HydroBOT: an integrated toolkit for assessment of hydrology-dependent outcomes

HydroBOT: an integrated toolkit for assessment of hydrology-dependent outcomes | 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 Research Article HydroBOT: an integrated toolkit for assessment of hydrology-dependent outcomes Galen Holt, Georgia K Dwyer, David Robertson, Martin Job, Rebeccal E Lester This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6793584/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 18 Jun, 2025 Read the published version in Environmental Modelling & Software → Version 1 posted You are reading this latest preprint version Abstract Water management rarely focuses only on water; instead, management targets values across many water-dependent responses. To assess past performance and plan future actions, water managers need to understand how changes to hydrology (over which they typically have most control) affect water-dependent values. Models relating values to hydrology can be difficult to integrate into management processes; they are often developed for other uses and target subsets of values. We describe and demonstrate HydroBOT (Hydrology-dependent Basin Outcomes Toolkit), a modeling toolkit co-designed with the primary federal water management agency in the Murray-Darling Basin, Australia. HydroBOT can integrate disparate response models and scale and synthesize those results across space, time, and groups of values. Outputs target water management needs and can be tailored to a range of questions, from local, short-term evaluation to basin-scale climate assessment. HydroBOT provides new capacity to move beyond hydrology to assess outcomes across diverse target values. Ecological Modeling Marine and Freshwater Ecology Environmental Policy 1) Environmental flows 2) Water management 3) Hydrology 4) Integrated management 5) R package 6) Synthesis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Highlights The HydroBOT toolkit provides new management modeling capability for data and model integration, demonstrated for the Murray-Darling Basin. HydroBOT synthesizes and assesses outcomes across disparate target values. Scaling and synthesis of outcomes in space, time, and groups of values unlocks improved interpretation and decision making. Scenario comparison, particularly impacts of climate and management adaptation, are primary uses. HydroBOT emphasizes automation, transparency, and error detection for consistent repeatable analyses. 1. Introduction Water management in large river systems typically targets a wide range of desired values (Holling, 2001 ; Brown, Keath & Wong, 2009 ; Jones et al., 2016 ). For example, in the Murray-Darling Basin (MDB), Australia, the objective of water management is to maintain a healthy working river that supports productive and resilient water-dependent industries, healthy and resilient ecosystems, and communities with access to sufficient and reliable water supplies (MDBA, 2011 ). These values, defined here as any social, economic, environmental or cultural asset or function of significance, importance, worth, or use, are not unique to Australia; water is managed to protect a diverse range of values worldwide (e.g. Kaye-Blake et al., 2014 ; Connor, 2015 ; Ziolkowska & Ziolkowski, 2016 ; Stern & Sheikh, 2019 ). That these values are the target of water management implies that each value depends on water in some way but, in many cases, dependencies are not well defined and so management targets water volumes and flow characteristics under the assumption dependent values will follow (e.g. Arthington et al., 2018 ; Mentzafou et al., 2023 ). Water managers typically have a limited range of potential actions (levers) at their disposal. While the particular levers vary across jurisdictions, those that directly affect hydrology dominate (e.g. flow releases from dams or inundation of wetlands, MDBA, 2011 ; Holt, Macqueen & Lester, 2024 ). When non-hydrologic levers are available, such as water trading rules in the Murray-Darling Basin, their impact is still often assessed by how they alter hydrologic conditions – that is, their impact on the spatio-temporal patterns of flow in the system (e.g. DPIE, 2020 ; DELWP, 2022 ). Hydrologic modeling is often better-integrated with management workflows than models of other values (e.g. King, Tharme & De Villiers, 2000 ; Tharme, 2003 ; Shenton et al., 2012 ). In part, this is due to the availability of large-scale physically-based numerical models that provide robust ability to simulate flows as they arise from natural drivers, such as rainfall, and describe infrastructure impacts including dams and diversions (e.g. Yang et al., 2017 ). Though these models tend to have high precision, describe complex spatio-temporal physical relationships, and produce detailed outputs of state variables across time and space, they target relatively few primary outcome variables, e.g. streamflow, runoff, or snowpack (Welsh et al., 2013 ; e.g. Regan et al., 2019 ; Cosgrove et al., 2024 ). In contrast, models of ecological or economic responses are often less detailed but must model a wide range of outcomes, including any number of species, ecosystem metabolism, or economic output from various sectors (e.g. Robson, 2014 ; Frota et al., 2021 ; Harrison et al., 2023 ). The impact of given management actions (e.g. dam releases) on hydrology is well understood and captured by the hydrograph, allowing accurate targeting of particular aspects of the flow regime by management actions (e.g. Singer, 2007 ; Loire et al., 2021 ). While these proximate models of system state and the impact of management actions are invaluable, they do not provide the necessary information to directly assess the status of the range of flow-dependent values. Assessing values as they respond to flow and management actions requires models of these relationships. Such models exist, though their quality and extent vary widely, ranging from unstated mental models to highly-detailed population dynamics or economic models (Lester, 2019 ; Holt et al., 2024 ). Quantitative computational models are often written in various languages by subject-matter experts, focus on one or a limited suite of target values, and return disparate outputs depending on their particular goals and approaches (Ryder et al., 2010 ). These models are often designed with the aim of studying the dynamics of highly specific variables (e.g. fish population responses to anthropogenic change, Scheuerell et al., 2021 ), not to produce the most useful analysis for management questions or to capture the breadth of target values. Moreover, such targeted models typically cannot individually provide information about larger-scale values; for example, single-species models are insufficient to assess the condition of all fish or the whole ecosystem (e.g. Olden et al., 2014 ; Gawne et al., 2018 ). Integrating distinct quantitative models into a holistic modeling approach is necessary to assess management-relevant outcomes of disparate, typically broad-scale, values under different hydrologic conditions and management actions. This integration gives each response model utility beyond its original purpose and identifies where new work is needed, such as where response models are limited or non-existent (Holt et al., 2024 ). The creation of an integrated toolkit provides the opportunity for robust decision-making (Harrison et al., 2023 ), the ability to prioritize water planning across a range of values, and to identify conditions that achieve disproportionately large (or small) impacts, as well as the explicit consideration of synergies and trade-offs among different values. Exemplifying these issues, the Murray-Darling Basin is centrally important to the economic, cultural, social and environmental wellbeing of Australia. Within the MDB, 70% of water is utilized for agricultural purposes (Hart et al., 2021 ). Agriculture and tourism in the MDB contribute nearly 2% to Australian gross domestic product (Hart et al., 2021 ). The MDB is home to 2.2 million people, including more than 50 Aboriginal Nations, and contains approximately 500,000 mapped water-dependent habitats (Hart et al., 2021 ; Brooks, 2021 ; MDBA, 2023 ). While balancing this range of values is difficult in any highly-utilized basin, the Murray-Darling is unusually dry and variable compared to other systems of similar importance (Bond et al., 2021 ). Contemporary management of the MDB arises from the Commonwealth (Water Act 2007, 2007), which established the Murray-Darling Basin Authority (MDBA) with requirements to develop and implement a Basin Plan governing water resource management (MDBA, 2011 ; Hart et al., 2021 ). The MDBA has obligations to manage water to maintain a “healthy working river” (MDBA, 2011 ). The state of the MDB is required to be assessed annually, with a major review of, and updates to, the Basin Plan made at legislated intervals (e.g. 15 years between development and first review, Hart et al., 2021 ). The MDBA does not itself deliver water; instead states within the basin have primary responsibility for water delivery and integrated catchment management, with the MDBA providing coordination and an overarching policy framework (MDBA, 2019 ; Hart et al., 2021 ). Other stakeholders (e.g. the Commonwealth Environmental Water Holder) have additional responsibilities for water management and coordination in the MDB. The MDBA and other water managers in the Murray-Darling Basin have an increasing focus on assessing system response to climate change and management actions including climate adaptations that may be taken by the MDBA itself or other stakeholders in the MDB (Neave et al., 2015 ). Taken together, the need for collaborative management across jurisdictions to deliver legislated requirements and manage into the future for a healthy working basin require a holistic modeling approach. Such an approach should integrate across values and be adaptable to these various management needs while also simplifying repeated, ongoing use. These needs are not unique to the MDBA; other water management groups, researchers and stakeholders have similar complex and competing needs (Campbell, 2016 ). The MDBA has access to robust hydrologic models describing flows in the system and responses to current management practices and these models are under continual use, development, and improvement (e.g. Yang et al., 2017 ). Despite the need for ongoing assessment and forward planning across a range of values, models of the responses to hydrologic conditions are patchy, only represent some values, and have been developed and used in a more ad hoc approach rather than integrated with each other or the hydrologic models (Olden et al., 2014 ; Gawne et al., 2018 ; Holt et al., 2024 ). In this paper, we describe an integrated modeling ‘toolkit’, hereafter referred to as HydroBOT (Hydrology-dependent Basin Outcomes Toolkit), to integrate these disparate models and assess the response of a wide range of water-dependent values. Decision making for water management in a complex basin such as the Murray-Darling requires multiple lines of evidence and balancing competing values and sources of information. HydroBOT greatly improves the capacity of the relevant manager (here, demonstrated for the MDBA) to assess outcomes across a range of values, provides the structure to adapt and include additional values as additional component models become available, and addresses a number of issues with current assessment practices and modeling approaches. To demonstrate how HydroBOT meets these identified needs, we develop a set of example scenarios capturing some qualitative aspects of one intended use: the assessment and comparison of outcomes under different climate scenarios and different management adaptations to those changes. These scenarios represent changes to the system due to processes over which managers have no control (i.e. a changing climate), and management actions which would typically be targeted interventions. These scenarios are not intended to represent real climate change or proposed management actions; instead we use this demonstration to illustrate the problem space and show how HydroBOT can aid in assessing potential system change and prioritising management actions to mitigate impacts. 2. HydroBOT description HydroBOT provides a flexible architecture with distinct components that can be run separately (in a modular fashion) or together to assess how various scenarios, such as climate change or management options, affect multiple sets of values spanning many scales and disciplines. This modularity enhances the iterative nature of decision making for water planning and policy and allows tuning for a particular set of management questions (Fig. 1 ). Scenarios are typically represented by their effect on the hydrograph, though other inputs could be included if relevant (e.g. time series of variables such as temperature). HydroBOT then synthesizes a wide range of outputs from individual response models into results that directly compare scenarios and are digestible and useful for management decision making (Fig. 1 ). Because the need for assessing outcomes is nearly universal in water management and not limited to a particular scenario or project (Medema, McIntosh & Jeffrey, 2008 ), HydroBOT is designed with strong scenario-comparison capabilities but is agnostic to what those scenarios represent. For a similar reason, the standardized synthesis and outputs are highly flexible, allowing the user to choose and tailor the most relevant outputs for different management needs. One common issue with modeling in general, and particularly integrated models spanning several tools, is that such tools can become a black box where the relationships modeled are opaque. This can yield mistrust by the public and other stakeholders influenced by the model, but also mistrust and misunderstanding by users and developers of the model (Ascough et al., 2008 ). To avoid these issues, HydroBOT has been continuously co-designed with the relevant management agency, the MDBA, with emphasis on public code, reproducibility, self-documentation and production of outputs that describe the model itself in addition to its results (e.g. causal relationships). In practice, modeling is almost never a one-off analysis of a single scenario. Instead, a wide range of scenarios are typically analyzed and compared, and those scenarios may be updated even within a single project (Harrison et al., 2023 ). Across multiple projects, the same modeling tools might be used to address vastly different questions, e.g. assessing response to potential future climate at continental scale vs. historical conditions at a small set of locations (Holt et al., 2024 ). The onward analysis of those scenarios is also rarely static, needing to be updated as new questions are asked or preliminary results are tailored more closely to the questions. As is common with large data analysis workflows or software, repeated small changes often introduce errors especially in complex scripts (Wilson et al., 2014 ). It is not unusual for a single project to include hundreds or thousands of repeated analyses over scenarios or sets of analysis options. HydroBOT provides facilities for automating these processes and documenting the code, critical for efficiency and reproducibility. Most importantly, HydroBOT shifts much of the complex data manipulation to tested software, while limiting user input to a small number of arguments related only to scenario configuration and analysis choices. This approach not only improves the ability to understand specific runs and reproducibility, but also dramatically reduces the incidence of inadvertent errors that arise from making small adjustments to large, complex scripts (Marwick, Boettiger & Mullen, 2018 ). 2.1 Architecture Taken holistically, the HydroBOT architecture comprises three primary processing and control components (Controller, Aggregator, Comparer), three components that integrate models and data from elsewhere (Response models, Spatial data, and Causal networks), and the links between them (Fig. 1 ). The processing flow through HydroBOT is represented by the boxes along the top of Fig. 1 , and in more detail in the Supplement (Figure S1). Input data (hydrographs) are ingested by the Controller, which packages and runs Response Models. Results from the Response Models are then processed by the Aggregator, and the analysis and final outputs are prepared by the Comparer. HydroBOT provides coordination of analyses through this workflow, as well as the algorithms within each component. Each step in HydroBOT is parameterized for a given analysis with arguments to a set of relevant functions, giving the user wide latitude to tailor an analysis to the input data, target values, and question at hand. As each stage of HydroBOT runs, it produces metadata files including the full set of parameters for that stage and all preceding stages. These files can be used as parameter files for subsequent runs to support reproducibility. The HydroBOT architecture emphasizes modularity. Outputs can be saved at each step in the processing flow, allowing users to run HydroBOT as a whole or re-run only needed components to update analysis. For example, if a user wished to change how they aggregate the results of the modules, they could re-run the aggregation step and the subsequent comparison step to match without re-running earlier steps. This ability to adjust intermediate steps allows rapid iteration of results to address a given management question, or adaptation of pre-existing results to new questions. Modularity also allows rapid, iterative development of HydroBOT itself. Any HydroBOT component can be updated, e.g. new aggregation functionality added, without affecting others. Obtaining new results utilising new capacity simply requires parameterising that capacity for a given analysis and re-running HydroBOT from that point forward. Here we describe the specific implementation of each component of HydroBOT (Fig. 1 ), with additional detail in Supplement. 2.1.1 HydroBOT Controller The HydroBOT Controller is the interface between the externally generated input data (i.e. scenarios), the chosen response model, and other external and internal HydroBOT components (SI Figure S1). This component initiates the downstream processing steps according to user-defined settings for a particular run. It includes arguments for locating the input data and the response model(s) to use, along with any necessary parameters for those models. The Controller can also control later components, allowing the full HydroBOT workflow to be run at once. These include defining aggregation steps as discussed in Section 2.1.4 and analysis of the results with the Comparer (Section 2.1.5 ). The Controller determines whether and where outputs are returned at each step. Having control over the full workflow enables large batched runs using parameter files (yaml) to specify the control arguments. This core functionality of the Controller is delivered via simple functions that apply to the input data for each scenario and can be parallelized over scenarios. The Controller can be accessed by the user by using Quarto notebooks to work interactively, R scripts, via the command line, or with yaml parameter files, depending on the use case. 2.1.2 Response models The impacts of climate and management options on social, cultural, environmental, or economic values are estimated based on causal relationships between drivers (e.g. hydrology) and responses (e.g. the state of values). Response models may exist in many different forms, ranging from binary achievement of hydrologic indicators to fully quantitative responses. To be included in HydroBOT, the minimal requirements are that they can be scripted, and return meaningful output values that can be tagged either internally or in post-processing with scenario information, though only the second is a requirement for their data to be used in later steps of HydroBOT processing (see Section 2.1.4 ). These tools are expected to be sourced from existing or in-development models developed by subject-matter experts and, as such, will be written in different languages and will target different outcomes. Thus, the Response models component is depicted as being defined externally to HydroBOT (Fig. 1 ; Figure S1), and new response models require development of lightweight wrappers to integrate them into compatible, modular HydroBOT components. Once integrated, HydroBOT provides a unified interface and ongoing analysis and modeling of the response model results. HydroBOT seeks to avoid changing the response models themselves, but instead wraps them with input and output management to fit into the workflow. Specifically, to incorporate a new model, HydroBOT includes a wrapper function that passes necessary data locations, as well as any necessary arguments from the user to that model. Each response model then has a distinct set of outputs, reflecting the captured responses and the structure of the model. HydroBOT includes cleanup functions that save or return the outputs in formats usable by onward processing in HydroBOT. These formats are quite flexible, requiring only a data frame (or csv file) with output values and columns reflecting the spatial, temporal, and value dimension of those outputs and often a column indicating the scenario. Once these wrappers are in place, HydroBOT can directly run the incorporated response models within its own consistent, unified workflow across disparate response models. By providing a single consistent interface to a range of response models, we avoid the need to manually run models separately and abstract their different interfaces, languages, and idiosyncrasies. Moreover, the design is highly modular, built to allow the integration of new response models with limited additional updates to HydroBOT itself. In some cases, e.g. proprietary software, HydroBOT may not itself run the response models, though this is less ideal. In that situation, the outputs of those models would need to be cleaned by the user, and the HydroBOT workflow could begin at the Aggregator step. At present, HydroBOT expects the user to provide input data in a format readable by the response model. Any data cleaning or formatting necessary to meet the response model requirements should be a preprocessing step. This allows HydroBOT to pass only paths to clean data, rather than the data itself, to the response models. Outputs can be saved to disk or retained in-memory for interactive use, depending on the user’s needs. The outcomes of the response models are then processed by the Aggregator to provide outcomes at scales relevant for management decision making (in time, space or value dimensions; Fig. 2 ). 