Effective Modes of Transboundary Water-Related Cooperation to Turn Future Climate Conflict Risks into Opportunities

Effective Modes of Transboundary Water-Related Cooperation to Turn Future Climate Conflict Risks into Opportunities | 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 Article Effective Modes of Transboundary Water-Related Cooperation to Turn Future Climate Conflict Risks into Opportunities Cong Cong, Haozhi Pan, Yiming Yang, Zipan Cai, Elisie Kåresdotter, and 7 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6847954/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Geopolitical conflicts can rapidly escalate over water resources, particularly in regions already facing vulnerabilities amplified by climate change. At the same time, environmental challenges can be viewed as opportunities for transboundary cooperation, offering a path to mutual understanding and collaboration. This study aims to identify and analyze the effectiveness of water-related cooperation modes in mitigating conflicts, particularly under climate change scenarios, using a framework that leverages large language models (LLMs), retrieval-augmented generation (RAG), and interpretable machine learning. We analyze over 2,000 global transboundary water cooperation and conflicts cases from 1951 to 2019, going beyond the limitations of small-scale qualitative studies and enabling granular evaluation of impacts. The study identifies six cooperation modes: Cross-Border Basin Agreements, Collaborative Planning and Adaptation Strategies, Joint Water Allocation Models, Joint Data-Sharing Systems, Transboundary Water Quality Standards, and Coordinated Hydropower Operations. The evaluation shows that integrating Cross-Border Basin Agreements, Collaborative Planning and Adaptation Strategies, and Joint Water Allocation Models are highly effective, with the potential to mitigate around 1.11 (95% CI: 0.32–1.90) conflicts over five years. Cross-Border Basin Agreements and Transboundary Water Quality are recommended for less affluent countries, while Joint Data-Sharing Systems are most effective for countries under high water stress. Climate change will intensify water-related conflicts in the future, doubling conflicts in South America, Middle Southern Africa, and Eastern Asia by 2050, while climate action could nearly eliminate conflicts in high-risk regions and cut them by half in many other regions. However, the effectiveness of cooperative strategies is limited in less developed regions, underscoring the urgent need for international support and inclusive diplomacy to ensure global water security and stability. Scientific community and society/Water resources Earth and environmental sciences/Environmental social sciences/Climate-change impacts/Governance Water-related Cooperation Large Language Models Water-related Conflicts Water Diplomacy Climate Change Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Triggered by the recent deadly Kashmir attack, India’s threat to restrict Indus water flows provides a stark, contemporary illustration of how geopolitical conflicts can rapidly escalate over water resources, particularly in regions already facing vulnerabilities amplified by climate change 1 . Yet within this crisis lies an overlooked opportunity: environmental interdependence has historically catalyzed cooperation when institutions employ inclusive, data-driven approaches 2-4 . This urgency aligns with the UN’s 2025 Water for Peace initiative and IPCC AR7’s call for conflict-sensitive adaptation. Water cooperation refers to collaborative efforts between nations, regions, or communities to manage shared water resources sustainably and equitably. Transboundary water cooperation, in particular, involves collaboration between countries that share rivers, lakes, or aquifers, aiming to balance competing demands and ensure the sustainable use of these resources 5,6 . This process not only mitigates disputes but also strengthens regional ties by emphasizing shared interest in the sustainable management of natural resources. Through environmental peacebuilding, nations can transcend competition over scarce resources and create lasting peace 7,8 . The urgency for transboundary water cooperation is underscored by a global rise in water-related conflicts, exacerbated by climate change, with over 60% of the world’s transboundary river basins lacking cooperative management frameworks 9 . As global water demand is projected to increase by 20-30% by 2050 10 , climate change will intensify water stress by altering precipitation patterns, reducing water availability, and increasing the frequency of extreme weather events, particularly in regions already facing scarcity. An example of this can already be seen with the Nile River Basin, where there has been escalating tensions between Ethiopia, Egypt, and Sudan over the Grand Ethiopian Renaissance Dam (GERD), with climate change exacerbating the situation by reducing river flows and increasing variability in seasonal water availability 11 . Despite over 200 water-sharing agreements signed globally since 1948 12 , many of these lack effective implementation mechanisms 13 . This growing threat necessitates a proactive, inclusive approach to transboundary water cooperation that not only addresses immediate disputes but also anticipates future challenges, such as equitable resource allocation and adaptive management strategies in the face of climate change 14 . Regions like the Nile, Indus, and other shared basins highlight the critical need for robust, climate-resilient agreements to prevent conflicts and ensure sustainable water access for all, even when there is a history strong interstate rivalry. Developing effective strategies and actionable guidance for water-related cooperation remains a complex task. Water diplomacy and environmental peacebuilding require the integration of diverse fields such as sustainability science, environmental policy, and political science. Each cooperation attempt involves a variety of stakeholders, including governmental and non-governmental actors 15,16 . The complexity of these efforts can make developing an effective, scalable, cohesive approach challenging. Prior studies have mapped water conflicts and cooperation at a descriptive or aggregated level 2,3,17 . Research is often time-consuming as it involves understanding diverse perspectives and combining interdisciplinary knowledge 18 . To address these challenges, our study demonstrates how Artificial Intelligence (AI) can facilitate such analysis and illuminate pathways to peace: transforming conflict triggers into trust-building mechanisms. Large language models (LLMs) and retrieval-augmented generation (RAG) have emerged as powerful tools for understanding and analyzing past water-related cooperation events. LLMs can systematically process vast amounts of historical data to identify patterns and classify the factors contributing to successful cooperation. Meanwhile, RAG helps connect theories from multiple disciplines, such as environmental peacebuilding and water diplomacy 19,20 . This study introduces the MaCdip (Multi-agent, Cross-disciplinary Integrative qualitative study Process) framework, leveraging large language models (LLMs), retrieval-augmented generation (RAG), and machine learning to analyze over 2,000 global transboundary water cooperation and conflicts cases from 1951 to 2019. It identifies six distinct modes of cooperation: Cross-Border Basin Agreements, Collaborative Planning and Adaptation Strategies, Joint Water Allocation Models, Joint Data-Sharing Systems, Transboundary Water Quality Standards, and Coordinated Hydropower Operations. The study finds that combining Cross-Border Basin Agreements, Collaborative Planning and Adaptation Strategies, and Joint Water Allocation Models are highly effective, with the potential to mitigate around 1.11 (95% CI: 0.32–1.90) conflicts over five years. Under 2050 climate change scenarios, particularly for countries with projected severe water stress (under 250 mm annual precipitation), a Collaborative and Communicative Planning integrative strategy, encompassing Joint Data-Sharing Systems, Cross-Border Basin Agreements, Collaborative Planning and Adaptation Strategies are recommended, projected to reduce conflicts by over 50% in North America, Europe, Eastern Asia, Southeastern Asia, and Oceania. Identifying Modes of Water-related Cooperation Figure 1b briefly shows our analytical approach by using the integration of large language models (LLMs) and interpretable machine learning to analyze over 1,700 transboundary water cooperation events from databases like TFDD and WCC. To ensure comprehensive coverage, the study fills in spatial and temporal gaps in historical cooperation event descriptions using human searching, semi-automated web scraping, and event augmentation through task-specific LLMs while employing a hallucination guardrail to maintain data accuracy. The study introduces the MaCdip (Multi-agent, Cross-disciplinary Integrative qualitative study Process) framework, in which LLM agents are “trained” to be equipped with professional tone and relevant knowledge from experts in the three interdisciplinary fields of International Relations, Sustainability Science, and Hydrology. This is realized through a Retrieval Augmented Generation (RAG) approach that provides literature and theoretical background from the three interdisciplinary field perspectives to cluster and label water-related events. All automated coding was validated through a rigorous two-stage process involving expert review and intercoder reliability tests, ensuring consistency between machine-generated and human-coded results. Following LLM automatic coding, two workshops were held between human field scholars to verify the theoretical robustness and clarity of the LLM-generated labels of cooperative modes. Suggested modifications and revisions were incorporated through human expert consensus. Figure 1a shows how water-related conflicts tend to cluster in regions with high water stress, such as Central and Southern Asia and the Middle East, where competition over shared resources has historically fueled tensions. The ongoing tensions between India and Pakistan over the Indus Waters Treaty highlight how rising water demand exacerbates political frictions. In Northern Africa and Sub-Saharan Africa, conflicts related to poverty and political tensions are also prevalent, with disputes over the Nile between Egypt, Sudan, and Ethiopia, especially after the construction of the Grand Ethiopian Renaissance Dam (GERD). Additionally, the shrinking waters of Lake Chad have heightened instability in Nigeria, Niger, Chad, and Cameroon, deepening the region's poverty and conflict. There has also been evidence of successful water-related cooperation that prevents conflicts for at least five years, with Europe and parts of Sub-Saharan and Western Africa showing positive progress. The Senegal River Basin Organization (OMVS) is a standout example, in which Senegal, Mauritania, and Mali share resources cooperatively, fostering long-term collaboration. However, successful cooperation is lacking in regions like Southern Asia, the Middle East, and the Balkans. The South Asian nations have struggled to develop a robust cooperative framework around the Ganges-Brahmaputra-Meghna Basin, while unresolved water-sharing disputes along the Drin River in the Balkans illustrate the difficulties of managing transboundary water resources in politically complex areas. Using large language models (LLMs) to process event texts, the events were clustered through K-means, forming six distinct modes of cooperation: Cross-Border Basin Agreements, Collaborative Planning and Adaptation Strategies, Joint Water Allocation Models, Joint Data-Sharing Systems, Transboundary Water Quality Standards, and Coordinated Hydropower Operations. These modes were further refined by training three multidisciplinary agents (assimilating scholars in International Relations, Sustainability Science, and Hydrology) with field-specific literature on water-related cooperation. The Retrieval-Augmented Generation (RAG) method was used to assign theoretically grounded meanings to each cluster, with inputs from the three scholars informing the LLM prompts. Each cooperation mode was then matched to specific countries and years by comparing descriptions and examples to event details (Figure 2) . Among these, Cross-Border Basin Agreements emerged as the most frequently identified mode. These agreements are formal treaties or frameworks established between countries sharing transboundary water resources, such as rivers or aquifers, to ensure equitable and sustainable management. These agreements often include provisions for dispute resolution, conservation, and flood control. An example of such an agreement can be seen in the Indus Waters Treaty (1960) between India and Pakistan, which has facilitated cooperation despite political tensions. The detailed descriptions and case examples of the modes are recorded in Supplementary Materials S1 . Other modes also play critical roles in transboundary water cooperation. For example, Joint Data-Sharing Systems involve collaborative frameworks for sharing hydrological and environmental data, as seen in the Mekong River Commission (MRC). Transboundary Water Quality Standards focus on maintaining water quality across borders, exemplified by the Great Lakes Water Quality Agreement (GLWQA) between the U.S. and Canada; Coordinated Hydropower Operations optimize shared water resources for energy production, as demonstrated by the management of the Kariba Dam by the Zambezi River Authority (ZRA); Joint Water Allocation Models ensure equitable water distribution, such as in the Zambezi Watercourse Commission (ZAMCOM) strategy; and Collaborative Planning and Adaptation Strategies involve multiple stakeholders to manage shared river basins sustainably and address climate change impacts, as seen in the Lake Victoria Basin Commission (LVBC) Climate Adaptation Program. Effectiveness Evaluation of Different Modes of Water-related Cooperation The second phase of MaCdip focuses on evaluating the effectiveness of these cooperation modes in preventing future water conflicts and recommending future modes of cooperation under climate change scenarios. After coding qualitative insights into one-hot encoded features, the cooperation modes are fed into an interpretable machine learning model, which, considering country-specific socioeconomic and climate variables, predicts the effects of different cooperation mode combinations. An “Interpreter Agent” translates these machine learning outcomes into recommendations for cooperation modes to mitigate water conflicts in scenarios projected beyond 2050. The interpretable machine learning results ( Figure 3 ) indicate varying levels of effectiveness for the identified water-related cooperation modes. Coordinated Hydropower Operations show the highest feature importance scores, followed by Cross-Border Basin Agreements , highlighting how these modes are highly correlated with occurrences of conflicts; Joint Data-Sharing Systems has the lowest correlation. This result is expected as Coordinated Hydropower Operations and Cross-Border Basin Agreements are usually linked to the causes and outcomes of water-related conflicts. In comparison, Joint Data-Sharing Systems can be established without imminent resource scarcity or