luu C, Phong T, Van, et al. Flood risk assessment using hybrid artificial intelligence models integrated with multi-criteria decision analysis in Quang Nam Province, Vietnam. J Hydrol (Amst). 2021;592. https://doi.org/10.1016/j.jhydrol.2020.125815 . Poli G, Cuntò S, Muccio E, Cerreta M. (2024) A spatial decision support system for multi-dimensional sustainability assessment of river basin districts: the case study of Sarno river, Italy. Land use policy 141:. https://doi.org/10.1016/j.landusepol.2024.107123 Poursaeid M. A Novel Approach for Water Resources Management to Accurate River Flow Prediction: Utilization of Linear Programming and Weighted Ensemble Learning. Water Resour Manage. 2025;39:5827–43. https://doi.org/10.1007/s11269-025-04230-6 . Rashidi Shikhteymour S, Borji M, Bagheri-Gavkosh M, et al. A novel approach for assessing flood risk with machine learning and multi-criteria decision-making methods. Appl Geogr. 2023;158. https://doi.org/10.1016/j.apgeog.2023.103035 . Romero-Ben L, Alves D, Blesa J, et al. Leak detection and localization in water distribution networks: Review and perspective. Annu Rev Control. 2023;55:392–419. https://doi.org/10.1016/j.arcontrol.2023.03.012 . Roostaei J, Wager YZ, Shi W, et al. IoT-based edge computing (IoTEC) for improved environmental monitoring. Sustainable Computing: Inf Syst. 2023;38. https://doi.org/10.1016/j.suscom.2023.100870 . Schabbauer G. Polymicrobial sepsis models: CLP versus CASP. Drug Discov Today Dis Models. 2012;9:e17–21. https://doi.org/10.1016/j.ddmod.2011.10.002 . Schiller J, Stiller S, Ryo M. Artificial intelligence in environmental and Earth system sciences: explainability and trustworthiness. Artif Intell Rev. 2025;58. https://doi.org/10.1007/s10462-025-11165-2 . Schuwirth N, Honti M, Logar I, Stamm C. Multi-criteria decision analysis for integrated water quality assessment and management support. Water Res X. 2018. https://doi.org/10.1016/j.wroa.2018.100010 . 1:. Serafeim AV, Fourniotis NTh, Deidda R, et al. Leakages in Water Distribution Networks: Estimation Methods, Influential Factors, and Mitigation Strategies—A Comprehensive Review. Water (Basel). 2024;16:1534. https://doi.org/10.3390/w16111534 . Shaheen N, Shaheen A, Ramadan A, et al. Appraising systematic reviews: a comprehensive guide to ensuring validity and reliability. Front Res Metr Anal. 2023;8. https://doi.org/10.3389/frma.2023.1268045 . Sharma SK. A novel approach on water resource management with Multi-Criteria Optimization and Intelligent Water Demand Forecasting in Saudi Arabia. Environ Res. 2022;208. https://doi.org/10.1016/j.envres.2021.112578 . Singh M, Ahmed S. IoT based smart water management systems: A systematic review. Mater Today Proc. 2021;46:5211–8. https://doi.org/10.1016/j.matpr.2020.08.588 . Snyder H. Literature review as a research methodology: An overview and guidelines. J Bus Res. 2019;104:333–9. https://doi.org/10.1016/j.jbusres.2019.07.039 . Sohrabi C, Franchi T, Mathew G, et al. PRISMA 2020 statement: What’s new and the importance of reporting guidelines. Int J Surg. 2021;88:105918. https://doi.org/10.1016/j.ijsu.2021.105918 . Taloma RJL, Cuomo F, Comminiello D, Pisani P. Machine learning for smart water distribution systems: exploring applications, challenges and future perspectives. Artif Intell Rev. 2025;58. https://doi.org/10.1007/s10462-024-11093-7 . Xi X, Poh KL. A Novel Integrated Decision Support Tool for Sustainable Water Resources Management in Singapore: Synergies Between System Dynamics and Analytic Hierarchy Process. Water Resour Manage. 2015;29:1329–50. https://doi.org/10.1007/s11269-014-0876-8 . Xu Q, Liu R, Chen Q, Li R. Review on water leakage control in distribution networks and the associated environmental benefits. J Environ Sci (China). 2014;26:955–61. https://doi.org/10.1016/S1001-0742(13)60569-0 . Zhu F, Zhong P, an, Cao Q, et al. A stochastic multi-criteria decision making framework for robust water resources management under uncertainty. J Hydrol (Amst). 2019;576:287–98. https://doi.org/10.1016/j.jhydrol.2019.06.049 . Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 17 Apr, 2026 Reviews received at journal 17 Apr, 2026 Reviews received at journal 15 Apr, 2026 Reviewers agreed at journal 09 Apr, 2026 Reviewers agreed at journal 09 Apr, 2026 Reviewers agreed at journal 09 Apr, 2026 Reviewers agreed at journal 07 Apr, 2026 Reviewers agreed at journal 07 Apr, 2026 Reviewers agreed at journal 07 Apr, 2026 Reviewers agreed at journal 07 Apr, 2026 Reviews received at journal 01 Apr, 2026 Reviewers agreed at journal 29 Mar, 2026 Reviewers invited by journal 29 Mar, 2026 Editor assigned by journal 29 Mar, 2026 Editor invited by journal 29 Mar, 2026 Submission checks completed at journal 28 Mar, 2026 First submitted to journal 28 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9145184","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Systematic Review","associatedPublications":[],"authors":[{"id":615010807,"identity":"67b2fb2b-6c10-4cee-b5e6-369f52f4404b","order_by":0,"name":"Murphy Bonkogia Lomboli","email":"data:image/png;base64,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","orcid":"","institution":"University of Johannesburg","correspondingAuthor":true,"prefix":"","firstName":"Murphy","middleName":"Bonkogia","lastName":"Lomboli","suffix":""},{"id":615010809,"identity":"ff5cc4b9-c231-4393-a28a-ba51fbcde01b","order_by":1,"name":"Opeyeolu Timothy Laseinde","email":"","orcid":"","institution":"University of Johannesburg","correspondingAuthor":false,"prefix":"","firstName":"Opeyeolu","middleName":"Timothy","lastName":"Laseinde","suffix":""}],"badges":[],"createdAt":"2026-03-17 07:08:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9145184/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9145184/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105982599,"identity":"157f5946-b44f-4c4a-9c21-671240e8cd42","added_by":"auto","created_at":"2026-04-02 07:03:19","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":433129,"visible":true,"origin":"","legend":"\u003cp\u003eConceptual overview of the complementary roles of machine learning and multi-criteria decision-making in water resource management\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-9145184/v1/69a9a8a8a0950443c8e1dc87.png"},{"id":106094140,"identity":"5987d6e8-57ea-44bc-9a55-bc05d8e63d29","added_by":"auto","created_at":"2026-04-03 11:41:16","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":201282,"visible":true,"origin":"","legend":"\u003cp\u003eFunctional role of MCDM within integrated MCDM-ML systems for water resource management.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-9145184/v1/6330f56a3dc2027cca074b94.png"},{"id":106093561,"identity":"c1d1d509-2919-471a-ae15-88e87d5855e0","added_by":"auto","created_at":"2026-04-03 11:38:02","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":215593,"visible":true,"origin":"","legend":"\u003cp\u003eFunctional role of ML within integrated MCDM-ML systems for water resource management.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-9145184/v1/19d4fc531fd5db8ee6a5866d.png"},{"id":106093803,"identity":"2da21ee9-3b8e-49ee-9925-78c573f6fcb3","added_by":"auto","created_at":"2026-04-03 11:39:17","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":279070,"visible":true,"origin":"","legend":"\u003cp\u003eMain integration patterns between MCDM and ML in water resource management\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-9145184/v1/8238ee9850e5445b45f57445.png"},{"id":106093698,"identity":"f4818891-66a4-45d9-9d9a-fdb8c3355506","added_by":"auto","created_at":"2026-04-03 11:38:41","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":572607,"visible":true,"origin":"","legend":"\u003cp\u003eGroundwater potential zones in the Kangsabati River basin.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-9145184/v1/b09213d3440f1d306c86d077.png"},{"id":105982601,"identity":"06ee1bce-cfbb-4ce4-817a-207bb2e48390","added_by":"auto","created_at":"2026-04-02 07:03:19","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":368093,"visible":true,"origin":"","legend":"\u003cp\u003eMethodological framework for groundwater potential assessment using geospatial and machine learning approaches.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-9145184/v1/1fffee6c3b00ce82759be1fb.png"},{"id":106093690,"identity":"596a929b-75f8-454e-82e6-d729e5edacb9","added_by":"auto","created_at":"2026-04-03 11:38:37","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":476023,"visible":true,"origin":"","legend":"\u003cp\u003eConceptual overview of how machine learning and multi-criteria decision-making approaches are integrated for flood risk and hydrological hazard management, highlighting the roles of hazard estimation, vulnerability assessment and model comparison\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-9145184/v1/253bfd34ef8072fc5c268986.png"},{"id":106093560,"identity":"dfb570ee-c07e-4140-a225-67b82a5dbf58","added_by":"auto","created_at":"2026-04-03 11:38:02","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":1248010,"visible":true,"origin":"","legend":"\u003cp\u003eSection 4 thematic synthesis across WRM application domains\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-9145184/v1/83206d569a82121e4946ccb2.png"},{"id":106095788,"identity":"c5f978e7-fce0-42b3-96e3-2fab04133fd5","added_by":"auto","created_at":"2026-04-03 11:51:09","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5515470,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9145184/v1/ec793568-5bb3-4c95-ac35-9d79949d05f8.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Structured Review and Critical Synthesis of Multi-Criteria Decision- Making Models Integrated with Machine Learning for Water Resource Management","fulltext":[{"header":"1. Introduction","content":"\u003cdiv id=\"Sec2\" class=\"Section2\"\u003e \u003ch2\u003e1.1 Water resource management as a multi-dimensional decision problem\u003c/h2\u003e \u003cp\u003eWater resource management (WRM) has increasingly become complex due to the combined effects of population growth, urbanisation, industrial expansion, climate variability, land-use change as well as rising pressure on finite freshwater systems (He et al. \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Li et al. \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). These pressures do not affect water systems in isolation, they fundamentally influence other major factors such as water availability, quality, allocation, infrastructure planning, ecosystem protection and long-term sustainability at the same time. As a result, WRM is no longer a matter of managing supply alone, but of balancing environmental, technical, social, economic and institutional considerations under conditions of uncertainty. Decision-makers are often therefore required to assess multiple and often conflicting criteria while responding to dynamic hydrological conditions as well as context-specific management priorities (Miller and Belton \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2014\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn practice, WRM decisions often involve trade-offs between short-term operational demands and long-term sustainability goals. For example, water allocation strategies may improve immediate service delivery while increasing ecological stress, and infrastructure interventions may strengthen supply reliability while introducing substantial cost and governance challenges (Zhu et al. \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Poli et al. \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). These trade-offs are further complicated by incomplete data, spatial and temporal variability as well as the need to incorporate both quantitative evidence and expert judgment into decision processes. Because of this, WRM is best understood as a multi-dimensional decision environment in which robust planning depends on methods that can capture complexity, uncertainty and competing priorities in a structured and defensible way.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e1.2 Why MCDM and ML are increasingly combined\u003c/h2\u003e \u003cp\u003eMulti-Criteria Decision-Making (MCDM) and Machine Learning (ML) have emerged as two important methodological approaches for addressing the complexity of WRM, but they contribute in different ways. MCDM methods are well suited to decision contexts where multiple criteria must be evaluated simultaneously, especially when factors such as stakeholder preferences, weighting of alternatives and transparent prioritisation are central to the problem (Kumar \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Kocaman and Asan \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). In contrast, ML methods are valuable for and in identifying hidden patterns within large datasets, forecasting future conditions, classifying system states and supporting adaptive decision-making in data-rich and uncertain environments. Their growing use in WRM reflects the need for methods that can move beyond conventional deterministic analysis and respond to increasingly complex water-related challenges (Drogkoula et al. \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe increasing integration of MCDM and ML is driven by their complementary strengths. ML can improve predictive capability by modelling non-linear relationships in hydrological, environmental, and demand-related datasets, while MCDM can translate those predictive outputs into structured decision support that reflects practical priorities and competing objectives. In this sense, ML helps answer what is likely to happen, while MCDM helps in evaluating what should be prioritised once that information is available (Schuwirth et al. \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Mohammadifar et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). This combination is particularly relevant in WRM applications such as groundwater potential mapping, flood risk assessment, water demand forecasting, water quality evaluation and infrastructure planning, where technical performance alone is not enough and decisions must also remain interpretable, actionable, and context sensitive, a summary of these is provided in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e1.3 Gap in the existing review literature\u003c/h2\u003e \u003cp\u003eAlthough the application of MCDM and ML in WRM has grown significantly over time, the review literature has not always kept pace with the way these methods are now being used together. Existing discussions often focus on MCDM methods or ML models separately, in other cases they summarise applications descriptively without critically examining how integrated approaches are structured, where they are most effective, what limitations they share and how comparable their results really are across studies (Hajkowicz and Collins \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Calizaya et al. \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). As a result, the literature still lacks a sufficiently focused synthesis of how MCDM-ML combinations are being applied across WRM domains and what methodological patterns can be identified from that body of work.\u003c/p\u003e \u003cp\u003eA further gap lies in the limited critical evaluation of performance reporting, interpretability, transferability and reproducibility in existing studies. Many published applications report strong results, but they often do so by using different datasets, validation strategies, criteria structures and performance metrics, which makes direct comparison difficult (Ghobadi and Kang \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Ahmed et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). In addition, there is limited consistency in how studies explain the interaction between predictive modelling and decision-ranking processes. This weakens the ability of researchers and practitioners to determine which integrated approaches are genuinely robust, which are context-dependent and which are difficult to generalise beyond individual case studies (Ali et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Costa et al. \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Drogkoula et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). These limitations create a clear need for a review that goes beyond method description and instead offers structured thematic synthesis, critical comparison and clearer guidance for future research and application\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e1.4 Aim and review questions\u003c/h2\u003e \u003cp\u003eThis review aims to critically examine how MCDM and ML have been integrated in WRM and to evaluate the extent to which these integrated approaches support more effective, transparent and context-appropriate decision-making. Rather than treating MCDM and ML as separate methodological streams, the review considers their joint use across major WRM application areas and assesses the strengths, limitations and practical implications of different combinations. The article is therefore positioned not only as a summary of existing studies, but as a synthesis of the conceptual patterns, application trends, methodological trade-offs as well as reporting gaps that currently shape the field.\u003c/p\u003e \u003cp\u003eTo guide this review, four main questions are addressed:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eHow are MCDM and ML being integrated in water resource management?\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eWhich combinations of MCDM and ML are most commonly applied across different WRM domains, and for what reasons?\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eWhat trade-offs emerge among predictive performance, interpretability, data requirements, scalability, and contextual suitability?\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eWhat methodological and reporting gaps continue to limit comparability, reproducibility and broader implementation of integrated MCDM-ML approaches in WRM?\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThese questions provide the foundation for the sections that follow as summarized in Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, including the review design, conceptual integration framework, thematic synthesis, critical comparison, and the proposed directions for future research and implementation.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eReview questions summary\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReview question\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFocus\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSection(s) addressing it\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eType of synthesis\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHow are MCDM and ML being integrated in water resource management?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIntegration logic, roles of MCDM and ML, and major integration patterns\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSection 3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eConceptual synthesis\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWhich combinations of MCDM and ML are most commonly applied across WRM domains, and for what reasons?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDomain-specific application patterns and commonly used method combinations\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSection 4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eThematic synthesis\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWhat trade-offs emerge among predictive performance, interpretability, data requirements, scalability, and contextual suitability?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStrengths, limitations, and comparative performance of integrated approaches\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSection 5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCritical comparative analysis\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWhat methodological and reporting gaps continue to limit comparability, reproducibility, and broader implementation?