2.1.3 Causal networks Causal networks describe the topology of dependence among many drivers and outcomes (Martínez, Toro & J. León, 2019 ; Peeters et al., 2022 ). These networks provide useful graphical representation of the relationships between nodes (state variables) and links between them (dependencies) (see Figure S2). Beyond graphical representation, they can also extend the ability to assess a range of values beyond the response models themselves. Each link represents the existence of a relationship, and so defining those relationships mathematically allows causal networks to translate from initial inputs (e.g. rainfall, flows, or management actions) through to all values of interest (nodes). Thus, such networks are ideal for systems with multiple interacting relationships (Martínez et al., 2019 ; Peeters et al., 2022 ). HydroBOT provides a causal network for included response models, where available, to describe how their outputs arise from hydrology and how they relate to various levels of management-relevant outcomes. Like their associated response models, these networks are typically sourced externally to the workflow, and so depicted outside the data flow in Fig. 1 . The causal networks enable: 1) visual representation of the complex inter-relationships between scenario inputs and outcomes across a range of objectives and 2) assessment of outcomes scaled along the value dimension (see Section 2.1.4 ). The former aids transparency, elucidating the intentions and causal relationships behind the response models and is a useful device for communication alongside other final outputs. The latter allows outcomes to be quantified for individual (or sets of) values at different levels of complexity or groupings (e.g. fish breeding, individual fish species, native fish overall, or target fish condition), or at overarching sets of values such as environmental or economic groupings. This quantification provides a powerful assessment tool and the ability to identify synergies and trade-offs across interrelated values. The structure of the causal networks define the (sometimes) complex inter-relationships defined in the response models and subsequent impacts on larger-scale values. Many response models capture only proximate responses, e.g. whether flow requirements were met. In these cases, causal networks describing the hypothesized links to target management values allows HydroBOT to model those management values. The network allows moving from those proximate outcomes to other relevant values by specifying the mathematical form of the links from the response model outputs to other nodes. This is not specific to HydroBOT; any quantitative assessment of those relevant values needs specified relationships from the outcomes of detailed response models to dependent target outcomes. The specifications of the mathematical relationships underlying each link can be highly variable, depending on the state of the best available science or expert opinion. Some may be highly-detailed models, including hydrologic models linking rainfall to flows or dynamical models linking flows to species’ populations. Other links might be specified with simpler linear models, averages, or even leaving the mathematical expression unspecified where relationships are hypothesized to exist but no information is available about the nature of the relationship. The availability and detail of causal networks is variable, but use is increasing for assessing ecological risk from stressors (e.g. Ohe et al., 2014 ; Martínez et al., 2019 ), social responses to hydrology (e.g. Frota et al., 2021 ) and hydrology itself (e.g. Pang et al., 2024 ). In cases where a causal network is not available for a particular location and response model, one can be created even if it is as simple as a conceptual diagram. Assessing the quality of knowledge around each link provides a powerful assessment of knowledge deficiencies and uncertainty in responses. In the absence of a causal network, HydroBOT can still analyse values at the scale returned by the response model. This analysis is likely beneficial; response model outputs are developed to be useful for some purposes and these can be assessed with HydroBOT. Where causal networks are available, even conceptually, they can be used to investigate outcomes at multiple value levels. The models implicit in the causal links can then be updated as more information is available. Indeed, in some cases with well-defined response models that link all the way from hydrology to desired management targets, the causal networks and the response models are the same, as they describe the pathways from driver (hydrology) through to relevant values. In these situations, describing the links in an explicit causal network is not necessary for the functioning of HydroBOT, though it is recommended to provide visualization of the causal relationships encoded in the response model. 2.1.4 HydroBOT Aggregator Response model output is typically very granular in many dimensions because the best response models operate near the scale of the processes being modeled (Levin, 1992 ; Holt & Chesson, 2016 ; Gawne et al., 2018 ). In many cases, those processes (e.g. fish breeding, crop planting) individually occur at small spatial and temporal scales. This is the scale of the process itself, though these processes may be repeated and connected over much larger scales. For example, fish may breed across a basin and migrate among breeding locations but each female breeds in only one location at a given time. Moreover, outcomes from response models are typically at small value scales as well, e.g. capturing portions of the life cycle of a single fish species, rather than an overall outcome for all fish, or representing planting of particular crops, rather than overall agricultural output. The consequence is that there are potentially thousands of different modeled outcomes across time, locations, and values. This plethora of outcomes must then be aggregated to extract meaning at larger scales and condense the information for digestibility in management decision making. The HydroBOT Aggregator aggregates and scales information to resolutions useful for interpretation and planning or that capture larger-scale processes. Depending on the use, the desired resolution(s) may range from local, short-term responses of fine-grained outcomes to large spatial scales over longer time periods for high-level outcomes such as environmental condition (Fig. 2 ). For example, a response model might assess a suite of hydrologic indicators at individual gauges, but outcomes may be desired for a range of ecological groups within larger management units or at the basin scale. Likewise, multiple indicators are typically required to meet objectives for proximate ecological values, such as fish breeding, of which many are required for each larger ecological value (such as native fish diversity). The HydroBOT Aggregator provides a robust, consistent aggregation approach along three dimensions (time, space, and value), while maintaining the ability to define those aggregation steps flexibly to meet the needs of the specific analysis and details of the inputs (Fig. 2 ). The Aggregator allows the user to define columns to aggregate (i.e. the outcomes from the response models), as well as any necessary grouping variables indicating e.g. scenarios, time, space, or value dimensions. As such, it is quite flexible, and can accept any dataframe where that information is specified. No internal changes are required to accept outputs from new response models. This flexibility is by design, enabling incorporation of new outputs from new response models or aggregating outputs from response models that are unable to be fully integrated. Moreover, this modularity means the Aggregator is run on the outputs of the response models, and so adjustments to the aggregation step do not require re-running response models, yielding large computational savings as aggregations are adjusted and tailored during the course of a project. The choice of aggregation steps and aggregation functions at each step are highly flexible. The Aggregator consists of a set of functions that scale along each of the three dimensions while handling dimensional idiosyncracies, such as time arithmetic, spatial awareness, and many-to-many links in the causal network. Overarching functions coordinate these dimensional scaling functions, allowing stepwise aggregation along a sequence of dimensions and scales (Fig. 2 ). These coordination functions maintain awareness of all dimensions; e.g. they ensure scaling in the spatial dimension does not accidentally collapse along the value or temporal dimensions. The structure of the Aggregator functions provides these safeguards and capability for sequential multidimensional scaling, while giving the user control over the sequence of those steps and the scaling functions used at each step. The primary arguments to the aggregation functions are lists of the sequence of aggregation steps and the function(s) to apply at each step. The choice of aggregation sequence and scaling functions will change depending on the question being asked and the nature of the processes being modeled. For example, a project focused on expected conditions might use means at most steps, while one focused on risk might use minima to capture the worst outcomes or essential species requirements. To give the user the ability to tailor these steps to different questions, each step can be along any dimension and multiple functions can be assigned at each step. The sequence of aggregation steps can interleave the dimensions, e.g. aggregate first through time to cover the period of the hydrograph, then along the value dimension to an intermediate level, then aggregate in space, then value again. For example, following Fig. 2 , the user might want to aggregate from hydrologic indicator to life-cycle values, such as fish breeding, to species at each gauge, then aggregate those species outcomes up to catchment to assess performance of each species over a larger area, followed by aggregating to values comprising ecological groups, e.g. native fish, waterbirds, native vegetation, or ecosystem function. HydroBOT provides a standard set of MDB-relevant spatial units for aggregation but can also accept any user-supplied polygons (Fig. 2 ). Aggregation along the value dimension follows the causal network, which is supplied by HydroBOT for included response models, though the user can specify other causal relationships if required. Supplied spatial units and causal networks are discussed in SI Section 9.2 with additional information about providing custom versions in the HydroBOT documentation. Whatever the spatial and causal relationships, HydroBOT provides the capacity to scale along them; it is the responsibility of the user to ensure those links and scaling functions are scientifically sound for the question at hand. The Aggregator allows the user to choose any aggregation function at each step. These aggregation functions should be considered carefully, as they should be appropriate to the processes being scaled and the outputs needed for management decision making. HydroBOT provides a standard set of functions (e.g. mean, max, min, median, geometric mean, harmonic mean, and spatially-weighted mean), but the user can also specify any other aggregation function, including custom-written functions tailored to particular analyses. For example, in some situations the user might want to know the proportion of indicators that pass across some area, or perhaps the average value of an abundance measure. In this case, a mean would be appropriate. In other situations, however, a single failure may be disproportionately important (e.g. ‘no loss’ requirements), or perhaps a single passing value is sufficient for an ecological process to occur (e.g. bird breeding can occur anywhere in a catchment). These might be captured with a minimum and maximum, respectively. Additional discussion of aggregation function choice is provided in the Supplement (Section 9.3). The choice of aggregation at each step has large influence on the outcomes, and so HydroBOT does not prescribe the ‘appropriate’ aggregation, instead giving that control to the user. Where aggregation choices are not clear, or to assess the impact of these choices, HydroBOT can be run with different sets of aggregation sequences, with the outcomes of each compared using the Comparer as if they were scenarios. The Aggregator self-documents into a yaml file, and its output datasets retain the full sequence and functions alongside each value, ensuring that values are always paired with their provenance and meaning. 2.1.5 HydroBOT Comparer The HydroBOT Comparer is designed primarily to make comparisons between scenarios, including assessment and visualization of their differences. This component also provides generalized capacity to produce plots and other outputs such as tables using a consistent approach, even when not directly comparing scenarios. While the primary use of the Comparer is comparing Aggregator output, hydrographs and direct Response model output can also use this functionality. The functions within the Comparer can be divided into two main categories, those for analysis and those for plotting. Although in some instances presenting the absolute outcome values can be useful across scenarios, explicitly calculating comparisons (e.g. the absolute or relative difference between scenarios) provides distinct advantages. Difficulty in accurately simulating a complex system means that comparisons between scenario outcomes can be more useful and accurate because any bias in assumptions applies to all outcomes and the focus moves from the total level to the change between scenarios (Holt et al., 2024 ). The best method for comparing will vary depending on the quantities being compared and the intended use of the comparison, so several common default options (e.g. differences, relative change) have been developed along with the flexibility for the user to define alternatives. The comparison functions provide the ability to choose a baseline level for comparison, which may be one of the scenarios, but also may be a reference dataset or a scalar value. For example, we might want to compare a set of outcomes for climate scenarios to a ‘no change’ climate scenario, to historical observations, or to a mean value. Output values are calculated relative to the defined baseline level using either default functions for the absolute or relative difference, or any other user-supplied function. These functions can be applied to any dataframe, but a major advantage of HydroBOT is including them within the plotting functions. This allows the generation of comparison plots from raw Aggregator outputs without the need for subsequent data calculations by the user and avoids potential errors that arise from repeated or forgotten data transformations in an analysis workflow. The plotting functions in the Comparer provide a wide range of possible plots, with standardization providing consistency across plot types and through the project, ensuring values plotted are robust and interpretable. Different purposes require different sets of outputs; for instance, maps are particularly useful for visualizing geographic patterns, tables and graphs typically provide more precise ability to assess impacts on values, and timeseries plots are useful for visualizing climate trajectories. While potential visualizations and comparisons vary widely depending on the intended use, the Comparer standardizes data cleaning and processing for each plot, as well as aesthetics and plotting approaches. Like the Aggregator, a key advantage of the Comparer is the standardization and dimensional safety over repeated data manipulations and processing. By standardizing data cleaning within the Comparer, we avoid losing information or performing unsupported data manipulations and so ensure the quality and meaning of the outputs. For example, because the plotting function was designed to incorporate temporal, spatial, and value dimensions, it identifies and prevents accidental overplotting or misleading lumping of data across dimensions. 2.2 Implementation HydroBOT is available as an R package, which provides a suite of functions representing the steps in the architecture. A documentation website provides additional worked examples and a fuller exploration of capacity and use-cases. Because translating from hydrologic scenarios to various responses is a general problem in water management, we expect the ways in which HydroBOT is used will be highly variable. Thus, by providing a general structure, users can target particular questions and analyses needed for a given question. Although the initial impetus for creating HydroBOT was to assess climate scenarios, its use in practice can be far more general. Any hydrograph can be assessed, and any set of scenarios represented by hydrographs can be compared, provided a response model exists. As a result of an initial scan of available response models in the Murray-Darling Basin, HydroBOT development proceeded with a single response model (written in Python), as this was the only module that met requirements (see Section 3.2 ), but with an architecture designed to allow modular integration of additional response models as they become available. The scan accentuated a critical feature of this modularity; HydroBOT is designed to incorporate and standardize models written in various languages and with a wide range of input needs and outputs. HydroBOT achieves this by wrapping those other tools so as to make their differences as invisible to the user as possible. In the case of Python modules, HydroBOT uses the reticulate R package (Ushey, Allaire & Tang, 2023 ) in combination with small amounts of internal Python code to call Python modules directly. The Python code internal to HydroBOT performs limited cleaning and translation to prevent passing large objects between languages while ensuring that any idiosyncrasies in module inputs and outputs are handled consistently. Python dependencies (and Python itself) are automatically installed on first use of Controller functions that call Python modules unless they already exist on the user’s system. This approach provides essentially invisible Python for most users, while providing flexibility for the user to provide their own Python environment if desired. Modules in other languages are not yet available, but the key requirement is that they be available in a format that is scriptable. In that case, HydroBOT will provide small setup and cleanup functions as with the Python modules and wrapper functions to call these modules. The modularity of HydroBOT means it can be run stepwise, with the user calling the relevant functions at each step (Figure S1). The outputs of each step can be saved or returned directly to an interactive session or both. Typically, selected outputs would be saved for reproducibility and speed unless the project is small enough to retain a comprehensive set in memory or re-run quickly in a notebook. There are also wrapper functions that allow running the HydroBOT flow from Controller through Aggregator, which are extremely useful for large runs or remote runs on batching services. One particularly useful wrapper provides the ability to run from a yaml parameter file providing function arguments. This function allows the use of a default file and a ‘modified run’ file, making it ideal for holding many parameters constant at a default for a particular set of analyses and only changing a subset on a per-run basis in a smaller file. It also takes command-line or R list arguments, making it a flexible solution to run HydroBOT from the command line, in scripting contexts, or notebooks including Databricks and Quarto (Allaire et al., 2022 ). The metadata saved at each step in the process is a yaml file with parameters that are a superset of those needed to run HydroBOT. Thus, an identical run can be produced by running HydroBOT using an output metadata file as an input parameter file, supporting reproducibility. Additional information is included in these exported yaml files documenting versions of HydroBOT and response models, as well as capturing any available metadata pertaining to the input