competition. With regards to the mitigative capacity of each mode, Cross-Border Basin Agreements shows the weakest impacts, as it is expected to reduce 0.78 (95% CI: 0.60, 0.97) conflicts within 5 years after the cooperative action. The other modes have similar impacts, ranging from 0.84 to 0.89, with Coordinated Hydropower Operations showing the strongest mitigative impacts 0.89 (95% CI: 0.67, 1.11). The effects of conflict mitigation are the most pronounced when multiple modes are combined. A combination of Cross-Border Basin Agreements , Collaborative Planning and Adaptation Strategies , and Joint Water Allocation Models is the most frequently used and has a stronger mitigative effect 1.11 (95% CI: 0.32–1.90) than individually implemented. The effectiveness of water-related cooperation modes is evaluated based on six criteria: Prescriptive, Preventive, Affordable, Facilitative, Stress-relieving, and Complementary ( Figures 4a-f ). Cross-Border Basin Agreements are the most prescriptive, which is essential for managing disputes, as evidenced by the 2010 meeting between Jordanian and Syrian ministers focused on fostering cooperation over shared water resources, including discussions of sustainable use, joint projects, and stakeholder participation. For preventive measures, Collaborative Planning and Adaptation Strategies stand out, with the 2019 EU-funded environmental projects in Romania, led by Hill International, serving as a modern example of effective conflict prevention through adaptive management frameworks and stakeholder engagement. Joint Water Allocation Models and Cross-Border Basin Agreements are usually affordable and can be implemented in less wealthy nations through international aid, such as the 2006 Nile Basin Initiative (NBI). Joint Data-Sharing Systems have emerged as the most complementary and are often integrated with other strategies, as can be seen in the 2001 joint flood defense agreement for the Tisza region signed by Ukraine, Slovakia, Romania, Yugoslavia, and Hungary, which combined data-sharing with technical assistance and stakeholder participation. In wealthier nations, Transboundary Water Quality Standards and Collaborative Planning and Adaptation Strategies are more facilitative, supporting trade and other non-water-related activities between countries, as seen in the 1999 Lake Victoria Initiative, which applied the Baltic Sea model for pollution clean-up and prevention. In high water-stress regions, Coordinated Hydropower Operations are the most effective at relieving such stress. Figure 4g shows the results of inter-coder reliability tests for all modes. Recommended Modes for Transboundary Water-related Cooperation Under an Uncertain Climate Future Under climate change scenarios projected beyond 2050, the recommended water-related cooperation modes for countries at high risk of future conflicts focus on addressing water conflicts in regions with extremely low precipitation ( Figure 5 ). Countries with annual precipitation under 1,000 mm are mostly projected to have a high likelihood of conflict, with high-risk countries including Palestine, Yemen, and Nigeria. For countries with extremely low precipitation (under 250 mm), a Collaborative and Communicative Planning integrative strategy, encompassing Joint Data-Sharing Systems , Cross-Border Basin Agreements , and Collaborative Planning and Adaptation Strategies , is recommended. Other modes, such as Joint Water Allocation Models , Coordinated Hydropower Operations , and Transboundary Water Quality Standards , are suggested for countries such as Ethiopia and Somalia. In 2015, Ethiopia demonstrated significant diplomatic intention to cooperate with Sudan and Egypt over the contentious Grand Ethiopian Renaissance Dam (GERD) issue, marking a shift from unilateral actions to a more collaborative approach to managing shared water resources. Somalia received assistance from international donors, non-governmental organizations, and many other countries through international water diplomacy and technical assistance amid its severe drought and civil war. The identification of countries with high conflict risks beyond 2050 and the corresponding cooperative actions are recorded in Supplementary Materials S4 . The projection of the mitigation effects under future climate scenarios from 2050 to 2100 highlights the growing challenge of water-related conflicts exacerbated by climate change. Increased water stress is expected to intensify conflicts in most regions, particularly in South America, central southern Africa, and East Asia, where water scarcity could more than double the number of conflicts. Conversely, East Africa and South Asia may experience fewer conflicts. The “time trend effect” further reveals that South America and central south Asia will face a significant rise in annual conflicts, while central southern Africa and North Africa are projected to see a decline, suggesting that historical and geopolitical factors influence conflict trajectories. Climate action, particularly transitioning from fossil-fueled development (SSP5) to sustainable practices (SSP1), emerge as a highly effective strategy for mitigating water-related conflicts ( Figure 6 ). This shift could nearly eliminate conflicts in South America and East Asia and reduce conflicts by over half in Europe, the Middle East, West Africa, and central south Africa. However, the effectiveness of water-related cooperation modes varies significantly across regions. Our identified and recommended cooperative actions can mitigate more than half of conflicts in regions like Eastern and Southeastern Asia, Europe, Eastern Asia, and North America. Despite these successes, the effectiveness of cooperative strategies is limited in less developed regions, such as the Middle East and Africa, where countries often lack the diplomatic influence and resources to implement and leverage such frameworks. This disparity highlights the need for international support, capacity-building, and inclusive, multi-stakeholder engagement to empower these nations. Discussion and Implications The study highlights the crucial role of transboundary water cooperation in mitigating conflicts, especially under climate change pressure, providing proof-of-evidence of how cooperative water management reduces disputes and fosters regional stability 17,21 . This study builds on these insights, demonstrating that integrated cooperation, through Cross-Border Basin Agreements , Collaborative Planning , and Joint Water Allocation Models , is more effective than isolated strategies in reducing conflicts and promoting long-term resilience. From a methodological perspective, our framework uses LLMs in a novel way to analyze complex transboundary conflict data. This underscores its relevance to computer scientists interested in domain-specific applications for peacebuilding and water governance. However, effectiveness varies by region. Less affluent, high water-stress regions like the Middle East and Africa face barriers due to resource and political constraints, limiting their ability to implement cooperative frameworks 22,23 . This disparity underscores the need for tailored interventions and international support. While wealthier nations benefit from integrated cooperation, less developed regions struggle, exacerbating vulnerabilities and hindering global sustainability efforts 24,25 . Furthermore, some regions have robust legal frameworks and high levels of trust, while others lack comprehensive treaties or enforcement mechanisms. Varying degrees of political stability and institutional capacity can also lead to very different outcomes when disputes arise over shared water resources. To address inequities, global institutions and wealthier nations should enhance diplomatic influence and resource access for developing countries. Inclusive multilateral platforms, investment in infrastructure, education, and governance, and regional water diplomacy hubs could support equitable agreements and climate resilience 26 . Climate action can further reduce water conflicts, particularly in South America and Eastern Asia. Aligning water cooperation with climate adaptation ensures resilience to future uncertainties 11,17 . Climate-resilient infrastructure, such as flood control and drought-resistant irrigation, should be included in agreements with international climate funds like the Green Climate Fund supporting these projects 27,28 . Policymakers must prioritize integrated cooperation strategies, especially in high-risk and less affluent regions. International organizations like the UN and World Bank could establish dedicated funds for cooperative water projects in areas like the Nile Basin and Lake Chad 29,30 . Strengthening institutional capacity and cross-border collaboration remains essential for sustainable water governance and conflict mitigation in the face of climate challenges. Finally, by leveraging LLMs to synthesize complex water and climate data, our approach can democratize access to diplomatic and climate-resilience insights for low-resource nations, empowering evidence-based planning even where expertise and resources are scarce. It is also noteworthy that the findings of the model, as well as practical evidence, suggest that choosing the relevant and appropriate cooperation modes and their combinations is important, as adopting more cooperation modes does not linearly reduce conflicts. While some cooperation initiatives can alleviate regional tension, they might also generate local disputes within individual countries. For example, if a national government agrees to certain water allocation terms with a neighboring state, local communities might feel their interests have been overlooked or threatened. In regions facing climate change impacts, a water management strategy that appears to be beneficial in the short term (e.g., building a dam for flood control) may unintentionally harm downstream communities or ecosystems over time. Such maladaptation risks can fuel local tensions and undermine the broader goal of peacebuilding. 31 Methods Study data and preprocessing with Large Language Model (LLM) Data Collection and Integration: This study employs a comprehensive data collection and integration strategy to analyze over 1,700 transboundary water cooperation events and 400 water conflict events. The primary data sources are described below, while more details about the data sources are recorded in the data availability statement. 1) Water-Related Cooperation and Conflict Event Database: Transboundary Freshwater Dispute Database (TFDD) and Water Cooperation and Conflict (WCC) Database are the repositories used to repositories provide historical records of water-related cooperation and conflict events. To enhance data coverage and reliability, an updated cooperation database was constructed by integrating newly gathered data with existing records. Similarly, water-related conflict datasets were merged to create a more comprehensive record of conflicts over an extended time span 32 . The resulting Extended Cooperation-Conflict Dataset (1951–2019) (provided with publicly accessible link in the Data Availability Section) covers all trigger conflicts from the WCC and severe conflicts from the TFDD (BAR rating -3 to -7). The dataset was further enriched by integrating available cooperation data, resulting in a consolidated database spanning 1951–2019. To facilitate analysis, events were geographically matched with global climate zones, regional precipitation patterns, evapotranspiration rates, elevation data, and socioeconomic indicators (e.g., population density). The resulting spatial data is shown in Figure 1a . The new database compiled in this study is compared to prior mainstream databases in Supplementary Materials 1 . 2) Demographic, Economic, and Climatic Variables: Key variables were sourced from internationally recognized databases. World Development Indicators (WDI) provides annual country-level data on population density, GDP per capita (constant 2015 US$), rural population percentage, and exports of goods and services as a percentage of GDP 33 . Precipitation data is extracted from the Water Balance Model (WBM) (Grogan, 2016; Grogan et al., 2022) and aggregated at the HydroBASINS level 6 scale 34 .Water Stress Indicators were measured using Aqueduct 3.0 (for events before 1985) and Aqueduct 4.0 (for events from 1985 onward) (World Resources Institute, 2023) 35 . For regional classifications data were structured according to the United Nations Statistics Division (UNSD) regional definitions, ensuring consistency in basin-level event aggregation. Historic values (previous 5 years) and projected future values (following 5 years) were computed for socioeconomic and precipitation variables. Future conflict events were estimated using predictive models incorporating climatic, economic, and demographic predictors. LLM-Based Event Augmentation and Hallucination Guardrail: Given the incomplete nature of historical event records, as many are short excerpt from historic newspaper and other hardcopy sources that are not as easily accessible as more recent online news pieces, task-specific LLMs were employed to generate plausible event descriptions where data gaps exist. These models were fine-tuned using historical event data to ensure contextual accuracy and domain relevance. Since LLMs are prone to generating hallucinated or factually incorrect information, a robust hallucination guardrail was implemented to ensure data reliability 36 . The methodology follows best practices in LLM-generated text evaluation 36 . First, a sample of 200 LLM-generated event descriptions was manually reviewed by domain experts. The assessment found that only 2% (4 events) exhibited signs of hallucination, indicating a high degree of reliability of Initial LLM-generated content. Second, A hallucination classification model was developed using human-verified data. Key features for detection included semantic consistency, factual alignment with known data, and contextual relevance. The model was trained to distinguish high-confidence, plausible events from hallucinated ones using a supervised learning approach. Third, for any event with a hallucination probability exceeding 20% was flagged for manual review. Events classified as high-risk hallucinations were either discarded or corrected using external data validation. Finally, a feedback loop was established to continuously update the hallucination detection model based on new data and expert evaluations. Hallucination probability was again then used to compare generated text with historical records. All LLM operations are executed using ChatGPT-4o on Azure Cloud. Cooperation Mode Identification by LLM with RAG in a MacDip Framework This study employs the MaCdip framework (Multi-agent, Cross-disciplinary Integrative Qualitative Study Process) to analyze over 2,000 global cases of transboundary water cooperation and conflcits. The methodology integrates Large Language Models (LLMs), advanced embedding techniques, clustering algorithms, and Retrieval-Augmented Generation (RAG) to classify cooperation modes. The prompts, pseudo-code, parameters, and literature search strategy in our MacDip framework is available in Supplementary Materials S2 . Embedding Generation and Initial Clustering: Event texts were transformed into dense numerical representations using ChatGPT-4o embeddings, selected for their superior performance in dialogue and QA tasks 37 . Key technical configurations include: 1) Model Architecture: Leveraged the transformer-based text-embedding-3-large variant, generating 3,072-dimensional embeddings; 2) Context Handling: Utilized a sliding window of 8,192 tokens to process long texts, with mean pooling for sentence-level aggregation. Initial Text Clustering: K-means clustering was applied to group embeddings, chosen for scalability and robustness in high-dimensional spaces. Critical implementation details include: 1) Parameterization: n_clusters determined via the elbow method (k=9 optimal), n_init=10 to mitigate initialization bias, and random_state=42 for reproducibility. 