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMetric inconsistency, validation gaps, reproducibility, and framework development\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSections 6 and 7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGap analysis and future-oriented synthesis\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"2. Review design and literature selection","content":"\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Review approach\u003c/h2\u003e \u003cp\u003eThis study adopts a structured and systematic review approach combined with critical synthesis to examine the integration of Multi-Criteria Decision-Making (MCDM) and Machine Learning (ML) in water resource management. While systematic search and screening procedures were applied to ensure transparency and relevance, the review is positioned as a critical thematic synthesis rather than a fully PRISMA-compliant systematic review. This was important because the objective of the study was not only to summarise existing work, but also to critically compare how integrated MCDM-ML approaches have been applied across different WRM domains, how they have been evaluated, and where important methodological gaps remain (Snyder \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Page et al. \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2021a\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eRather than treating the review as a purely narrative survey, the study followed a structured evidence-synthesis logic in which the review questions, eligibility criteria, screening steps, appraisal approach and synthesis categories were defined in advance. This approach made it possible to organise the literature around explicit themes such as integration patterns, WRM application domains, interpretability, scalability, and reporting quality, which is more consistent with a critical review than a descriptive listing of studies.\u003c/p\u003e \u003cp\u003eThe review process follows key principles of systematic literature reviews, including defined search strategies, eligibility criteria, and structured synthesis; however, a formal PRISMA flow diagram and quantitative study selection reporting were not included, as the focus of this work is on conceptual integration, thematic synthesis and critical comparison across studies.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Search sources and search strategy\u003c/h2\u003e \u003cp\u003eThe literature search was conducted across five major academic databases namely: Scopus, Web of Science, Google Scholar, IEEE Xplore and SpringerLink. These sources were selected because together they provide broad coverage of peer-reviewed work in water resource management, environmental modelling, artificial intelligence, machine learning as well as decision-support research. The search strategy was designed to retrieve studies that explicitly addressed the integration of MCDM and ML rather than studies focused on either method family in isolation.\u003c/p\u003e \u003cp\u003eA combination of keywords and Boolean operators was used to identify relevant studies. The main search terms included \u0026ldquo;Multi-Criteria Decision-Making\u0026rdquo; or \u0026ldquo;MCDM\u0026rdquo;, \u0026ldquo;Machine Learning\u0026rdquo; or \u0026ldquo;Artificial Intelligence\u0026rdquo;, \u0026ldquo;Water Resource Management\u0026rdquo; or \u0026ldquo;Water Availability\u0026rdquo; and \u0026ldquo;Integration\u0026rdquo; or \u0026ldquo;Combined Approach\u0026rdquo;. These terms were combined using AND/OR operators to improve search sensitivity while retaining topical relevance.\u003c/p\u003e \u003cp\u003eSearches were limited to studies published in English, and the formal study selection for the reviewed articles covered literature up to 2024, while additional supporting references published up to 2025 were used where relevant to strengthen the conceptual framing, emerging directions, and discussion. A representative search string used across databases was: (\u0026lsquo;Multi-Criteria Decision-Making\u0026rsquo; OR MCDM) AND (\u0026lsquo;Machine Learning\u0026rsquo; OR \u0026lsquo;Artificial Intelligence\u0026rsquo;) AND (\u0026lsquo;Water Resource Management\u0026rsquo; OR \u0026lsquo;Water Availability\u0026rsquo;) AND (\u0026lsquo;Integration\u0026rsquo; OR \u0026lsquo;Combined Approach\u0026rsquo;).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Eligibility criteria\u003c/h2\u003e \u003cp\u003eTo maintain relevance and comparability, the review applied explicit inclusion and exclusion criteria during study selection. Studies were included if they addressed water resource management or a closely related water-sector application, explicitly integrated at least one MCDM method with at least one ML technique, reported a clear methodological workflow and presented sufficient information on model application, outputs or performance evaluation. In contrast, studies were excluded if they used MCDM or ML as standalone approaches without integration, fell outside the WRM context, lacked sufficient methodological detail, or were not peer-reviewed academic contributions.\u003c/p\u003e \u003cp\u003eThe eligibility process was also guided by the need to preserve the analytical focus of the review. Since the manuscript is concerned with the comparative value of integrated decision-support systems, studies were screened not only for topical relevance but also for their ability to contribute to cross-study synthesis in areas such as integration logic, data requirements, interpretability, and reported strengths or limitations. This helped avoid an overly broad review corpus and ensured that the final set of studies remained aligned with the core review questions.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eInclusion and exclusion criteria applied in study selection\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCriterion\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInclusion\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eExclusion\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy focus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExplicit integration of at least one MCDM method with at least one ML technique\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUse of MCDM only or ML only without integrated application\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eApplication area\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWater resource management or a directly related water-sector decision/problem\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eStudies outside WRM or only loosely related to water issues\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePublication type\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePeer-reviewed journal articles and conference papers with sufficient methodological detail\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEditorials, opinion pieces, theses without accessible review status, and non-scholarly material\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMethodological clarity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClear description of workflow, inputs, methods, and outputs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eInsufficient methodological detail to support comparison\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePerformance or decision output\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eReports model outputs, ranking outcomes, validation results, or comparable analytical findings\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNo usable results, no evaluation basis, or purely conceptual discussion without application\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLanguage and time frame\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEnglish-language studies published up to 2024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNon-English studies and material outside the defined search period\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Screening and study selection\u003c/h2\u003e \u003cp\u003eThe screening and study selection process followed the PRISMA logic of identification, screening, eligibility assessment and final inclusion. The records retrieved from the selected databases were first screened by title and abstract to remove clearly irrelevant studies and duplicates. Full-text screening was then conducted on the remaining records using the predefined inclusion and exclusion criteria (Page et al. \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2021b\u003c/span\u003e; Sohrabi et al. \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). This staged approach improved consistency in study selection and provided a clear basis for documenting how the final review dataset was assembled.\u003c/p\u003e \u003cp\u003eAt the full-text stage, particular attention was given to whether a study genuinely presented an integrated MCDM-ML workflow, rather than simply discussing both methods in the same paper. This distinction was particularly important because several water-related studies apply predictive ML models and decision tools in parallel without formally linking them in a single analytical framework. Only those studies that contributed meaningfully to the review\u0026rsquo;s comparative objective were retained for the final synthesis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Data extraction and synthesis framework\u003c/h2\u003e \u003cp\u003eA structured extraction template was used to record comparable information from each included study. The extracted items included the WRM application area, study objective, MCDM technique used, ML technique used, mode of integration, type of data employed, performance metrics reported, validation approach, interpretability features and the main strengths as well as limitations identified in the study. Recording these categories consistently was necessary for the later thematic and comparative synthesis developed in Sections 4 to 6.\u003c/p\u003e \u003cp\u003eThe synthesis framework was designed to go beyond summary description. Instead of reviewing each article in isolation, extracted studies were grouped according to WRM application domain and then compared using recurring analytical dimensions such as predictive performance, transparency, weighting sensitivity, computational demand, data intensity and transferability (Snyder \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). This structure supports the critical orientation of the manuscript and directly addresses the need for thematic rather than purely procedural synthesis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Quality appraisal and risk-of-bias considerations\u003c/h2\u003e \u003cp\u003eTo strengthen methodological consistency, the included studies were appraised using an adapted Critical Appraisal Skills Programme (CASP) logic (Schabbauer \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Kolaski et al. \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The appraisal focused on whether each study had a clear objective, an appropriate design, transparent data and modelling procedures, a justifiable integration workflow, sufficient reporting of outputs or performance, and a discussion of limitations relevant to interpretation and use. Although CASP was originally developed for broader evidence appraisal, its structured question-based approach provided a practical basis for judging methodological soundness and risk of bias across heterogeneous WRM studies.\u003c/p\u003e \u003cp\u003eRisk of bias was considered in relation to common weaknesses observed in the reviewed literature, including incomplete reporting of data sources, unclear validation procedures, insufficient explanation of criteria weighting, weak justification for method selection and limited treatment of uncertainty (Long et al. \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Shaheen et al. \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). These issues do not necessarily invalidate the studies, but they do affect how confidently their findings can be compared and generalised. For that reason, appraisal in this review was used not to exclude studies mechanically, but to support a more careful interpretation of the evidence base.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e2.7 Limitations of the review process\u003c/h2\u003e \u003cp\u003eAlthough the review was designed to be systematic and transparent, several limitations should be acknowledged. First, the search was restricted to selected academic databases and English-language publications, which may have excluded relevant studies published in other languages or in less visible outlets. Secondly, the emphasis on peer-reviewed literature improved quality control but may have reduced visibility of technical reports, policy documents and grey literature that could contain useful implementation insights, especially in developing-country WRM contexts. Thirdly, despite the use of structured screening and appraisal procedures, some degree of interpretive judgement remained necessary during eligibility assessment and comparative synthesis.\u003c/p\u003e \u003cp\u003eA further limitation arises from the heterogeneity of the reviewed studies themselves. Differences in datasets, performance metrics, validation strategies, decision criteria and reporting depth constrained the extent to which studies could be compared on a fully standardised basis. This is not only a limitation of the present review, but also an indication of a broader reporting problem in the MCDM-ML literature for WRM, which is addressed later in Section 6.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Conceptual framework for MCDM-ML integration in water resource management","content":"\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Roles of MCDM in integrated systems\u003c/h2\u003e \u003cp\u003eIn integrated water resource management systems, MCDM plays an important role in structuring decisions that involve multiple, and often conflicting, objectives. Water management decisions are rarely guided by a single consideration. They often involve several factors such as economic, environmental, technical and social factors that must be assessed together. In this context, MCDM provides a practical and systematic way of identifying relevant criteria, assigning their relative importance, and evaluating alternatives in a transparent manner. This makes it particularly useful in WRM, where the quality of a decision depends not only on technical evidence, but also on how competing priorities are balanced (Mendoza and Martins \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2006\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAnother important role of MCDM in integrated systems is that it helps convert technical outputs into decision-ready results. In many WRM applications, modelling and forecasting tools generate large amounts of information, but those outputs still need to be interpreted in a way that supports real planning choices. MCDM therefore helps organise that process by comparing alternatives against selected criteria and producing a structured basis for ranking or prioritisation. This is especially useful in water resource planning, where decision-makers often need to choose between competing interventions, management scenarios, or infrastructure options under uncertain conditions (Lai et al. \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2008\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn integrated MCDM-ML systems, MCDM often functions as the decision layer that receives and interprets outputs from predictive or simulation-based models. In other words, ML may be used to estimate future conditions, classify risks, or predict performance, while MCDM is used to determine which alternative should be preferred once those results are available. This role is valuable because predictive performance on its own does not resolve the wider decision problem. Water resource decisions still require trade-offs, judgment, and preference-based evaluation. In basin-scale water planning, for example, MCDM has been used to combine environmental, social and economic criteria in order to support more balanced and defensible decisions (Calizaya et al. \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2010\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eMCDM also strengthens integrated systems by improving transparency and making the final decision process easier to explain. Compared with purely data-driven approaches, MCDM makes it possible to show how criteria were selected, how alternatives were assessed, and why one option performed better than another. This is important in WRM because many decisions affect public systems, long-term planning, and policy implementation. For this reason, the role of MCDM in integrated systems as summarised in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, is not only analytical, but also interpretive and practical. It helps ensure that technically informed decisions remain understandable, defensible, and aligned with broader management objectives (Xi and Poh \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Across the reviewed literature, MCDM is most frequently used as the final decision-structuring layer rather than as a predictive tool. However, its effectiveness depends heavily on how criteria are selected and weighted, which introduces subjectivity into otherwise data-driven workflows. This creates a recurring trade-off in integrated systems: while MCDM improves transparency and interpretability, it can also introduce variability in outcomes when weighting structures are not robustly justified or validated.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Roles of ML in integrated systems\u003c/h2\u003e \u003cp\u003eMachine Learning (ML) primarily contributes to integrated systems for water resource management by identifying patterns in complicated datasets and producing predictions that are challenging to acquire through traditional analytical techniques alone. Several interrelated factors such as rainfall, temperature, streamflow, land use, groundwater behaviour, water demand and water quality conditions, have an impact on water systems. ML is helpful for finding patterns, predicting future conditions, and strengthening the analytical foundation upon which management decisions are made because these interactions are frequently nonlinear and dynamic. By transforming massive and diverse water-related datasets into data that may aid in planning, risk assessment, and operational response, machine learning (ML) thereby enhances integrated systems (Ahmed et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eOne of the roles of ML in integrated systems is its ability to support classification, clustering, anomaly detection as well as other forms of data-driven interpretation that go beyond simple prediction. In WRM, this is important because decision support often depends on recognising system states, identifying high-risk areas or distinguishing between management conditions that are not immediately visible through descriptive statistics alone (Ghobadi and Kang \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Reviews of ML in WRM have shown that its applications commonly extend across three broad functions, namely prediction, clustering and reinforcement-based optimisation, which makes it a flexible component within integrated decision-support systems.\u003c/p\u003e \u003cp\u003eML also plays an important role in improving responsiveness and adaptability in integrated systems. Because many water-related problems evolve over time including drought conditions, water demand shifts and water quality deterioration, decision support cannot rely only on static analysis. ML models can be trained on historical and real-time data to update forecasts, detect changes early, and improve situational awareness under changing environmental conditions (Liu et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This makes them particularly useful in integrated systems where the purpose is not only to understand present conditions, but also to support forward-looking and adaptive management.