scenarios. Even if a run is started in the middle, such as to re-run aggregation in a different way, the metadata for that run captures the metadata for the previous steps, ensuring that outputs at every stage are tagged with full provenance information. In practice, HydroBOT functions are primarily run in one of two ways. Interactive investigation of relatively small sets of hydrographs and small numbers of scenarios can be done in notebooks (typically Quarto) or simple R scripts. Larger investigations typically would be run on remote computers as part of batching systems, whether HPC, Azure, AWS, Databricks, or similar, with the outputs at the end of the aggregation (and potentially each step) returned. Through the Aggregator, all operations can proceed in parallel over scenarios, and HydroBOT provides internal capacity for this using the future package (Bengtsson, 2021 ). The Comparer step would usually not be run as part of this larger batching, but considered interactively for two reasons. First, the Comparer necessarily looks across scenarios, and so breaks parallelization. Conducting comparisons interactively is typically not excessively CPU- or memory-intensive provided aggregation is sufficient to yield interpretable figures. Second, the primary goal of the Comparer is to produce usable, interpretable outputs. Arriving at a set of meaningful outputs is an iterative process that rarely will be precisely known a priori , and so working through this step interactively is usually necessary. If, however, HydroBOT is being used for ongoing monitoring of the same analyses (e.g. a ‘dashboard’), then the first iteration could be done interactively, with subsequent uses incorporating those settings written into a script, Shiny app, or similar to auto-generate the same figures. If the outputs of the Aggregator step (and so the inputs to the Comparer) are large enough to require high processing or memory, it is unlikely to be simplified enough to make interpretable figures. Additional aggregation steps to better target desired figures (or subsetting of the data to focus on a particular aspect) should be undertaken. If, after careful consideration, a user does need to make processing or memory-intensive figures, these issues are due to the amount of data being plotted and so will not be unique to HydroBOT. Methods used for any other situation involving plotting large amounts of data with ggplot2 should then be considered, such as remote, high performance, or cloud computing. The first step, though, should be to ask whether additional aggregation is needed. HydroBOT was originally designed for analyses of management questions in the Murray-Darling Basin and so, along with the functions to perform the analyses, it also provides a standard set of spatial data, including the boundary of the MDB itself, river lines, gauge locations (at which hydrographs may be available), and various management units (see SI Section 9.2). All geographic information in HydroBOT uses spatial dataframes using the sf package implementing simple feature geometries in R (Pebesma & Bivand, 2023 ). Thus, any user-supplied sf objects can be used for spatial aggregations. Some example hydrograph data are also included for testing and demonstration. HydroBOT provides a clean causal network for included modules, described in more detail in Section 2.1.3 and SI Section 9.2. Analogous elements can be constructed for other spatial units or responses to extend functionality within or outside the Murray-Darling Basin. Supplying new causal networks only requires providing dataframes or lists of dataframes with the needed levels and their relationships. All plotting functions in the Comparer are built on the on the ggplot2 R package (Wickham, 2016 ). While the argument structure is modified to provide the necessary dimensional safety, processing consistency, and simplification of some common complex data processing, much of the ggplot functionality is available. Most importantly, the plotting functions produce ggplot objects that can be further modified by the user. This approach gives the user great flexibility in tailoring plots to the visualization needs of a particular project. 2.3 Co-design and scoping for management relevance To ensure HydroBOT meets management needs and is trusted by management users, a team was created at project inception consisting of primary HydroBOT developers, ecologists, hydrologic modelers, and managers. Collaboration among this team ensured decisions about HydroBOT construction reflected the needs of the end-users. As implementation proceeded, the collaboration provided a mechanism by which goals could be adjusted or the implications of decisions discussed and understood. By having this window into HydroBOT development, the managers obtained a much more granular view of how it operates, along with its capabilities and limitations. This view into HydroBOT development and reasoning was crucial to avoid becoming a ‘black box’, and enhances trust, interpretability and usability of the HydroBOT outputs. Key to the collaborative development was identification of the response models to include. These models needed to provide information about values relevant to water management decisions. Early in the development process, a wide scan was taken to identify candidate response models. This scan considered models being used or developed within the MDBA itself, other government agencies (both federal and state), and externally. A number of response models were found to be available, but many were outdated or highly manual, and their use was patchy both within and across management agencies (Holt et al., 2022 ). Ultimately, only one was suitably modern and well-developed to include in HydroBOT, the Environmental Water Requirements (EWR) Tool (MDBA, 2024 ). Other in-development modules were identified as candidates to include during the development of HydroBOT, including economic and social models. Further, this scan identified values that are mandated management targets for which there were no available or in-development response models, highlighting areas needing future work. 2.4 Advantages and tradeoffs Integrating new modules and shifting workflows into a new package always comes at a cost, and using HydroBOT is no exception. Repeatedly doing the same operations suggests that those operations should be done in a software package, not a script (Wickham & Bryan, 2025 ). This is particularly true if many of those operations are checks or data manipulations to meet certain conditions, e.g. to handle multi-dimensional aggregation or compare scenarios in multiple ways (Wilson et al., 2014 ), as they are here. Thus, if a response model already produces outputs that are essentially finished, or if an analysis is a one-off that is unlikely to be repeated or adjusted, HydroBOT may not be needed. However, in our experience, if a response model produces outputs that provide useful management information, the running and onward processing are likely to be desired across a range of projects, as they have been for the EWR tool discussed here. It is not unusual for a single project to require hundreds to thousands of runs, particularly when small tweaks and changes to the analysis are considered. These adjustments are a frequent source of error, particularly when multi-dimensional aggregation is involved, with quality control rapidly becoming unwieldy or impossible (Wilson et al., 2014 ). Indeed, it is just these issues, compounded across different sets of scenarios and different management questions, that led to creation of HydroBOT as an end-to-end package. Allowing HydroBOT to handle that processing in tested, documented functions while keeping the analysis script short and clear provides advantages for readability, time, consistency, and correctness. Identical processing of a set of scenarios, as well as internal parallelization, provides similar advantages in terms of consistency and speed. Integrating new response models into HydroBOT requires wrapper functions that send relevant arguments to the response model, and potentially (as is the case with the EWR tool), clean up the output of those models for onward processing by the Aggregator. These wrappers should typically be lightweight, performing only simple data formatting. However, it is also possible to forego complete integration and use the Aggregator and Comparer on the outputs of such response models. This approach would lose some advantages (e.g. scripting, automated metadata, parallelization) but can be useful if the input integration is difficult, as might be the case with proprietary models or those with only GUI (non-scriptable) interfaces. The outputs would still need to be cleaned into an appropriate format for the Aggregator, though this process is rarely more onerous than would need to happen for analysis outside HydroBOT. The Aggregator simply needs dataframes with columns of output values, any required grouping variables (e.g. scenarios), and those that link the values to the relevant spatial, temporal, or value dimension. This same information would be needed by most analyses. Similarly, while a new response model is likely to need its own causal network this would be needed for any analysis that aggregated along the value dimension. Spatial units are likely to be shared across models unless HydroBOT is being used in wholly-different locations. If they are needed, that would again be true of any spatial analyses with or without HydroBOT. Both causal networks and spatial data can be passed as arguments rather than needing to be incorporated into HydroBOT itself. HydroBOT is not the answer for all projects and there are certainly tradeoffs. Our belief and experience is that the marginal cost of its implementation, combined with the advantages HydroBOT brings in terms of safety, reproducibility, and speed, make those tradeoffs fall in favor of HydroBOT quite quickly, when compared with conducting the same analyses outside HydroBOT (e.g. with a set of bespoke scripts). This investment is particularly worthwhile for informative response models with clear management importance such as the EWR tool demonstrated here. Once HydroBOT is set up, using it for new scenarios and new questions is straightforward and robust, enabling rapid exploration of new avenues of analysis not otherwise possible. 3. HydroBOT demonstration Here, we present example results produced using HydroBOT, a set of demonstration hydrological scenarios, and the EWR tool for the response model (see Section 3.2 for a description of the EWR module and its associated causal network). These results demonstrate the flexibility and robustness of HydroBOT, which give it the capability to produce rigorous but interpretable and management-relevant results in a reproducible way. The scenarios used here are developed to illustrate HydroBOT functionality with perturbed hydrographs that do not represent real flow scenarios. The plots shown here demonstrate several of the plot types designed to illustrate some potential visualizations relevant to these scenarios. Many more visualizations are possible due to the flexibility of the plotting functions, giving users the ability to generate plots relevant to different projects and questions. 3.1 Demonstration scenarios Our input data consists of hypothetical flow timeseries generated from historical hydrographs at 43 gauges. These gauges fall within the Lachlan, Macquarie–Castlereagh, and Namoi Sustainable Diversion Limit (SDL) units of the Murray-Darling Basin, Australia. These SDL units have detailed Long Term Watering Plans, and so well-specified environmental water requirements and causal networks (described in Section 3.2 ). We develop a set of simple scenarios that capture two sorts of changes that may be commonly represented in management analyses. First, we consider scaled flow throughout the period of the hydrograph, representing overall increases or decreases in flow as might occur from changes in large-scale climate patterns. Second, we consider short-duration additions to flow, representing periodic pulses of change as might happen from targeted interventions. Each scenario characterizes all water in the system including natural inflows, extraction, and release of environmental water, yielding a complete hydrograph (SI Section 9.4 shows a selected subset). These do not reflect realistic future scenarios but provide an avenue to test and illustrate the capabilities of HydroBOT. To scale flow, we apply nine flow multipliers, ranging from 0.5 to 2.0, to the historical hydrographs (Table 1). We refer to these as ‘climate’ scenarios, reflecting a common representation where entire hydrographs might shift to represent climate change. Though these scenarios are not derived from climate models and do not represent hypothesized climate change, they do fall within the range of predictions from GCMs (Devanand et al., 2025 ). To achieve pulsed change within each of the ‘climate’ scenarios, four flow additions were applied including 1) no addition, 2) addition of 250 ML/d, 3) addition of 6500 ML/d, and 4) addition of 12000 ML/d (Table 1). These additional flows were added throughout the period of September to December to target the characteristics of specific environmental water requirements, as might be done by water managers. For example, there are 202 EWRs in the three catchments used here with flow thresholds above 250 ML/d starting in September, which the smallest addition would ensure are met. We refer to these scenarios as ‘climate adaptations’ because management options are often available in the form of altering water availability for short time periods through mechanisms like water releases, though the options here do not represent proposed actions. These scenarios should not be interpreted as potential climate impacts or adaptations, but instead as different ways flows might change (multiplicative or additive) and different magnitudes of change. The ‘E1’ scenario (Table 1) is the observed historical data with no climate change or adaptation options applied, and so is a natural baseline for comparisons. Table 1: Demonstration scenarios are a factorial combination of ‘climate’ (scaled flow) and ‘adaptation’ (pulsed additions). ‘Climate’ scenarios included in this demonstration were produced by applying a flow multiplier to historical flows. ‘Adaptation’ options were applied to each climate scenario with additional flows added throughout the period of September to December. For ease of display in some figures, we provide alpha codes for the ‘climate’ changes and numeric codes for the ‘adaptations’. In most plots, we present the ‘A’ (0.5 x historical flow), ‘E’ (1.0 x historical flow and ‘I’ (2.0 x historical flow) scenarios, with adaptions 1 (no flow addition), 2 (addition of 250 ML/d) and 3 (addition of 6500 ML/d), with ‘adaptation’ scenario 4 and the intermediate ‘climate’ multipliers only appearing in heatmaps and smoothed fits. Climate code Flow multiplier Adaptation code Flow addition (ML/d) A 0.50 1 0 B 0.67 2 250 C 0.80 3 6,500 D 0.91 4 12,000 E 1.00 F 1.10 G 1.20 H 1.50 I 2.00 3.2 Environmental Water Requirements (EWR) Tool The EWR tool, forming the core of this demonstration, is a response model for hydrologic indicators of ecological outcomes. The EWR tool is written in Python and is used internally by the MDBA, the states of New South Wales and Queensland for water planning, and other interested stakeholders (Sheldon et al., 2024 ; MDBA, 2024 ). It models the response of hydrologic indicators established in Long Term Watering Plans (LTWPs) and Environmental Water Management Plans (e.g. DEW, 2020 ; DPIE, 2020 ; DRDMW, 2022 ; DELWP, 2022 ). The EWR tool holds databases of the environmental water requirements (EWRs; the indicators), characterized by the discharge, frequency, timing and duration of flows (SI Table S3). These indicators were developed based on hypothesized relationships to the ecological objectives of the MDB, which protect or enhance environmental assets and ecosystem functions that are valued based on ecological significance (DPIE, 2020 ; Sheldon et al., 2024 ). The EWR tool assesses hydrologic indicators, determining whether spatially explicit flow timeseries meet each EWR (indicator) at each gauge (illustrated in Fig. 3 ). The precise definitions for each EWR differ at each gauge, due to the unique hydrology and channel morphology. For example, a small fresh for 10 days is required every year between October and April to meet indicator EWR SF1 (small fresh 1), but the discharge defined as a ‘small fresh’ differs between gauges. Moreover, the expected ecological outcome for achieving SF1 might vary between gauges or catchments due to different impacts of small freshes on local species or ecological processes, as defined in the LTWPs. The EWR tool returns binary responses of hydrologic indicators. However, the LTWPs describe not only the indicator requirements assessed by the EWR tool, but also how those hydrologic indicators are expected to influence both proximate and larger-scale ecological outcomes according to the best available information from water managers, ecologists, scientific publications, and analysis of gauged and modeled flows (e.g. DPIE, 2020 ; Lobegeiger & Prior, 2022 ). We extracted these expected links from tables within the LTWPs for each planning unit, producing causal networks defining the expected relationships between the EWRs and several higher levels of ecological values. These clean causal networks are provided in HydroBOT (all links for a single gauge are shown in SI Figure S2). While extracting these causal networks was a significant undertaking, it may not be for other response models that capture ecological values or if the hypothesized links are developed from the outset to be conceptualized causally and are machine-readable. Causal networks link the hydrologic indicators produced by the EWR tool to the values for which the system is being managed. Thus, these networks greatly expand the utility of the EWR tool by allowing modeling of a range of ecological values dependent on those indicators. These values are defined at several levels, from components of the life cycle of single species to broad ecological groupings to whole-community outcomes at 10- or 20-year target dates (see SI Section 9.1 and Figure S2). This increases the transparency in the causal relationships that underpin HydroBOT results and builds understanding and trust in the outcomes. For example, the SF1 indicator described above might contribute to multiple ecological objectives pertaining to native fish (ecological objectives NF1-9; e.g. NF1 = No loss of native fish species), native vegetation (NV1), and ecosystem functions (EF1-5). These ecological objectives are defined to support the completion of all elements of a life cycle of an organism or group of organisms (taxonomic or spatial) to achieve a “defined goal for a state, condition or characteristic of an ecological asset or function” (DPIE, 2020 glossary). Outcomes for ecological objectives are then linked in the causal network to five large-scale ecological groups (native fish, native vegetation, waterbirds, other species, and ecosystem functions, MDBA, 2019 ) and are associated with long-term targets (5-, 10-, and 20-year). This structure not only provides a visual definition of the links in the LTWP, it also enables assessment of outcomes that are directly equivalent to the LTWP management strategies. 