2) Validation Metrics: Assessed cluster quality using Silhouette Score; 3) Cluster Refinement: Merged overlapping clusters using cosine similarity thresholds (<0.85) and re-ran K-means to ensure distinctiveness. Literature Database Construction for 3 Scholars with Interdisciplinary Insights: The initial clusters lack theoretical robustness and meanings to adequately identify modes of cooperation from the tasks. Thus, this study constructs a literature database from interdisciplinary insights from International Relations, Sustainability Science, and Hydrology to refine cluster labels, using PRISMA literature review technology. Search strategy for the literature uses query from Web of Science (2010–2024) with Boolean logic: TS=(water resource AND (cooperation OR conflict OR governance)) AND PY=(2010–2024) AND DT=(ARTICLE) Inclusion/Exclusion criteria is set for the screened 1,513 articles to identify 73 high-relevance papers, focusing on peacebuilding mechanisms and transboundary case studies. The details of the literature search and PRISMA flow chart are available in Supplementary Materials S2 . RAG-Driven Label Refinement: The RAG system integrated interdisciplinary insights from the Literature Database built from 3 scholarly perspectives. In the first step, we split texts into 1,000-character chunks with 30-character overlaps using CharacterTextSplitter. Then, we embedded chunks via sentence-transformers/all-mpnet-base-v2 (768d), optimized for semantic similarity and stored vectors in FAISS with IVF indexing for efficient retrieval 38 . Then, we deployed three discipline-specific retrievers, each fine-tuned on domain literature. Each scholar receives equal voting. We prompted ChatGPT 4-o to generate labels (e.g., Joint Water Allocation Models from raw “water allocation”) for each piece of collaborative event, with temperature=0 to ensure deterministic, theory-grounded, and theoretically robust outputs. For each final label, detailed descriptions and real-world examples are provided by LLM with retrieval both relevant information from the event database as well as the literature database. These descriptions clarify the meaning of each label and offer practical illustrations of how countries collaborate in these ways. For instance, the Cross-Border Basin Agreements label includes examples such as the Nile Basin Initiative and the Mekong River Commission. Validation and Verification: All automated coding is validated through a rigorous two-stage process: 1) Expert review: Following the LLM automatic coding, a workshop is held among human field scholars (16 scholars) on 11.17.2024 and (5 scholars) 03.11.2025 to verify the results Domain experts review the machine-generated labels to ensure their accuracy, relevance, and theoretical robustness. Some suggestions are made to each label title, its content and examples. The research team then synthesizes these comments and finalizes the identified modes of water-related cooperation. 2) Intercoder Reliability Tests: Consistency between machine-generated and human-coded results is assessed through intercoder reliability tests of 20 human coders, ensuring the labeling process is both reliable and valid. The details of the inter-coder reliability test are recorded Supplementary Materials S3 . The results demonstrate strong agreement (overall score: 0.712) between human and LLM coding, with five of six modes scoring 0.7–0.9. The exception, Cross-Border Basin Agreements (0.583), reflects a divergence in interpretation: human coders focused on explicit treaty events, while the LLM included implicit agreements not captured manually. Evaluating Effectiveness of Water-related Cooperation Mode This study employs a multi-criteria evaluation framework 39 powered by interpretable machine learning (ML) and natural language processing (NLP) framework to evaluate the effectiveness of transboundary water cooperation modes in mitigating conflicts. The methodology integrates quantitative ML analysis with qualitative NLP-driven multi-criteria evaluation, ensuring both statistical rigor and contextual depth. The methodology details of mode evaluation and future mitigative effect projection is recorded in Supplementary Materials S4 . Interpretable Machine Learning Framework: Random Forest regression with SHAP (SHapley Additive exPlanations) is chosen for its interpretability and robustness in handling complex, non-linear relationships 40 . Bayesian optimization is used to fine-tune hyperparameters such as the number of trees, maximum depth, and minimum samples per leaf. Data from 2011-2019 is used as test data to validate the model. Performance metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-squared are used to evaluate the model. And the MAE metric is utilized to determine the optimal combination of cooperation modes that maximize conflict mitigation effects. SHAP values are used to quantify the contribution of each cooperation mode to conflict mitigation. The SHAP summary plot (Figure 3d) visualizes the impact of cooperation mode on the model's output, providing a global interpretation of the model. The SHAP values are aggregated to rank the importance of each feature, highlighting which cooperation modes and socioeconomic factors are most influential in mitigating conflicts. Mitigative effects are calculated for hypothetical scenarios where all conflict events are provided with one certain mode of collaboration or a highly frequent combination of modes. This allows for a detailed understanding of how specific modes or combinations of modes influence conflict mitigation. Predictors and independent variables are selected as the following. The dependent variable is the count of water-related conflicts within a 5-year period (integer). The independent variables include Event type, annotating whether the event is a conflict or a certain mode of cooperation. Climate variables include annual precipitation and water stress levels, socioeconomic controls include population density, GDP per capita, rural population, and export dependency ratio. The variables are integrated into a country-year panel dataset, ensuring temporal and spatial consistency. NLP-Driven Multi-Criteria Evaluation: The evaluation is based on six dimensions, each scored on a scale of 0.5 to 6.0. Prescriptive: The predictive power (MAE) of the mode in relation to conflict occurrence. This dimension evaluates the correlation between each mode and the occurrence of conflict. Preventive: The capacity of the mode to mitigate conflicts. Affordable: The cost-effectiveness of the mode for low-GDP nations. Facilitative: The alignment of the mode with trade-dependent economies. Stress-Relieving: The impact of the mode on water scarcity. Complementary: The ability of the mode to synergize with other modes. The evaluation is conducted using inputs from SHAP values and ML metrics for each mode. The NLP model uses GPT-4 via Azure OpenAI API, fine-tuned on labeled events. By integrating event texts, the NLP model provides a qualitative layer to the quantitative ML analysis, offering insights into why certain modes are more effective than others in specific contexts. The following prompt template is used to generate scores and justifications: "Score [Mode] on [Criterion] (0.5 – 6 scale) using: - Machine Learning Output - Case study [Excerpt from Nile Basin Initiative 2006] Justify scoring." For output, the NLP model generates evaluation metric scores (in 0.5 increments) with expert-style rationales. For example: Cross-Border Basin Agreements scored 5.0 on “Prescriptive” due to a high MAE and feature importance among all modes. SSP Future Climate Scenario and Recommendation of Future Cooperation Modes SSP Scenarios: To project future water-related conflict risks under climate change, we adopt the S hared Socioeconomic Pathways (SSPs) framework, which integrates socio-economic narratives with climate projections 41 . This approach aligns with methodologies from Chen et al. 42 , where SSPs are used to simulate urban land patterns by combining population, GDP, and urbanization rate predictions. The scenarios include: 1)Benchmark Scenario: SSP5 (Fossil-Fueled Development), characterized by high greenhouse gas emissions and minimal climate policy; 2) Climate Action Scenario: SSP1 (Sustainability), emphasizing low-carbon transitions and sustainable resource management; 3) Increased Water Stress Scenario: Future basin-level water stress under SSP1 and SSP5 is derived from Aqueduct 4.0 toaccount for increased water stress under climate change; 4) Time trend effect: Use time series model to account for conflict trend for each country till 2050; 5) Implementing Water-related Cooperation: Estimate the mitigative effects for future conflicts with recommended modes of water-related cooperation implemented. For other socioeconomic and climate data used in the machine learning model, we find replacement data from different SSP projections. Precipitation and temperature projections for SSP1 and SSP5 are sourced from the World Data Center for Climate (CMIP5 downscaled datasets) and CEDA, ensuring consistency with IPCC AR6 regional extremes. Future basin-level water stress under SSP1 and SSP5 is derived from Aqueduct 4.0 (World Resources Institute), which models hydrological risks under varying socio-economic conditions. Population, GDP, and urbanization trends are extracted from the IIASA SSP Database, aggregated into 32 macro-regions as per SSP regional classifications. Inter-country trade data to estimate future export is obtained from trading economics and World Bank. Since there are huge uncertainties to project to a certain year beyond 2025, we collect future projection data with the closest time to 2050, but acknowledging that the projection could occur in any time within 2050 to 2100. Time Series Analysis and Conflict Risk Projection : Building on methodologies from Ge et al. (2022) 43 , we employ a hybrid machine learning and time series model to project conflict risks. Adopted from Ge et al. (2022) 43 , An ARIMA model integrates historical conflict trends (1951–2019) with SSP-driven climate and socio-economic projections to simulate annual conflict likelihoods for 2050–2100. Agglomerative Hierarchical Clustering (AHC) is employed to stratify national risk levels through conflict probability metrics, enabling systematic identification of high-risk regions. Recommendation System for Future Cooperation Modes: Drawing from Lyu et al (2023) 44 , we design an AI-assisted decision framework combining NLP, quantile clustering, and reinforcement learning. First, country-specific variables (GDP per capita, baseline water stress) are transformed into textual prompts. For example: "Country X exhibits high transboundary dependency (85%), low GDP ($1,200/capita), and extreme water stress (score: 4.8). Recommend cooperative strategies aligned with SSP1 sustainability pathways." Second, Countries are grouped into quantiles for their future projected variables. Clusters reflect geopolitical contexts (e.g., "Low-Income/High Stress") to tailor recommendations like Joint Data-Sharing Systems for Somalia or Transboundary Water Standards for Ethiopia. A feedback loop with iterative prompting, updates model weights based on historical cooperation outcomes. For instance, successful dam negotiations in the Nile Basin (2015–2023) increase the confidence score for Cross-Border Basin Agreements in similar contexts. Mitigation Effect Evaluation: The coupled machine learning and time series model re-projects conflict counts for 2050–2100 under SSP5 and SSP1, with recommended cooperation modes activated (binary flag: 1). Conflict counts under SSP5 are normalized to 1 for comparative analysis. Results are aggregated into IPCC AR6 regions. Standard deviations from 20 ensemble machine learning runs generate uncertainty maps, ensuring robustness in policy prescriptions. References Koubi, V. Climate change and conflict. Annual Review of Political Science 22 , 343-360 (2019). Bernauer, T. & Böhmelt, T. International conflict and cooperation over freshwater resources. Nature Sustainability 3 , 350-356 (2020). Turgul, A. et al. Reflections on transboundary water conflict and cooperation trends. Water International 49 , 274-288 (2024). Wolf, A. T. Shared waters: Conflict and cooperation. Annu. Rev. Environ. Resour. 32 , 241-269 (2007). Dinar, A., Dinar, S., Mckinney, D. C. & Mccaffrey, S. C. Bridges over water: Understanding transboundary water conflict, negotiation and cooperation . Vol. 3 (World Scientific Publishing Company, 2007). Wolf, A. T. Transboundary water conflicts and cooperation. In Search of Sustainable Water Management , 131-154 (2005). Krampe, F., Hegazi, F. & VanDeveer, S. D. Sustaining peace through better resource governance: Three potential mechanisms for environmental peacebuilding. World Development 144 , 105508 (2021). Keskinen, M., Häkkinen, E., Haapala, J. & Sharipova, B. Trust in Transboundary Waters: Identifying Trust-Building in Water Diplomacy Literature. Water Alternatives 16 (2023). UN-Water. Transboundary Waters: Sharing Benefits, Sharing Responsibilities., (2021). IPCC. Climate change 2022: Impacts, adaptation and vulnerability. (2022). Basheer, M. et al. Cooperative adaptive management of the Nile River with climate and socio-economic uncertainties. Nature Climate Change 13 , 48-57 (2023). Katz, D. Hydro-political hyperbole: Examining incentives for overemphasizing the risks of water wars. Global Environmental Politics 11 , 12-35 (2011). WHO/UNICEF. Progress on household drinking water, sanitation and hygiene 2000-2020: five years into the SDGs . (World Health Organization, 2021). de Bruin, S. P., Schmeier, S., van Beek, R. & Gulpen, M. Projecting conflict risk in transboundary river basins by 2050 following different ambition scenarios. International Journal of Water Resources Development 40 , 7-32 (2024). Islam, S. & Susskind, L. Using complexity science and negotiation theory to resolve boundary-crossing water issues. Journal of Hydrology 562 , 589-598 (2018). Sommer, U. & Fassbender, F. Environmental Peacebuilding: Moving beyond resolving Violence-Ridden conflicts to sustaining peace. World Development 178 , 106555 (2024). Dinar, S., Katz, D., De Stefano, L. & Blankespoor, B. Climate change, conflict, and cooperation: Global analysis of the effectiveness of international river treaties in addressing water variability. Political geography 45 , 55-66 (2015). Munia, H. A. et al. Future transboundary water stress and its drivers under climate change: a global study. Earth's future 8 , e2019EF001321 (2020). Qiu, L. et al. Phenomenal yet puzzling: Testing inductive reasoning capabilities of language models with hypothesis refinement. arXiv preprint arXiv (2023). MacAvaney, S. & Soldaini, L. in Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2230-2235. Ide, T. & Detges, A. International water cooperation and environmental peacemaking. Global Environmental Politics 18 , 63-84 (2018). Swain, A. Challenges for water sharing in the Nile basin: changing geo-politics and changing climate. Hydrological Sciences Journal 56 , 687-702 (2011). Galvez, V., Rojas, R., Bennison, G., Prats, C. & Claro, E. Collaborate or perish: Water resources management under contentious water use in a semiarid basin. International Journal of River Basin Management 18 , 421-437 (2020). Eriksen, S. et al. Adaptation interventions and their effect on vulnerability in developing countries: Help, hindrance or irrelevance? World development 141 , 105383 (2021). Nguyen, T. T., Grote, U., Neubacher, F., Do, M. H. & Paudel, G. P. Security risks from climate change and environmental degradation: Implications for sustainable land use transformation in the Global South. Current Opinion in Environmental Sustainability 63 , 101322 (2023). Klimes, M., Michel, D., Yaari, E. & Restiani, P. Water diplomacy: The intersect of science, policy and practice. Journal of Hydrology 575 , 1362-1370 (2019). Bowman, M. & Minas, S. Resilience through interlinkage: the green climate fund and climate finance governance. Climate policy 19 , 342-353 (2019). Glemarec, Y. Financing green and climate resilient infrastructure in ASEAN countries. Environmental Progress Sustainable Energy 42 , e14097 (2023). Biswas, A. K. in Asian Perspectives on Water Policy 1-12 (Routledge, 2013). Hirji, R. F. & Duda, A. Integrated management of lakes, reservoirs, and their basins is critical for a climate-resilient planet: an urgent wake-up call from collective amnesia. Water Policy 27 , 66-87 (2025). Krampe, F., Smith, E. S. & Hamidi, M. D. J. P. G. Security implications of climate development in conflict-affected states: implications of local-level effects of rural hydropower development on farmers in Herat. Political Geography 90 , 102454 (2021). Kåresdotter, E., Skoog, G., Pan, H. & Kalantari, Z. Water-related conflict and cooperation events worldwide: A new dataset on historical and change trends with potential drivers. Science of the Total Environment 868 , 161555 (2023). Singh, J. Does economic misery stifle human development? empirical evidence from Asian countries. GeoJournal 89 , 1-16 (2024). Grogan, D. S. et al. Water balance model (WBM) v. 1.0. 0: a scalable gridded global hydrologic model with water-tracking functionality. Geoscientific Model Development 15 , 7287-7323 (2022). Kuzma, S. et al. Aqueduct 4.0: Updated decision-relevant global water risk indicators . (World Resources Institute Washington, DC, USA, 2023). van Schaik, T. A. & Pugh, B. in Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2832-2836. Petukhova, A., Matos-Carvalho, J. P. & Fachada, N. Text Clustering with Large Language Model Embeddings. arXiv preprint arXiv , 15112 (2024). Arslan, M., Mahdjoubi, L. & Munawar, S. J. E. Driving sustainable energy transitions with a multi-source RAG-LLM system. Energy & Buildings 324 , 114827 (2024). Chambers, J. M. et al. Six modes of co-production for sustainability. Nature Sustainability 4 , 983-996 (2021). Shevchenko, V. et al. Climate change impact on agricultural land suitability: An interpretable machine learning-based Eurasia case study. IEEE Access 12 , 15748-15763 (2024). O’Neill, B. C. et al. The roads ahead: Narratives for shared socioeconomic pathways describing world futures in the 21st century. Global environmental change 42 , 169-180 (2017). Chen, G. et al. Global projections of future urban land expansion under shared socioeconomic pathways. Nature communications 11 , 537 (2020). Ge, Q. et al. Modelling armed conflict risk under climate change with machine learning and time-series data. Nature communications 13 , 2839 (2022). Lyu, H. et al. Llm-rec: Personalized recommendation via prompting large language models. arXiv preprint arXiv , 15780 (2023). Additional Declarations There is NO Competing Interest. Supplementary Files S1FinalModesDescriptionv250418.docx Supplementary Materials 1. Description of Identified Water-Related Cooperation Modes S2PromptPseudoCodeMacDIP.docx Supplementary Materials S2. Parameters, Prompt, Psuedo-Code, and Literature Search Strategy for the MacDip Framework S3InterCoderReliabilityTests.docx Supplementary Materials S3. Inter-Coder Reliability Tests Process S4evaluationcriteriasspstorylineZC250416.docx Supplementary Materials S4. Evaluation of Water-related Cooperation Modes and Projections of Conflict Mitigation Effects under Future Climate Uncertainty Cite Share Download PDF Status: Posted Version 1 posted 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-6847954","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":473069322,"identity":"a0c53b57-3325-46fc-85bc-c48101f76438","order_by":0,"name":"Cong Cong","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAp0lEQVRIiWNgGAWjYNCCChsGNhDNQ7yWM2kMbGwkaWFsO8zAQLQW/vbDxx5+YTsfzSffwPjgbRsRWiTOpKUby/Dczm1jY2A2nEuMFgMJHjNpCQmwFjZpXuK1GJwDaWH/TbQWyQ8JB8C2MBOlBeiXNGmGA8lALYnNknPOEaEFFGKSP//Z5c5vPnzww5syIrSAADMkOhgbiFQPUvuDeLWjYBSMglEwEgEA9y0uYPQtpZAAAAAASUVORK5CYII=","orcid":"","institution":"MIT","correspondingAuthor":true,"prefix":"","firstName":"Cong","middleName":"","lastName":"Cong","suffix":""},{"id":473069323,"identity":"1fc1bf1e-b2bd-4066-a21a-c998bdfff583","order_by":1,"name":"Haozhi Pan","email":"","orcid":"https://orcid.org/0000-0002-0709-632X","institution":"Shanghai Jiao Tong University","correspondingAuthor":false,"prefix":"","firstName":"Haozhi","middleName":"","lastName":"Pan","suffix":""},{"id":473069324,"identity":"cde0bacf-dcca-4ee3-b0f1-0852626941a1","order_by":2,"name":"Yiming Yang","email":"","orcid":"","institution":"Shanghai Jiao Tong University","correspondingAuthor":false,"prefix":"","firstName":"Yiming","middleName":"","lastName":"Yang","suffix":""},{"id":473069325,"identity":"4538c3ae-7b5e-4486-9ec6-23f34aa3402c","order_by":3,"name":"Zipan Cai","email":"","orcid":"","institution":"Southeast University","correspondingAuthor":false,"prefix":"","firstName":"Zipan","middleName":"","lastName":"Cai","suffix":""},{"id":473069326,"identity":"b5fadd20-1562-4512-afae-c56f63f8be0f","order_by":4,"name":"Elisie Kåresdotter","email":"","orcid":"https://orcid.org/0000-0003-3424-3847","institution":"KTH Royal Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Elisie","middleName":"","lastName":"Kåresdotter","suffix":""},{"id":473069327,"identity":"21b33b3d-2477-4fef-ad98-7fa6e7020662","order_by":5,"name":"Jay Bahn","email":"","orcid":"","institution":"MIT","correspondingAuthor":false,"prefix":"","firstName":"Jay","middleName":"","lastName":"Bahn","suffix":""},{"id":473069328,"identity":"6a79d61c-ef25-4a61-b997-55c32d51a8d8","order_by":6,"name":"Hong Zhou","email":"","orcid":"","institution":"Shanghai Jiao Tong University","correspondingAuthor":false,"prefix":"","firstName":"Hong","middleName":"","lastName":"Zhou","suffix":""},{"id":473069329,"identity":"cb959f6e-b288-40fa-9fb5-4bdb2fc60260","order_by":7,"name":"Jessica Page","email":"","orcid":"","institution":"Stockholm University","correspondingAuthor":false,"prefix":"","firstName":"Jessica","middleName":"","lastName":"Page","suffix":""},{"id":473069330,"identity":"8573478a-149e-48d7-9311-e846eb546bbe","order_by":8,"name":"Stefan Döring","email":"","orcid":"https://orcid.org/0000-0001-6949-7084","institution":"Uppsala university","correspondingAuthor":false,"prefix":"","firstName":"Stefan","middleName":"","lastName":"Döring","suffix":""},{"id":473069331,"identity":"d5673df6-5c75-406d-85f0-dab6fd42546d","order_by":9,"name":"Florian Krampe","email":"","orcid":"https://orcid.org/0000-0002-2208-794X","institution":"Stockholm International Peace Research Institute (SIPRI)","correspondingAuthor":false,"prefix":"","firstName":"Florian","middleName":"","lastName":"Krampe","suffix":""},{"id":473069332,"identity":"11eefe34-0145-4f16-8ede-617f4d8a0536","order_by":10,"name":"Ashok Swain","email":"","orcid":"","institution":"Uppsala University","correspondingAuthor":false,"prefix":"","firstName":"Ashok","middleName":"","lastName":"Swain","suffix":""},{"id":473069333,"identity":"2f41e60c-b712-4331-9254-3e113dad3e6f","order_by":11,"name":"Zahra Kalantari","email":"","orcid":"https://orcid.org/0000-0002-7978-0040","institution":"KTH Royal Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Zahra","middleName":"","lastName":"Kalantari","suffix":""}],"badges":[],"createdAt":"2025-06-08 14:15:46","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6847954/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6847954/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":85071549,"identity":"0aa800f7-2a20-4188-b324-d8134c57ea2a","added_by":"auto","created_at":"2025-06-20 15:40:40","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":2436316,"visible":true,"origin":"","legend":"\u003cp\u003ea. Global transboundary water-related conflicts and effective cooperation. b. The method flows of combining Large Language Model and human coding to identify effective transboundary water-related cooperation.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6847954/v1/b96a1533825ec0bdce0cfe94.png"},{"id":85071548,"identity":"bde9f3b8-f813-4e19-9289-33d5a1b389c8","added_by":"auto","created_at":"2025-06-20 15:40:40","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1834007,"visible":true,"origin":"","legend":"\u003cp\u003eWater-related cooperation modes identification, clustering, and labeling for involving actors (countries) at specific years.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6847954/v1/7bb9fe96010711d963f853e1.png"},{"id":85071551,"identity":"7a04b6d6-c2f3-47f0-8d3e-de9f4b7d5325","added_by":"auto","created_at":"2025-06-20 15:40:40","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1060974,"visible":true,"origin":"","legend":"\u003cp\u003eEffectiveness projection of water-related cooperation modes. 3a. Feature importance of different cooperation modes; 3b. Pearson correlation of each mode with other variables; c. Mean Absolute Error (MAE) of out-of-bag prediction performance of different modes combinations; d. SHAP visualization of conflict mitigation effects of each mode; e. Average conflict mitigation effects of each cooperation mode; f. Conflict mitigation effect of the best MAE combination of cooperation modes.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6847954/v1/4bccf276250d3ede843f5b69.png"},{"id":85071550,"identity":"9d87ebee-cf16-43e3-9d36-f72547b0561a","added_by":"auto","created_at":"2025-06-20 15:40:40","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1574738,"visible":true,"origin":"","legend":"\u003cp\u003ea-f. Multi-dimensional effectiveness evaluation of water-related cooperation modes. g. Inter-coder reliability assessment of machine coding for each cooperation mode.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6847954/v1/77dcf9c8f34d5d2c040ad1d1.png"},{"id":85071547,"identity":"79710be1-f0e7-4255-852a-9f28abc8ddf5","added_by":"auto","created_at":"2025-06-20 15:40:40","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1377972,"visible":true,"origin":"","legend":"\u003cp\u003eRecommended modes for water-related cooperation for conflict mitigation beyond 2050: 5a is recommendations and projected conflicts for all countries; 5b zooms in on conflict-prone countries; 5c focuses on countries with extremely high water stress; and 5d lists recommendations for each country.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6847954/v1/63f00a7641d7a27c81853523.png"},{"id":85071553,"identity":"dee397c0-3ad1-420f-b622-73661bf398fe","added_by":"auto","created_at":"2025-06-20 15:40:40","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1700320,"visible":true,"origin":"","legend":"\u003cp\u003eProjections of the mitigation effects for each mode under future climate scenarios from 2050 to 2100. The projected number of conflicts in the baseline scenario (SSP5) is normalized to 1.00.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-6847954/v1/9869e9bf70b2114c625af469.png"},{"id":88517644,"identity":"7918ce91-0ff9-4460-9d74-53e00b3401de","added_by":"auto","created_at":"2025-08-07 09:10:33","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":15017431,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6847954/v1/ca8c2a1e-cd08-4483-9138-2b965e20b026.pdf"},{"id":85072890,"identity":"8febb9cd-e174-493f-a60f-04f7ba3b7772","added_by":"auto","created_at":"2025-06-20 15:56:40","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":26167,"visible":true,"origin":"","legend":"Supplementary Materials 1. Description of Identified Water-Related Cooperation Modes","description":"","filename":"S1FinalModesDescriptionv250418.docx","url":"https://assets-eu.researchsquare.com/files/rs-6847954/v1/9098305ca7c665f0d072c8f7.docx"},{"id":85072114,"identity":"2cb7030c-6ac9-45b5-8355-81d44fd0d2d2","added_by":"auto","created_at":"2025-06-20 15:48:40","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":284901,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Materials S2. Parameters, Prompt, Psuedo-Code, and Literature Search Strategy for the MacDip Framework\u003c/p\u003e","description":"","filename":"S2PromptPseudoCodeMacDIP.docx","url":"https://assets-eu.researchsquare.com/files/rs-6847954/v1/75188c7c924d5f6c19a48c3c.docx"},{"id":85071543,"identity":"b5bed1e6-3224-459e-aba3-ead2482192fb","added_by":"auto","created_at":"2025-06-20 15:40:40","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":27109,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Materials S3. Inter-Coder Reliability Tests Process\u003c/p\u003e","description":"","filename":"S3InterCoderReliabilityTests.docx","url":"https://assets-eu.researchsquare.com/files/rs-6847954/v1/f72591d82b970d8480d2fe6e.docx"},{"id":85071552,"identity":"51b0e61b-731e-4775-8d9e-88f06b91738e","added_by":"auto","created_at":"2025-06-20 15:40:40","extension":"docx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":386269,"visible":true,"origin":"","legend":"Supplementary Materials S4. Evaluation of Water-related Cooperation Modes and Projections of Conflict Mitigation Effects under Future Climate Uncertainty","description":"","filename":"S4evaluationcriteriasspstorylineZC250416.docx","url":"https://assets-eu.researchsquare.com/files/rs-6847954/v1/97a282390a67877f804a814d.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Effective Modes of Transboundary Water-Related Cooperation to Turn Future Climate Conflict Risks into Opportunities","fulltext":[{"header":"Introduction","content":"\u003cp\u003eTriggered by the recent deadly Kashmir attack, India\u0026rsquo;s threat to restrict Indus water flows provides a stark, contemporary illustration of how geopolitical conflicts can rapidly escalate over water resources, particularly in regions already facing vulnerabilities amplified by climate change\u003csup\u003e1\u003c/sup\u003e. Yet within this crisis lies an overlooked opportunity: environmental interdependence has historically catalyzed cooperation when institutions employ inclusive, data-driven approaches\u003csup\u003e2-4\u003c/sup\u003e. This urgency aligns with the UN\u0026rsquo;s 2025 Water for Peace initiative and IPCC AR7\u0026rsquo;s call for conflict-sensitive adaptation.