\u003c/p\u003e \u003cp\u003eAt the same time, the role of ML in integrated systems is not limited to its predictive strength. It also affects how useful the system becomes in practice. Recent work in hydroclimatic and water-related modelling has shown that interpretability and explainability are becoming increasingly important, especially where model outputs are expected to inform public decisions or policy-facing applications (Başağaoğlu et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). This means that, within integrated MCDM-ML systems, the value of ML lies not only in its ability to improve estimation accuracy, but also in its ability to produce outputs that can be meaningfully incorporated into transparent and defensible decision processes, a standard procedure in shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. However, across the reviewed WRM studies, the strength of ML in prediction does not automatically translate into improved decision-making. In many cases, high-performing models are used without sufficient explanation of how their outputs influence final decisions. This highlights a key limitation: predictive accuracy alone is insufficient unless it is explicitly linked to structured decision frameworks such as MCDM.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Main integration patterns\u003c/h2\u003e \u003cp\u003eThe integration of MCDM and ML in water resource management does not follow a single fixed structure. Across the literature, the two methods are combined in different ways depending on the decision problem, the type of data available, and the role expected from each method in the overall workflow. In most cases, the integration is not simply about combining two techniques for novelty (Gebre et al. \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Ali et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). It is done because ML and MCDM solve different parts of the problem. ML is generally used to learn from data, detect patterns, or generate predictions, while MCDM is used to evaluate alternatives, weigh criteria, and support final prioritisation. For that reason, the structure of integration usually reflects the order in which prediction and decision support are needed in practice.\u003c/p\u003e \u003cp\u003eOne of the most common patterns is sequential integration. In this arrangement, ML is applied first to produce outputs such as forecasts, classifications, susceptibility estimates, or scenario results, and these outputs are then passed into an MCDM framework for ranking or decision-making. This pattern is common because it separates analytical prediction from final prioritisation in a clear and practical way. In water-related applications, this is especially useful where the first task is to estimate future conditions or identify suitable zones, and the second task is to choose among alternatives using environmental, economic, or policy-related criteria. A recent systematic review of integrated multicriteria and data-driven methods also shows that many hybrid systems are built around this staged logic, where one method produces the informational base and the other supports the final decision layer (Costa et al. \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Nasiri Khiavi et al. \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eA second pattern is parallel integration, where MCDM and ML are applied alongside each other rather than in a strict sequence. In this case, the two methods may be used to analyse the same problem from different angles, after which their outputs are compared, combined, or jointly interpreted. This pattern is often used where researchers want to evaluate whether data-driven and criteria-based methods lead to similar conclusions, or where each method captures a different dimension of the problem. In environmental mapping studies, for example, MCDM may be used to represent expert-based weighting while ML is used to test predictive performance from data, allowing both perspectives to inform the final assessment (He et al. \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Khalil et al. \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eA third pattern is fully coupled or hybrid integration, where MCDM and ML are embedded within a single decision-support framework rather than treated as separate stages. In such systems, the interaction between the two methods is tighter. MCDM outputs may be used to support feature weighting, rule development, or optimisation, while ML outputs may directly shape the criteria structure, scenario evaluation, or final recommendation process. This type of integration is usually more complex, but it can be more useful in situations where prediction, optimisation, and decision support must interact continuously rather than sequentially. Water demand planning studies that combine intelligent forecasting with multi-criteria optimisation illustrate this more tightly linked form of integration (Sharma \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Poursaeid \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTaken together, these patterns show that integration is best understood as a design choice rather than a standard formula. Sequential integration tends to be clearer and easier to interpret. Parallel integration is useful where comparison and triangulation are important. Fully coupled integration is more ambitious and can be more powerful, but it usually demands stronger methodological justification and clearer reporting. For this reason, understanding the main integration patterns is important not only for describing the literature, but also for evaluating which integrated systems are actually suitable for different WRM contexts. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e summarises the main ways in which MCDM and ML are combined in WRM, namely sequential, parallel, and fully coupled integration patterns. Despite these distinctions, the literature does not consistently demonstrate that more complex integration leads to better decision outcomes, as fully coupled systems often lack transparency and reproducibility compared to simpler sequential approaches.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Thematic synthesis by WRM application domain","content":"\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Groundwater potential and recharge assessment\u003c/h2\u003e \u003cp\u003eGroundwater potential and recharge assessment is one of the strongest application areas for integrated geospatial, MCDM, and ML approaches in WRM. A study conducted by (Kanji and Das \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e) in the Kangsabati Upper Catchment showed how this integration can move beyond simple potential mapping toward recharge feasibility assessment. Their study combined XGBoost with AHP and considered both hydrogeological conditions and water quality constituents. The results showed that XGBoost outperformed SVM, RF, and ANN, achieving 81% accuracy, while the integrated workflow also identified suitable artificial recharge zones and proposed recharge structures for different lithological groups. This is important because it shows that integrated systems can support both resource identification and management action, rather than stopping at spatial classification alone.\u003c/p\u003e \u003cp\u003eAnother study conducted by (Ghosh et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) in the Kangsabati River basin (shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e) illustrates the continued relevance of MCDM-based groundwater zoning where structured weighting of conditioning factors is required. Using MCDA and AHP, the authors integrated geology, geomorphology, elevation, slope, drainage, lineament, curvature, topographic wetness, land use/land cover, and soil to delineate groundwater potential zones. Their results showed that 14.62% of the basin fell within high potential zones and 6.11% within very high potential zones. More importantly, the mapped zones were checked against pre-monsoon and post-monsoon groundwater depth data and ROC-based validation, which strengthened confidence in the applicability of the method. This study shows that MCDM remains particularly useful where the main goal is transparent spatial prioritisation grounded in hydrogeological reasoning.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAdapted from (Ghosh et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Assessment of groundwater potential zone using MCDA and AHP techniques: case study from a tropical river basin of India. Applied Water Science, 12:37, under the Creative Commons Attribution (CC BY 4.0) license.\u003c/p\u003e \u003cp\u003e(Anand et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) shows a more recent shift toward ensemble ML models for groundwater potential assessment, especially in data-rich and urban settings. Their study using the methodology in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, compared six ML models and found that Random Forest was the most reliable, with an AUC of 0.91, while AdaBoost achieved the highest MCC.\u003c/p\u003e \u003cp\u003eThe sensitivity analysis further showed that geomorphology, elevation, and rainfall were among the most influential predictors. A key observation is that Which suggests that the strongest direction Which suggests that the strongest directionthe study did not just compare models by accuracy but also examined how changes in specific input layers affected prediction outcomes. This gives the assessment more interpretive value and suggests that recent ML-based groundwater studies are increasingly moving toward stronger model diagnostics instead of relying only on final performance scores.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAdapted from (Anand et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Evaluating groundwater potential with the synergistic use of geospatial methods and advanced machine learning approaches. Discover Cities, 2:56, under the Creative Commons Attribution (CC BY 4.0) license.\u003c/p\u003e \u003cp\u003e(Akbari et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) adds another useful dimension by comparing different MCDM weighting frameworks for groundwater recharge potential mapping. Using AHP, BWM, and FUCOM, the authors showed that groundwater recharge zones can be highly sensitive to how weights are assigned to conditioning factors. Across their models, land use and lithology emerged as the most important variables, while slope angle had the lowest influence. Their findings suggest that recharge assessment is not only a mapping problem, but also a weighting problem, because different decision structures can influence the final spatial pattern even when the same environmental layers are used. This is important for the present review because it highlights that, in groundwater studies, methodological transparency is just as important as predictive performance.\u003c/p\u003e \u003cp\u003eTaken together, these studies show a clear pattern in the groundwater literature. MCDM-based approaches remain strong where transparent weighting, hydrogeological reasoning, and practical prioritisation are central. ML-based approaches are increasingly preferred where the data structure is more complex and higher predictive discrimination is required. The most useful direction, however, appears to be integration. Studies that combine predictive ML with structured MCDM are better placed to support groundwater management decisions that need both analytical strength and practical interpretability. This is especially important for recharge assessment, where the objective is not only to identify where groundwater potential exists, but also where intervention is feasible and justifiable.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e summarises the main groundwater potential and recharge assessment studies discussed in this subsection, with emphasis on their methodological approach, key inputs, and main contribution to the field.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparative overview of groundwater potential and recharge assessment studies reviewed in this subsection\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStudy area\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eApproach\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eKey conditioning factors\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMain result\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMain contribution\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Kanji and Das \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eKangsabati Upper Catchment, India\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eXGBoost, SVM, RF, and ANN integrated with AHP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHydrogeological conditions and water quality constituents\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eXGBoost achieved the highest performance with 81% accuracy, and the framework identified suitable artificial recharge zones\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eDemonstrated that integrated ML-MCDM systems can support both groundwater potential mapping and recharge feasibility planning\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Ghosh et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eKangsabati River basin, India\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMCDA integrated with AHP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGeology, geomorphology, elevation, slope, drainage density, lineament density, curvature, topographic wetness index, land use/land cover, and soil\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14.62% of the basin was classified as high groundwater potential and 6.11% as very high potential, with validation using groundwater depth data and ROC analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eShowed the continued value of transparent factor weighting for groundwater potential zoning and practical spatial prioritisation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Anand et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUrban groundwater assessment context\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eComparative geospatial ML modelling using six ML algorithms\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGeomorphology, elevation, rainfall, and other groundwater conditioning factors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRandom Forest achieved the best predictive performance with an AUC of 0.91, while AdaBoost produced the highest MCC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHighlighted the strength of ensemble ML models and the importance of sensitivity analysis in groundwater potential assessment\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Akbari et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGroundwater recharge assessment context\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eComparative MCDM weighting using AHP, BWM, and FUCOM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLand use, lithology, slope angle, and other recharge-related environmental layers\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLand use and lithology were the most influential variables, while slope angle had the least influence across the models\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eShowed that recharge mapping outcomes are sensitive to the choice of weighting framework and that methodological transparency is critical\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Water demand forecasting and supply planning\u003c/h2\u003e \u003cp\u003eIn a review conducted by (Niknam et al. \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) it was shown that water demand forecasting has become a core planning tool rather than only an operational exercise. Their review of more than 100 studies found that traditional time-series models and artificial neural networks were among the most widely used methods, but they also made it clear that no single forecasting method performs best in every system. Instead, model choice depends on the time horizon, the type of data available, and the management objective, whether that is pumping efficiency, leakage control, pressure management, or long-term supply planning. This is an important point for this review because it shows that forecasting models only become useful in practice when they are aligned with a clear planning purpose.\u003c/p\u003e \u003cp\u003e(Sharma \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) moved this discussion further by explicitly linking water demand forecasting to supply-side decision-making in Saudi Arabia, as shown in their workflow. The study proposed a combined framework based on Multi-Criteria Optimization and Intelligent Water Demand Forecasting, where monthly demand data were used to compare alternative water management responses, including interbasin transfer, rainwater harvesting, greywater recycling, water recycling, and irrigation-related options. The study reported a prediction accuracy of 98.96% and an optimization ratio of 97.87% and argued that forecasting should not end with demand estimation alone but should feed directly into the evaluation of practical supply and reuse strategies. Even though the paper is more application-driven than critical, it is useful here because it shows how hybrid ML-MCDM frameworks can connect demand prediction to actual planning choices.\u003c/p\u003e \u003cp\u003e(Khalilzadeh et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) further adds a different but equally relevant planning dimension by showing that supply planning is not only about matching demand and source options, but also about managing project risk. Their framework combined Fuzzy DEMATEL, Fuzzy ANP, and ANN to assess environmental risks in water supply projects and showed that technical risks received the highest weight in both the ANP and ANN assessments, while supplier-related and communication-related risks also emerged as important decision variables. Although the study is not a demand forecasting paper in the narrow sense, it strengthens this subsection because it shows that supply planning becomes more robust when prediction and prioritisation are combined with formal risk assessment. In other words, planning future water supply requires not only knowing how much water may be needed, but also understanding which project and implementation risks are most likely to undermine delivery.\u003c/p\u003e \u003cp\u003eEssentially, these studies show that water demand forecasting and supply planning are most effective when treated as linked parts of the same decision problem. Forecasting improves the timing and direction of planning, while MCDM-based optimisation and risk assessment help decision-makers compare options, account for uncertainty, and justify final choices. This suggests that the strongest direction in this domain is not forecasting alone, and not planning alone, but integrated systems that connect predictive outputs to structured, defensible supply decisions. Taken together, these studies show that water demand forecasting becomes significantly more valuable when it is directly linked to decision-making frameworks. Forecasting alone provides limited practical benefit unless it is embedded within optimisation or risk-based evaluation structures. This distinguishes the water demand domain from others, where prediction is often treated as an end rather than a means to structured planning.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Flood risk and hydrological hazard management\u003c/h2\u003e \u003cp\u003eFlood risk and hydrological hazard management are some of the clearest areas where integrated MCDM-ML systems have shown practical value. A study conducted by Pham et al. in Quang Nam Province, Vietnam, developed a flood risk framework that combined hybrid artificial intelligence models with MCDA by linking a flood susceptibility map to a flood consequences map. Their results showed that the Bagging-DT model performed best, with an AUC of 0.96, outperforming AdaBoost-DT and the single DT model. What makes this study especially useful is that it did not stop at susceptibility mapping (Pham et al. \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). It moved further into risk assessment by incorporating human health and financial consequences, which makes the final output more relevant for flood management than hazard mapping alone.