3.3 Demonstration aggregations The aggregation step is critically important for outputs and should be tailored to the values being aggregated and the questions being asked, as described in Section 2.1.4 . The examples used here are designed for a simple demonstration; many other aggregations are possible, with the capacity explored in-depth on the documentation website. For any analysis, aggregation choices should be considered carefully (see SI Section 9.3). Since our purpose here is a simple demonstration, we use arithmetic means for nearly all aggregations. The exception is the first value dimension step (following temporal aggregation) due to the structure of the EWRs. In the EWR tool, the returned outcomes (and so finest scale on value dimension) gives achievement of multiple versions of EWR indicators, typically representing different leniency in flow conditions required to pass. Here, we consider that if any of the versions of an EWR is achieved, that EWR is achieved. We thus use a maximum for that aggregation step; first we obtain a time-average of each EWR version, and then we take the maximum of each of those versions (typically 1–4 values) to obtain the most-lenient outcome for that EWR. Specifying a different aggregation sequence would yield a different result and have different meaning (e.g. a minimum would require the most-strict version to pass, or we might skip this step entirely and treat the versions as fundamentally distinct indicators). Though we use means for other steps, their interpretation changes depending on the level of aggregation. For simplicity, we assume the value of each outcome represents some ‘condition’, and the ‘condition’ at a particular level can be assessed as the mean of the conditions contributing to it. For example, the condition of detailed ecological values such as fish spawning or bird fledging might simply be the proportion of EWRs that contribute to each value that are achieved. Then, the condition of ecological groups, e.g. native fish, might be the mean condition over all the NF1…n values, despite those depending on different numbers and sets of EWRs. This captures the idea that native fish condition improves when the life-cycle components captured by constituent lower-level values are met, whether it takes 1 or 10 EWRs. 3.4 Demonstration outputs Here, we provide example illustrations of HydroBOT outputs for the demonstration scenarios and aggregations described above. Just as the aggregation steps demonstrate a small part of the capacity of HydroBOT, tailored for this example, so too do these figures demonstrate example capacity for a small subset of the potential outputs. Many more figures might be generated to address other questions or needs, with further detail available on the documentation website for HydroBOT. Moreover, because the outputs of the plotting functions are ggplot objects, additional bespoke tailoring is straightforward using the most common plotting package in R. 3.4.1 Spatial relationships Comparison of the direct outputs from the response model (in this example, pass/fail of EWRs) without any aggregation along the value dimension gives an overview of ecologically-relevant hydrologic outcomes at different locations (e.g. gauges) or areas (e.g. SDL units) to different scenarios (Fig. 4 ; noting that these scenarios do not represent real climate simulations or proposed adaptations). Three focal ‘climate’ scenarios (A, E and I; 0.5x, no change, and 2x the historical base level, respectively) and three adaptation options (1, 2 and 3; additions of 0, 250 and 6500 ML/d, respectively) affect each SDL unit differently, with each catchment having some scenarios in which they have higher EWR achievement than the others. In all situations, the ‘climate’ (A, E, I) scenarios have less of an impact than the ‘adaptation’ (1, 2, 3) scenarios, though neither is reflective of expected change in these dimensions. The point here is simply that different types of changes to the hydrograph can affect locations differently. These outcomes are not linked to ecological values here (e.g. ecosystem functions or environmental assets) and so each hydrologic indicator (EWR) at each gauge is represented with equal importance, whether it is required for all ecological values or just one. 3.4.2 High-level scenario comparisons We use the causal networks to link response model outputs to ecological outcomes and investigate how changes in flow affect ecological values at several levels of the ‘value’ dimension. Thus, we can compare how native fish, native vegetation, waterbirds, other species, and ecosystem function targets are likely to respond to our hypothetical set of multiplicative and additive changes to the hydrograph (‘climate’ and ‘adaptation’ scenarios; Fig. 5 ). Indeed, one of the primary purposes of HydroBOT is to provide this assessment of how high-level values respond to different scenarios. Such a summary collapses thousands of datapoints (EWR outcomes or proximate ecological responses at every gauge) across many scenarios into digestible yet scientifically-robust visualizations. At a glance, we can see in Fig. 5 panel a that Ecosystem function outperforms Native vegetation in the halving (A) and no-change (E; historical hydrograph) scenarios, but is less sensitive to changes between scenarios. Halving water (moving from E to A) reduces Native fish outcomes more than doubling water (E to I) improves them, while the opposite is true for Waterbirds. The choice of aggregation sequence can be interrogated with the Comparer to better understand their implications. For example, aggregating the ‘versions’ of the EWRs using the minimum (i.e. the strictest requirement for a given EWR) reduces the condition of all outcomes (Fig. 5 panel b) compared to the case with the maximum in panel a. Using the median for aggregations at every step following an initial time average (panel c) yields much lower outcomes for the historical (E) and halving (A) climate scenarios, but idiosyncratic patterns in the doubling (I) scenario, with Waterbirds and Native Vegetation much lower than when using the means, while Other species is higher. These are relatively crude changes for the purpose of illustration; more complex aggregation functions or different functions used at different stages are likely appropriate in many situations. These choices should reflect the underlying biology (e.g. how do fish outcomes arise from the different set of flow conditions) and analysis questions (e.g. risk vs. resilience) and the consequences of these choices can be investigated as illustrated in Fig. 5 or with other Comparer visualizations (see SI Section 9.3). 3.4.3 Values at multiple levels of detail High-level outcomes are essential, but managers need to drill down to understand how they arise, including comparing lower-level values, spatial areas, and the scenarios themselves. While there are many ways to display this information (See SI Section 9.6.1 for spatial analysis of each step), Fig. 6 consolidates a large amount of information across scenarios, planning units, and multiple value scales. Each set of colors (and row of panels) represents outcomes at the scale of large ecological value groups. Within each bar, the different shades represent single smaller-scale ecological values (e.g. WB1; Maintain the number and type of waterbird species) contributing to the larger value group, with the heights of these shades being the proportion of the constituent hydrologic requirements that are met (EWRs). Because the overall bar heights are a sum of these proportions, we provide a dashed line as reference that shows the situation where all relevant EWRs are met. This reference is necessary because the number of EWRs contributing to each small-scale ecological value varies, as does the number of small-scale ecological values contributing to the larger ecological value groups and spatial units. Dividing total bar heights by this line would thus give the proportional ‘condition’ score for the larger groupings. Figure 6 provides a useful overview of the impact of different scenarios on large ecological value groups across space, while also retaining the ability to assess more proximate constituent ecological values. Each shade represents a unique smaller-scale ecological value, and so this view shows how those smaller values contribute to the overall condition at the larger scale without undue complexity. If an analysis needed to examine these contributions more explicitly, a different figure might be used to accentuate the lower-level environmental objectives more clearly (e.g. Figure S14, Figure S15). Though the scenarios here are only demonstrations, they illustrate the sorts of comparisons that are possible. For example, WB1 (Maintain the number and type of waterbird species) and WB2 (Increase total waterbird abundance across all functional groups) are only achieved to any appreciable extent in the Macquarie–Castlereagh in the high water addition (3) scenarios, while they are met more generally in the Namoi, and show an interaction with the ‘climate’ scenarios in the Lachlan (met in ‘adaptation’ scenarios 1 and 2 only in the doubled water ‘climate’ scenario). At the larger ecological value group scale, we can see that the effect of halving water (moving from the baseline E1 scenario to the A1 scenario) has a disproportionately greater impact on Native fish than doubling water (the I1 scenario) in the Macquarie–Castlereagh, while the Lachlan and Namoi show slightly larger impacts of doubling than halving. 3.4.4 Quantitative scenario descriptions Scenarios can often be described quantitatively, most obviously here in the flow multipliers used for ‘climate’ and flow additions used for ‘adaptation’, though these can also be combined to yield total changes in daily flow volumes. This quantification makes additional visualizations possible, as well as quantitative analyses to explore the impact changing water has on changing ecological values. The next three figures provide examples of different ways this might be done. Calculating comparisons from a baseline clarifies disproportionate impacts and thresholds where management actions could be most effective. Here, we represent both our ‘climate’ (multiplicative) and ‘adaptation’ (additive) scenarios on a single axis by quantifying the difference in mean flow from historical hydrograph (E1) using the baselining functionality provided by HydroBOT (see Section 9.5). Plotting the proportion of EWRs achieved for each of the Native fish ecological values shows how that EWR achievement is related to changes in the overall levels of flow. One striking feature of Fig. 7 is that the lines do not smoothly increase; EWR achievement is not simply a direct relationship to discharge. Instead, the timing matters. There are particularly steep slopes between the ‘1’ (no-addition baseline) and ‘2’ (+ 250 ML/d) ‘adaptations’. The application of water at those particular times is thus disproportionately impactful in these example scenarios. Indeed, the ‘A2’ scenarios yield better outcomes than the ‘E1’ for nearly all objectives even though they have less water. The same is true for ‘E2’ compared to ‘I1’ in the Macquarie–Castlereagh and Namoi. Though these scenarios are not developed from real adaptation proposals, they identify real impacts of the timing of water additions. Such analyses can identify highly efficacious points to add water to the system, provided they represent real responses rather than model artefacts. With a large number of scenarios, we can characterize how values respond along axes relevant to how those scenarios change. Using smoothed fits, we show trends in the performance of ecological values (and groups thereof) in response to changes in flow from both ‘climate’ and ‘adaptation’ scenarios (Fig. 8 , SI Figure S12). We use baselining (see Section 9.5) to show the relative change on both the x (flow) and y (response) axes. Considering these relative changes on a log-log axis then illuminates whether doubling flow has the same impact as halving, and whether doubling (or halving) flow doubles (or halves) response. Thus, any slope other than 1:1 on the log-log axes represents a disproportionate impact of increases versus decreases in flow, and whether ecological condition shifts proportionately with flow. The responses seen in Fig. 8 vary between SDL unit (columns) and ecological values (rows), indicating different sensitivity to both ‘climate’ and ‘adaptation’ flow conditions across space and among ecological groupings. For example, Ecosystem function responds disproportionately positively on a multiplicative scale to flow across the range in the Lachlan (and so doubling flow more than doubles ecological condition). Ecosystem function responds approximately proportionately to flow changes in the Namoi. Ecosystem function and Native fish respond more strongly to decreased than increased flows in the Macquarie–Castlereagh, while all other groups are both more sensitive to increases. By using smoothed fits, we can identify thresholds or areas of flow change that are disproportionately more important (i.e. where small changes in flow can yield large shifts in condition). Though the scenarios here do not represent real changes, these results do show disproportionate change across a range of flows; whether that is also true for realistic ranges would require a similar analysis of meaningful scenarios. The response to ‘adaptation’ options (simulated seasonal water additions; difference between the lines in Fig. 8 panel b) tends to be much stronger than the response to ‘climate’ (multiplicative changes to the whole hydrograph; slopes), even though the shift in total water may be lower (particularly for the baseline ‘1’ and + 250 ML/d ‘2’ adaptations; see x-axis of Fig. 7 ). Moreover, the adaptation options are not equally sensitive to the climate scenarios; adaptation options ‘2’ and ‘3’ show less change in response to climate than ‘1’ (no adaptation) (Figure S12). Thus, these ‘adaptations’ are increasing resilience to holistic ‘climate’ flow changes. Although these scenarios do not represent true adaptation options or climate scenarios, this shows that such changes to resilience are possible with targeted interventions, and HydroBOT provides the capacity to investigate them. Scenarios will often be defined along more than one axis. In this demonstration, we define a flow multiplication axis as a proxy for ‘climate’ shifts, and a flow addition axis as a proxy for ‘adaptations’. Similarly, values often respond to several different aspects of the flow regime, for example the mean flow and the flow variance or the return interval of floods or low-flow periods. Heatmaps or contour surfaces provide a powerful tool for visualizing these interacting responses (Fig. 9 , SI Figure S13). By examining multiple driver axes, such plots can highlight important interactions and identify where thresholds occur and where change along one axis (e.g. management actions, illustrated with our ‘adaptation’ options) can mitigate change along the other (e.g. ‘climate’ shifts). This ability to identify the axes that provide resilience or sensitivity to changes in others will be critically important for management uses of HydroBOT targeting climate change adaptation or efficient use of limited water. Such assessments must also take spatial and value differences into account; the different ecological values and geographical SDL units show very different patterns in how they respond to the interaction between the ‘climate’ and ‘adaptation’ scenarios (Fig. 9 ). Perhaps most usefully, if scenarios are carefully developed to explore the range of potential change in the system, then the heatmap would represent the response surface of values over this range. Developing the response surface can be iterative, with subsequent scenarios developed to increase resolution near areas of rapid change in outcome. This approach differs from the usual approach of assessing only a small set of scenarios targeting specific proposed actions or environmental conditions (Marchau et al., 2019 ; Helgeson, 2020 ). Instead, by covering the range to generate a response surface, it can provide hypotheses about how proposed scenarios might perform and assess how sensitive the outcomes are to uncertainty in single scenarios. Specific proposed scenarios could, of course, also be assessed and mapped onto the surface. For example, we might think of the flow multipliers here as defining the entire range of plausible ‘climate’ changes, and the additive changes as the entire potential range of ‘adaptation’ options. Then, specific proposed adaptations or climate sequences could be mapped onto the heatmaps in Fig. 9 at particular points. The reverse is also possible; identification of areas of high sensitivity in the heatmaps could be used to choose a set of potential management actions designed to increase climate resilience. This idea extends to additional dimensions, which can then be collapsed in various ways (e.g. via principal component analysis, or by identifying dominant hydrometrics) to visualize with 2d heatmaps even when the response surface itself is best defined and studied in more dimensions. 3.4.5 Outcomes on the causal network Plotting the condition of each value directly on the network shows how condition changes across values at different scales (illustrated here for the historical [E1] scenario, Fig. 10 ). Perhaps most usefully, tracing through the causal network shows how higher-level outcomes result from the complex sets of dependencies on lower levels. For example, in the E1 scenario, the EWRs themselves (leftmost nodes) either nearly always fail (LF1 and above), or nearly always pass (SF2 and below). Many detailed ecological values (middle column) depend only on those EWRs that frequently fail, and so have very low condition (EF4 and above). In contrast, ecological values that depend on a range of EWRs (EF5 and below) show variable condition, depending not only on the values of those EWRs, but the mix of EWRs contributing to each. For example, EF5 has low condition because only two of its eight dependencies had high condition, while EF1 has high condition because only one of its six dependencies had low condition. These outcomes then combine to yield larger-scale ecological value groups (rightmost column). Here, we can see that no detailed values contributing to Waterbirds has good condition, and so neither does the Waterbird group. In contrast, the other groups show a range of conditions reflective of the number and condition of their constituent lower-level values. This network view provides a clear visualization of the importance of the aggregation functions. By calculating the ‘condition’ at each step as the mean of the constituent values at earlier steps, the network connections act to smooth out impacts. While the EWRs themselves tend to have extreme values (leftmost nodes), they combine to yield more-similar values in the ecological groupings as they incorporate multiple contributing lower-scale values (Fig. 10 ). This smoothing is a feature of averaging and occurs with spatial and temporal aggregation as well; it is simply more evident here because the networks show the steps in the process (see SI Section 9.6.1 for a spatial example). Whether the mean is appropriate is an empirical question relevant for each particular use, and may vary within and between sets of analyses. Using a mean implies that adding more constituent nodes would increase resilience, as good condition for some would compensate for poor condition of others. Using minima or maxima, for example, would have different biological meaning and management interpretation and would yield more persistent differences through the network (see SI Section 9.3). These decisions are critical for extracting meaning and interpretation and should be interrogated thoroughly to ensure they are appropriate for the particular analysis. HydroBOT encourages comparison of different aggregation choices to assess their impact, as we illustrate above in Fig. 5 . Visualizing the causal network can also show the holistic consequences of the different scenarios across the complex and interrelated sets of values, as well as identify nodes that are unusually resilient or sensitive to change using comparisons with a baseline scenario (SI Section 9.6.5). While the relative changes are a useful way of understanding change in condition for each node, they should not themselves be propagated through the network because they are nonlinear transformations and do not capture the meaning of the causal relationships. Instead, the condition scores (Figure S16) capture the causal relationships between conditions at nodes as a consequence of conditions at constituent nodes within a scenario. Then, these condition scores for each node can be compared across scenarios to visualize differences (Figure S17). Analysis of network structure, combined with aggregated values at each node, could identify critical nodes with outsized importance for outcomes at other nodes in the network. Any such nodes should be investigated to determine whether they capture key aspects of the system or are instead an artifact of model structure (possibly resulting from the way the network was developed) that may bias results. If their sensitivity does in fact capture the system, these nodes may be ideal targets for management intervention. The edges also identify the dependence between values, where we can see that different environmental outcomes depend on not only different hydrologic indicators (EWRs), but also different numbers or sets of those indicators. HydroBOT provides functionality to extract parts of the causal network relevant to a node or nodes, allowing more targeted visualization and investigation of these effects. 