\u003c/p\u003e\n\u003cp\u003eWater cooperation refers to collaborative efforts between nations, regions, or communities to manage shared water resources sustainably and equitably. Transboundary water cooperation, in particular, involves collaboration between countries that share rivers, lakes, or aquifers, aiming to balance competing demands and ensure the sustainable use of these resources\u003csup\u003e5,6\u003c/sup\u003e. This process not only mitigates disputes but also strengthens regional ties by emphasizing shared interest in the sustainable management of natural resources. Through environmental peacebuilding, nations can transcend competition over scarce resources and create lasting peace\u003csup\u003e7,8\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eThe urgency for transboundary water cooperation is underscored by a global rise in water-related conflicts, exacerbated by climate change, with over 60% of the world\u0026rsquo;s transboundary river basins lacking cooperative management frameworks\u003csup\u003e9\u003c/sup\u003e. As global water demand is projected to increase by 20-30% by 2050\u003csup\u003e10\u003c/sup\u003e, climate change will intensify water stress by altering precipitation patterns, reducing water availability, and increasing the frequency of extreme weather events, particularly in regions already facing scarcity. An example of this can already be seen with the Nile River Basin, where there has been escalating tensions between Ethiopia, Egypt, and Sudan over the Grand Ethiopian Renaissance Dam (GERD), with climate change exacerbating the situation by reducing river flows and increasing variability in seasonal water availability\u003csup\u003e11\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDespite over 200 water-sharing agreements signed globally since 1948\u003csup\u003e12\u003c/sup\u003e, many of these lack effective implementation mechanisms\u003csup\u003e13\u003c/sup\u003e. This growing threat necessitates a proactive, inclusive approach to transboundary water cooperation that not only addresses immediate disputes but also anticipates future challenges, such as equitable resource allocation and adaptive management strategies in the face of climate change\u003csup\u003e14\u003c/sup\u003e. Regions like the Nile, Indus, and other shared basins highlight the critical need for robust, climate-resilient agreements to prevent conflicts and ensure sustainable water access for all, even when there is a history strong interstate rivalry.\u003c/p\u003e\n\u003cp\u003eDeveloping effective strategies and actionable guidance for water-related cooperation remains a complex task. Water diplomacy and environmental peacebuilding require the integration of diverse fields such as sustainability science, environmental policy, and political science. Each cooperation attempt involves a variety of stakeholders, including governmental and non-governmental actors\u003csup\u003e15,16\u003c/sup\u003e. The complexity of these efforts can make developing an effective, scalable, cohesive approach challenging. Prior studies have mapped water conflicts and cooperation at a \u003cem\u003edescriptive or aggregated\u003c/em\u003e level\u003csup\u003e2,3,17\u003c/sup\u003e. Research is often time-consuming as it involves understanding diverse perspectives and combining interdisciplinary knowledge\u003csup\u003e18\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eTo address these challenges, our study demonstrates how Artificial Intelligence (AI) can facilitate such analysis and illuminate pathways to peace: transforming conflict triggers into trust-building mechanisms. Large language models (LLMs) and retrieval-augmented generation (RAG) have emerged as powerful tools for understanding and analyzing past water-related cooperation events. LLMs can systematically process vast amounts of historical data to identify patterns and classify the factors contributing to successful cooperation. Meanwhile, RAG helps connect theories from multiple disciplines, such as environmental peacebuilding and water diplomacy\u003csup\u003e19,20\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis study introduces the MaCdip (Multi-agent, Cross-disciplinary Integrative qualitative study Process) framework, leveraging large language models (LLMs), retrieval-augmented generation (RAG), and machine learning to analyze over 2,000 global transboundary water cooperation and conflicts cases from 1951 to 2019. It identifies six distinct modes of cooperation: \u003cem\u003eCross-Border Basin Agreements, Collaborative Planning and Adaptation Strategies, Joint Water Allocation Models, Joint Data-Sharing Systems, Transboundary Water Quality Standards, and Coordinated Hydropower Operations.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe study finds that combining \u003cem\u003eCross-Border Basin Agreements, Collaborative Planning and Adaptation Strategies, and Joint Water Allocation Models\u003c/em\u003e are highly effective, with the potential to mitigate around 1.11 (95% CI: 0.32\u0026ndash;1.90) conflicts over five years. Under 2050 climate change scenarios, particularly for countries with projected severe water stress (under 250 mm annual precipitation), a \u003cem\u003eCollaborative and Communicative Planning\u003c/em\u003e integrative strategy, encompassing \u003cem\u003eJoint Data-Sharing Systems, Cross-Border Basin Agreements, Collaborative Planning and Adaptation Strategies\u003c/em\u003e are recommended, projected to reduce conflicts by over 50% in North America, Europe, Eastern Asia, Southeastern Asia, and Oceania.\u003c/p\u003e\n\u003cp\u003eIdentifying Modes of Water-related Cooperation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eFigure 1b\u003c/em\u003e\u003c/strong\u003e briefly shows our analytical approach by using the integration of large language models (LLMs) and interpretable machine learning to analyze over 1,700 transboundary water cooperation events from databases like TFDD and WCC. To ensure comprehensive coverage, the study fills in spatial and temporal gaps in historical cooperation event descriptions using human searching, semi-automated web scraping, and event augmentation through task-specific LLMs while employing a hallucination guardrail to maintain data accuracy.\u003c/p\u003e\n\u003cp\u003eThe study introduces the MaCdip (Multi-agent, Cross-disciplinary Integrative qualitative study Process) framework, in which LLM agents are \u0026ldquo;trained\u0026rdquo; to be equipped with professional tone and relevant knowledge from experts in the three interdisciplinary fields of International Relations, Sustainability Science, and Hydrology. This is realized through a Retrieval Augmented Generation (RAG) approach that provides literature and theoretical background from the three interdisciplinary field perspectives to cluster and label water-related events. All automated coding was validated through a rigorous two-stage process involving expert review and intercoder reliability tests, ensuring consistency between machine-generated and human-coded results. Following LLM automatic coding, two workshops were held between human field scholars to verify the theoretical robustness and clarity of the LLM-generated labels of cooperative modes. Suggested modifications and revisions were incorporated through human expert consensus.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eFigure 1a\u003c/em\u003e\u003c/strong\u003e shows how water-related conflicts tend to cluster in regions with high water stress, such as Central and Southern Asia and the Middle East, where competition over shared resources has historically fueled tensions. The ongoing tensions between India and Pakistan over the Indus Waters Treaty highlight how rising water demand exacerbates political frictions. In Northern Africa and Sub-Saharan Africa, conflicts related to poverty and political tensions are also prevalent, with disputes over the Nile between Egypt, Sudan, and Ethiopia, especially after the construction of the Grand Ethiopian Renaissance Dam (GERD). Additionally, the shrinking waters of Lake Chad have heightened instability in Nigeria, Niger, Chad, and Cameroon, deepening the region\u0026apos;s poverty and conflict.\u003c/p\u003e\n\u003cp\u003eThere has also been evidence of successful water-related cooperation that prevents conflicts for at least five years, with Europe and parts of Sub-Saharan and Western Africa showing positive progress. The Senegal River Basin Organization (OMVS) is a standout example, in which Senegal, Mauritania, and Mali share resources cooperatively, fostering long-term collaboration. However, successful cooperation is lacking in regions like Southern Asia, the Middle East, and the Balkans. The South Asian nations have struggled to develop a robust cooperative framework around the Ganges-Brahmaputra-Meghna Basin, while unresolved water-sharing disputes along the Drin River in the Balkans illustrate the difficulties of managing transboundary water resources in politically complex areas.\u003c/p\u003e\n\u003cp\u003eUsing large language models (LLMs) to process event texts, the events were clustered through K-means, forming six distinct modes of cooperation: \u003cem\u003eCross-Border Basin Agreements,\u0026nbsp;Collaborative Planning and Adaptation Strategies, Joint Water Allocation Models, Joint Data-Sharing Systems, Transboundary Water Quality Standards,\u003c/em\u003e and \u003cem\u003eCoordinated Hydropower Operations.\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThese modes were further refined by training three multidisciplinary agents (assimilating scholars in International Relations, Sustainability Science, and Hydrology) with field-specific literature on water-related cooperation. The Retrieval-Augmented Generation (RAG) method was used to assign theoretically grounded meanings to each cluster, with inputs from the three scholars informing the LLM prompts.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eEach cooperation mode was then matched to specific countries and years by comparing descriptions and examples to event details \u003cstrong\u003e\u003cem\u003e(Figure 2)\u003c/em\u003e\u003c/strong\u003e. Among these, \u003cem\u003eCross-Border Basin Agreements\u0026nbsp;\u003c/em\u003eemerged as the most frequently identified mode. These agreements are formal treaties or frameworks established between countries sharing transboundary water resources, such as rivers or aquifers, to ensure equitable and sustainable management. These agreements often include provisions for dispute resolution, conservation, and flood control. An example of such an agreement can be seen in the Indus Waters Treaty (1960) between India and Pakistan, which has facilitated cooperation despite political tensions. The detailed descriptions and case examples of the modes are recorded in \u003cstrong\u003e\u003cem\u003eSupplementary Materials S1\u003c/em\u003e.\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\u003cp\u003eOther modes also play critical roles in transboundary water cooperation. For example, \u003cem\u003eJoint Data-Sharing Systems\u003c/em\u003e involve collaborative frameworks for sharing hydrological and environmental data, as seen in the Mekong River Commission (MRC). \u003cem\u003eTransboundary Water Quality Standards\u003c/em\u003e focus on maintaining water quality across borders, exemplified by the Great Lakes Water Quality Agreement (GLWQA) between the U.S. and Canada; \u003cem\u003eCoordinated Hydropower Operations\u003c/em\u003e optimize shared water resources for energy production, as demonstrated by the management of the Kariba Dam by the Zambezi River Authority (ZRA); \u0026nbsp;\u003cem\u003eJoint Water Allocation Models\u003c/em\u003e ensure equitable water distribution, such as in the Zambezi Watercourse Commission (ZAMCOM) strategy; and \u003cem\u003eCollaborative Planning and Adaptation Strategies\u003c/em\u003e involve multiple stakeholders to manage shared river basins sustainably and address climate change impacts, as seen in the Lake Victoria Basin Commission (LVBC) Climate Adaptation Program.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eEffectiveness Evaluation of Different Modes of Water-related Cooperation\u003c/p\u003e\n\u003cp\u003eThe second phase of MaCdip focuses on evaluating the effectiveness of these cooperation modes in preventing future water conflicts and recommending future modes of cooperation under climate change scenarios. After coding qualitative insights into one-hot encoded features, the cooperation modes are fed into an interpretable machine learning model, which, considering country-specific socioeconomic and climate variables, predicts the effects of different cooperation mode combinations. An \u0026ldquo;Interpreter Agent\u0026rdquo; translates these machine learning outcomes into recommendations for cooperation modes to mitigate water conflicts in scenarios projected beyond 2050.\u003c/p\u003e\n\u003cp\u003eThe interpretable machine learning results \u003cem\u003e(\u003cstrong\u003eFigure 3\u003c/strong\u003e)\u003c/em\u003e indicate varying levels of effectiveness for the identified water-related cooperation modes. \u003cem\u003eCoordinated Hydropower Operations\u003c/em\u003e show the highest feature importance scores, followed by \u003cem\u003eCross-Border Basin Agreements\u003c/em\u003e, highlighting how these modes are highly correlated with occurrences of conflicts; \u003cem\u003eJoint Data-Sharing Systems\u003c/em\u003e has the lowest correlation. This result is expected as \u003cem\u003eCoordinated Hydropower Operations\u003c/em\u003e and \u003cem\u003eCross-Border Basin Agreements\u003c/em\u003e are usually linked to the causes and outcomes of water-related conflicts. In comparison, \u003cem\u003eJoint Data-Sharing Systems\u003c/em\u003e can be established without imminent resource scarcity or competition.\u003c/p\u003e\n\u003cp\u003eWith regards to the mitigative capacity of each mode, \u003cem\u003eCross-Border Basin Agreements\u0026nbsp;\u003c/em\u003eshows the weakest impacts, as it is expected to reduce 0.78 (95% CI: 0.60, 0.97) conflicts within 5 years after the cooperative action. The other modes have similar impacts, ranging from 0.84 to 0.89, with \u003cem\u003eCoordinated Hydropower Operations\u003c/em\u003e showing the strongest mitigative impacts 0.89 (95% CI: 0.67, 1.11). The effects of conflict mitigation are the most pronounced when multiple modes are combined. A combination of \u003cem\u003eCross-Border Basin Agreements\u003c/em\u003e,\u003cem\u003e\u0026nbsp;Collaborative Planning and Adaptation Strategies\u003c/em\u003e, and \u003cem\u003eJoint Water Allocation Models\u003c/em\u003e is the most frequently used and has a stronger mitigative effect 1.11 (95% CI: 0.32\u0026ndash;1.90) than individually implemented.\u003c/p\u003e\n\u003cp\u003eThe effectiveness of water-related cooperation modes is evaluated based on six criteria: \u003cem\u003ePrescriptive, Preventive, Affordable, Facilitative, Stress-relieving, and Complementary\u003c/em\u003e (\u003cstrong\u003e\u003cem\u003eFigures 4a-f\u003c/em\u003e\u003c/strong\u003e). \u003cem\u003eCross-Border Basin Agreements\u003c/em\u003e are the most prescriptive, which is essential for managing disputes, as evidenced by the 2010 meeting between Jordanian and Syrian ministers focused on fostering cooperation over shared water resources, including discussions of sustainable use, joint projects, and stakeholder participation. For preventive measures, \u003cem\u003eCollaborative Planning and Adaptation Strategies\u003c/em\u003e stand out, with the 2019 EU-funded environmental projects in Romania, led by Hill International, serving as a modern example of effective conflict prevention through adaptive management frameworks and stakeholder engagement. \u003cem\u003eJoint Water Allocation Models\u003c/em\u003e and \u003cem\u003eCross-Border Basin Agreements\u003c/em\u003e are usually affordable and can be implemented in less wealthy nations through international aid, such as the 2006 Nile Basin Initiative (NBI).