\u003c/p\u003e \u003cp\u003e(Gholami et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) then showed how this application area is now moving beyond conventional machine learning into deep learning. In southern Iran, they combined a bidirectional LSTM (bLSTM) model with the COPRAS MCDM method to produce a flood risk map that integrated hazard and vulnerability. Their results showed that 15.8% of the study area fell within the high-risk class and 14.6% within the very high-risk class. The study is important because it shows that deep learning can improve the hazard side of flood analysis, while MCDM still remains necessary for vulnerability weighting and final risk interpretation. It also makes clear that topographic variables such as TWI, river density, TPI, SPI, slope, elevation, and distance to river remain dominant even in more advanced modelling workflows.\u003c/p\u003e \u003cp\u003e(Rashidi Shikhteymour et al. \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) adds an important social dimension to this domain. Their work in Abarkuh County, Iran, integrated SVM-based flood hazard mapping with ANP-DEMATEL-based vulnerability assessment. The SVM model performed best among the tested ML models, and the final integrated results showed that 6% of the study area was classified as high or very high flood risk. A key strength of this study is that it treated flood risk as the interaction between physical hazard and social vulnerability rather than as a purely hydrological problem. This makes the approach more useful for local authorities because it helps identify where flood exposure coincides with weak adaptive capacity and socio-economic sensitivity.\u003c/p\u003e \u003cp\u003eFurthermore, (Asiri et al. \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) extended this integration logic into the coastal flood context by combining AHP with SVM and Decision Tree models under RCP 2.6 and RCP 8.5 scenarios. Their results showed that the AHP-DT model achieved the best performance, with an AUC of 0.95. This study is useful because it brings climate-change-induced sea-level rise and coastal flood drivers into the flood risk workflow. It also shows that integrated MCDM-ML approaches are not limited to inland flood susceptibility studies but can be adapted to exposure-prone coastal environments where scenario-based planning is essential.\u003c/p\u003e \u003cp\u003eA different perspective is provided by (Akay \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), who compared statistical, fuzzy logic, and MCDM methods for flood hazard susceptibility mapping in Turkey. Although the study did not build a full integrated ML-MCDM risk model in the same way as the other papers, it remains useful for this subsection because it showed that model choice strongly affects mapping outcomes. The IoE model performed best under ROC-based evaluation, while PCA and AHP performed better under SCAI-based assessment. This is an important reminder that, in flood and hazard management, strong predictive performance does not always align perfectly with spatial reliability or interpretive usefulness. That point matters for this review because it reinforces the need for careful model selection and transparent validation when integrated systems are used for planning and intervention.\u003c/p\u003e \u003cp\u003eThese studies show that the flood literature has moved well beyond single-model hazard mapping. A clear pattern has emerged in which ML or DL models are increasingly used to estimate hazard or susceptibility, while MCDM methods are used to handle consequences, vulnerability, exposure, weighting, and final prioritisation. The strongest studies are those that treat flood risk as a combined physical and socio-economic problem rather than a prediction problem alone. This is particularly important in hydrological hazard management, where decisions about intervention, infrastructure, and resource allocation depend on more than identifying where flooding is likely to occur. Figure\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e summarises how the reviewed studies integrate machine learning and multi-criteria decision-making approaches for flood risk and hydrological hazard management. The diagram highlights how predictive models are used to estimate flood susceptibility or hazard, while MCDM frameworks support vulnerability weighting, risk classification, and final decision-oriented flood risk mapping.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e4.4 Water quality assessment and pollution control\u003c/h2\u003e \u003cp\u003e(Das \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e) conducted a study on the Mahanadi River and showed how water quality assessment can be strengthened when GIS, MCDM, and machine learning are used together in a single framework. The study combined DEMATEL-based WQI with Random Forest, Decision Tree, Multilayer Perceptron, Na\u0026iuml;ve Bayes, and a Borda Scoring Algorithm, and used IDW interpolation to map spatial variation in pollution.\u003c/p\u003e \u003cp\u003eThe results showed that 31.58% of the tested locations were classified as poor water, 15.79% as very poor, and 5.26% as unsuitable, while the Random Forest model produced the best overall classification performance, with 90.50% accuracy, 99.87% sensitivity, and 74.56% specificity. The study is particularly useful because it did not only classify water quality status, but also moved into pollution source apportionment, identifying domestic organic debris and expanding human activity as major contributors to degradation. This makes the framework more useful for pollution control than WQI reporting alone.\u003c/p\u003e \u003cp\u003eA second study conducted by (Das \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025b\u003c/span\u003e) in the same catchment approached the problem through a different MCDM structure by integrating CILOS-based WQI, EDAS, ANN, and GIS for river water quality management. The study found that 31.58% of sites fell under poor water quality and 5.26% under very poor water quality, while the ANN model achieved an R\u0026sup2; of 0.96 across training, cross-validation, and testing.\u003c/p\u003e \u003cp\u003eIt also identified total coliform, total Kjeldahl nitrogen, electrical conductivity, total dissolved solids, and chloride as major contributors to water quality decline. What makes this study important for the present review is that it shows how integrated systems can move beyond broad quality classification toward ranking polluted sites, identifying influential pollutants, and supporting more targeted water quality control strategies.(Schuwirth et al. \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) adds an important management perspective to this domain by showing that integrated water quality assessment should not only identify current degradation but also evaluate future management alternatives. Their work combined multi-criteria decision analysis, integrated water quality modelling, and scenario planning to compare the future cost-effectiveness of management options under uncertainty.\u003c/p\u003e \u003cp\u003eThis is valuable for the present subsection because it shows that water quality assessment becomes more useful for pollution control when model outputs are linked to structured decision support rather than treated as stand-alone environmental indicators. In practical terms, this means integrated ML-MCDM type systems are most effective when they help decision-makers move from diagnosis to intervention. Taken together, these studies show that water quality applications are strongest when ML supports classification, pollutant identification, and spatial pattern recognition, while MCDM structures site ranking, intervention prioritisation, and management trade-offs. The most useful integrated systems are those that move beyond water quality diagnosis and support pollution control decisions under uncertainty.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003e4.5 Sediment, watershed, and catchment management\u003c/h2\u003e \u003cp\u003eIn a work conducted by (Ghosh and Mukhopadhyay \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), it was shown that sediment and watershed management can be strengthened when sub-watershed prioritisation is treated as a multi-model decision problem rather than a single-method exercise. In the Dwarkeswar River basin, they used an MCDM-based ensemble approach that combined SAW, COPRAS, ARAS, TOPSIS, and MOORA, while AHP was used to assign weights to the morphometric parameters. Their results showed that the COPRAS model produced the highest prediction accuracy, but the final prioritisation was based on an adjusted ensemble ranking rather than a single model output.\u003c/p\u003e \u003cp\u003eThis is important because it addresses a recurring problem in catchment management, namely that different prioritisation models can produce different rankings for the same watershed. By averaging statistically meaningful model outputs, the study moved toward a more defensible basis for identifying erosion-prone sub-watersheds and allocating management effort.\u003c/p\u003e \u003cp\u003e(Kanji and Das \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2025b\u003c/span\u003e) expanded this line of work by integrating MCDM and ensemble machine learning to prioritise sub-watersheds based on surface runoff potentialities in the upper Kangsabati catchment. Their study considered morpho-tectonic, hydrological, and physical characteristics together, and then linked TOPSIS and VIKOR with Extra Trees, XGBoost, Bagging, and Voting models. The results showed that TOPCOM-ETR, VIKHYP-ETR, and TOPMPT-ETR were the most effective combinations, with low RMSE values and strong discrimination of runoff-generating sub-watersheds. What makes this study especially useful for this subsection is that it shows watershed and catchment management is no longer limited to erosion ranking based on morphometry alone. It is now moving toward more integrated frameworks that combine terrain structure, hydrological behaviour, and machine learning-based prioritisation to guide runoff control and water resource management at sub-watershed scale.\u003c/p\u003e \u003cp\u003eTaken together, these studies show that sediment, watershed, and catchment management is increasingly shifting from single-factor prioritisation toward integrated decision-support systems. The older strength of MCDM in this domain lies in transparent ranking and management prioritisation, especially where morphometric and erosion-related indicators dominate. The newer contribution of ML is that it improves the ability to handle non-linear watershed behaviour and more diverse input structures. The strongest direction therefore appears to be integration, where MCDM supports prioritisation and interpretability, while ML strengthens predictive discrimination across sub-watersheds.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003e4.6 Urban water systems and loss management\u003c/h2\u003e \u003cp\u003e(Ayad et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) showed that urban water loss management becomes more effective when field activities, hydraulic modelling, optimisation, and GIS are treated as part of one workflow rather than as separate utility functions. Their approach combined EPANET, genetic algorithms, SCE-UA, field leak detection, and GIS-based data handling to calibrate the network, estimate physical losses, identify faulty meters, and improve pipe roughness estimation. What makes this study important for this subsection is that it framed non-revenue water reduction as an integrated operational problem, where leak quantification, network calibration, and spatial decision support must work together if utilities are to move beyond reactive repair practices.\u003c/p\u003e \u003cp\u003eIn a review by (Romero-Ben et al. \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) it was shown that leak detection and localization in urban water distribution networks have shifted from purely model-based approaches toward broader combinations of model-based, data-driven, and mixed strategies. Their review explained that utilities are increasingly relying on software-based methods linked to SCADA, flow and pressure sensing, and district-metered structures to improve how quickly leaks are detected and located. This is important because it shows that urban loss management is no longer only about identifying a leak after major failure. It is now tied to continuous monitoring, network partitioning, and automated decision support, all of which make integrated MCDM-ML type systems more relevant for urban utilities.\u003c/p\u003e \u003cp\u003eA review conducted by (Serafeim et al. \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) reinforced this by showing that leakage control in urban systems depends not only on detection technology, but also on how utilities estimate losses, understand influencing factors, and choose mitigation strategies. Their review covered the water balance, BABE, and minimum night flow approaches, while also showing that leakage severity is shaped by infrastructure condition, maintenance quality, monitoring capability, and broader water stress conditions. For this review, that is important because it shows that loss management in urban systems is both a technical and managerial decision problem. In other words, reducing leakage is not simply a matter of better sensors, but of combining estimation, prioritisation, and intervention in a structured way. (Xu et al. \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2014\u003c/span\u003e) adds a useful environmental perspective by showing that leakage control in urban distribution networks produces benefits that extend beyond direct water savings. Their review concluded that leakage can be further reduced through better detection capability, improved pipe maintenance strategies, and pressure regulation, while also noting that these actions reduce energy use and greenhouse gas emissions.\u003c/p\u003e \u003cp\u003eThis makes the urban water loss problem especially relevant in WRM because it connects infrastructure efficiency with broader environmental sustainability. It also supports the argument that urban loss management should be treated as a strategic part of integrated water system planning rather than only as a maintenance task. Compared to other WRM domains, urban water loss management shows a stronger emphasis on operational integration rather than methodological innovation. The focus is less on developing new hybrid models and more on combining monitoring, modelling, and optimisation into actionable workflows. This suggests that, in practice, integration effectiveness is often determined more by implementation design than by algorithmic complexity.\u003c/p\u003e \u003cp\u003eTo summarize on this, section 4 shows that the integration of MCDM and ML has been applied across a wide range of water resource management domains, including groundwater potential and recharge assessment, water demand forecasting and supply planning, flood risk and hydrological hazard management, water quality assessment and pollution control, sediment and catchment prioritisation, and urban water loss management. Across these domains, a clear pattern emerges: ML strengthens prediction, classification, and pattern detection, while MCDM improves weighting, prioritisation, and the translation of technical outputs into practical decisions, Table \u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e provides a cross-domain quantitative comparison across the studies. The strongest studies are those that do not treat prediction and decision support as separate tasks but integrate them into coherent decision-support frameworks. However, the depth of this integration varies significantly across domains, with many studies still relying on loosely connected workflows where predictive outputs are not fully embedded into decision structures. This limits the practical applicability of otherwise high-performing models and highlights a key gap in the literature: the need for more tightly integrated, transparent, and reproducible MCDM-ML frameworks that explicitly link prediction to actionable decision-making.\u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e summarises the thematic pattern across all WRM domains discussed in Section 4.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCross-domain quantitative comparison of integrated MCDM\u0026ndash;ML studies reviewed\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDomain\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMCDM\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eML\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eIntegration pattern\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMetric(s) reported\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMain result\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eValidation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eComparative note\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Ghosh et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGroundwater potential\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMCDA, AHP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNo ML (MCDM-only approach)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMCDM-led spatial prioritisation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eROC, spatial validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHigh potential 14.62%, very high 6.11%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003ePre-/post-monsoon depth\u0026thinsp;+\u0026thinsp;ROC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eStrong transparency, limited direct comparison with ML-heavy studies\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Anand et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGroundwater potential\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNot explicitly used\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF, AdaBoost, etc.