4. Implications HydroBOT provides consistent, scientifically robust, and repeatable capacity to model responses to flow, producing management-relevant outcomes at multiple dimensions. This capacity provides a step change for management agencies such as the Murray-Darling Basin Authority to move beyond assessment of hydrologic outcomes and towards assessing the values that depend on that hydrology. Water managers must make decisions with the tools and information currently available, while also seeking to identify knowledge gaps to address for future decisions. Even where HydroBOT is lacking (e.g. additional response models, better knowledge of aggregation choices), it serves to accentuate this missing information, leading to improvements and acknowledgment of limitations. Attempting these models and identifying weaknesses is a far more robust basis for decision making than being unaware of or ignoring these issues. Requirements such as those for a “healthy working basin” (MDBA, 2011 ) that demand attention across socio-economic, cultural, and environmental values are commonly included in management plans for large basins (e.g. Reclamation, 2012 ; DSC, 2013 ; Connor, 2015 ; Ziolkowska & Ziolkowski, 2016 ), and reflect the interconnections of a range of values provided by these systems. Key to assessing this range of values is modeling their dependence on hydrology and other management levers, and requires the capacity to make decisions relating to multiple spatial, temporal, and value dimensions. Management is most successful when multiple scales within those dimensions are considered (Moore, 2021 ). The HydroBOT aggregation and scaling framework can navigate those dimensions even when the underlying response models are highly detailed. Combining such disparate information in a standard and comparable manner can illuminate synergies and trade-offs, which could be critical for decision making. Moreover, the scaling provided by HydroBOT condenses and synthesizes large numbers of complex outcomes to results tailored for interpretation for management questions, and provides consistency, rigour, and reproducibility. Our demonstration concerns the ecological values of the Murray-Darling Basin; however, the framework is equally applicable to social, economic, and cultural values and to other locations. HydroBOT itself has a modular design to incorporate disparate response models and spatial units. Water management faces an increasing need and expectation for scientifically-sound and interpretable modeling from a range of stakeholders (Ryder et al., 2010 ; Gawne et al., 2018 ). These issues are magnified when facing major long-range decision making to understand the impacts of climate change and potential policy or operational adaptations management agencies might undertake (Neave et al., 2015 ; Tonkin et al., 2019 ). Co-design of HydroBOT with a diverse team within the Murray-Darling Basin Authority yielded distinct benefits to achieving these goals, with developers and researchers better understanding and building in the needs of the management agency, while the relevant managers better understand the limitations, uses, and potential of the tool. The focus on communication both within management agencies and to external partners, including model provenance, production of tailored synthesis figures, and causal networks, is a key outcome to build trust and understanding in HydroBOT as a tool and the results it produces. HydroBOT provides new capacity to assess a wide range of target values across scenarios and an exciting opportunity for continued advances and improvement with the development of new response models and climate scenarios. Declarations 5. CRediT authorship contribution statement Galen Holt*: Writing – original draft, review & editing, Visualization, Validation, Software, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Georgia Dwyer*: Writing – original draft, Visualization, Methodology, Investigation, Formal analysis, Conceptualization. David Robertson: Writing – review & editing, Supervision, Conceptualization, Funding acquisition. Martin Job: Writing – review & editing, Supervision, Conceptualization. Rebecca E Lester Writing – review & editing, Supervision, Investigation, Conceptualization, Funding acquisition. *Holt and Dwyer co-first authors. 6. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. 7. Acknowledgements This research was supported through funding from the Australian Government Murray–Darling Water and Environment Research Program (MD-WERP). We thank Murray-Darling Basin Authority collaborators including Lara Palmer, Elisha Freedman, Tumi Bjornsson, Matt Coleman, and Joel Bailey for their assistance and support. 8. Data availability All data and their sources have been explained in the manuscript. HydroBOT code is available at github.com/galenholt/HydroBOT, and code to create this manuscript is available at github.com/galenholt/toolkit-demo-paper. Documentation with examples is available at github.com/MDBAuth/HydroBOT_website. All are archived at 10.6084/m9.figshare.27605157. Links are public. HydroBOT and paper code at submission are archived at https://figshare.com/s/699c18f781df7ed2f39d. References Allaire JJ, Teague C, Scheidegger C, Xie Y, Dervieux C (2022) Quarto Arthington AH, Kennen JG, Stein ED, Webb JA (2018) Recent advances in environmental flows science and water management – Innovation in the Anthropocene. Freshw Biol 63:1022–1034. https://doi.org/10.1111/fwb.13108 Ascough JC, Maier HR, Ravalico JK, Strudley MW (2008) Future research challenges for incorporation of uncertainty in environmental and ecological decision-making. Ecol Model 219:383–399. https://doi.org/10.1016/j.ecolmodel.2008.07.015 Bengtsson H (2021) A unifying framework for parallel and distributed processing in R using futures. R J 13:273–291. https://doi.org/10.32614/RJ-2021-048 Bond NR, Brooks S, Capon S, Hale J, Kennard M, McGinness H (2021) Chapter 4 - Water-based assets of the Murray–Darling Basin and their ecological condition. In: Murray-Darling Basin, Australia: Its future management . Ecohydrology from Catchment to Coast, (Eds B.T. Hart, N.R. Bond, N. Byron, C.A. Pollino & M.J. Stewardson), pp. 75–93. Elsevier Brooks S (2021) ANAE Classification of the Murray-Darling Basin Technical Report, Revision 3.0 . Commonwealth Environmental Water Office, Department of Agriculture, Water; the Environment, Australia Brown RR, Keath N, Wong THF (2009) Urban water management in cities: Historical, current and future regimes. Water Sci Technol 59:847–855. https://doi.org/10.2166/wst.2009.029 Campbell IC (2016) Integrated management of large rivers and their basins. Ecohydrol Hydrobiol 16:203–214. https://doi.org/10.1016/j.ecohyd.2016.09.006 Connor R (2015) The United Nations world water development report 2015: Water for a sustainable world. UNESCO, France Cosgrove B, Gochis D, Flowers T, Dugger A, Ogden F, Graziano T et al (2024) NOAA’s National Water Model: Advancing operational hydrology through continental-scale modeling. JAWRA J Am Water Resour Association 60:247–272. https://doi.org/10.1111/1752-1688.13184 DELWP (2022) Environmental water management plan guidelines for rivers and wetlands. Department of Environment, Land, Water, and Planning, State of Victoria, Melbourne Devanand A, Khan Z, Chiew F, Zheng H, Potter N, Robertson D et al (2025) Hydroclimate projections for the murray-darling basin DEW (2020) Long-term environmental watering plan for the South Australian River Murray water resource plan area. Department for Environment and Water, South Australian Government, Adelaide DPIE (2020) Macquarie-Castlereagh long term water plan part A: Macquarie-Castlereagh catchment. NSW Department of Planning, Industry; Environment, Parramatta DRDMW (2022) Warrego Paroo Bulloo and Nebine long-term watering plan. Department of Regional Development, Manufacturing; Water, Queensland Government, Brisbane DSC (2013) The Delta Plan. Delta Stewardship Council, California Frota RL, Souza Filho F, de Barros A, Silva LS, Porto SMO V.C., Rocha RV (2021) Network socio-hydrology: A case study of causal factors that shape the Jaguaribe River Basin, Ceará-Brazil. Hydrol Sci J 66:935–950. https://doi.org/10.1080/02626667.2021.1913282 Gawne B, Capon SJ, Hale J, Brooks SS, Campbell C, Stewardson MJ et al (2018) Different conceptualizations of river basins to inform management of environmental flows. Front Environ Sci 6. https://doi.org/https://doi.org/10.3389/fenvs.2018.00111 Harrison PA, Beauchamp K, Cooper J, Dickie I, Fitch A, Gooday R et al (2023) An adaptable integrated modelling platform to support rapidly evolving agricultural and environmental policy. Environ Model Softw 169:105821. https://doi.org/10.1016/j.envsoft.2023.105821 Hart BT, Bond NR, Byron N, Pollino CA, Stewardson MJ (2021) Chapter 1 - Introduction to the Murray–Darling Basin system, Australia. In: Murray-Darling Basin, Australia . Ecohydrology from Catchment to Coast, (Eds B.T. Hart, N.R. Bond, N. Byron, C.A. Pollino & M.J. Stewardson), pp. 1–17. Elsevier Helgeson C (2020) Structuring decisions under deep uncertainty. Topoi 39:257–269. https://doi.org/10.1007/s11245-018-9584-y Holling CS (2001) Understanding the complexity of economic, ecological, and social systems. Ecosystems 4:390–405. https://doi.org/10.1007/s10021-001-0101-5 Holt G, Chesson P (2016) Scale-dependent community theory for streams and other linear habitats. Am Nat 188:E59–E73. https://doi.org/10.1086/687525 Holt G, Dwyer GK, Robertson D, Bailey J, Job M, Palmer L et al (2022) Architecture design for the Climate Adaptation Toolkit workflow. Murray-Darling Basin Authority, Canberra Holt G, Macqueen A, Lester RE (2024) A flexible consistent framework for modelling multiple interacting environmental responses to management in space and time. J Environ Manage 367:122054. https://doi.org/10.1016/j.jenvman.2024.122054 Jones NA, Shaw S, Ross H, Witt K, Pinner B (2016) The study of human values in understanding and managing social-ecological systems. Ecol Soc 21. https://doi.org/http://dx.doi.org/10.5751/ES-07977-210115 Kaye-Blake B, Schilling C, Nixon C, Destremau K (2014) Water management in New Zealand: A road map for understanding water value. New Zealand Institute of Economic Research King JM, Tharme RE, De Villiers M (2000) Environmental flow assessments for rivers: Manual for the building block methodology. Water Research Commission Pretoria Lester RE (2019) Wise use: Using ecological models to understand and manage aquatic ecosystems. Mar Freshw Res 71:46–55. https://doi.org/10.1071/MF18402 Levin SA (1992) The problem of pattern and scale in ecology: The Robert H. MacArthur Award Lecture. Ecology 73, 1943–1967. https://doi.org/https://doi-org/10.2307/1941447 Lobegeiger J, Prior A (2022) Border Rivers and Moonie long-term watering plan. Queensland Department of Regional Development, Manufacturing; Water, Brisbane Loire R, Piégay H, Malavoi J-R, Kondolf GM, Bêche LA (2021) From flushing flows to (eco)morphogenic releases: Evolving terminology, practice, and integration into river management. Earth Sci Rev 213:103475. https://doi.org/10.1016/j.earscirev.2020.103475 Marchau VAWJ, Walker WE, Bloemen PJTM, Popper SW (eds) (2019) Decision Making under Deep Uncertainty: From Theory to Practice. Springer International Publishing, Cham Martínez LF, Toro J, León J C (2019) A complex network approach to environmental impact assessment. Impact Assess Project Appraisal 37:407–420. https://doi.org/10.1080/14615517.2018.1552442 Marwick B, Boettiger C, Mullen L (2018) Packaging data analytical work reproducibly using R (and friends). Am Stat 72:80–88. https://doi.org/10.1080/00031305.2017.1375986 MDBA (2023) A guide to Traditional Owner. groups for Water Resource Plan areas MDBA (2019) Basin-wide environmental watering strategy. Murray-Darling Basin Authority, Canberra MDBA (2011) Delivering a healthy working basin. Murray-Darling Basin Authority, Canberra MDBA (2024) EWR tool Medema W, McIntosh BS, Jeffrey PJ (2008) From premise to practice: A critical assessment of integrated water resources management and adaptive management approaches in the water sector. Ecol Soc 13 Mentzafou A, Katsafados P, Papadopoulos A, Dimitriou E (2023) Hydrological regime alteration assessment in the context of WFD 2000/60: A European and global review. Sustainability 15:15704. https://doi.org/10.3390/su152215704 Moore S (2021) Toward effective river basin management (RBM): The politics of cooperation, sustainability, and collaboration in the Delaware River basin. J Environ Manage 298:113421. https://doi.org/10.1016/j.jenvman.2021.113421 Neave I, McLeod A, Raisin G, Swirepik J (2015) Managing water in the Murray-Darling Basin under a variable and changing climate. Water: J Australian Water Association 42:102–107. https://doi.org/10.3316/informit.269194666182946 von der Ohe PC, Apitz SE, Arbačiauskas K, Beketov MA, Borchardt D et al (2014) In: Brils J, Brack W, Müller-Grabherr D, Négrel P, Vermaat JE (eds) Status and Causal Pathway Assessments Supporting River Basin Management. Springer, Berlin, Heidelberg, pp 53–149. de Olden JD, Konrad CP, Melis TS, Kennard MJ, Freeman MC, Mims MC et al (2014) Are large-scale flow experiments informing the science and management of freshwater ecosystems? Front Ecol Environ 12:176–185. https://doi.org/10.1890/130076 Pang J, Chen S, Men B, Wang H (2024) Causal networks and spatiotemporal patterns of different droughts from the perspective of hydrological cycle - A case study of the Nenjiang River Basin, China. Ecol Ind 166:112326. https://doi.org/10.1016/j.ecolind.2024.112326 Pebesma E, Bivand R (2023) Spatial Data Science: With applications in R. Chapman and Hall/CRC Peeters LJM, Holland KL, Huddlestone-Holmes C, Boulton AJ (2022) A spatial causal network approach for multi-stressor risk analysis and mapping for environmental impact assessments. Sci Total Environ 802:149845. https://doi.org/10.1016/j.scitotenv.2021.149845 Reclamation B (2012) of Colorado River basin water supply and demand study technical report D — System reliability metrics . US Department of the Interior, Bureau of Reclamation Regan RS, Juracek KE, Hay LE, Markstrom SL, Viger RJ, Driscoll JM et al (2019) The U.S. Geological Survey National Hydrologic Model infrastructure: Rationale, description, and application of a watershed-scale model for the conterminous United States. Environ Model Softw 111:192–203. https://doi.org/10.1016/j.envsoft.2018.09.023 Robson BJ (2014) When do aquatic systems models provide useful predictions, what is changing, and what is next? Environ Model Softw 61:287–296. https://doi.org/10.1016/j.envsoft.2014.01.009 Ruel JJ, Ayres MP (1999) Jensen’s inequality predicts effects of environmental variation. Trends Ecol Evol 14:361–366. https://doi.org/10.1016/S0169-5347(99)01664-X Ryder DS, Tomlinson M, Gawne B, Likens GE (2010) Defining and using ‘best available science’: a policy conundrum for the management of aquatic ecosystems. Mar Freshw Res 61:821–828. https://doi.org/10.1071/MF10113 Scheuerell MD, Ruff CP, Anderson JH, Beamer EM (2021) An integrated population model for estimating the relative effects of natural and anthropogenic factors on a threatened population of steelhead trout. J Appl Ecol 58:114–124. https://doi.org/10.1111/1365-2664.13789 Sheldon F, Rocheta E, Steinfeld C, Colloff MJ, Moggridge B, Carmody E et al (2024) Are environmental water requirements being met in the Murray-Darling Basin, Australia? Mar Freshw Res 75. https://doi.org/10.1071/MF23172 Shenton W, Bond NR, Yen JDL, MacNally R (2012) Putting the ecology into environmental flows: Ecological dynamics and demographic modelling. Environ Manage 50:1–10. https://doi.org/10.1007/s00267-012-9864-z Singer MB (2007) The influence of major dams on hydrology through the drainage network of the Sacramento River basin, California. River Res Appl 23:55–72. https://doi.org/10.1002/rra.968 Stern CV, Sheikh PA (2019) Management of the Colorado River: Water Allocations, Drought, and the Federal Role. Congressional Research Service, Washington, D.C. Tharme RE (2003) A global perspective on environmental flow assessment: emerging trends in the development and application of environmental flow methodologies for rivers. River Res Appl 19:397–441. https://doi.org/10.1002/rra.736 Tonkin JD, Poff NL, Bond NR, Horne A, Reynolds MDM L.V., et al (2019) Prepare river ecosystems for an uncertain future. Nature 570:301–303. https://doi.org/10.1038/d41586-019-01877-1 Ushey K, Allaire JJ, Tang Y (2023) Reticulate: Interface to ’python’ Water Act (2007) (2007) Welsh WD, Vaze J, Dutta D, Rassam D, Rahman JM, Jolly ID et al (2013) An integrated modelling framework for regulated river systems. Environ Model Softw 39:81–102. https://doi.org/10.1016/j.envsoft.2012.02.022 Wickham H (2016) ggplot2: Elegant graphics for data analysis. Springer-, New York Wickham H, Bryan J (2025) R packages, 2nd edn. O’Reilly Wilson G, Aruliah DA, Brown CT, Hong NPC, Davis M, Guy RT et al (2014) Best practices for scientific computing. PLoS Biol 12:e1001745. https://doi.org/10.1371/journal.pbio.1001745 Yang A, Dutta D, Vaze J, Kim S, Podger G (2017) An integrated modelling framework for building a daily river system model for the Murray–Darling Basin, Australia. Int J River Basin Manage 15:373–384. https://doi.org/10.1080/15715124.2017.1315814 Ziolkowska JR, Ziolkowski B (2016) Effectiveness of water management in Europe in the 21st century. Water Resour Manage 30:2261–2274. https://doi.org/10.1007/s11269-016-1287-9 Additional Declarations The authors declare potential competing interests as follows: Holt, Dwyer, Robertson, and Lester received funding from the Murray-Darling Basin Authority for the Murray-Darling Water and Environment Research Program. Job is an employee of the Murray-Darling Basin Authority. Supplementary Files GA.png Graphical abstract Supplementaryinformation.pdf Supplementary Information for HydroBOT: an integrated toolkit for assessment of hydrology-dependent outcomes Cite Share Download PDF Status: Published Journal Publication published 18 Jun, 2025 Read the published version in Environmental Modelling & Software → 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-6793584","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":464692928,"identity":"70058c3f-94c6-49c0-93c3-ec789a927744","order_by":0,"name":"Galen Holt","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3UlEQVRIie2OsQrCMBCGTwp1OZ27hL5CJCAOpfgoCYW4Ci6O3VzEru1b1CcwEtDNrgWXTroKgrhpdHOJdXPIBwf/wX38B+Bw/CUI0EDUScEHZVYT2iicy98V/VaglRKOe9tG3CuSdXdCTyEipfJOgU0ZqH5COT+yYimVzkGyUvlDu5LiMDCKKNUk1QjaBGilHMSmOr+Uh1G6N6sSwltRogzMYwgmKLS3UA8Z5TJheX3iGmnCCo2zkbVlsRw0lygmWSbZFecxWe0X69raoj9XM57t/tWSfjlwOBwOBzwB1CtMfoSE2tsAAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0002-7455-9275","institution":"Deakin University","correspondingAuthor":true,"prefix":"","firstName":"Galen","middleName":"","lastName":"Holt","suffix":""},{"id":464692929,"identity":"208d4ab2-4b75-4e79-9e7b-debc9ec6eedb","order_by":1,"name":"Georgia K Dwyer","email":"","orcid":"https://orcid.org/0000-0002-3579-3819","institution":"Deakin University","correspondingAuthor":false,"prefix":"","firstName":"Georgia","middleName":"K","lastName":"Dwyer","suffix":""},{"id":464779196,"identity":"c038bd67-4244-4364-93af-2b89a97e934d","order_by":2,"name":"David Robertson","email":"","orcid":"https://orcid.org/0000-0003-4230-8006","institution":"CSIRO","correspondingAuthor":false,"prefix":"","firstName":"David","middleName":"","lastName":"Robertson","suffix":""},{"id":464779197,"identity":"3bd3daa8-dfa2-4b12-98d5-1484285559c1","order_by":3,"name":"Martin Job","email":"","orcid":"","institution":"Murray-Darling Basin Authority","correspondingAuthor":false,"prefix":"","firstName":"Martin","middleName":"","lastName":"Job","suffix":""},{"id":464779198,"identity":"a849d0e3-10b9-4f26-9198-c0e72c8b6442","order_by":4,"name":"Rebeccal E Lester","email":"","orcid":"https://orcid.org/0000-0003-2682-6495","institution":"Deakin University","correspondingAuthor":false,"prefix":"","firstName":"Rebeccal","middleName":"E","lastName":"Lester","suffix":""}],"badges":[],"createdAt":"2025-06-01 03:51:38","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":true,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-6793584/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6793584/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1016/j.envsoft.2025.106579","type":"published","date":"2025-06-19T00:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":83835392,"identity":"7c9a897d-63be-4dd6-a60b-fc4970068a6f","added_by":"auto","created_at":"2025-06-03 12:57:44","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":416858,"visible":true,"origin":"","legend":"\u003cp\u003eConceptualisation of HydroBOT architecture (blue arrows) and links to management decision making (yellow arrows). Scenarios, represented by hydrographs, reflect observed flows or modeled scenarios (produced via hydrological modeling outside HydroBOT). These hydrographs are fed in a consistent way to various response models via the HydroBOT Controller. These response models could assess response of any values of interest, here illustrated as a range of value types across environmental and human dimensions. Additional inputs of spatial data and causal networks provide grouping information for scaling via the HydroBOT Aggregator. The HydroBOT Comparer synthesizes the results into management-relevant outputs that aid decision making. This decision making process is also supported by feedback from stakeholders, other expert advice and the social and political landscape. Water planning and policy decisions can then guide the development of scenarios to assess potential management options which, in turn, can be assessed with HydroBOT.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/552506dde22beaf3f7efb351.png"},{"id":83834256,"identity":"7ad3c9af-2b4c-4004-8e81-23aa7ed0e79f","added_by":"auto","created_at":"2025-06-03 12:41:44","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":232057,"visible":true,"origin":"","legend":"\u003cp\u003eAggregation along multiple dimensions. HydroBOT provides flexible capability to aggregate along spatial, temporal and value scales. Boxes here show example scales, which may not all be used and which can be user-defined if others are needed. Users can control the sequence of steps along these axes, with the capability to switch between axes at different steps.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/7f26668c51d398e9e5a213f6.png"},{"id":83834451,"identity":"4270a05e-eaa4-40f3-a1ba-aeb0650ea848","added_by":"auto","created_at":"2025-06-03 12:49:44","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":77704,"visible":true,"origin":"","legend":"\u003cp\u003eThe flow levels of various EWRs are illustrated at two example gauges (412002 and 419001) on historical hydrographs. Water levels crossing the lines indicate that the immediate flow level requirements are met. Flow requirements for some EWRs (OB4 and OB5) are never met for these hydrographs because flows do not get high enough, and so ecological responses dependent on these overbank conditions would not be expected to occur. This does not illustrate the required frequency, duration or timing for the EWRs; some of the lines crossed by the hydrograph here would not count as achieving the EWR if the flows did not stay above the relevant EWR line long enough or did not cross it again within the defined timeframe. See SI Table S3.