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eJoint Data-Sharing Systems\u003c/em\u003e have emerged as the most complementary and are often integrated with other strategies, as can be seen in the 2001 joint flood defense agreement for the Tisza region signed by Ukraine, Slovakia, Romania, Yugoslavia, and Hungary, which combined data-sharing with technical assistance and stakeholder participation. In wealthier nations, \u003cem\u003eTransboundary Water Quality Standards\u003c/em\u003e and \u003cem\u003eCollaborative Planning and Adaptation Strategies\u003c/em\u003e are more facilitative, supporting trade and other non-water-related activities between countries, as seen in the 1999 Lake Victoria Initiative, which applied the Baltic Sea model for pollution clean-up and prevention. In high water-stress regions, \u003cem\u003eCoordinated Hydropower Operations\u003c/em\u003e are the most effective at relieving such stress. \u003cstrong\u003e\u003cem\u003eFigure 4g\u003c/em\u003e\u003c/strong\u003e shows the results of inter-coder reliability tests for all modes.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eRecommended Modes for Transboundary Water-related Cooperation Under an Uncertain Climate Future\u003c/p\u003e\n\u003cp\u003eUnder climate change scenarios projected beyond 2050, the recommended water-related cooperation modes for countries at high risk of future conflicts focus on addressing water conflicts in regions with extremely low precipitation\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e(\u003cstrong\u003e\u003cem\u003eFigure 5\u003c/em\u003e\u003c/strong\u003e). Countries with annual precipitation under 1,000 mm are mostly projected to have a high likelihood of conflict, with high-risk countries including Palestine, Yemen, and Nigeria.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFor countries with extremely low precipitation (under 250 mm), a \u003cem\u003eCollaborative and Communicative Planning\u003c/em\u003e integrative strategy, encompassing \u003cem\u003eJoint Data-Sharing Systems\u003c/em\u003e, \u003cem\u003eCross-Border Basin Agreements\u003c/em\u003e, and \u003cem\u003eCollaborative Planning and Adaptation Strategies\u003c/em\u003e, is recommended. Other modes, such as \u003cem\u003eJoint Water Allocation Models\u003c/em\u003e, \u003cem\u003eCoordinated Hydropower Operations\u003c/em\u003e, and \u003cem\u003eTransboundary Water Quality Standards\u003c/em\u003e, are suggested for countries such as Ethiopia and Somalia. In 2015, Ethiopia demonstrated significant diplomatic intention to cooperate with Sudan and Egypt over the contentious Grand Ethiopian Renaissance Dam (GERD) issue, marking a shift from unilateral actions to a more collaborative approach to managing shared water resources. Somalia received assistance from international donors, non-governmental organizations, and many other countries through international water diplomacy and technical assistance amid its severe drought and civil war. The identification of countries with high conflict risks beyond 2050 and the corresponding cooperative actions are recorded in \u003cstrong\u003e\u003cem\u003eSupplementary Materials S4\u003c/em\u003e\u003c/strong\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe projection of the mitigation effects under future climate scenarios from 2050 to 2100 highlights the growing challenge of water-related conflicts exacerbated by climate change. Increased water stress is expected to intensify conflicts in most regions, particularly in South America, central southern Africa, and East Asia, where water scarcity could more than double the number of conflicts. Conversely, East Africa and South Asia may experience fewer conflicts. The \u0026ldquo;time trend effect\u0026rdquo; further reveals that South America and central south Asia will face a significant rise in annual conflicts, while central southern Africa and North Africa are projected to see a decline, suggesting that historical and geopolitical factors influence conflict trajectories.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eClimate action, particularly transitioning from fossil-fueled development (SSP5) to sustainable practices (SSP1), emerge as a highly effective strategy for mitigating water-related conflicts (\u003cstrong\u003e\u003cem\u003eFigure 6\u003c/em\u003e\u003c/strong\u003e). This shift could nearly eliminate conflicts in South America and East Asia and reduce conflicts by over half in Europe, the Middle East, West Africa, and central south Africa. However, the effectiveness of water-related cooperation modes varies significantly across regions. Our identified and recommended cooperative actions can mitigate more than half of conflicts in regions like Eastern and Southeastern Asia, Europe, Eastern Asia, and North America. Despite these successes, the effectiveness of cooperative strategies is limited in less developed regions, such as the Middle East and Africa, where countries often lack the diplomatic influence and resources to implement and leverage such frameworks. This disparity highlights the need for international support, capacity-building, and inclusive, multi-stakeholder engagement to empower these nations.\u0026nbsp;\u003c/p\u003e"},{"header":"Discussion and Implications","content":"\u003cp\u003eThe study highlights the crucial role of transboundary water cooperation in mitigating conflicts, especially under climate change pressure, providing proof-of-evidence of how cooperative water management reduces disputes and fosters regional stability\u003csup\u003e17,21\u003c/sup\u003e. This study builds on these insights, demonstrating that integrated cooperation, through\u0026nbsp;\u003cem\u003eCross-Border Basin Agreements\u003c/em\u003e, \u003cem\u003eCollaborative Planning\u003c/em\u003e, and \u003cem\u003eJoint Water Allocation Models\u003c/em\u003e,\u0026nbsp;is more effective than isolated strategies in reducing conflicts and promoting long-term resilience. From a methodological perspective, our framework uses LLMs in a novel way to analyze complex transboundary conflict data. This underscores its relevance to computer scientists interested in domain-specific applications for peacebuilding and water governance.\u003c/p\u003e\n\u003cp\u003eHowever, effectiveness varies by region. Less affluent, high water-stress regions like the Middle East and Africa face barriers due to resource and political constraints, limiting their ability to implement cooperative frameworks\u003csup\u003e22,23\u003c/sup\u003e. This disparity underscores the need for tailored interventions and international support. While wealthier nations benefit from integrated cooperation, less developed regions struggle, exacerbating vulnerabilities and hindering global sustainability efforts\u003csup\u003e24,25\u003c/sup\u003e. Furthermore, some regions have robust legal frameworks and high levels of trust, while others lack comprehensive treaties or enforcement mechanisms. Varying degrees of political stability and institutional capacity can also lead to very different outcomes when disputes arise over shared water resources.\u003c/p\u003e\n\u003cp\u003eTo address inequities, global institutions and wealthier nations should enhance diplomatic influence and resource access for developing countries. Inclusive multilateral platforms, investment in infrastructure, education, and governance, and regional water diplomacy hubs could support equitable agreements and climate resilience\u003csup\u003e26\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eClimate action can further reduce water conflicts, particularly in South America and Eastern Asia. Aligning water cooperation with climate adaptation ensures resilience to future uncertainties\u003csup\u003e11,17\u003c/sup\u003e. Climate-resilient infrastructure, such as flood control and drought-resistant irrigation, should be included in agreements with international climate funds like the Green Climate Fund supporting these projects\u003csup\u003e27,28\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003ePolicymakers must prioritize integrated cooperation strategies, especially in high-risk and less affluent regions. International organizations like the UN and World Bank could establish dedicated funds for cooperative water projects in areas like the Nile Basin and Lake Chad\u003csup\u003e29,30\u003c/sup\u003e. Strengthening institutional capacity and cross-border collaboration remains essential for sustainable water governance and conflict mitigation in the face of climate challenges. Finally, by leveraging LLMs to synthesize complex water and climate data, our approach can democratize access to diplomatic and climate-resilience insights for low-resource nations, empowering evidence-based planning even where expertise and resources are scarce.\u003c/p\u003e\n\u003cp\u003eIt is also noteworthy that the findings of the model, as well as practical evidence, suggest that choosing the relevant and appropriate cooperation modes and their combinations is important, as adopting more cooperation modes does not linearly reduce conflicts. While some cooperation initiatives can alleviate regional tension, they might also generate local disputes within individual countries. For example, if a national government agrees to certain water allocation terms with a neighboring state, local communities might feel their interests have been overlooked or threatened. In regions facing climate change impacts, a water management strategy that appears to be beneficial in the short term (e.g., building a dam for flood control) may unintentionally harm downstream communities or ecosystems over time. Such maladaptation risks can fuel local tensions and undermine the broader goal of peacebuilding.\u003csup\u003e31\u003c/sup\u003e\u003c/p\u003e"},{"header":"Methods","content":"\u003ch2\u003eStudy data and preprocessing with Large Language Model (LLM)\u003c/h2\u003e\n\u003cp\u003e\u003cstrong\u003eData Collection and Integration:\u0026nbsp;\u003c/strong\u003eThis study employs a comprehensive data collection and integration strategy to analyze over 1,700 transboundary water cooperation events and 400 water conflict events. The primary data sources are described below, while more details about the data sources are recorded in the data availability statement.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e1) Water-Related Cooperation and Conflict Event Database:\u0026nbsp;\u003c/em\u003eTransboundary Freshwater Dispute Database (TFDD) and Water Cooperation and Conflict (WCC) Database are the repositories used to repositories provide historical records of water-related cooperation and conflict events. To enhance data coverage and reliability, an updated cooperation database was constructed by integrating newly gathered data with existing records. Similarly, water-related conflict datasets were merged to create a more comprehensive record of conflicts over an extended time span\u003csup\u003e32\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eThe resulting \u003cem\u003eExtended Cooperation-Conflict Dataset\u003c/em\u003e (1951\u0026ndash;2019) (provided with publicly accessible link in the Data Availability Section) covers all trigger conflicts from the WCC and severe conflicts from the TFDD (BAR rating -3 to -7). The dataset was further enriched by integrating available cooperation data, resulting in a consolidated database spanning 1951\u0026ndash;2019. To facilitate analysis, events were geographically matched with global climate zones, regional precipitation patterns, evapotranspiration rates, elevation data, and socioeconomic indicators (e.g., population density). The resulting spatial data is shown in\u003cstrong\u003e\u0026nbsp;\u003cem\u003eFigure 1a\u003c/em\u003e.\u0026nbsp;\u003c/strong\u003eThe\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003enew database compiled in this study is compared to prior mainstream databases in\u003cstrong\u003e\u0026nbsp;Supplementary Materials 1\u003c/strong\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e2) Demographic, Economic, and Climatic Variables:\u0026nbsp;\u003c/em\u003eKey variables were sourced from internationally recognized databases. World Development Indicators (WDI) provides annual country-level data on population density, GDP per capita (constant 2015 US$), rural population percentage, and exports of goods and services as a percentage of GDP\u003csup\u003e33\u003c/sup\u003e. Precipitation data is extracted from the Water Balance Model (WBM) (Grogan, 2016; Grogan et al., 2022) and aggregated at the HydroBASINS level 6 scale\u003csup\u003e34\u003c/sup\u003e.Water Stress Indicators were measured using Aqueduct 3.0 (for events before 1985) and Aqueduct 4.0 (for events from 1985 onward) (World Resources Institute, 2023)\u003csup\u003e35\u003c/sup\u003e. For regional classifications data were structured according to the United Nations Statistics Division (UNSD) regional definitions, ensuring consistency in basin-level event aggregation. Historic values (previous 5 years) and projected future values (following 5 years) were computed for socioeconomic and precipitation variables. Future conflict events were estimated using predictive models incorporating climatic, economic, and demographic predictors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLLM-Based Event Augmentation and Hallucination Guardrail:\u003c/strong\u003e Given the incomplete nature of historical event records, as many are short excerpt from historic newspaper and other hardcopy sources that are not as easily accessible as more recent online news pieces, task-specific LLMs were employed to generate plausible event descriptions where data gaps exist. These models were fine-tuned using historical event data to ensure contextual accuracy and domain relevance.\u003c/p\u003e\n\u003cp\u003eSince LLMs are prone to generating hallucinated or factually incorrect information, a robust hallucination guardrail was implemented to ensure data reliability\u003csup\u003e36\u003c/sup\u003e. The methodology follows best practices in LLM-generated text evaluation\u003csup\u003e36\u003c/sup\u003e. First, a sample of 200 LLM-generated event descriptions was manually reviewed by domain experts. The assessment found that only 2% (4 events) exhibited signs of hallucination, indicating a high degree of reliability of Initial LLM-generated content. Second, A hallucination classification model was developed using human-verified data. Key features for detection included semantic consistency, factual alignment with known data, and contextual relevance. The model was trained to distinguish high-confidence, plausible events from hallucinated ones using a supervised learning approach.\u003c/p\u003e\n\u003cp\u003eThird, for any event with a hallucination probability exceeding 20% was flagged for manual review. Events classified as high-risk hallucinations were either discarded or corrected using external data validation. Finally, a feedback loop was established to continuously update the hallucination detection model based on new data and expert evaluations. Hallucination probability was again then used to compare generated text with historical records. All LLM operations are executed using ChatGPT-4o on Azure Cloud.