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eParallel / comparative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC, MCC, CA, F1, Precision, Recall\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eRF AUC 0.91; AdaBoost highest MCC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eHeld-out validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eStrong predictive comparison, weaker decision-layer transparency\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Kanji and Das \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGroundwater recharge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAHP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXGBoost, SVM, RF, ANN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSequential\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eXGBoost 81%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eModel comparison\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eUseful integrated recharge planning, but limited metric breadth\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Sharma \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2022\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDemand forecasting and supply planning\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMulti-criteria optimisation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eIntelligent forecasting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFully coupled\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePrediction accuracy, optimisation ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e98.96% accuracy; 97.87% optimisation ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSimulation-based\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eStrong application claim, but metric not directly comparable across domains\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Pham et al. \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2021\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFlood risk\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMCDA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBagging-DT, AdaBoost-DT, DT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSequential\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eBagging-DT AUC 0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eROC/AUC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eStrong hazard discrimination, but risk depends on consequence integration\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Gholami et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFlood risk\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCOPRAS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ebLSTM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFully coupled\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eRisk-class distribution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e15.8% high risk, 14.6% very high risk\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eSpatial risk mapping\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eGood hybrid logic, limited cross-study comparability\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Das \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWater quality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDEMATEL, WQI, Borda\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF, DT, MLP, NB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFully coupled\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAccuracy, sensitivity, specificity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eRF 90.50% accuracy, 99.87% sensitivity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eClassification validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eStrong diagnostic framework, but metrics differ from other domains\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Das \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025b\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWater quality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCILOS, EDAS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eANN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSequential\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eR\u0026sup2;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eANN R\u0026sup2; 0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eTrain/cross-val/test\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eGood predictive fit, weak comparability with AUC-based studies\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAlthough the reviewed studies frequently report strong case-specific performance, direct cross-study comparison remains limited because different WRM domains rely on different evaluation logics and reporting conventions. Groundwater and flood studies most commonly report discrimination-based metrics such as AUC, ROC, and MCC, whereas water quality and demand forecasting studies more often report accuracy, sensitivity, specificity, R\u0026sup2; or optimisation ratios. This makes it difficult to identify a single best integrated MCDM\u0026ndash;ML configuration across domains. Instead, the quantitative evidence suggests a domain-dependent pattern: ensemble ML models often provide strong predictive discrimination in groundwater and flood applications, while MCDM components add the greatest value in weighting, prioritisation, and intervention structuring. The comparison also shows that studies using integrated frameworks often report stronger practical relevance than studies relying on prediction alone, but this practical advantage is rarely evaluated using standardised metrics.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMost commonly reported quantitative metrics across WRM domains\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDomain\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMost common metrics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTypical output type\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGroundwater potential/recharge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAUC, ROC, Accuracy, MCC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSpatial suitability / classification\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFlood risk/hazard\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAUC, ROC, risk-class distribution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHazard/risk discrimination\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWater demand forecasting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRMSE, MAE, R\u0026sup2;, optimisation ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eForecasting / planning\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWater quality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy, sensitivity, specificity, R\u0026sup2;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClassification / pollutant prioritisation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCatchment/watershed prioritisation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRanking outputs, RMSE in hybrid studies\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrioritisation / runoff discrimination\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUrban water loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOperational indicators, calibration outputs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSystem optimisation / intervention support\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"5. Critical comparison of integrated approaches","content":"\u003cdiv id=\"Sec26\" class=\"Section2\"\u003e \u003ch2\u003e5.1 Predictive performance versus transparency\u003c/h2\u003e \u003cp\u003eOne of the clearest trade-offs across the studies reviewed in Section 4 is between predictive performance and decision transparency. In several domains, ML-driven models produced stronger predictive or classificatory performance than more traditional rule-based or purely weighted approaches. This was especially visible in groundwater assessment, where XGBoost and Random Forest showed stronger discrimination than several competing models, and in flood risk studies where hybrid or ensemble models such as Bagging-DT, bLSTM, and AHP-DT achieved high predictive performance.\u003c/p\u003e \u003cp\u003eThese results suggest that integrated approaches benefit when ML is used to capture non-linear relationships in complex environmental datasets. However, the gain in predictive strength does not automatically make the overall system easier to interpret, especially when the decision pathway between input variables, intermediate outputs, and final recommendations is not clearly explained.\u003c/p\u003e \u003cp\u003eBy contrast, MCDM-led approaches were generally more transparent in showing how criteria were selected, weighted, and translated into final rankings or priority zones. This was particularly evident in groundwater recharge assessment and watershed prioritisation studies, where methods such as AHP, FUCOM, BWM, TOPSIS, and VIKOR made the logic of prioritisation easier to follow.\u003c/p\u003e \u003cp\u003eIn these cases, the main strength was not necessarily superior predictive accuracy, but rather the ability to justify why one area, option, or intervention was ranked above another. This is an important advantage in WRM because many planning decisions need to be explained to stakeholders, utilities, or public agencies, not just technically optimised.\u003c/p\u003e \u003cp\u003eThe most effective integrated approaches were therefore not the ones that maximised predictive performance alone, but the ones that balanced predictive strength with a decision structure that remained understandable and practically useful. This was evident in studies on water demand planning, water quality management, and urban water systems, where forecasting or classification became more valuable once linked to optimisation, ranking, source identification, or operational prioritisation.\u003c/p\u003e \u003cp\u003eIn such cases, ML improved the analytical depth of the system, while MCDM or related decision structures improved how the results could be acted on. In other words, predictive accuracy became more meaningful when it was embedded in a framework that supported intervention rather than only diagnosis.\u003c/p\u003e \u003cp\u003eTaken together, the comparison suggests that predictive performance and transparency should not be treated as competing goals in absolute terms, but as design variables that need to be balanced according to the WRM context. Where the main need is strong spatial discrimination or forecasting accuracy, more advanced ML components may be justified. Where the main need is defensible prioritisation, stakeholder communication, or policy-facing justification, transparent MCDM structures remain especially valuable. The strongest integrated approaches are those that combine both, allowing technically strong results to remain interpretable enough for real planning and management use.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec27\" class=\"Section2\"\u003e \u003ch2\u003e5.2 Sensitivity to weighting and expert judgment\u003c/h2\u003e \u003cp\u003eAnother important issue across the integrated approaches reviewed in Section 4 is the extent to which final outputs remain sensitive to weighting structures and expert judgment. This is especially visible in groundwater and recharge assessment, where the same environmental layers can produce different priority zones depending on how criteria are weighted. A study conducted by Akbari et al. showed that groundwater recharge maps changed across AHP, BWM, and FUCOM, even though the underlying conditioning factors remained broadly the same. Their results make it clear that the final pattern is not determined only by environmental data, but also by how decision importance is assigned to factors such as land use, lithology, and slope. This matters because it shows that integrated systems are not purely objective, even when they include strong predictive or spatial components. The decision structure itself can shape the output in meaningful ways.\u003c/p\u003e \u003cp\u003eA similar issue appears in watershed and catchment management. The study conducted by Ghosh and Mukhopadhyay showed that different MCDM models can generate different prioritisation results for the same sub-watersheds, even when the same morphometric parameters are used. In their study, SAW, COPRAS, ARAS, TOPSIS, and MOORA did not produce identical rankings, which is why the authors moved toward an ensemble-adjusted result instead of relying on a single model. This is important because it suggests that weighting and ranking sensitivity is not only a groundwater issue. It also affects sediment and catchment management, where management decisions may shift depending on which prioritisation structure is selected. In such cases, expert judgment remains influential because it affects both the assignment of weights and the choice of ranking logic.\u003c/p\u003e \u003cp\u003eIn flood risk and hazard management, the role of expert judgment becomes even more significant because vulnerability, consequences, and exposure are often harder to quantify than physical hazard variables. Rashidi Shikhteymour et al. showed that flood risk results depended not only on the hazard map produced by SVM, but also on the ANP-DEMATEL structure used to assess social vulnerability.\u003c/p\u003e \u003cp\u003eThis means that even where the predictive model performs well, the final risk outcome still depends on how social and environmental criteria are framed and weighted. The same issue is visible in coastal flood assessment, where AHP was used alongside SVM and Decision Tree models to incorporate scenario-driven flood risk factors. These examples show that integrated approaches often remain partly judgment-driven, especially when they move from hazard estimation into final risk interpretation.\u003c/p\u003e \u003cp\u003eIn practical terms, this comparison suggests that sensitivity to weighting and expert judgment should be treated as a normal feature of integrated MCDM-ML systems rather than a weakness to be ignored. The issue is not that expert input exists, but whether it is applied transparently, justified clearly, and tested for robustness. Where weighting choices are left unexplained, the credibility of the final output becomes weaker, even if predictive performance appears strong. By contrast, studies that make their weighting logic explicit, compare alternative structures, or use ensemble ranking approaches provide more defensible results. For WRM, this is important because many interventions are not chosen only on technical grounds. They are also shaped by planning priorities, institutional preferences, and local decision contexts.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSensitivity of integrated MCDM-ML approaches to weighting structures and expert judgment across WRM domains\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWRM domain\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIntegrated approach\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWhere weighting or judgment enters\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMain sensitivity observed\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePractical implication\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGroundwater recharge assessment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAHP, BWM, FUCOM with spatial evaluation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFactor weighting of recharge-related variables\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDifferent weighting frameworks produced different recharge zone patterns\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eWeighting structure should be reported clearly and, where possible, compared across methods\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWatershed and catchment management\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAHP with SAW, COPRAS, ARAS, TOPSIS, MOORA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eParameter weighting and ranking logic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDifferent MCDM models produced different prioritisation outputs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eEnsemble or comparative ranking can improve robustness\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFlood risk management\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSVM with ANP-DEMATEL; AHP with SVM/DT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eVulnerability weighting, consequence assessment, and scenario framing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFinal risk classes depended on social and environmental weighting choices\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHazard prediction should be complemented by transparent vulnerability weighting\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWater demand and supply planning\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eForecasting with multi-criteria optimisation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRanking of supply and reuse options\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFinal planning choices depended on prioritisation rules beyond forecast output\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eForecast accuracy alone is not enough for defensible supply planning\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWater quality and pollution control\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWQI with DEMATEL, CILOS, EDAS, ML models\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eParameter weighting, site ranking, and pollutant importance\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePriority sites and pollutant rankings changed with decision structure\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePollution control frameworks should explain how criteria importance is assigned\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec28\" class=\"Section2\"\u003e \u003ch2\u003e5.3 Data requirements, computational cost, and scalability\u003c/h2\u003e \u003cp\u003eThe studies reviewed in Section 4 show that integrated MCDM-ML approaches differ significantly in their data requirements, computational burden, and scalability. In general, MCDM-led approaches tend to be easier to apply where datasets are limited but expert knowledge and spatial layers are available. This was particularly clear in groundwater zoning, recharge assessment, and watershed prioritisation studies, where methods such as AHP, FUCOM, BWM, TOPSIS, and related ranking models worked effectively with mapped conditioning factors, expert-derived weights, and GIS-based overlays. These approaches are usually less computationally demanding than ensemble ML or deep learning models, and they are often easier to implement in settings where technical infrastructure or long-term historical datasets are limited. Their main advantage is that they remain workable under relatively modest data conditions, although that often comes at the cost of lower adaptive or predictive capacity.\u003c/p\u003e \u003cp\u003eBy contrast, the ML-heavy and ensemble-based approaches reviewed in Section 4 generally required richer datasets and greater computational effort, but they also offered stronger predictive discrimination in more complex environments. This was evident in groundwater studies using Random Forest, XGBoost, AdaBoost, and other ensemble models, as well as in flood risk studies that used Bagging-DT and bLSTM. These methods were better able to capture non-linear interactions and variable importance across multiple predictors, but their performance depended more strongly on data volume, input diversity, and validation design. In practical terms, this means that high-performing integrated models are often more suitable in data-rich environments where computing capability, preprocessing capacity, and technical expertise are available. Their scalability is potentially strong, but only where the supporting data ecosystem is sufficiently developed.\u003c/p\u003e \u003cp\u003eA similar contrast appears in application domains linked to urban systems, water quality management, and demand forecasting. In these areas, the integration of prediction, optimisation, and spatial or operational decision support can become computationally demanding because the workflow extends beyond one analytical step. For example, water demand forecasting linked to multi-criteria optimisation, urban non-revenue water reduction linked to hydraulic calibration, and water quality classification linked to pollutant ranking all require more than simple model execution. They involve repeated data handling, scenario evaluation, or system-level integration across forecasting, optimisation, and prioritisation layers. This increases their practical value, but it also raises the demand for reliable databases, technical calibration, and implementation capacity. In this sense, scalability is not only about whether a model can process more data, but also whether the integrated workflow can be maintained under real operational conditions.\u003c/p\u003e \u003cp\u003eOverall, the comparison suggests that there is no single best level of complexity for integrated MCDM-ML systems. Simpler MCDM-oriented frameworks may be more scalable in low-resource contexts because they are easier to apply, explain, and reproduce. More advanced ML-integrated systems may offer stronger analytical performance, but they usually require more extensive data, greater computational support, and stronger implementation capacity. For WRM practice, this means the choice of approach should be guided not only by expected accuracy, but also by whether the available data, computational infrastructure, and institutional capacity can support the model at the scale required.