\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/d25b72878f74031a7478b6c8.png"},{"id":83834252,"identity":"3d3caa51-5685-40f8-bb7a-067469faa230","added_by":"auto","created_at":"2025-06-03 12:41:44","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":108911,"visible":true,"origin":"","legend":"\u003cp\u003e(a) Location of the Murray-Darling Basin (MDB; light blue) in eastern Australia. Example SDL units within the MDB shown in red, dark blue, and yellow. (b) Gauge locations within the three example SDL units (Namoi, yellow; Macquarie–Castlereagh, dark blue; and Lachlan, red). (c-d) Scenario comparison for different Sustainable Diversion Limit areas emphasizing (c) quantitative differences and (d) spatial patterns. The proportion of environmental watering requirements (EWRs) that are achieved under each scenario for each SDL unit are shown. Climate scenarios and adaptation options as in Table 1.\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/9587463e7e9ded6ee94181b5.png"},{"id":83834452,"identity":"bc995451-f721-44e0-b910-21a2a1419b9a","added_by":"auto","created_at":"2025-06-03 12:49:44","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":76364,"visible":true,"origin":"","legend":"\u003cp\u003eCondition results for large ecological groupings and three ‘climate’ scenarios (color) and three different aggregation options (a-c). visualizing as a radar plot allows comparing outcomes of disparate thematic variables (as illustrated in the graphical abstract) and is useful for quick broad-scale representation of the results. Outcomes here for Macquarie–Castlereagh SDL unit. Dotted lines provide axis values, with the outer line corresponding to a condition of 0.5 and the center of the circle being a condition of 0. One advantage of HydroBOT is the ability to specify aggregations relevant to a question, or to test the impact of different aggregation choices, reflected here in the different panels; a) Aggregation as described in the text, with means at each stage except where the ‘versions’ of EWRs use the maximum to get the EWR itself; b) As in a, except using a minimum of those versions and so the most stringent requirement for the EWR itself; c) Using the median at all steps other than the initial time-average.\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/9a4b3d8ad5e0b101c9e3afcf.png"},{"id":83834259,"identity":"47597745-ec2a-4241-b099-9a1e9e75f312","added_by":"auto","created_at":"2025-06-03 12:41:44","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":83024,"visible":true,"origin":"","legend":"\u003cp\u003eEWR achievement for broader ecological values, including: ecosystem function (EF), native fish (NF), native vegetation (NV), other species (OS) and waterbirds (WB) (color groups \u0026amp; rows). Columns stack all small-scale ecological values (e.g. WB1-5) (each with its own shade) that contribute to each broader ecological value (e.g. Waterbirds), shown with differen colors. This provides an assessment of the performance of individual small-scale values, as well as how those contribute to the overall sensitivity of the broader groups. These small-scale values (and so shades) are distinct within each group; e.g. WB1 is not the same as NF1. Each small-scale value depends on a set of constituent hydrologic requirements (EWRs), with the heights of each individual shade being the proportion of these EWRs that are met. The full bar heights are then the sum of these proportions. Dashed lines represent full achievement of all constituent EWRs for all small-scale values. If all EWRs contributing to small-scale values were met, the columns would reach the lines. Not all EWRs are applicable in all locations and there are not equal number of EWR in each category, hence different values of the dashed lines. If the details of the smaller-scale ecological values were the focus of an investigation, more granular plots might be created that better differentiate these values for target larger-scale values, as illustrated in Figure S14 and Figure S15. This illustration includes three ‘climate’ scenarios: A (0.5x), E (historical base level/no change), and I (2x); and three ‘adaptation’ options: 1 (no adaptation), 2 (+250 ML/d), and 3 (+6500 ML/d). There are no EWRs contributing to Other species in the Namoi. Columns for scenarios A1, E1, and I1 for Macquarie–Castlereagh correspond to values shown in Figure 5 panel a.\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/fc6e4ec0a2896392b49ce84f.png"},{"id":83834255,"identity":"c7cc5ff4-5fa6-4684-bdc2-b8f757e0a2d9","added_by":"auto","created_at":"2025-06-03 12:41:44","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":70541,"visible":true,"origin":"","legend":"\u003cp\u003eQuantitative scenario comparison for Native fish objectives for three SDL units. Scenarios defined here as the difference in mean flow (GL/d) from the baseline historical scenario (E1). The lines are not smooth because EWR achievement is not simply a function of the amount of water. Instead, the ‘adaptation’ options, which add water at particular times, particularly the ‘2’ scenarios of adding 250 ML/d of water, have disproportionately large impact (steep slopes). Shading indicates different ecological objectives within the native fish ecological value. Three ‘climate’ scenarios are shown: A (0.5x), E (historical base level/no change), and I (2x) and three ‘adaptation’ options: 1 (no adaptation), 2 (+250 ML/d), and 3 (+6500 ML/d). HydroBOT provides in-built capacity for baselining with differences, relative change, or user-defined comparison functions. Difference was used here due to some additive flow changes, though relative would be a reasonable choice as well.\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/cace9d3c79dc43f22e79de20.png"},{"id":83834454,"identity":"52cc8e78-6fd2-4aa8-88d3-8593c528bfe9","added_by":"auto","created_at":"2025-06-03 12:49:44","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":65591,"visible":true,"origin":"","legend":"\u003cp\u003eSmoothed fits of shifts in condition (proportion of EWR achieved relative to the scenario with no climate or adaptation changes [E1]) of broad ecological values to relative changes in water availability (also relative to E1). (a) All 9 ‘climate’ scenarios (without adaptation; adaptation code 1) are included, ranging from halving to doubling flows (Table 1). (b) The full combination of scenarios (all climate and adaptations options) are shown for the example of Native fish in the Namoi (with each represented as a dot). The green points and line are the same as the Namoi Native fish panel in a. Panel b thus illustrates the larger impact of the seasonal additions (‘adaptation’ options) used here compared to the ‘climate’ shifts; differences between lines are greater than the change along the lines. The ‘adaptations’ also make the outcome less sensitive to ‘climate’ (the yellow and grey lines have lower slope than green). In all plots, both axes are presented on the log scale so multiplicative changes are correctly represented (i.e. halving yields the same axis distance as doubling). The multiplicative ‘climate’ scenarios in a and b range from halving to doubling flows, but outcomes over the range of flow changes varied between approximately 1/3x and 3x. Thus, the x-axis is extended to help visualize multiplicative proportionality of outcome changes relative to flow changes. We also add dashed lines at y = 0.5 and 2, which are the minumum and maximum multiplicative flow changes, to better see where outcomes have multiplicative changes outside these values. Slopes steeper than 1:1 and values outside these dashed lines indicate that outcomes multiply faster than flows, while shallower slopes indicate changes in flows result in less than proportional changes in outcomes. Nonlinear relationships mean that a multiplicative increase in flows (e.g. doubling) does not have the same impact as an equivalent multiplicative decrease (e.g. halving). In all plots, fitted lines are loess smooths. Y-axes are restricted to better visualize the majority of the data; thus, a small amount of data is not plotted but is accounted for in the loess fits. The full set of SDL units and groups as in b are shown in SI Figure S12.\u003c/p\u003e","description":"","filename":"image9.png","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/8c2cc854464bd6ee6d550053.png"},{"id":83834261,"identity":"51ef8977-3546-4de0-be06-997142d6f3c5","added_by":"auto","created_at":"2025-06-03 12:41:44","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":81785,"visible":true,"origin":"","legend":"\u003cp\u003eCondition results visualized as a surface with the two scenario definitions on the axes. This approach allows visualizing changes in outcome as the result of multiple axes on which scenarios might differ. These axes might be different aspects of scenario creation (as here), or they might be different axes describing the outcome of scenarios (e.g. two different hydrometrics such as mean flow and flow variance). There are no EWRs contributing to Other species in the Namoi.\u003c/p\u003e","description":"","filename":"image10.png","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/9e475f9ff8b79702b464f580.png"},{"id":83834455,"identity":"752c1524-7dff-4bcc-bed6-2196167423e2","added_by":"auto","created_at":"2025-06-03 12:49:45","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":545962,"visible":true,"origin":"","legend":"\u003cp\u003eConceptualisation of HydroBOT architecture (blue arrows) and links to management decision making (yellow arrows). Scenarios, represented by hydrographs, reflect observed flows or modeled scenarios (produced via hydrological modeling outside HydroBOT). These hydrographs are fed in a consistent way to various response models via the HydroBOT Controller. These response models could assess response of any values of interest, here illustrated as a range of value types across environmental and human dimensions. Additional inputs of spatial data and causal networks provide grouping information for scaling via the HydroBOT Aggregator. The HydroBOT Comparer synthesizes the results into management-relevant outputs that aid decision making. This decision making process is also supported by feedback from stakeholders, other expert advice and the social and political landscape. Water planning and policy decisions can then guide the development of scenarios to assess potential management options which, in turn, can be assessed with HydroBOT.\u003c/p\u003e","description":"","filename":"image11.png","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/3defebcf600ff577d40cbe43.png"},{"id":85221261,"identity":"fdb27456-f1e0-4094-931b-45b5ac9d0351","added_by":"auto","created_at":"2025-06-23 14:19:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2439381,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/97e13491-9751-45ed-93b6-bc467b8d502d.pdf"},{"id":83834263,"identity":"13c310a1-7d5c-418e-807b-31793400da65","added_by":"auto","created_at":"2025-06-03 12:41:45","extension":"png","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":231649,"visible":true,"origin":"","legend":"\u003cp\u003eGraphical abstract\u003c/p\u003e","description":"","filename":"GA.png","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/e5bd0d6a95bce050ab4d172f.png"},{"id":83834260,"identity":"03547687-3e31-4660-85a9-b81c7a617fe3","added_by":"auto","created_at":"2025-06-03 12:41:44","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":8254559,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Information for HydroBOT: an integrated toolkit for assessment of hydrology-dependent outcomes\u003c/p\u003e","description":"","filename":"Supplementaryinformation.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6793584/v1/607944c46b1a7a0ddbbbdebf.pdf"}],"financialInterests":"The authors declare potential competing interests as follows: Holt, Dwyer, Robertson, and Lester received funding from the Murray-Darling Basin Authority for the Murray-Darling Water and Environment Research Program. Job is an employee of the Murray-Darling Basin Authority. ","formattedTitle":"\u003cp\u003eHydroBOT: an integrated toolkit for assessment of hydrology-dependent outcomes\u003c/p\u003e","fulltext":[{"header":"Highlights","content":"\u003cul type=\"disc\"\u003e\n \u003cli\u003eThe HydroBOT toolkit provides new management modeling capability for data and model integration, demonstrated for the Murray-Darling Basin.\u003c/li\u003e\n \u003cli\u003eHydroBOT synthesizes and assesses outcomes across disparate target values.\u003c/li\u003e\n \u003cli\u003eScaling and synthesis of outcomes in space, time, and groups of values unlocks improved interpretation and decision making.\u003c/li\u003e\n \u003cli\u003eScenario comparison, particularly impacts of climate and management adaptation, are primary uses.\u003c/li\u003e\n \u003cli\u003eHydroBOT emphasizes automation, transparency, and error detection for consistent repeatable analyses.\u003c/li\u003e\n\u003c/ul\u003e"},{"header":"1. Introduction","content":"\u003cp\u003eWater management in large river systems typically targets a wide range of desired values (Holling, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2001\u003c/span\u003e; Brown, Keath \u0026amp; Wong, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Jones et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). For example, in the Murray-Darling Basin (MDB), Australia, the objective of water management is to maintain a healthy working river that supports productive and resilient water-dependent industries, healthy and resilient ecosystems, and communities with access to sufficient and reliable water supplies (MDBA, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). These values, defined here as any social, economic, environmental or cultural asset or function of significance, importance, worth, or use, are not unique to Australia; water is managed to protect a diverse range of values worldwide (e.g. Kaye-Blake et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Connor, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Ziolkowska \u0026amp; Ziolkowski, \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Stern \u0026amp; Sheikh, \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). That these values are the target of water management implies that each value depends on water in some way but, in many cases, dependencies are not well defined and so management targets water volumes and flow characteristics under the assumption dependent values will follow (e.g. Arthington et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Mentzafou et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWater managers typically have a limited range of potential actions (levers) at their disposal. While the particular levers vary across jurisdictions, those that directly affect hydrology dominate (e.g. flow releases from dams or inundation of wetlands, MDBA, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Holt, Macqueen \u0026amp; Lester, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). When non-hydrologic levers are available, such as water trading rules in the Murray-Darling Basin, their impact is still often assessed by how they alter hydrologic conditions \u0026ndash; that is, their impact on the spatio-temporal patterns of flow in the system (e.g. DPIE, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; DELWP, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Hydrologic modeling is often better-integrated with management workflows than models of other values (e.g. King, Tharme \u0026amp; De Villiers, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2000\u003c/span\u003e; Tharme, \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; Shenton et al., \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). In part, this is due to the availability of large-scale physically-based numerical models that provide robust ability to simulate flows as they arise from natural drivers, such as rainfall, and describe infrastructure impacts including dams and diversions (e.g. Yang et al., \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2017\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThough these models tend to have high precision, describe complex spatio-temporal physical relationships, and produce detailed outputs of state variables across time and space, they target relatively few primary outcome variables, e.g. streamflow, runoff, or snowpack (Welsh et al., \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; e.g. Regan et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Cosgrove et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). In contrast, models of ecological or economic responses are often less detailed but must model a wide range of outcomes, including any number of species, ecosystem metabolism, or economic output from various sectors (e.g. Robson, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Frota et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Harrison et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The impact of given management actions (e.g. dam releases) on hydrology is well understood and captured by the hydrograph, allowing accurate targeting of particular aspects of the flow regime by management actions (e.g. Singer, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Loire et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). While these proximate models of system state and the impact of management actions are invaluable, they do not provide the necessary information to directly assess the status of the range of flow-dependent values.\u003c/p\u003e \u003cp\u003eAssessing values as they respond to flow and management actions requires models of these relationships. Such models exist, though their quality and extent vary widely, ranging from unstated mental models to highly-detailed population dynamics or economic models (Lester, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Holt et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Quantitative computational models are often written in various languages by subject-matter experts, focus on one or a limited suite of target values, and return disparate outputs depending on their particular goals and approaches (Ryder et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). These models are often designed with the aim of studying the dynamics of highly specific variables (e.g. fish population responses to anthropogenic change, Scheuerell et al., \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), not to produce the most useful analysis for management questions or to capture the breadth of target values. Moreover, such targeted models typically cannot individually provide information about larger-scale values; for example, single-species models are insufficient to assess the condition of all fish or the whole ecosystem (e.g. Olden et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Gawne et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIntegrating distinct quantitative models into a holistic modeling approach is necessary to assess management-relevant outcomes of disparate, typically broad-scale, values under different hydrologic conditions and management actions. This integration gives each response model utility beyond its original purpose and identifies where new work is needed, such as where response models are limited or non-existent (Holt et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The creation of an integrated toolkit provides the opportunity for robust decision-making (Harrison et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), the ability to prioritize water planning across a range of values, and to identify conditions that achieve disproportionately large (or small) impacts, as well as the explicit consideration of synergies and trade-offs among different values.\u003c/p\u003e \u003cp\u003eExemplifying these issues, the Murray-Darling Basin is centrally important to the economic, cultural, social and environmental wellbeing of Australia. Within the MDB, 70% of water is utilized for agricultural purposes (Hart et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Agriculture and tourism in the MDB contribute nearly 2% to Australian gross domestic product (Hart et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The MDB is home to 2.2\u0026nbsp;million people, including more than 50 Aboriginal Nations, and contains approximately 500,000 mapped water-dependent habitats (Hart et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Brooks, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; MDBA, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). While balancing this range of values is difficult in any highly-utilized basin, the Murray-Darling is unusually dry and variable compared to other systems of similar importance (Bond et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eContemporary management of the MDB arises from the Commonwealth (Water Act 2007, 2007), which established the Murray-Darling Basin Authority (MDBA) with requirements to develop and implement a Basin Plan governing water resource management (MDBA, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Hart et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The MDBA has obligations to manage water to maintain a \u0026ldquo;healthy working river\u0026rdquo; (MDBA, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). The state of the MDB is required to be assessed annually, with a major review of, and updates to, the Basin Plan made at legislated intervals (e.g. 15 years between development and first review, Hart et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The MDBA does not itself deliver water; instead states within the basin have primary responsibility for water delivery and integrated catchment management, with the MDBA providing coordination and an overarching policy framework (MDBA, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Hart et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Other stakeholders (e.g. the Commonwealth Environmental Water Holder) have additional responsibilities for water management and coordination in the MDB.