\u003c/p\u003e\n\u003ch2\u003eCooperation Mode Identification by LLM with RAG in a MacDip Framework\u003c/h2\u003e\n\u003cp\u003eThis study employs the MaCdip framework (Multi-agent, Cross-disciplinary Integrative Qualitative Study Process) to analyze over 2,000 global cases of transboundary water cooperation and conflcits. The methodology integrates Large Language Models (LLMs), advanced embedding techniques, clustering algorithms, and Retrieval-Augmented Generation (RAG) to classify cooperation modes. The prompts, pseudo-code, parameters, and literature search strategy in our MacDip framework is available in \u003cstrong\u003e\u003cem\u003eSupplementary Materials S2\u003c/em\u003e\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEmbedding Generation and Initial Clustering:\u003c/strong\u003e Event texts were transformed into dense numerical representations using ChatGPT-4o embeddings, selected for their superior performance in dialogue and QA tasks\u003csup\u003e37\u003c/sup\u003e. Key technical configurations include: 1) Model Architecture: Leveraged the transformer-based text-embedding-3-large variant, generating 3,072-dimensional embeddings; 2) Context Handling: Utilized a sliding window of 8,192 tokens to process long texts, with mean pooling for sentence-level aggregation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInitial Text Clustering:\u003c/strong\u003e K-means clustering was applied to group embeddings, chosen for scalability and robustness in high-dimensional spaces. Critical implementation details include: 1) Parameterization: n_clusters determined via the elbow method (k=9 optimal), n_init=10 to mitigate initialization bias, and random_state=42 for reproducibility. 2) Validation Metrics: Assessed cluster quality using Silhouette Score; 3) Cluster Refinement: Merged overlapping clusters using cosine similarity thresholds (\u0026lt;0.85) and re-ran K-means to ensure distinctiveness.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLiterature Database Construction for 3 Scholars with Interdisciplinary Insights:\u0026nbsp;\u003c/strong\u003eThe initial clusters lack theoretical robustness and meanings to adequately identify modes of cooperation from the tasks. Thus, this study constructs a literature database from interdisciplinary insights from International Relations, Sustainability Science, and Hydrology to refine cluster labels, using PRISMA literature review technology. Search strategy for the literature uses query from Web of Science (2010\u0026ndash;2024) with Boolean logic:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTS=(water resource AND (cooperation OR conflict OR governance)) AND PY=(2010\u0026ndash;2024) AND DT=(ARTICLE) \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eInclusion/Exclusion criteria is set for the screened 1,513 articles to identify 73 high-relevance papers, focusing on peacebuilding mechanisms and transboundary case studies. The details of the literature search and PRISMA flow chart are available in \u003cstrong\u003e\u003cem\u003eSupplementary Materials S2\u003c/em\u003e\u003c/strong\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRAG-Driven Label Refinement:\u0026nbsp;\u003c/strong\u003eThe RAG system integrated interdisciplinary insights from the Literature Database built from 3 scholarly perspectives. In the first step, we split texts into 1,000-character chunks with 30-character overlaps using CharacterTextSplitter. Then, we embedded chunks via sentence-transformers/all-mpnet-base-v2 (768d), optimized for semantic similarity and stored vectors in FAISS with IVF indexing for efficient retrieval\u003csup\u003e38\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eThen, we deployed three discipline-specific retrievers, each fine-tuned on domain literature. Each scholar receives equal voting. We prompted ChatGPT 4-o to generate labels (e.g., \u003cem\u003eJoint Water Allocation Models\u003c/em\u003e from raw\u0026nbsp;\u0026ldquo;water allocation\u0026rdquo;) for each piece of collaborative event, with temperature=0 to ensure deterministic, theory-grounded, and theoretically robust outputs.\u003c/p\u003e\n\u003cp\u003eFor each final label, detailed descriptions and real-world examples are provided by LLM with retrieval both relevant information from the event database as well as the literature database. These descriptions clarify the meaning of each label and offer practical illustrations of how countries collaborate in these ways. For instance, the \u003cem\u003eCross-Border Basin Agreements\u003c/em\u003e label includes examples such as the Nile Basin Initiative and the Mekong River Commission.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eValidation and Verification:\u0026nbsp;\u003c/strong\u003eAll automated coding is validated through a rigorous two-stage process: 1) Expert review: Following the LLM automatic coding, a workshop is held among human field scholars (16 scholars) on 11.17.2024 and (5 scholars) 03.11.2025 to verify the results Domain experts review the machine-generated labels to ensure their accuracy, relevance, and theoretical robustness. Some suggestions are made to each label title, its content and examples. The research team then synthesizes these comments and finalizes the identified modes of water-related cooperation. 2) Intercoder Reliability Tests: Consistency between machine-generated and human-coded results is assessed through intercoder reliability tests of 20 human coders, ensuring the labeling process is both reliable and valid. The details of the inter-coder reliability test are recorded\u003cstrong\u003e\u0026nbsp;\u003cem\u003eSupplementary Materials S3\u003c/em\u003e.\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe results demonstrate strong agreement (overall score: 0.712) between human and LLM coding, with five of six modes scoring 0.7\u0026ndash;0.9. The exception, \u003cem\u003eCross-Border Basin Agreements\u003c/em\u003e (0.583), reflects a divergence in interpretation: human coders focused on explicit treaty events, while the LLM included implicit agreements not captured manually.\u003c/p\u003e\n\u003ch2\u003eEvaluating Effectiveness of Water-related Cooperation Mode\u003c/h2\u003e\n\u003cp\u003eThis study employs a multi-criteria evaluation framework\u003csup\u003e39\u003c/sup\u003e powered by interpretable machine learning (ML) and natural language processing (NLP) framework to evaluate the effectiveness of transboundary water cooperation modes in mitigating conflicts. The methodology integrates quantitative ML analysis with qualitative NLP-driven multi-criteria evaluation, ensuring both statistical rigor and contextual depth. The methodology details of mode evaluation and future mitigative effect projection is recorded in \u003cstrong\u003e\u003cem\u003eSupplementary Materials S4\u003c/em\u003e.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInterpretable Machine Learning Framework:\u003c/strong\u003e Random Forest regression with SHAP (SHapley Additive exPlanations) is chosen for its interpretability and robustness in handling complex, non-linear relationships\u003csup\u003e40\u003c/sup\u003e. Bayesian optimization is used to fine-tune hyperparameters such as the number of trees, maximum depth, and minimum samples per leaf. Data from 2011-2019 is used as test data to validate the model. Performance metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-squared are used to evaluate the model. And the MAE metric is utilized to determine the optimal combination of cooperation modes that maximize conflict mitigation effects.\u003c/p\u003e\n\u003cp\u003eSHAP values are used to quantify the contribution of each cooperation mode to conflict mitigation. The SHAP summary plot (Figure 3d) visualizes the impact of cooperation mode on the model\u0026apos;s output, providing a global interpretation of the model. The SHAP values are aggregated to rank the importance of each feature, highlighting which cooperation modes and socioeconomic factors are most influential in mitigating conflicts.\u003c/p\u003e\n\u003cp\u003eMitigative effects are calculated for hypothetical scenarios where all conflict events are provided with one certain mode of collaboration or a highly frequent combination of modes. This allows for a detailed understanding of how specific modes or combinations of modes influence conflict mitigation.\u003c/p\u003e\n\u003cp\u003ePredictors and independent variables are selected as the following. The dependent variable is the count of water-related conflicts within a 5-year period (integer). The independent variables include Event type, annotating whether the event is a conflict or a certain mode of cooperation. Climate variables include annual precipitation and water stress levels, socioeconomic controls include population density, GDP per capita, rural population, and export dependency ratio. The variables are integrated into a country-year panel dataset, ensuring temporal and spatial consistency.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNLP-Driven Multi-Criteria Evaluation:\u003c/strong\u003e The evaluation is based on six dimensions, each scored on a scale of 0.5 to 6.0. Prescriptive: The predictive power (MAE) of the mode in relation to conflict occurrence. This dimension evaluates the correlation between each mode and the occurrence of conflict. Preventive: The capacity of the mode to mitigate conflicts. Affordable: The cost-effectiveness of the mode for low-GDP nations. Facilitative: The alignment of the mode with trade-dependent economies. Stress-Relieving: The impact of the mode on water scarcity. Complementary: The ability of the mode to synergize with other modes.\u003c/p\u003e\n\u003cp\u003eThe evaluation is conducted using inputs from SHAP values and ML metrics for each mode. The NLP model uses GPT-4 via Azure OpenAI API, fine-tuned on labeled events. By integrating event texts, the NLP model provides a qualitative layer to the quantitative ML analysis, offering insights into why certain modes are more effective than others in specific contexts. The following prompt template is used to generate scores and justifications:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e \u0026quot;Score [Mode] on [Criterion] (0.5\u003c/em\u003e\u003cem\u003e\u0026ndash;\u003c/em\u003e\u003cem\u003e6 scale) using:\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e - Machine Learning Output\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e - Case study [Excerpt from Nile Basin Initiative 2006]\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e Justify scoring.\u0026quot;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eFor output, the NLP model generates evaluation metric scores (in 0.5 increments) with expert-style rationales. For example: \u003cem\u003eCross-Border Basin Agreements\u003c/em\u003e scored 5.0 on \u0026ldquo;Prescriptive\u0026rdquo; due to a high MAE and feature importance among all modes.\u003c/p\u003e\n\u003ch2\u003eSSP Future Climate Scenario and Recommendation of Future Cooperation Modes\u003c/h2\u003e\n\u003cp\u003e\u003cstrong\u003eSSP Scenarios: \u003c/strong\u003eTo project future water-related conflict risks under climate change, we adopt the \u003cstrong\u003eS\u003c/strong\u003ehared Socioeconomic Pathways (SSPs) framework, which integrates socio-economic narratives with climate projections\u003csup\u003e41\u003c/sup\u003e. This approach aligns with methodologies from Chen et al.\u003csup\u003e42\u003c/sup\u003e, where SSPs are used to simulate urban land patterns by combining population, GDP, and urbanization rate predictions. The scenarios include: 1)Benchmark Scenario: SSP5 (Fossil-Fueled Development), characterized by high greenhouse gas emissions and minimal climate policy; 2) Climate Action Scenario: SSP1 (Sustainability), emphasizing low-carbon transitions and sustainable resource management; 3) Increased Water Stress Scenario: Future basin-level water stress under SSP1 and SSP5 is derived from Aqueduct 4.0 toaccount for increased water stress under climate change; 4) Time trend effect: Use time series model to account for conflict trend for each country till 2050; 5) Implementing Water-related Cooperation: Estimate the mitigative effects for future conflicts with recommended modes of water-related cooperation implemented.\u003c/p\u003e\n\u003cp\u003eFor other socioeconomic and climate data used in the machine learning model, we find replacement data from different SSP projections. Precipitation and temperature projections for SSP1 and SSP5 are sourced from the World Data Center for Climate (CMIP5 downscaled datasets) and CEDA, ensuring consistency with IPCC AR6 regional extremes. Future basin-level water stress under SSP1 and SSP5 is derived from Aqueduct 4.0 (World Resources Institute), which models hydrological risks under varying socio-economic conditions. Population, GDP, and urbanization trends are extracted from the IIASA SSP Database, aggregated into 32 macro-regions as per SSP regional classifications. Inter-country trade data to estimate future export is obtained from trading economics and World Bank. Since there are huge uncertainties to project to a certain year beyond 2025, we collect future projection data with the closest time to 2050, but acknowledging that the projection could occur in any time within 2050 to 2100.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTime Series Analysis and Conflict Risk Projection\u003c/strong\u003e: Building on methodologies from Ge et al. (2022)\u003csup\u003e43\u003c/sup\u003e, we employ a hybrid machine learning and time series model to project conflict risks. Adopted from Ge et al. (2022)\u003csup\u003e43\u003c/sup\u003e, An ARIMA model integrates historical conflict trends (1951\u0026ndash;2019) with SSP-driven climate and socio-economic projections to simulate annual conflict likelihoods for 2050\u0026ndash;2100. Agglomerative Hierarchical Clustering (AHC) is employed to stratify national risk levels through conflict probability metrics, enabling systematic identification of high-risk regions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRecommendation System for Future Cooperation Modes: \u003c/strong\u003eDrawing from Lyu et al (2023)\u003csup\u003e44\u003c/sup\u003e, we design an AI-assisted decision framework combining NLP, quantile clustering, and reinforcement learning. First, country-specific variables (GDP per capita, baseline water stress) are transformed into textual prompts. For example:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026quot;Country X exhibits high transboundary dependency (85%), low GDP ($1,200/capita), and extreme water stress (score: 4.8). Recommend cooperative strategies aligned with SSP1 sustainability pathways.\u0026quot;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eSecond, Countries are grouped into quantiles for their future projected variables. Clusters reflect geopolitical contexts (e.g., \u0026quot;Low-Income/High Stress\u0026quot;) to tailor recommendations like \u003cem\u003eJoint Data-Sharing Systems\u003c/em\u003e for Somalia or \u003cem\u003eTransboundary Water Standards\u003c/em\u003e for Ethiopia. A feedback loop with iterative prompting, updates model weights based on historical cooperation outcomes. For instance, successful dam negotiations in the Nile Basin (2015\u0026ndash;2023) increase the confidence score for \u003cem\u003eCross-Border Basin Agreements\u003c/em\u003e in similar contexts.