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec29\" class=\"Section2\"\u003e \u003ch2\u003e5.4 Suitability under data scarcity and uncertainty\u003c/h2\u003e \u003cp\u003eSuitability under data scarcity and uncertainty is one of the most important points of comparison across the integrated approaches reviewed in Section 4. In general, MCDM-oriented frameworks appear more workable where measured data are limited, because they can still operate using mapped conditioning factors, expert judgment, and GIS-derived variables. This was especially visible in groundwater zoning and recharge assessment studies, where prioritisation could still be performed even when the system depended more on weighted environmental layers than on long continuous monitoring records. In such contexts, MCDM provides a practical way of supporting decisions where uncertainty is high and observational data are incomplete.\u003c/p\u003e \u003cp\u003eBy contrast, ML-dominant approaches are usually more sensitive to data scarcity. Their predictive advantage depends on the availability of enough representative input data for training, calibration, and validation. This was clear in groundwater and flood studies where ensemble models and deep learning methods performed strongly, but only because the workflows were supported by multiple input layers and relatively rich datasets. Where such data are sparse, the robustness of the model becomes harder to defend, even if the final outputs appear accurate.\u003c/p\u003e \u003cp\u003eA similar issue appears in operational domains such as water demand forecasting, urban water loss management, and water quality control. These systems are useful when continuous or regularly updated datasets are available, but their practical value weakens when monitoring is inconsistent, field measurements are sparse, or uncertainty in implementation conditions is high. In such cases, integrated frameworks still remain valuable, but simpler and more transparent structures may be more suitable than highly data-hungry predictive systems.\u003c/p\u003e \u003cp\u003eOverall, the comparison suggests that integrated MCDM-ML systems are most suitable under data scarcity when the decision framework can tolerate uncertainty explicitly. Approaches that combine moderate predictive support with transparent ranking or prioritisation are often more defensible than highly complex models built on weak data foundations. For WRM practice, this means that model suitability should be judged not only by accuracy, but also by whether the available data can support stable, credible, and repeatable decisions under uncertain conditions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec30\" class=\"Section2\"\u003e \u003ch2\u003e5.5 Transferability across contexts\u003c/h2\u003e \u003cp\u003eTransferability across contexts remains limited in many of the integrated approaches reviewed above. Most studies were developed for specific basins, catchments, cities, or hazard-prone regions, and their performance was closely tied to local data structure, environmental conditions, and decision priorities. This means that a model or framework that performs well in one area cannot automatically be assumed to perform equally well in another.\u003c/p\u003e \u003cp\u003eThis issue is especially clear in groundwater, flood, and watershed studies. In these domains, the importance of factors such as lithology, slope, rainfall, drainage density, land use, and terrain structure changes from one location to another. As a result, both the predictive model and the weighting structure may need to be recalibrated when transferred to a new setting. The challenge is not only technical. It also affects whether the final rankings or risk classes remain meaningful under different physical conditions.\u003c/p\u003e \u003cp\u003eA similar limitation appears in urban and operational domains. Water demand forecasting, leakage control, and water quality assessment depend heavily on local infrastructure, monitoring systems, consumption behaviour, and management practice. This reduces direct transferability because even if the integration logic remains useful, the final model still depends on context-specific operational conditions. In such cases, what transfers best is often the framework rather than the exact calibrated model.\u003c/p\u003e \u003cp\u003eOverall, this suggests that transferability is stronger at the level of integration design than at the level of exact model output. Sequential, parallel, or hybrid MCDM-ML structures can often be adapted to new settings, but their criteria weights, predictor importance, validation results, and final priorities usually require local adjustment. For WRM practice, this means that integrated approaches should be viewed as adaptable frameworks rather than universally transferable solutions.\u003c/p\u003e \u003c/div\u003e"},{"header":"6. Reporting gaps and a comparative framework for future studies","content":"\u003cdiv id=\"Sec32\" class=\"Section2\"\u003e \u003ch2\u003e6.1 Metric inconsistency and benchmarking problems\u003c/h2\u003e \u003cp\u003eOne of the clearest problems across the studies reviewed in Section 4 is the lack of consistency in how performance is measured and reported. Different studies used different validation metrics, and in some cases the reported outputs were not directly comparable. This makes it difficult to judge whether one integrated approach is genuinely stronger than another, or whether it only appears better because it was evaluated differently.\u003c/p\u003e \u003cp\u003eThis inconsistency was especially visible in groundwater and flood studies. Some studies reported AUC, others used accuracy, MCC, or ROC-based validation, while some MCDM-led studies relied more on spatial agreement or ranking consistency than on predictive metrics. As a result, studies addressing similar WRM problems often cannot be compared on a common performance basis, even when they appear to target the same type of output.\u003c/p\u003e \u003cp\u003eA similar issue appears in water quality, urban water, and supply-planning studies. In these domains, some studies reported classification accuracy, others reported R\u0026sup2;, while others emphasised optimisation ratios, ranking outputs, or operational usefulness. This broad variation weakens cross-study benchmarking because predictive quality, decision quality, and practical utility are often mixed together without a shared reporting structure.\u003c/p\u003e \u003cp\u003eBenchmarking is also weakened by the absence of common datasets, standard validation protocols, and shared comparison baselines. Many studies are designed around local case data, which is valuable for practical relevance, but it means that performance is tested under different data conditions, scales, and assumptions. In that setting, even strong results are hard to generalise because the benchmark itself is unstable. Overall, the problem is not only that metrics differs, but that the literature still lacks a common benchmarking logic for integrated MCDM-ML systems in WRM.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec33\" class=\"Section2\"\u003e \u003ch2\u003e6.2 Validation and reproducibility gaps\u003c/h2\u003e \u003cp\u003eAnother issue across the reviewed studies is the limited consistency in validation procedures. Many studies reported strong model performance, but the validation approaches differed widely. Some studies relied on ROC curves or AUC values, while others used accuracy, R\u0026sup2;, or confusion matrix measures. Because validation procedures vary, it becomes difficult to determine whether results reflect genuine methodological strength or simply differences in evaluation design.\u003c/p\u003e \u003cp\u003eValidation gaps are also visible in spatial applications such as groundwater potential mapping and flood risk assessment. In several cases, models were validated using historical data or a limited number of field observations. While this approach can demonstrate internal consistency, it does not always confirm whether the model will perform reliably under different environmental conditions or future scenarios. This is particularly important in WRM, where models are often used to support long-term planning decisions rather than only retrospective analysis.\u003c/p\u003e \u003cp\u003eReproducibility also remains a challenge. Many studies provide a description of their workflow but do not include enough detail about data preprocessing, parameter settings, or weighting procedures for the results to be easily replicated. This is especially noticeable in integrated systems where multiple techniques are combined, such as GIS analysis, machine learning models, and MCDM ranking frameworks. Without clear documentation of each step, reproducing the results in another study area becomes difficult.\u003c/p\u003e \u003cp\u003eThe issue becomes even more complex in operational domains such as water demand forecasting, water quality management, and urban water loss reduction. These applications often depend on local datasets, monitoring systems, and institutional practices that are not always accessible to other researchers. As a result, even when the methodological framework is well described, the data required to reproduce the analysis may not be available.\u003c/p\u003e \u003cp\u003eOverall, the literature shows that stronger validation protocols and clearer methodological reporting are needed for integrated MCDM-ML systems in WRM. Transparent validation procedures, accessible datasets, and well-documented workflows would improve the credibility of future studies and make it easier to test whether integrated approaches perform consistently across different contexts.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec34\" class=\"Section2\"\u003e \u003ch2\u003e6.3 Proposed minimum reporting framework\u003c/h2\u003e \u003cp\u003eBased on the issues identified in Sections 6.1 and 6.2, a minimum reporting framework is needed for future integrated MCDM-ML studies in WRM. The purpose of such a framework is not to force all studies into one methodological design. Rather, it is to ensure that studies report enough information for readers to understand how the system was built, how it was validated, and how the final decision outputs were generated. Without this, cross-study comparison will remain weak, and reproducibility will continue to be limited.\u003c/p\u003e \u003cp\u003eAt a minimum, future studies should clearly report the WRM application domain, study objective, data sources, input variables, temporal and spatial scale, and the exact MCDM and ML methods used. This is necessary because many integrated studies appear similar at the title level, but differ substantially in their data structure, modelling depth, and decision logic. Clear reporting at this level would make it easier to compare studies across domains such as groundwater, flood risk, water quality, and urban water management.\u003c/p\u003e \u003cp\u003eStudies should also report how integration actually occurs. This includes whether the approach is sequential, parallel, or fully coupled, what outputs are passed from one method to another, and how the final recommendation or prioritisation is produced. This is especially important because integrated systems are often presented as hybrid frameworks, but the interaction between the ML and MCDM components is not always explained in enough detail to show how the final decision support output was constructed.\u003c/p\u003e \u003cp\u003eA further requirement is transparent reporting of validation, weighting, uncertainty, and sensitivity procedures. Future studies should state the validation metrics used, the reason for selecting them, the source of weights or preferences, and whether sensitivity analysis was performed. Where expert judgment is used, the basis for that judgment should be made explicit. Where uncertainty is present, it should be acknowledged and, where possible, tested rather than left implicit.\u003c/p\u003e \u003cp\u003eFinally, studies should report enough practical information to support reuse and adaptation. This includes software or platform details, preprocessing steps, calibration choices, and any constraints that may affect implementation in other settings. For WRM, this matters because the usefulness of an integrated system depends not only on academic performance, but also on whether the framework can be interpreted, replicated, and adapted for real planning and management use. This framework can serve as a baseline checklist for future integrated WRM studies and as a basis for improving cross-study comparison, transparency, and reproducibility.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eProposed minimum reporting framework for integrated MCDM-ML studies in water resource management\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReporting element\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMinimum information that should be reported\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWhy it matters\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy context\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWRM domain, study objective, and decision problem\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClarifies what the integrated system is intended to solve\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eData description\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eData sources, variables, sample size, spatial scale, temporal scale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSupports interpretation of model suitability and comparability\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMCDM component\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMethod used, criteria, weighting approach, source of weights\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMakes the decision structure transparent\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eML component\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eModel used, input features, preprocessing steps, calibration or training details\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eImproves reproducibility and technical clarity\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntegration structure\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSequential, parallel, or fully coupled design; how outputs move between methods\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eShows how the hybrid system actually works\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eValidation approach\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMetrics used, validation procedure, benchmark or comparison basis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAllows fairer performance assessment across studies\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSensitivity and uncertainty\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSensitivity testing, uncertainty treatment, robustness checks\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eStrengthens confidence in the reported results\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOutput format\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eType of final output, such as map, ranking, forecast, classification, or intervention priority\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClarifies how the system supports decision-making\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eImplementation relevance\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSoftware/tools used, data availability, operational constraints, transferability limits\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHelps others assess practical applicability\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLimitations\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMain methodological, data, or contextual limitations\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrevents overstatement and improves interpretive caution\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"7. Emerging research and implementation directions in MCDM-ML for WRM","content":"\u003cdiv id=\"Sec36\" class=\"Section2\"\u003e \u003ch2\u003e7.1 Explainable and trustworthy AI\u003c/h2\u003e \u003cp\u003eExplainable and trustworthy AI is becoming an important direction in WRM because strong predictive performance alone is no longer enough for practical decision support. The review conducted by (Başağaoğlu et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) showed that interpretable and explainable AI methods are increasingly being applied across hydroclimatic domains such as groundwater, streamflow, water quality, floods, and droughts. Their review makes it clear that explainability is valuable because it helps reveal how predictors influence model outputs, which is especially important where AI results may affect planning, regulation, or public-facing decisions. For this review, that is important because integrated MCDM-ML systems depend not only on accurate outputs, but also on decision pathways that users can understand and justify.\u003c/p\u003e \u003cp\u003e(Infant et al. \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) further showed that explainable AI is becoming increasingly relevant in urban water systems, especially in hydrological modelling, demand prediction, and leak detection. Their review discussed tools such as SHAP, LIME, and counterfactual analysis, and argued that explainability improves both model understanding and practical confidence in deployment. This matters for the present manuscript because future MCDM-ML frameworks in WRM will need to do more than generate outputs. They will also need to show why certain variables, risks, or options are driving the final recommendation.\u003c/p\u003e \u003cp\u003eA much broader review conducted by (Schiller et al. \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) on AI in environmental and Earth system sciences added an important caution. In their review it was found that although XAI methods are often presented as a way to increase trust, very few studies explicitly examine trustworthiness itself. In other words, explainability and trust are related, but they are not the same. This is an important distinction for WRM because a model may be partially interpretable while still raising concerns about robustness, fairness, uncertainty, or governance suitability. For that reason, future integrated systems will need to treat trustworthiness as a design requirement rather than assuming that explanation alone is enough.\u003c/p\u003e \u003cp\u003eA practical example is provided by (Mau\u0026szlig;ner et al. \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) who developed explainable AI for water demand forecasting and linked it to the idea of reliable water supply decision-making. Their work is useful because it moves the discussion from principle to application. It shows that explainability can help utilities understand model behaviour in operational settings rather than treating AI as a black box. For the future of integrated MCDM-ML systems, this suggests that explainable AI is most valuable when it supports better human judgment, clearer prioritisation, and more defensible implementation decisions.\u003c/p\u003e \u003cp\u003eThe issue is no longer only whether AI can improve prediction. It is whether its outputs can be interpreted, questioned, trusted, and integrated into transparent decision-support structures. In this sense, explainability strengthens the ML side of integrated systems, while trustworthiness determines whether those systems can be responsibly used in real water management settings.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec37\" class=\"Section2\"\u003e \u003ch2\u003e7.2 Real-time and IoT-enabled decision support\u003c/h2\u003e \u003cp\u003eReal-time monitoring has also shown to improve the ability of water utilities to manage infrastructure and reduce operational losses. (Singh and Ahmed \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Roostaei et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) demonstrated that IoT-enabled sensing systems can support continuous monitoring of distribution networks and improve leak detection, pressure management, and demand tracking. By combining sensor data with predictive analytics, utilities can detect anomalies and system failures earlier than would be possible with manual inspection alone. This development is particularly relevant for integrated MCDM-ML frameworks because real-time data streams provide the dynamic inputs needed for predictive models while decision-support layers help prioritise operational responses.