\u003c/p\u003e \u003cp\u003eThe MDBA and other water managers in the Murray-Darling Basin have an increasing focus on assessing system response to climate change and management actions including climate adaptations that may be taken by the MDBA itself or other stakeholders in the MDB (Neave et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Taken together, the need for collaborative management across jurisdictions to deliver legislated requirements and manage into the future for a healthy working basin require a holistic modeling approach. Such an approach should integrate across values and be adaptable to these various management needs while also simplifying repeated, ongoing use. These needs are not unique to the MDBA; other water management groups, researchers and stakeholders have similar complex and competing needs (Campbell, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2016\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe MDBA has access to robust hydrologic models describing flows in the system and responses to current management practices and these models are under continual use, development, and improvement (e.g. Yang et al., \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Despite the need for ongoing assessment and forward planning across a range of values, models of the responses to hydrologic conditions are patchy, only represent some values, and have been developed and used in a more \u003cem\u003ead hoc\u003c/em\u003e approach rather than integrated with each other or the hydrologic models (Olden et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Gawne et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Holt et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). In this paper, we describe an integrated modeling \u0026lsquo;toolkit\u0026rsquo;, hereafter referred to as \u003cem\u003eHydroBOT\u003c/em\u003e (Hydrology-dependent Basin Outcomes Toolkit), to integrate these disparate models and assess the response of a wide range of water-dependent values. Decision making for water management in a complex basin such as the Murray-Darling requires multiple lines of evidence and balancing competing values and sources of information. HydroBOT greatly improves the capacity of the relevant manager (here, demonstrated for the MDBA) to assess outcomes across a range of values, provides the structure to adapt and include additional values as additional component models become available, and addresses a number of issues with current assessment practices and modeling approaches.\u003c/p\u003e \u003cp\u003eTo demonstrate how HydroBOT meets these identified needs, we develop a set of example scenarios capturing some qualitative aspects of one intended use: the assessment and comparison of outcomes under different climate scenarios and different management adaptations to those changes. These scenarios represent changes to the system due to processes over which managers have no control (i.e. a changing climate), and management actions which would typically be targeted interventions. These scenarios are not intended to represent real climate change or proposed management actions; instead we use this demonstration to illustrate the problem space and show how HydroBOT can aid in assessing potential system change and prioritising management actions to mitigate impacts.\u003c/p\u003e"},{"header":"2. HydroBOT description","content":"\u003cp\u003eHydroBOT provides a flexible architecture with distinct components that can be run separately (in a modular fashion) or together to assess how various scenarios, such as climate change or management options, affect multiple sets of values spanning many scales and disciplines. This modularity enhances the iterative nature of decision making for water planning and policy and allows tuning for a particular set of management questions (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Scenarios are typically represented by their effect on the hydrograph, though other inputs could be included if relevant (e.g. time series of variables such as temperature). HydroBOT then synthesizes a wide range of outputs from individual response models into results that directly compare scenarios and are digestible and useful for management decision making (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Because the need for assessing outcomes is nearly universal in water management and not limited to a particular scenario or project (Medema, McIntosh \u0026amp; Jeffrey, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2008\u003c/span\u003e), HydroBOT is designed with strong scenario-comparison capabilities but is agnostic to what those scenarios represent. For a similar reason, the standardized synthesis and outputs are highly flexible, allowing the user to choose and tailor the most relevant outputs for different management needs.\u003c/p\u003e \u003cp\u003eOne common issue with modeling in general, and particularly integrated models spanning several tools, is that such tools can become a black box where the relationships modeled are opaque. This can yield mistrust by the public and other stakeholders influenced by the model, but also mistrust and misunderstanding by users and developers of the model (Ascough et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2008\u003c/span\u003e). To avoid these issues, HydroBOT has been continuously co-designed with the relevant management agency, the MDBA, with emphasis on public code, reproducibility, self-documentation and production of outputs that describe the model itself in addition to its results (e.g. causal relationships).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn practice, modeling is almost never a one-off analysis of a single scenario. Instead, a wide range of scenarios are typically analyzed and compared, and those scenarios may be updated even within a single project (Harrison et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Across multiple projects, the same modeling tools might be used to address vastly different questions, e.g. assessing response to potential future climate at continental scale \u003cem\u003evs.\u003c/em\u003e historical conditions at a small set of locations (Holt et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The onward analysis of those scenarios is also rarely static, needing to be updated as new questions are asked or preliminary results are tailored more closely to the questions. As is common with large data analysis workflows or software, repeated small changes often introduce errors especially in complex scripts (Wilson et al., \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). It is not unusual for a single project to include hundreds or thousands of repeated analyses over scenarios or sets of analysis options. HydroBOT provides facilities for automating these processes and documenting the code, critical for efficiency and reproducibility. Most importantly, HydroBOT shifts much of the complex data manipulation to tested software, while limiting user input to a small number of arguments related only to scenario configuration and analysis choices. This approach not only improves the ability to understand specific runs and reproducibility, but also dramatically reduces the incidence of inadvertent errors that arise from making small adjustments to large, complex scripts (Marwick, Boettiger \u0026amp; Mullen, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Architecture\u003c/h2\u003e \u003cp\u003eTaken holistically, the HydroBOT architecture comprises three primary processing and control components (Controller, Aggregator, Comparer), three components that integrate models and data from elsewhere (Response models, Spatial data, and Causal networks), and the links between them (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The processing flow through HydroBOT is represented by the boxes along the top of Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, and in more detail in the Supplement (Figure S1). Input data (hydrographs) are ingested by the Controller, which packages and runs Response Models. Results from the Response Models are then processed by the Aggregator, and the analysis and final outputs are prepared by the Comparer.\u003c/p\u003e \u003cp\u003eHydroBOT provides coordination of analyses through this workflow, as well as the algorithms within each component. Each step in HydroBOT is parameterized for a given analysis with arguments to a set of relevant functions, giving the user wide latitude to tailor an analysis to the input data, target values, and question at hand. As each stage of HydroBOT runs, it produces metadata files including the full set of parameters for that stage and all preceding stages. These files can be used as parameter files for subsequent runs to support reproducibility.\u003c/p\u003e \u003cp\u003eThe HydroBOT architecture emphasizes modularity. Outputs can be saved at each step in the processing flow, allowing users to run HydroBOT as a whole or re-run only needed components to update analysis. For example, if a user wished to change how they aggregate the results of the modules, they could re-run the aggregation step and the subsequent comparison step to match without re-running earlier steps. This ability to adjust intermediate steps allows rapid iteration of results to address a given management question, or adaptation of pre-existing results to new questions. Modularity also allows rapid, iterative development of HydroBOT itself. Any HydroBOT component can be updated, e.g. new aggregation functionality added, without affecting others. Obtaining new results utilising new capacity simply requires parameterising that capacity for a given analysis and re-running HydroBOT from that point forward.\u003c/p\u003e \u003cp\u003eHere we describe the specific implementation of each component of HydroBOT (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), with additional detail in Supplement.\u003c/p\u003e \u003cdiv id=\"Sec4\" class=\"Section3\"\u003e \u003ch2\u003e2.1.1 HydroBOT Controller\u003c/h2\u003e \u003cp\u003eThe HydroBOT Controller is the interface between the externally generated input data (i.e. scenarios), the chosen response model, and other external and internal HydroBOT components (SI Figure S1). This component initiates the downstream processing steps according to user-defined settings for a particular run. It includes arguments for locating the input data and the response model(s) to use, along with any necessary parameters for those models. The Controller can also control later components, allowing the full HydroBOT workflow to be run at once. These include defining aggregation steps as discussed in Section \u003cspan refid=\"Sec7\" class=\"InternalRef\"\u003e2.1.4\u003c/span\u003e and analysis of the results with the Comparer (Section \u003cspan refid=\"Sec8\" class=\"InternalRef\"\u003e2.1.5\u003c/span\u003e). The Controller determines whether and where outputs are returned at each step. Having control over the full workflow enables large batched runs using parameter files (yaml) to specify the control arguments. This core functionality of the Controller is delivered via simple functions that apply to the input data for each scenario and can be parallelized over scenarios. The Controller can be accessed by the user by using Quarto notebooks to work interactively, R scripts, via the command line, or with yaml parameter files, depending on the use case.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section3\"\u003e \u003ch2\u003e2.1.2 Response models\u003c/h2\u003e \u003cp\u003eThe impacts of climate and management options on social, cultural, environmental, or economic values are estimated based on causal relationships between drivers (e.g. hydrology) and responses (e.g. the state of values). Response models may exist in many different forms, ranging from binary achievement of hydrologic indicators to fully quantitative responses. To be included in HydroBOT, the minimal requirements are that they can be scripted, and return meaningful output values that can be tagged either internally or in post-processing with scenario information, though only the second is a requirement for their data to be used in later steps of HydroBOT processing (see Section \u003cspan refid=\"Sec7\" class=\"InternalRef\"\u003e2.1.4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThese tools are expected to be sourced from existing or in-development models developed by subject-matter experts and, as such, will be written in different languages and will target different outcomes. Thus, the Response models component is depicted as being defined externally to HydroBOT (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e; Figure S1), and new response models require development of lightweight wrappers to integrate them into compatible, modular HydroBOT components. Once integrated, HydroBOT provides a unified interface and ongoing analysis and modeling of the response model results.\u003c/p\u003e \u003cp\u003eHydroBOT seeks to avoid changing the response models themselves, but instead wraps them with input and output management to fit into the workflow. Specifically, to incorporate a new model, HydroBOT includes a wrapper function that passes necessary data locations, as well as any necessary arguments from the user to that model. Each response model then has a distinct set of outputs, reflecting the captured responses and the structure of the model. HydroBOT includes cleanup functions that save or return the outputs in formats usable by onward processing in HydroBOT. These formats are quite flexible, requiring only a data frame (or csv file) with output values and columns reflecting the spatial, temporal, and value dimension of those outputs and often a column indicating the scenario. Once these wrappers are in place, HydroBOT can directly run the incorporated response models within its own consistent, unified workflow across disparate response models.\u003c/p\u003e \u003cp\u003eBy providing a single consistent interface to a range of response models, we avoid the need to manually run models separately and abstract their different interfaces, languages, and idiosyncrasies. Moreover, the design is highly modular, built to allow the integration of new response models with limited additional updates to HydroBOT itself. In some cases, e.g. proprietary software, HydroBOT may not itself run the response models, though this is less ideal. In that situation, the outputs of those models would need to be cleaned by the user, and the HydroBOT workflow could begin at the Aggregator step.\u003c/p\u003e \u003cp\u003eAt present, HydroBOT expects the user to provide input data in a format readable by the response model. Any data cleaning or formatting necessary to meet the response model requirements should be a preprocessing step. This allows HydroBOT to pass only paths to clean data, rather than the data itself, to the response models. Outputs can be saved to disk or retained in-memory for interactive use, depending on the user\u0026rsquo;s needs. The outcomes of the response models are then processed by the Aggregator to provide outcomes at scales relevant for management decision making (in time, space or value dimensions; Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e \u003ch2\u003e2.1.3 Causal networks\u003c/h2\u003e \u003cp\u003eCausal networks describe the topology of dependence among many drivers and outcomes (Mart\u0026iacute;nez, Toro \u0026amp; J. Le\u0026oacute;n, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Peeters et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). These networks provide useful graphical representation of the relationships between nodes (state variables) and links between them (dependencies) (see Figure S2). Beyond graphical representation, they can also extend the ability to assess a range of values beyond the response models themselves. Each link represents the existence of a relationship, and so defining those relationships mathematically allows causal networks to translate from initial inputs (e.g. rainfall, flows, or management actions) through to all values of interest (nodes). Thus, such networks are ideal for systems with multiple interacting relationships (Mart\u0026iacute;nez et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Peeters et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eHydroBOT provides a causal network for included response models, where available, to describe how their outputs arise from hydrology and how they relate to various levels of management-relevant outcomes. Like their associated response models, these networks are typically sourced externally to the workflow, and so depicted outside the data flow in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The causal networks enable: 1) visual representation of the complex inter-relationships between scenario inputs and outcomes across a range of objectives and 2) assessment of outcomes scaled along the value dimension (see Section \u003cspan refid=\"Sec7\" class=\"InternalRef\"\u003e2.1.4\u003c/span\u003e). The former aids transparency, elucidating the intentions and causal relationships behind the response models and is a useful device for communication alongside other final outputs. The latter allows outcomes to be quantified for individual (or sets of) values at different levels of complexity or groupings (e.g. fish breeding, individual fish species, native fish overall, or target fish condition), or at overarching sets of values such as environmental or economic groupings. This quantification provides a powerful assessment tool and the ability to identify synergies and trade-offs across interrelated values.\u003c/p\u003e \u003cp\u003eThe structure of the causal networks define the (sometimes) complex inter-relationships defined in the response models and subsequent impacts on larger-scale values. Many response models capture only proximate responses, e.g. whether flow requirements were met. In these cases, causal networks describing the hypothesized links to target management values allows HydroBOT to model those management values. The network allows moving from those proximate outcomes to other relevant values by specifying the mathematical form of the links from the response model outputs to other nodes. This is not specific to HydroBOT; any quantitative assessment of those relevant values needs specified relationships from the outcomes of detailed response models to dependent target outcomes.\u003c/p\u003e \u003cp\u003eThe specifications of the mathematical relationships underlying each link can be highly variable, depending on the state of the best available science or expert opinion. Some may be highly-detailed models, including hydrologic models linking rainfall to flows or dynamical models linking flows to species\u0026rsquo; populations. Other links might be specified with simpler linear models, averages, or even leaving the mathematical expression unspecified where relationships are hypothesized to exist but no information is available about the nature of the relationship.\u003c/p\u003e \u003cp\u003eThe availability and detail of causal networks is variable, but use is increasing for assessing ecological risk from stressors (e.g. Ohe et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Mart\u0026iacute;nez et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), social responses to hydrology (e.g. Frota et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) and hydrology itself (e.g. Pang et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). In cases where a causal network is not available for a particular location and response model, one can be created even if it is as simple as a conceptual diagram. Assessing the quality of knowledge around each link provides a powerful assessment of knowledge deficiencies and uncertainty in responses.\u003c/p\u003e \u003cp\u003eIn the absence of a causal network, HydroBOT can still analyse values at the scale returned by the response model. This analysis is likely beneficial; response model outputs are developed to be useful for some purposes and these can be assessed with HydroBOT. Where causal networks are available, even conceptually, they can be used to investigate outcomes at multiple value levels. The models implicit in the causal links can then be updated as more information is available. Indeed, in some cases with well-defined response models that link all the way from hydrology to desired management targets, the causal networks and the response models are the same, as they describe the pathways from driver (hydrology) through to relevant values. In these situations, describing the links in an explicit causal network is not necessary for the functioning of HydroBOT, though it is recommended to provide visualization of the causal relationships encoded in the response model.