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMitigation Effect Evaluation: \u003c/strong\u003eThe coupled machine learning and time series model re-projects conflict counts for 2050\u0026ndash;2100 under SSP5 and SSP1, with recommended cooperation modes activated (binary flag: 1). Conflict counts under SSP5 are normalized to 1 for comparative analysis. Results are aggregated into IPCC AR6 regions. Standard deviations from 20 ensemble machine learning runs generate uncertainty maps, ensuring robustness in policy prescriptions.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eKoubi, V. Climate change and conflict. \u003cem\u003eAnnual Review of Political Science\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 343-360 (2019).\u003c/li\u003e\n\u003cli\u003eBernauer, T. \u0026amp; B\u0026ouml;hmelt, T. International conflict and cooperation over freshwater resources. \u003cem\u003eNature Sustainability\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, 350-356 (2020).\u003c/li\u003e\n\u003cli\u003eTurgul, A.\u003cem\u003e et al.\u003c/em\u003e Reflections on transboundary water conflict and cooperation trends. \u003cem\u003eWater International\u003c/em\u003e \u003cstrong\u003e49\u003c/strong\u003e, 274-288 (2024).\u003c/li\u003e\n\u003cli\u003eWolf, A. T. Shared waters: Conflict and cooperation. \u003cem\u003eAnnu. Rev. Environ. Resour.\u003c/em\u003e \u003cstrong\u003e32\u003c/strong\u003e, 241-269 (2007).\u003c/li\u003e\n\u003cli\u003eDinar, A., Dinar, S., Mckinney, D. C. \u0026amp; Mccaffrey, S. C. \u003cem\u003eBridges over water: Understanding transboundary water conflict, negotiation and cooperation\u003c/em\u003e. Vol. 3 (World Scientific Publishing Company, 2007).\u003c/li\u003e\n\u003cli\u003eWolf, A. T. Transboundary water conflicts and cooperation. \u003cstrong\u003eIn Search of Sustainable Water Management\u003c/strong\u003e, 131-154 (2005).\u003c/li\u003e\n\u003cli\u003eKrampe, F., Hegazi, F. \u0026amp; VanDeveer, S. D. Sustaining peace through better resource governance: Three potential mechanisms for environmental peacebuilding. \u003cem\u003eWorld Development\u003c/em\u003e \u003cstrong\u003e144\u003c/strong\u003e, 105508 (2021).\u003c/li\u003e\n\u003cli\u003eKeskinen, M., H\u0026auml;kkinen, E., Haapala, J. \u0026amp; Sharipova, B. Trust in Transboundary Waters: Identifying Trust-Building in Water Diplomacy Literature. \u003cem\u003eWater Alternatives\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e (2023).\u003c/li\u003e\n\u003cli\u003eUN-Water. Transboundary Waters: Sharing Benefits, Sharing Responsibilities., (2021).\u003c/li\u003e\n\u003cli\u003eIPCC. Climate change 2022: Impacts, adaptation and vulnerability. (2022).\u003c/li\u003e\n\u003cli\u003eBasheer, M.\u003cem\u003e et al.\u003c/em\u003e Cooperative adaptive management of the Nile River with climate and socio-economic uncertainties. \u003cem\u003eNature Climate Change\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 48-57 (2023).\u003c/li\u003e\n\u003cli\u003eKatz, D. Hydro-political hyperbole: Examining incentives for overemphasizing the risks of water wars. \u003cem\u003eGlobal Environmental Politics \u003cstrong\u003e11\u003c/strong\u003e, 12-35 (2011).\u003c/em\u003e\u003c/li\u003e\n\u003cli\u003eWHO/UNICEF. \u003cem\u003eProgress on household drinking water, sanitation and hygiene 2000-2020: five years into the SDGs\u003c/em\u003e. (World Health Organization, 2021).\u003c/li\u003e\n\u003cli\u003ede Bruin, S. P., Schmeier, S., van Beek, R. \u0026amp; Gulpen, M. Projecting conflict risk in transboundary river basins by 2050 following different ambition scenarios. \u003cem\u003eInternational Journal of Water Resources Development\u003c/em\u003e \u003cstrong\u003e40\u003c/strong\u003e, 7-32 (2024).\u003c/li\u003e\n\u003cli\u003eIslam, S. \u0026amp; Susskind, L. Using complexity science and negotiation theory to resolve boundary-crossing water issues. \u003cem\u003eJournal of Hydrology\u003c/em\u003e \u003cstrong\u003e562\u003c/strong\u003e, 589-598 (2018).\u003c/li\u003e\n\u003cli\u003eSommer, U. \u0026amp; Fassbender, F. Environmental Peacebuilding: Moving beyond resolving Violence-Ridden conflicts to sustaining peace. \u003cem\u003eWorld Development\u003c/em\u003e \u003cstrong\u003e178\u003c/strong\u003e, 106555 (2024).\u003c/li\u003e\n\u003cli\u003eDinar, S., Katz, D., De Stefano, L. \u0026amp; Blankespoor, B. Climate change, conflict, and cooperation: Global analysis of the effectiveness of international river treaties in addressing water variability. \u003cem\u003ePolitical geography\u003c/em\u003e \u003cstrong\u003e45\u003c/strong\u003e, 55-66 (2015).\u003c/li\u003e\n\u003cli\u003eMunia, H. A.\u003cem\u003e et al.\u003c/em\u003e Future transboundary water stress and its drivers under climate change: a global study. \u003cem\u003eEarth\u0026apos;s future\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, e2019EF001321 (2020).\u003c/li\u003e\n\u003cli\u003eQiu, L.\u003cem\u003e et al.\u003c/em\u003e Phenomenal yet puzzling: Testing inductive reasoning capabilities of language models with hypothesis refinement. \u003cem\u003earXiv preprint arXiv\u003c/em\u003e (2023).\u003c/li\u003e\n\u003cli\u003eMacAvaney, S. \u0026amp; Soldaini, L. in \u003cem\u003eProceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval.\u003c/em\u003e 2230-2235.\u003c/li\u003e\n\u003cli\u003eIde, T. \u0026amp; Detges, A. International water cooperation and environmental peacemaking. \u003cem\u003eGlobal Environmental Politics\u003c/em\u003e \u003cstrong\u003e18\u003c/strong\u003e, 63-84 (2018).\u003c/li\u003e\n\u003cli\u003eSwain, A. Challenges for water sharing in the Nile basin: changing geo-politics and changing climate. \u003cem\u003eHydrological Sciences Journal\u003c/em\u003e \u003cstrong\u003e56\u003c/strong\u003e, 687-702 (2011).\u003c/li\u003e\n\u003cli\u003eGalvez, V., Rojas, R., Bennison, G., Prats, C. \u0026amp; Claro, E. Collaborate or perish: Water resources management under contentious water use in a semiarid basin.\u003cem\u003e International Journal of River Basin Management\u003c/em\u003e \u003cstrong\u003e18\u003c/strong\u003e, 421-437 (2020).\u003c/li\u003e\n\u003cli\u003eEriksen, S.\u003cem\u003e et al.\u003c/em\u003e Adaptation interventions and their effect on vulnerability in developing countries: Help, hindrance or irrelevance? \u003cem\u003eWorld development\u003c/em\u003e \u003cstrong\u003e141\u003c/strong\u003e, 105383 (2021).\u003c/li\u003e\n\u003cli\u003eNguyen, T. T., Grote, U., Neubacher, F., Do, M. H. \u0026amp; Paudel, G. P. Security risks from climate change and environmental degradation: Implications for sustainable land use transformation in the Global South. \u003cem\u003eCurrent Opinion in Environmental Sustainability\u003c/em\u003e \u003cstrong\u003e63\u003c/strong\u003e, 101322 (2023).\u003c/li\u003e\n\u003cli\u003eKlimes, M., Michel, D., Yaari, E. \u0026amp; Restiani, P. Water diplomacy: The intersect of science, policy and practice. \u003cem\u003eJournal of Hydrology\u003c/em\u003e \u003cstrong\u003e575\u003c/strong\u003e, 1362-1370 (2019).\u003c/li\u003e\n\u003cli\u003eBowman, M. \u0026amp; Minas, S. Resilience through interlinkage: the green climate fund and climate finance governance. \u003cem\u003eClimate policy\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, 342-353 (2019).\u003c/li\u003e\n\u003cli\u003eGlemarec, Y. Financing green and climate resilient infrastructure in ASEAN countries. \u003cem\u003eEnvironmental Progress Sustainable Energy\u003c/em\u003e \u003cstrong\u003e42\u003c/strong\u003e, e14097 (2023).\u003c/li\u003e\n\u003cli\u003eBiswas, A. K. in \u003cem\u003eAsian Perspectives on Water Policy\u003c/em\u003e 1-12 (Routledge, 2013).\u003c/li\u003e\n\u003cli\u003eHirji, R. F. \u0026amp; Duda, A. Integrated management of lakes, reservoirs, and their basins is critical for a climate-resilient planet: an urgent wake-up call from collective amnesia. \u003cem\u003eWater Policy\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 66-87 (2025).\u003c/li\u003e\n\u003cli\u003eKrampe, F., Smith, E. S. \u0026amp; Hamidi, M. D. J. P. G. Security implications of climate development in conflict-affected states: implications of local-level effects of rural hydropower development on farmers in Herat. \u003cem\u003ePolitical Geography\u003c/em\u003e \u003cstrong\u003e90\u003c/strong\u003e, 102454 (2021).\u003c/li\u003e\n\u003cli\u003eK\u0026aring;resdotter, E., Skoog, G., Pan, H. \u0026amp; Kalantari, Z. Water-related conflict and cooperation events worldwide: A new dataset on historical and change trends with potential drivers. \u003cem\u003eScience of the Total Environment\u003c/em\u003e \u003cstrong\u003e868\u003c/strong\u003e, 161555 (2023).\u003c/li\u003e\n\u003cli\u003eSingh, J. Does economic misery stifle human development? empirical evidence from Asian countries. \u003cem\u003eGeoJournal\u003c/em\u003e \u003cstrong\u003e89\u003c/strong\u003e, 1-16 (2024).\u003c/li\u003e\n\u003cli\u003eGrogan, D. S.\u003cem\u003e et al.\u003c/em\u003e Water balance model (WBM) v. 1.0. 0: a scalable gridded global hydrologic model with water-tracking functionality. \u003cem\u003eGeoscientific Model Development\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 7287-7323 (2022).\u003c/li\u003e\n\u003cli\u003eKuzma, S.\u003cem\u003e et al.\u003c/em\u003e \u003cem\u003eAqueduct 4.0: Updated decision-relevant global water risk indicators\u003c/em\u003e. (World Resources Institute Washington, DC, USA, 2023).\u003c/li\u003e\n\u003cli\u003evan Schaik, T. A. \u0026amp; Pugh, B. in \u003cem\u003eProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval.\u003c/em\u003e 2832-2836.\u003c/li\u003e\n\u003cli\u003ePetukhova, A., Matos-Carvalho, J. P. \u0026amp; Fachada, N. Text Clustering with Large Language Model Embeddings. \u003cem\u003earXiv preprint arXiv\u003c/em\u003e, 15112 (2024).\u003c/li\u003e\n\u003cli\u003eArslan, M., Mahdjoubi, L. \u0026amp; Munawar, S. J. E. Driving sustainable energy transitions with a multi-source RAG-LLM system. \u003cem\u003eEnergy \u0026amp; Buildings\u003c/em\u003e \u003cstrong\u003e324\u003c/strong\u003e, 114827 (2024).\u003c/li\u003e\n\u003cli\u003eChambers, J. M.\u003cem\u003e et al.\u003c/em\u003e Six modes of co-production for sustainability. \u003cem\u003eNature Sustainability\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 983-996 (2021).\u003c/li\u003e\n\u003cli\u003eShevchenko, V.\u003cem\u003e et al.\u003c/em\u003e Climate change impact on agricultural land suitability: An interpretable machine learning-based Eurasia case study. \u003cem\u003eIEEE Access\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 15748-15763 (2024).\u003c/li\u003e\n\u003cli\u003eO\u0026rsquo;Neill, B. C.\u003cem\u003e et al.\u003c/em\u003e The roads ahead: Narratives for shared socioeconomic pathways describing world futures in the 21st century. \u003cem\u003eGlobal environmental change\u003c/em\u003e \u003cstrong\u003e42\u003c/strong\u003e, 169-180 (2017).\u003c/li\u003e\n\u003cli\u003eChen, G.\u003cem\u003e et al.\u003c/em\u003e Global projections of future urban land expansion under shared socioeconomic pathways. \u003cem\u003eNature communications\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 537 (2020).\u003c/li\u003e\n\u003cli\u003eGe, Q.\u003cem\u003e et al.\u003c/em\u003e Modelling armed conflict risk under climate change with machine learning and time-series data. \u003cem\u003eNature communications\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 2839 (2022).\u003c/li\u003e\n\u003cli\u003eLyu, H.\u003cem\u003e et al.\u003c/em\u003e Llm-rec: Personalized recommendation via prompting large language models. \u003cem\u003earXiv preprint arXiv\u003c/em\u003e, 15780 (2023).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Water-related Cooperation, Large Language Models, Water-related Conflicts, Water Diplomacy, Climate Change","lastPublishedDoi":"10.21203/rs.3.rs-6847954/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6847954/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eGeopolitical conflicts can rapidly escalate over water resources, particularly in regions already facing vulnerabilities amplified by climate change. At the same time, environmental challenges can be viewed as opportunities for transboundary cooperation, offering a path to mutual understanding and collaboration. This study aims to identify and analyze the effectiveness of water-related cooperation modes in mitigating conflicts, particularly under climate change scenarios, using a framework that leverages large language models (LLMs), retrieval-augmented generation (RAG), and interpretable machine learning. We analyze over 2,000 global transboundary water cooperation and conflicts cases from 1951 to 2019, going beyond the limitations of small-scale qualitative studies and enabling granular evaluation of impacts.\u003c/p\u003e\n\u003cp\u003eThe study identifies six cooperation modes:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eCross-Border Basin Agreements, Collaborative Planning and Adaptation Strategies, Joint Water Allocation Models, Joint Data-Sharing Systems, Transboundary Water Quality Standards, and Coordinated Hydropower Operations.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe evaluation shows that integrating \u003cem\u003eCross-Border Basin Agreements, Collaborative Planning and Adaptation Strategies,\u003c/em\u003e and \u003cem\u003eJoint Water Allocation Models\u003c/em\u003e are highly effective, with the potential to mitigate around 1.11 (95% CI: 0.32–1.90) conflicts over five years. \u003cem\u003eCross-Border Basin Agreements\u003c/em\u003e and \u003cem\u003eTransboundary Water Quality\u003c/em\u003e are recommended for less affluent countries, while \u003cem\u003eJoint Data-Sharing Systems\u003c/em\u003e are most effective for countries under high water stress. Climate change will intensify water-related conflicts in the future, doubling conflicts in South America, Middle Southern Africa, and Eastern Asia by 2050, while climate action could nearly eliminate conflicts in high-risk regions and cut them by half in many other regions. However, the effectiveness of cooperative strategies is limited in less developed regions, underscoring the urgent need for international support and inclusive diplomacy to ensure global water security and stability.\u003c/p\u003e","manuscriptTitle":"Effective Modes of Transboundary Water-Related Cooperation to Turn Future Climate Conflict Risks into Opportunities","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-20 15:40:35","doi":"10.21203/rs.3.rs-6847954/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c9712e55-bf51-4491-aaeb-c004aa497e9b","owner":[],"postedDate":"June 20th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":50239727,"name":"Scientific community and society/Water resources"},{"id":50239728,"name":"Earth and environmental sciences/Environmental social sciences/Climate-change impacts/Governance"}],"tags":[],"updatedAt":"2025-08-26T11:50:52+00:00","versionOfRecord":[],"versionCreatedAt":"2025-06-20 15:40:35","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6847954","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6847954","identity":"rs-6847954","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}