\u003c/p\u003e \u003cp\u003eIoT technologies also play an important role in environmental monitoring and early warning systems. (Kumar et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) showed that IoT-enabled hydrological monitoring can support flood forecasting and water quality monitoring through continuous data acquisition from distributed sensors. These systems improve situational awareness by allowing models to update predictions as new observations become available. In this context, real-time data streams strengthen the predictive capacity of ML models while enabling decision-support frameworks to react to changing conditions more quickly than static planning systems.\u003c/p\u003e \u003cp\u003eContinuous monitoring improves the availability and timeliness of data, while integrated analytical frameworks allow this information to be translated into actionable decisions. For integrated MCDM-ML systems, this means that future implementations will likely operate in more dynamic environments where models are updated regularly and decisions are supported by live system data rather than static datasets.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec38\" class=\"Section2\"\u003e \u003ch2\u003e7.3 Cloud and edge deployment for scalable WRM\u003c/h2\u003e \u003cp\u003eAs integrated WRM systems move from offline analysis toward real-time monitoring and operational control, cloud and edge deployment is becoming increasingly important. Large-scale sensor networks in water distribution and monitoring systems now generate data volumes that are difficult to manage through fully centralised architectures alone. For this reason, cloud platforms remain useful for storage, historical analysis, coordination, and large-model training, while edge computing is increasingly valuable for local preprocessing, near-real-time inference, and faster operational response (Pagano et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). In practical terms, this means future MCDM-ML systems are likely to operate across layered architectures in which cloud services support system-wide intelligence and edge nodes support immediate field-level decisions.\u003c/p\u003e \u003cp\u003eThe value of edge deployment becomes even clearer in time-sensitive WRM applications such as leak detection, anomaly detection, water quality alerts, and distributed environmental monitoring. By processing part of the data closer to the source, edge architectures can reduce latency, lower data transmission load, improve energy efficiency, and reduce operating cost while still supporting machine learning functions at local nodes. This is especially relevant for water systems that depend on rapid detection and response, because delayed transmission to remote servers can weaken the practical value of otherwise strong analytical models (Roostaei et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). For integrated MCDM-ML systems, edge deployment therefore improves not only technical speed, but also the timeliness of prioritisation and intervention.\u003c/p\u003e \u003cp\u003eAt the same time, scalable deployment is not only a matter of adding sensors or moving models to the cloud. It also depends on communication reliability, data quality, infrastructure change, privacy protection, and the ability to keep models updated as network conditions evolve. Recent reviews of smart water distribution systems show that real-time monitoring, digital twins, uncertainty-aware forecasting, and explainable AI are likely to become more important as utilities move toward more connected and adaptive systems (Taloma et al. \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). This suggests that future MCDM-ML frameworks in WRM should be designed not only for analytical performance, but also for operational scalability, maintainability, and secure long-term deployment.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec39\" class=\"Section2\"\u003e \u003ch2\u003e7.4 Policy-aligned and stakeholder-centred systems\u003c/h2\u003e \u003cp\u003eThe future usefulness of integrated MCDM-ML systems in WRM will depend not only on methodological sophistication, but also on whether they align with governance realities and stakeholder needs. Water decisions affect multiple users whose priorities, risks, and values are not always the same, and this makes technically strong models insufficient on their own. Evidence from participatory water governance shows that community engagement improves the legitimacy and durability of water interventions, especially when local actors are involved in planning, monitoring, and implementation rather than being treated only as end users of expert-generated outputs (Ahmadi et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Moreira et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). For integrated WRM systems, this means decision-support tools should be designed to support participation, communication, and shared understanding, not just optimisation.\u003c/p\u003e \u003cp\u003eThis is especially important where water systems are characterised by conflicting interests, institutional fragmentation, or unequal influence among users. Stakeholder-based decision support research has shown that management scenarios become more realistic and more sustainable when the characteristics, interests, and influence of different actors are explicitly built into the analytical process. In practice, this means future MCDM-ML systems should increasingly move toward structures that do not only predict outcomes or rank options, but also make room for negotiated priorities, decentralised decisions, and trade-off awareness across environmental, social, and economic dimensions.\u003c/p\u003e \u003cp\u003ePolicy alignment is equally important. Water management tools are more likely to be adopted when they fit local legal frameworks, institutional mandates, allocation rules, and practitioner realities. Studies on water allocation planning and decision support have shown that even well-developed tools are often underused when they do not match operational needs or when they are introduced without sufficient stakeholder participation during development (Pearson et al. \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Nel et al. \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). This means the next generation of integrated WRM systems should be designed with implementation in mind from the beginning, including how they fit existing governance structures, regulatory requirements, and the decision habits of practitioners.\u003c/p\u003e \u003cp\u003eA further shift is needed from technically correct systems toward socially workable systems. Sustainable decision support in water management increasingly depends on adaptive processes, social learning, and the ability to evaluate options in ways that remain understandable across professional and community boundaries. In that sense, policy-aligned and stakeholder-centred systems are not a soft addition to MCDM-ML integration. They are part of what makes these systems usable in practice. The strongest future frameworks will therefore be those that connect predictive intelligence with transparent prioritisation, institutional fit, and meaningful stakeholder engagement.\u003c/p\u003e \u003c/div\u003e"},{"header":"8. Conclusion","content":"\u003cp\u003eThis review examined how MCDM and ML have been integrated across major WRM domains and showed that the combination is now being used far beyond isolated experimental studies. Across groundwater assessment, demand forecasting, flood risk management, water quality control, watershed prioritisation, and urban water systems, a consistent pattern emerged. ML contributes most strongly where prediction, classification, and pattern detection are required, while MCDM contributes most strongly where weighting, prioritisation, and final decision structuring are necessary. The literature therefore suggests that the value of integration lies not in combining methods for novelty, but in linking analytical prediction with decision-ready interpretation in ways that better reflect the complexity of WRM.\u003c/p\u003e \u003cp\u003eAt the same time, the review also shows that the field has not yet reached methodological maturity. Strong case-specific results are now common, but cross-study comparison remains weakened by inconsistent metrics, varied validation procedures, limited transparency in weighting structures, uneven treatment of uncertainty, and weak reproducibility. These limitations reduce the extent to which current findings can be generalised across settings and make it difficult to identify which integrated approaches are most robust under different WRM conditions. For that reason, future work should move beyond reporting good outcomes and focus more deliberately on validation design, sensitivity analysis, reporting consistency, and practical transferability.\u003c/p\u003e \u003cp\u003eOverall, the strongest direction for future research and implementation is clear. Integrated MCDM-ML systems in WRM should become more explainable, more adaptive, more operationally scalable, and more closely aligned with stakeholder and policy contexts. In practice, this means combining predictive strength with transparent decision logic, real-time or near-real-time data support, clearer reporting frameworks, and implementation pathways that are credible to both technical experts and decision-makers. When these conditions are met, integrated systems are better positioned to support water management decisions that are not only accurate, but also defensible, context-appropriate, and useful in practice. A key immediate priority for the field is the adoption of clearer reporting and validation standards so that integrated WRM studies become easier to compare, reproduce, and implement.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eFunding:\u003c/h2\u003e \u003cp\u003eThis research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.\u003c/p\u003e \u003cp\u003eCompeting interests: The authors declare that they have no competing interests.\u003c/p\u003e \u003cp\u003eClinical trial number: not applicable.\u003c/p\u003e \u003cp\u003eEthics approval: not applicable.\u003c/p\u003e \u003cp\u003eConsent to participate: not applicable.\u003c/p\u003e \u003cp\u003eConsent for publication: not applicable.\u003c/p\u003e \u003cp\u003eData Availability Statement: Not applicable.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eMurphy Bonkogia Lomboli - conceptualization, methodology, literature search, screening, analysis, synthesis, visualisation, and writing of the original draft.Opeyeolu Timothy Laseinde - supervision, methodological guidance, critical review, validation, and editing of the manuscript. Both authors read and approved the final manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAhmadi A, Kerachian R, Skardi MJE, Abdolhay A. A stakeholder-based decision support system to manage water resources. J Hydrol (Amst). 2020;589. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jhydrol.2020.125138\u003c/span\u003e\u003cspan address=\"10.1016/j.jhydrol.2020.125138\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhmed AA, Sayed S, Abdoulhalik A, et al. Applications of machine learning to water resources management: A review of present status and future opportunities. J Clean Prod. 2024;441:140715. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jclepro.2024.140715\u003c/span\u003e\u003cspan address=\"10.1016/j.jclepro.2024.140715\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAkay H. Flood hazards susceptibility mapping using statistical, fuzzy logic, and MCDM methods. Soft comput. 2021;25:9325\u0026ndash;46. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00500-021-05903-1\u003c/span\u003e\u003cspan address=\"10.1007/s00500-021-05903-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAkbari M, Meshram SG, Krishna RS, et al. Identification of the Groundwater Potential Recharge Zones Using MCDM Models: Full Consistency Method (FUCOM), Best Worst Method (BWM) and Analytic Hierarchy Process (AHP). Water Resour Manage. 2021;35:4727\u0026ndash;45. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11269-021-02924-1\u003c/span\u003e\u003cspan address=\"10.1007/s11269-021-02924-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAli R, Hussain A, Nazir S, et al. Intelligent Decision Support Systems\u0026mdash;An Analysis of Machine Learning and Multicriteria Decision-Making Methods. Appl Sci. 2023;13:12426. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/app132212426\u003c/span\u003e\u003cspan address=\"10.3390/app132212426\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAnand V, Rajput VD, Minkina T et al. (2025) Evaluating groundwater potential with the synergistic use of geospatial methods and advanced machine learning approaches. Discover Cities 2:. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s44327-025-00095-x\u003c/span\u003e\u003cspan address=\"10.1007/s44327-025-00095-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAsiri MM, Aldehim G, Alruwais N, et al. Coastal Flood risk assessment using ensemble multi-criteria decision-making with machine learning approaches. Environ Res. 2024;245. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.envres.2023.118042\u003c/span\u003e\u003cspan address=\"10.1016/j.envres.2023.118042\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAyad A, Khalifa A, Fawy MEL, Moawad A. An integrated approach for non-revenue water reduction in water distribution networks based on field activities, optimisation, and GIS applications. Ain Shams Eng J. 2021;12:3509\u0026ndash;20. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.asej.2021.04.007\u003c/span\u003e\u003cspan address=\"10.1016/j.asej.2021.04.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaşağaoğlu H, Chakraborty D, Lago C, Do, et al. A Review on Interpretable and Explainable Artificial Intelligence in Hydroclimatic Applications. Water (Basel). 2022;14:1230. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/w14081230\u003c/span\u003e\u003cspan address=\"10.3390/w14081230\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCalizaya A, Meixner O, Bengtsson L, Berndtsson R. Multi-criteria decision analysis (MCDA) for integrated water resources management (IWRM) in the Lake Poopo basin, Bolivia. Water Resour Manage. 2010;24:2267\u0026ndash;89. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11269-009-9551-x\u003c/span\u003e\u003cspan address=\"10.1007/s11269-009-9551-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCosta AP, de Choren A, de Pereira R M, et al. Integrating multicriteria decision making and principal component analysis: a systematic literature review. Cogent Eng. 2024;11. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/23311916.2024.2374944\u003c/span\u003e\u003cspan address=\"10.1080/23311916.2024.2374944\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDas A. (2025a) An optimization based framework for water quality assessment and pollution source apportionment employing GIS and machine learning techniques for smart surface water governance. Discover Environment 3:. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s44274-025-00327-2\u003c/span\u003e\u003cspan address=\"10.1007/s44274-025-00327-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDas A. (2025b) Water pollution and water quality assessment and application of criterion impact loss (CILOS), geographical information system (GIS), artificial neural network (ANN) and decision-learning technique in river water quality management: An experiment on the Mahanadi catchment, Odisha, India. Desalination Water Treat 321:. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.dwt.2024.100969\u003c/span\u003e\u003cspan address=\"10.1016/j.dwt.2024.100969\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDrogkoula M, Kokkinos K, Samaras N. A Comprehensive Survey of Machine Learning Methodologies with Emphasis in Water Resources Management. Appl Sci. 2023;13:12147. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/app132212147\u003c/span\u003e\u003cspan address=\"10.3390/app132212147\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDrogkoula M, Samaras N, Iatrellis O, et al. Systematic review and topic classification of soft computing and machine learning in water resources management. Discover Sustain. 2025;6:860. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s43621-025-01832-3\u003c/span\u003e\u003cspan address=\"10.1007/s43621-025-01832-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGebre SL, Cattrysse D, Van Orshoven J. Multi-Criteria Decision-Making Methods to Address Water Allocation Problems: A Systematic Review. Water (Basel). 2021;13:125. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/w13020125\u003c/span\u003e\u003cspan address=\"10.3390/w13020125\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGhobadi F, Kang D. Application of Machine Learning in Water Resources Management: A Systematic Literature Review. Water (Basel). 2023;15:620. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/w15040620\u003c/span\u003e\u003cspan address=\"10.3390/w15040620\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGholami H, Mohammadifar A, Golzari S, et al. Mapping flood risk using a workflow including deep learning and MCDM\u0026ndash; Application to southern Iran. Urban Clim. 2025;59. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.uclim.2024.102272\u003c/span\u003e\u003cspan address=\"10.1016/j.uclim.2024.102272\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGhosh A, Adhikary PP, Bera B, et al. Assessment of groundwater potential zone using MCDA and AHP techniques: case study from a tropical river basin of India. Appl Water Sci. 2022;12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s13201-021-01548-5\u003c/span\u003e\u003cspan address=\"10.1007/s13201-021-01548-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGhosh B, Mukhopadhyay S. Erosion susceptibility mapping of sub-watersheds for management prioritization using MCDM-based ensemble approach. Arab J Geosci. 2021;14:36. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s12517-020-06297-4\u003c/span\u003e\u003cspan address=\"10.1007/s12517-020-06297-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHajkowicz S, Collins K. A Review of Multiple Criteria Analysis for Water Resource Planning and Management. Water Resour Manage. 2007;21:1553\u0026ndash;66. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11269-006-9112-5\u003c/span\u003e\u003cspan address=\"10.1007/s11269-006-9112-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHe C, Huang G, Liu L, et al. Multi-dimensional diagnosis model for the sustainable development of regions facing water scarcity problem: A case study for Guangdong, China. Sci Total Environ. 2020. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.scitotenv.2020.139394\u003c/span\u003e\u003cspan address=\"10.1016/j.scitotenv.2020.139394\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. 734:.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eInfant SS, Vickram S, Saravanan A, et al. Explainable artificial intelligence for sustainable urban water systems engineering. Results Eng. 2025;25:104349. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.rineng.2025.104349\u003c/span\u003e\u003cspan address=\"10.1016/j.rineng.2025.104349\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKanji S, Das S. Assessing groundwater potentialities and replenishment feasibility using machine learning and MCDM models considering hydro-geological aspects and water quality constituents. Environ Earth Sci. 2025a;84. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s12665-024-11996-2\u003c/span\u003e\u003cspan address=\"10.1007/s12665-024-11996-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKanji S, Das S. Exploring the morpho-tectonic nature, hydrological and physical characteristics of a watershed and prioritizing sub-watersheds surface runoff potentialities by integrating MCDM and ensemble machine learning models. J Environ Manage. 2025b;386:125772. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jenvman.2025.125772\u003c/span\u003e\u003cspan address=\"10.1016/j.jenvman.2025.125772\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhalil U, Imtiaz I, Aslam B, et al. Comparative analysis of machine learning and multi-criteria decision making techniques for landslide susceptibility mapping of Muzaffarabad district. Front Environ Sci. 2022;10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fenvs.2022.1028373\u003c/span\u003e\u003cspan address=\"10.3389/fenvs.2022.1028373\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhalilzadeh M, Banihashemi SA, Heidari A, et al. Risk Analysis and Assessment of Water Supply Projects Using the Fuzzy DEMATEL-ANP and Artificial Neural Network Methods. Water (Switzerland). 2025;17. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/w17131995\u003c/span\u003e\u003cspan address=\"10.3390/w17131995\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKocaman H, Asan U. Integration Modes Between MCDM Methods and Machine Learning Algorithms: A Structured Approach for Framework Development. Mathematics. 2025;14:33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/math14010033\u003c/span\u003e\u003cspan address=\"10.3390/math14010033\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKolaski K, Logan LR, Ioannidis JPA. Guidance to best tools and practices for systematic reviews. Syst Rev. 2023;12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s13643-023-02255-9\u003c/span\u003e\u003cspan address=\"10.1186/s13643-023-02255-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKumar N, Kumar S, Yadav S. Smart Water Resource Management using IoT and Data Analytics. J Water Eng Manage. 2025;6:45\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.47884/jweam.v6i3pp45\u003c/span\u003e\u003cspan address=\"10.47884/jweam.v6i3pp45\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKumar R. A Comprehensive Review of MCDM Methods, Applications, and Emerging Trends. Decis Mak Adv. 2025;3:185\u0026ndash;99. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.31181/dma31202569\u003c/span\u003e\u003cspan address=\"10.31181/dma31202569\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLai E, Lundie S, Ashbolt NJ. Review of multi-criteria decision aid for integrated sustainability assessment of urban water systems. Urban Water J. 2008;5:315\u0026ndash;27. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/15730620802041038\u003c/span\u003e\u003cspan address=\"10.1080/15730620802041038\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi M, Cao X, Liu D, et al. Sustainable management of agricultural water and land resources under changing climate and socio-economic conditions: A multi-dimensional optimization approach. Agric Water Manag. 2022;259. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.agwat.2021.107235\u003c/span\u003e\u003cspan address=\"10.1016/j.agwat.2021.107235\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu C, Xu J, Li X, et al. Water resource forecasting with machine learning and deep learning: A scientometric analysis. Artif Intell Geosci. 2024;5:100084. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.aiig.2024.100084\u003c/span\u003e\u003cspan address=\"10.1016/j.aiig.2024.100084\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLong HA, French DP, Brooks JM. Optimising the value of the critical appraisal skills programme (CASP) tool for quality appraisal in qualitative evidence synthesis. Res Methods Med Health Sci. 2020;1:31\u0026ndash;42. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/2632084320947559\u003c/span\u003e\u003cspan address=\"10.1177/2632084320947559\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMau\u0026szlig;ner C, Oberascher M, Autengruber A, et al. Explainable artificial intelligence for reliable water demand forecasting to increase trust in predictions. Water Res. 2025;268. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.watres.2024.122779\u003c/span\u003e\u003cspan address=\"10.1016/j.watres.2024.122779\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMendoza GA, Martins H. Multi-criteria decision analysis in natural resource management: A critical review of methods and new modelling paradigms. Ecol Manage. 2006;230:1\u0026ndash;22. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.foreco.2006.03.023\u003c/span\u003e\u003cspan address=\"10.1016/j.foreco.2006.03.023\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiller KA, Belton V. Water resource management and climate change adaptation: a holistic and multiple criteria perspective. Mitig Adapt Strateg Glob Chang. 2014;19:289\u0026ndash;308. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11027-013-9537-0\u003c/span\u003e\u003cspan address=\"10.1007/s11027-013-9537-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMohammadifar A, Gholami H, Golzari S. Novel integrated modelling based on multiplicative long short-term memory (mLSTM) deep learning model and ensemble multi-criteria decision making (MCDM) models for mapping flood risk. J Environ Manage. 2023;345. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jenvman.2023.118838\u003c/span\u003e\u003cspan address=\"10.1016/j.jenvman.2023.118838\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoreira FD, Fonseca PRS, Miranda RM, et al. Stakeholder engagement for inclusive water governance in a rural community in Brazil. Front Water. 2024;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/frwa.2024.1378514\u003c/span\u003e\u003cspan address=\"10.3389/frwa.2024.1378514\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNasiri Khiavi A, Vafakhah M, Kim D, et al. Integrating Remote Sensing, Machine Learning, and Local Knowledge for Innovative Flood Susceptibility and Vulnerability Mapping. J Flood Risk Manag. 2025;18. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/jfr3.70149\u003c/span\u003e\u003cspan address=\"10.1111/jfr3.70149\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNel JB, Mativenga PT, Marnewick AL. A Framework to Support the Selection of an Appropriate Water Allocation Planning and Decision Support Scheme. Water (Switzerland). 2022;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/w14121854\u003c/span\u003e\u003cspan address=\"10.3390/w14121854\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNiknam A, Zare HK, Hosseininasab H, et al. A Critical Review of Short-Term Water Demand Forecasting Tools\u0026mdash;What Method Should I Use? Sustainability. 2022;14:5412. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/su14095412\u003c/span\u003e\u003cspan address=\"10.3390/su14095412\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePagano A, Garlisi D, Tinnirello I, et al. A survey on massive IoT for water distribution systems: Challenges, simulation tools, and guidelines for large-scale deployment. Ad Hoc Netw. 2025;168:103714. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.adhoc.2024.103714\u003c/span\u003e\u003cspan address=\"10.1016/j.adhoc.2024.103714\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePage MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021a;372:n71. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/bmj.n71\u003c/span\u003e\u003cspan address=\"10.1136/bmj.n71\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePage MJ, Moher D, Bossuyt PM, et al. PRISMA 2020 explanation and elaboration: updated guidance and exemplars for reporting systematic reviews. BMJ. 2021b;372:n160. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/bmj.n160\u003c/span\u003e\u003cspan address=\"10.1136/bmj.n160\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePearson LJ, Coggan A, Proctor W, Smith TF. A sustainable decision support framework for Urban water management. Water Resour Manage. 2010;24:363\u0026ndash;76. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11269-009-9450-1\u003c/span\u003e\u003cspan address=\"10.1007/s11269-009-9450-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePham BT, Luu C, Phong T, Van, et al. Flood risk assessment using hybrid artificial intelligence models integrated with multi-criteria decision analysis in Quang Nam Province, Vietnam. J Hydrol (Amst). 2021;592. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jhydrol.2020.125815\u003c/span\u003e\u003cspan address=\"10.1016/j.jhydrol.2020.125815\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePoli G, Cunt\u0026ograve; S, Muccio E, Cerreta M. (2024) A spatial decision support system for multi-dimensional sustainability assessment of river basin districts: the case study of Sarno river, Italy. Land use policy 141:. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.landusepol.2024.107123\u003c/span\u003e\u003cspan address=\"10.1016/j.landusepol.2024.107123\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePoursaeid M. A Novel Approach for Water Resources Management to Accurate River Flow Prediction: Utilization of Linear Programming and Weighted Ensemble Learning. Water Resour Manage. 2025;39:5827\u0026ndash;43. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11269-025-04230-6\u003c/span\u003e\u003cspan address=\"10.1007/s11269-025-04230-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRashidi Shikhteymour S, Borji M, Bagheri-Gavkosh M, et al. A novel approach for assessing flood risk with machine learning and multi-criteria decision-making methods. Appl Geogr. 2023;158. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.apgeog.2023.103035\u003c/span\u003e\u003cspan address=\"10.1016/j.apgeog.2023.103035\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRomero-Ben L, Alves D, Blesa J, et al. Leak detection and localization in water distribution networks: Review and perspective. Annu Rev Control. 2023;55:392\u0026ndash;419. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.arcontrol.2023.03.012\u003c/span\u003e\u003cspan address=\"10.1016/j.arcontrol.2023.03.012\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRoostaei J, Wager YZ, Shi W, et al. IoT-based edge computing (IoTEC) for improved environmental monitoring. Sustainable Computing: Inf Syst. 2023;38. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.suscom.2023.100870\u003c/span\u003e\u003cspan address=\"10.1016/j.suscom.2023.100870\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchabbauer G. Polymicrobial sepsis models: CLP versus CASP. Drug Discov Today Dis Models. 2012;9:e17\u0026ndash;21. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ddmod.2011.10.002\u003c/span\u003e\u003cspan address=\"10.1016/j.ddmod.2011.10.002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchiller J, Stiller S, Ryo M. Artificial intelligence in environmental and Earth system sciences: explainability and trustworthiness. Artif Intell Rev. 2025;58. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10462-025-11165-2\u003c/span\u003e\u003cspan address=\"10.1007/s10462-025-11165-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchuwirth N, Honti M, Logar I, Stamm C. Multi-criteria decision analysis for integrated water quality assessment and management support. Water Res X. 2018. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.wroa.2018.100010\u003c/span\u003e\u003cspan address=\"10.1016/j.wroa.2018.100010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. 1:.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSerafeim AV, Fourniotis NTh, Deidda R, et al. Leakages in Water Distribution Networks: Estimation Methods, Influential Factors, and Mitigation Strategies\u0026mdash;A Comprehensive Review. Water (Basel). 2024;16:1534. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/w16111534\u003c/span\u003e\u003cspan address=\"10.3390/w16111534\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShaheen N, Shaheen A, Ramadan A, et al. Appraising systematic reviews: a comprehensive guide to ensuring validity and reliability. Front Res Metr Anal. 2023;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/frma.2023.1268045\u003c/span\u003e\u003cspan address=\"10.3389/frma.2023.1268045\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSharma SK. A novel approach on water resource management with Multi-Criteria Optimization and Intelligent Water Demand Forecasting in Saudi Arabia. Environ Res. 2022;208. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.envres.2021.112578\u003c/span\u003e\u003cspan address=\"10.1016/j.envres.2021.112578\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSingh M, Ahmed S. IoT based smart water management systems: A systematic review. Mater Today Proc. 2021;46:5211\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.matpr.2020.08.588\u003c/span\u003e\u003cspan address=\"10.1016/j.matpr.2020.08.588\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSnyder H. Literature review as a research methodology: An overview and guidelines. J Bus Res. 2019;104:333\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jbusres.2019.07.039\u003c/span\u003e\u003cspan address=\"10.1016/j.jbusres.2019.07.039\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSohrabi C, Franchi T, Mathew G, et al. PRISMA 2020 statement: What\u0026rsquo;s new and the importance of reporting guidelines. Int J Surg. 2021;88:105918. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ijsu.2021.105918\u003c/span\u003e\u003cspan address=\"10.1016/j.ijsu.2021.105918\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTaloma RJL, Cuomo F, Comminiello D, Pisani P. Machine learning for smart water distribution systems: exploring applications, challenges and future perspectives. Artif Intell Rev. 2025;58. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10462-024-11093-7\u003c/span\u003e\u003cspan address=\"10.1007/s10462-024-11093-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXi X, Poh KL. A Novel Integrated Decision Support Tool for Sustainable Water Resources Management in Singapore: Synergies Between System Dynamics and Analytic Hierarchy Process. Water Resour Manage. 2015;29:1329\u0026ndash;50. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11269-014-0876-8\u003c/span\u003e\u003cspan address=\"10.1007/s11269-014-0876-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu Q, Liu R, Chen Q, Li R. Review on water leakage control in distribution networks and the associated environmental benefits. J Environ Sci (China). 2014;26:955\u0026ndash;61. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S1001-0742(13)60569-0\u003c/span\u003e\u003cspan address=\"10.1016/S1001-0742(13)60569-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhu F, Zhong P, an, Cao Q, et al. A stochastic multi-criteria decision making framework for robust water resources management under uncertainty. J Hydrol (Amst). 2019;576:287\u0026ndash;98. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jhydrol.2019.06.049\u003c/span\u003e\u003cspan address=\"10.1016/j.jhydrol.2019.06.049\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"discover-water","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diwa","sideBox":"Learn more about [Discover Water](https://www.springer.com/43832)","snPcode":"","submissionUrl":"","title":"Discover Water","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Water resource management, multi-criteria decision-making, machine learning, decision support systems, systematic literature review, explainable artificial intelligence","lastPublishedDoi":"10.21203/rs.3.rs-9145184/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9145184/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eWater resource management (WRM) is increasingly shaped by interconnected environmental, technical, social, and institutional pressures that make single-method analysis insufficient for robust decision support. The study follows a structured review methodology with systematic search and screening, combined with critical thematic synthesis to identify and synthesise studies on integrated MCDM-ML applications in WRM, with emphasis on conceptual integration patterns, domain-specific applications, methodological trade-offs, reporting gaps, and future implementation directions. Critically examining how multi-criteria decision-making (MCDM) models and machine learning (ML) techniques are being integrated to support more effective, transparent and context-appropriate WRM. The review identifies three broad integration patterns, namely sequential, parallel, and fully coupled frameworks. Across groundwater potential and recharge assessment, water demand forecasting and supply planning, flood risk and hydrological hazard management, water quality assessment and pollution control, sediment and catchment management, and urban water loss management, a consistent pattern emerges. ML contributes most strongly to prediction, classification, and pattern detection, while MCDM strengthens criteria weighting, prioritisation, and final decision structuring. The review also shows that although integrated systems often report strong case-specific results, cross-study comparison remains limited by inconsistent performance metrics, uneven validation procedures, weak transparency in weighting structures, limited uncertainty treatment, and poor reproducibility. In response, the paper proposes a minimum reporting framework and highlights key future directions, including explainable and trustworthy AI, real-time and IoT-enabled decision support, cloud and edge deployment, and policy-aligned stakeholder-centred systems. Overall, integrated MCDM-ML approaches show strong potential for WRM, but their future value will depend on clearer reporting, stronger validation, and closer alignment with real decision contexts.\u003c/p\u003e","manuscriptTitle":"A Structured Review and Critical Synthesis of Multi-Criteria Decision- Making Models Integrated with Machine Learning for Water Resource Management","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-02 07:03:11","doi":"10.21203/rs.3.rs-9145184/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-04-17T06:01:34+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-17T05:26:24+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-15T13:14:08+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"284931208437346176679657787710826549817","date":"2026-04-09T15:37:02+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"167514813488718357173732918741461042108","date":"2026-04-09T14:14:40+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"93888186160213059179824727065045075196","date":"2026-04-09T14:01:18+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"113720605276854288902984926576148647431","date":"2026-04-07T17:04:42+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"243891433430435383234409735541738505638","date":"2026-04-07T15:56:58+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"35476636197919429912583196229286408736","date":"2026-04-07T11:40:45+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"106522260016815213389009103286488748269","date":"2026-04-07T11:37:53+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-01T15:26:17+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"56388641167667038041446084139041415193","date":"2026-03-29T15:16:07+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-29T11:09:25+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-29T06:07:36+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-03-29T05:52:46+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-28T14:31:39+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Water","date":"2026-03-28T14:26:11+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-water","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diwa","sideBox":"Learn more about [Discover Water](https://www.springer.com/43832)","snPcode":"","submissionUrl":"","title":"Discover Water","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"53865318-fa3e-4ede-9f07-8383b0295758","owner":[],"postedDate":"April 2nd, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-04T12:24:50+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-02 07:03:11","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9145184","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9145184","identity":"rs-9145184","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}