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003e2.1.4 HydroBOT Aggregator\u003c/h2\u003e \u003cp\u003eResponse model output is typically very granular in many dimensions because the best response models operate near the scale of the processes being modeled (Levin, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e1992\u003c/span\u003e; Holt \u0026amp; Chesson, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Gawne et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). In many cases, those processes (e.g. fish breeding, crop planting) individually occur at small spatial and temporal scales. This is the scale of the process itself, though these processes may be repeated and connected over much larger scales. For example, fish may breed across a basin and migrate among breeding locations but each female breeds in only one location at a given time. Moreover, outcomes from response models are typically at small value scales as well, e.g. capturing portions of the life cycle of a single fish species, rather than an overall outcome for all fish, or representing planting of particular crops, rather than overall agricultural output. The consequence is that there are potentially thousands of different modeled outcomes across time, locations, and values. This plethora of outcomes must then be aggregated to extract meaning at larger scales and condense the information for digestibility in management decision making.\u003c/p\u003e \u003cp\u003eThe HydroBOT Aggregator aggregates and scales information to resolutions useful for interpretation and planning or that capture larger-scale processes. Depending on the use, the desired resolution(s) may range from local, short-term responses of fine-grained outcomes to large spatial scales over longer time periods for high-level outcomes such as environmental condition (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). For example, a response model might assess a suite of hydrologic indicators at individual gauges, but outcomes may be desired for a range of ecological groups within larger management units or at the basin scale. Likewise, multiple indicators are typically required to meet objectives for proximate ecological values, such as fish breeding, of which many are required for each larger ecological value (such as native fish diversity).\u003c/p\u003e \u003cp\u003eThe HydroBOT Aggregator provides a robust, consistent aggregation approach along three dimensions (time, space, and value), while maintaining the ability to define those aggregation steps flexibly to meet the needs of the specific analysis and details of the inputs (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The Aggregator allows the user to define columns to aggregate (i.e. the outcomes from the response models), as well as any necessary grouping variables indicating e.g. scenarios, time, space, or value dimensions. As such, it is quite flexible, and can accept any dataframe where that information is specified. No internal changes are required to accept outputs from new response models. This flexibility is by design, enabling incorporation of new outputs from new response models or aggregating outputs from response models that are unable to be fully integrated. Moreover, this modularity means the Aggregator is run on the outputs of the response models, and so adjustments to the aggregation step do not require re-running response models, yielding large computational savings as aggregations are adjusted and tailored during the course of a project.\u003c/p\u003e \u003cp\u003eThe choice of aggregation steps and aggregation functions at each step are highly flexible. The Aggregator consists of a set of functions that scale along each of the three dimensions while handling dimensional idiosyncracies, such as time arithmetic, spatial awareness, and many-to-many links in the causal network. Overarching functions coordinate these dimensional scaling functions, allowing stepwise aggregation along a sequence of dimensions and scales (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). These coordination functions maintain awareness of all dimensions; e.g. they ensure scaling in the spatial dimension does not accidentally collapse along the value or temporal dimensions. The structure of the Aggregator functions provides these safeguards and capability for sequential multidimensional scaling, while giving the user control over the sequence of those steps and the scaling functions used at each step.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe primary arguments to the aggregation functions are lists of the sequence of aggregation steps and the function(s) to apply at each step. The choice of aggregation sequence and scaling functions will change depending on the question being asked and the nature of the processes being modeled. For example, a project focused on expected conditions might use means at most steps, while one focused on risk might use minima to capture the worst outcomes or essential species requirements. To give the user the ability to tailor these steps to different questions, each step can be along any dimension and multiple functions can be assigned at each step. The sequence of aggregation steps can interleave the dimensions, e.g. aggregate first through time to cover the period of the hydrograph, then along the value dimension to an intermediate level, then aggregate in space, then value again. For example, following Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, the user might want to aggregate from hydrologic indicator to life-cycle values, such as fish breeding, to species at each gauge, then aggregate those species outcomes up to catchment to assess performance of each species over a larger area, followed by aggregating to values comprising ecological groups, e.g. native fish, waterbirds, native vegetation, or ecosystem function.\u003c/p\u003e \u003cp\u003eHydroBOT provides a standard set of MDB-relevant spatial units for aggregation but can also accept any user-supplied polygons (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Aggregation along the value dimension follows the causal network, which is supplied by HydroBOT for included response models, though the user can specify other causal relationships if required. Supplied spatial units and causal networks are discussed in SI Section 9.2 with additional information about providing custom versions in the HydroBOT documentation. Whatever the spatial and causal relationships, HydroBOT provides the capacity to scale along them; it is the responsibility of the user to ensure those links and scaling functions are scientifically sound for the question at hand.\u003c/p\u003e \u003cp\u003eThe Aggregator allows the user to choose any aggregation function at each step. These aggregation functions should be considered carefully, as they should be appropriate to the processes being scaled and the outputs needed for management decision making. HydroBOT provides a standard set of functions (e.g. mean, max, min, median, geometric mean, harmonic mean, and spatially-weighted mean), but the user can also specify any other aggregation function, including custom-written functions tailored to particular analyses. For example, in some situations the user might want to know the proportion of indicators that pass across some area, or perhaps the average value of an abundance measure. In this case, a mean would be appropriate. In other situations, however, a single failure may be disproportionately important (e.g. \u0026lsquo;no loss\u0026rsquo; requirements), or perhaps a single passing value is sufficient for an ecological process to occur (e.g. bird breeding can occur anywhere in a catchment). These might be captured with a minimum and maximum, respectively. Additional discussion of aggregation function choice is provided in the Supplement (Section 9.3).\u003c/p\u003e \u003cp\u003eThe choice of aggregation at each step has large influence on the outcomes, and so HydroBOT does not prescribe the \u0026lsquo;appropriate\u0026rsquo; aggregation, instead giving that control to the user. Where aggregation choices are not clear, or to assess the impact of these choices, HydroBOT can be run with different sets of aggregation sequences, with the outcomes of each compared using the Comparer as if they were scenarios.\u003c/p\u003e \u003cp\u003eThe Aggregator self-documents into a yaml file, and its output datasets retain the full sequence and functions alongside each value, ensuring that values are always paired with their provenance and meaning.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section3\"\u003e \u003ch2\u003e2.1.5 HydroBOT Comparer\u003c/h2\u003e \u003cp\u003eThe HydroBOT Comparer is designed primarily to make comparisons between scenarios, including assessment and visualization of their differences. This component also provides generalized capacity to produce plots and other outputs such as tables using a consistent approach, even when not directly comparing scenarios. While the primary use of the Comparer is comparing Aggregator output, hydrographs and direct Response model output can also use this functionality.\u003c/p\u003e \u003cp\u003eThe functions within the Comparer can be divided into two main categories, those for analysis and those for plotting. Although in some instances presenting the absolute outcome values can be useful across scenarios, explicitly calculating comparisons (e.g. the absolute or relative difference between scenarios) provides distinct advantages. Difficulty in accurately simulating a complex system means that comparisons between scenario outcomes can be more useful and accurate because any bias in assumptions applies to all outcomes and the focus moves from the total level to the change between scenarios (Holt et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The best method for comparing will vary depending on the quantities being compared and the intended use of the comparison, so several common default options (e.g. differences, relative change) have been developed along with the flexibility for the user to define alternatives.\u003c/p\u003e \u003cp\u003eThe comparison functions provide the ability to choose a baseline level for comparison, which may be one of the scenarios, but also may be a reference dataset or a scalar value. For example, we might want to compare a set of outcomes for climate scenarios to a \u0026lsquo;no change\u0026rsquo; climate scenario, to historical observations, or to a mean value. Output values are calculated relative to the defined baseline level using either default functions for the absolute or relative difference, or any other user-supplied function. These functions can be applied to any dataframe, but a major advantage of HydroBOT is including them within the plotting functions. This allows the generation of comparison plots from raw Aggregator outputs without the need for subsequent data calculations by the user and avoids potential errors that arise from repeated or forgotten data transformations in an analysis workflow.\u003c/p\u003e \u003cp\u003eThe plotting functions in the Comparer provide a wide range of possible plots, with standardization providing consistency across plot types and through the project, ensuring values plotted are robust and interpretable. Different purposes require different sets of outputs; for instance, maps are particularly useful for visualizing geographic patterns, tables and graphs typically provide more precise ability to assess impacts on values, and timeseries plots are useful for visualizing climate trajectories.\u003c/p\u003e \u003cp\u003eWhile potential visualizations and comparisons vary widely depending on the intended use, the Comparer standardizes data cleaning and processing for each plot, as well as aesthetics and plotting approaches. Like the Aggregator, a key advantage of the Comparer is the standardization and dimensional safety over repeated data manipulations and processing. By standardizing data cleaning within the Comparer, we avoid losing information or performing unsupported data manipulations and so ensure the quality and meaning of the outputs. For example, because the plotting function was designed to incorporate temporal, spatial, and value dimensions, it identifies and prevents accidental overplotting or misleading lumping of data across dimensions.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Implementation\u003c/h2\u003e \u003cp\u003eHydroBOT is available as an R package, which provides a suite of functions representing the steps in the architecture. A documentation website provides additional worked examples and a fuller exploration of capacity and use-cases. Because translating from hydrologic scenarios to various responses is a general problem in water management, we expect the ways in which HydroBOT is used will be highly variable. Thus, by providing a general structure, users can target particular questions and analyses needed for a given question. Although the initial impetus for creating HydroBOT was to assess climate scenarios, its use in practice can be far more general. Any hydrograph can be assessed, and any set of scenarios represented by hydrographs can be compared, provided a response model exists.\u003c/p\u003e \u003cp\u003eAs a result of an initial scan of available response models in the Murray-Darling Basin, HydroBOT development proceeded with a single response model (written in Python), as this was the only module that met requirements (see Section \u003cspan refid=\"Sec14\" class=\"InternalRef\"\u003e3.2\u003c/span\u003e), but with an architecture designed to allow modular integration of additional response models as they become available. The scan accentuated a critical feature of this modularity; HydroBOT is designed to incorporate and standardize models written in various languages and with a wide range of input needs and outputs. HydroBOT achieves this by wrapping those other tools so as to make their differences as invisible to the user as possible.\u003c/p\u003e \u003cp\u003eIn the case of Python modules, HydroBOT uses the \u003cem\u003ereticulate\u003c/em\u003e R package (Ushey, Allaire \u0026amp; Tang, \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) in combination with small amounts of internal Python code to call Python modules directly. The Python code internal to HydroBOT performs limited cleaning and translation to prevent passing large objects between languages while ensuring that any idiosyncrasies in module inputs and outputs are handled consistently. Python dependencies (and Python itself) are automatically installed on first use of Controller functions that call Python modules unless they already exist on the user\u0026rsquo;s system. This approach provides essentially invisible Python for most users, while providing flexibility for the user to provide their own Python environment if desired. Modules in other languages are not yet available, but the key requirement is that they be available in a format that is scriptable. In that case, HydroBOT will provide small setup and cleanup functions as with the Python modules and wrapper functions to call these modules.\u003c/p\u003e \u003cp\u003eThe modularity of HydroBOT means it can be run stepwise, with the user calling the relevant functions at each step (Figure S1). The outputs of each step can be saved or returned directly to an interactive session or both. Typically, selected outputs would be saved for reproducibility and speed unless the project is small enough to retain a comprehensive set in memory or re-run quickly in a notebook. There are also wrapper functions that allow running the HydroBOT flow from Controller through Aggregator, which are extremely useful for large runs or remote runs on batching services. One particularly useful wrapper provides the ability to run from a yaml parameter file providing function arguments. This function allows the use of a default file and a \u0026lsquo;modified run\u0026rsquo; file, making it ideal for holding many parameters constant at a default for a particular set of analyses and only changing a subset on a per-run basis in a smaller file. It also takes command-line or R list arguments, making it a flexible solution to run HydroBOT from the command line, in scripting contexts, or notebooks including Databricks and Quarto (Allaire et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe metadata saved at each step in the process is a yaml file with parameters that are a superset of those needed to run HydroBOT. Thus, an identical run can be produced by running HydroBOT using an output metadata file as an input parameter file, supporting reproducibility. Additional information is included in these exported yaml files documenting versions of HydroBOT and response models, as well as capturing any available metadata pertaining to the input scenarios. Even if a run is started in the middle, such as to re-run aggregation in a different way, the metadata for that run captures the metadata for the previous steps, ensuring that outputs at every stage are tagged with full provenance information.\u003c/p\u003e \u003cp\u003eIn practice, HydroBOT functions are primarily run in one of two ways. Interactive investigation of relatively small sets of hydrographs and small numbers of scenarios can be done in notebooks (typically Quarto) or simple R scripts. Larger investigations typically would be run on remote computers as part of batching systems, whether HPC, Azure, AWS, Databricks, or similar, with the outputs at the end of the aggregation (and potentially each step) returned. Through the Aggregator, all operations can proceed in parallel over scenarios, and HydroBOT provides internal capacity for this using the \u003cem\u003efuture\u003c/em\u003e package (Bengtsson, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe Comparer step would usually not be run as part of this larger batching, but considered interactively for two reasons. First, the Comparer necessarily looks across scenarios, and so breaks parallelization. Conducting comparisons interactively is typically not excessively CPU- or memory-intensive provided aggregation is sufficient to yield interpretable figures. Second, the primary goal of the Comparer is to produce usable, interpretable outputs. Arriving at a set of meaningful outputs is an iterative process that rarely will be precisely known \u003cem\u003ea priori\u003c/em\u003e, and so working through this step interactively is usually necessary. If, however, HydroBOT is being used for ongoing monitoring of the same analyses (e.g. a \u0026lsquo;dashboard\u0026rsquo;), then the first iteration could be done interactively, with subsequent uses incorporating those settings written into a script, Shiny app, or similar to auto-generate the same figures.\u003c/p\u003e \u003cp\u003eIf the outputs of the Aggregator step (and so the inputs to the Comparer) are large enough to require high processing or memory, it is unlikely to be simplified enough to make interpretable figures. Additional aggregation steps to better target desired figures (or subsetting of the data to focus on a particular aspect) should be undertaken. If, after careful consideration, a user does need to make processing or memory-intensive figures, these issues are due to the amount of data being plotted and so will not be unique to HydroBOT. Methods used for any other situation involving plotting large amounts of data with \u003cem\u003eggplot2\u003c/em\u003e should then be considered, such as remote, high performance, or cloud computing. The first step, though, should be to ask whether additional aggregation is needed.\u003c/p\u003e \u003cp\u003eHydroBOT was originally designed for analyses of management questions in the Murray-Darling Basin and so, along with the functions to perform the analyses, it also provides a standard set of spatial data, including the boundary of the MDB itself, river lines, gauge locations (at which hydrographs may be available), and various management units (see SI Section 9.2). All geographic information in HydroBOT uses spatial dataframes using the \u003cem\u003esf\u003c/em\u003e package implementing simple feature geometries in R (Pebesma \u0026amp; Bivand, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Thus, any user-supplied \u003cem\u003esf\u003c/em\u003e objects can be used for spatial aggregations.\u003c/p\u003e \u003cp\u003eSome example hydrograph data are also included for testing and demonstration. HydroBOT provides a clean causal network for included modules, described in more detail in Section \u003cspan refid=\"Sec6\" class=\"InternalRef\"\u003e2.1.3\u003c/span\u003e and SI Section 9.2. Analogous elements can be constructed for other spatial units or responses to extend functionality within or outside the Murray-Darling Basin. Supplying new causal networks only requires providing dataframes or lists of dataframes with the needed levels and their relationships.\u003c/p\u003e \u003cp\u003eAll plotting functions in the Comparer are built on the on the \u003cem\u003eggplot2\u003c/em\u003e R package (Wickham, \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). While the argument structure is modified to provide the necessary dimensional safety, processing consistency, and simplification of some common complex data processing, much of the ggplot functionality is available. Most importantly, the plotting functions produce ggplot objects that can be further modified by the user. This approach gives the user great flexibility in tailoring plots to the visualization needs of a particular project.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Co-design and scoping for management relevance\u003c/h2\u003e \u003cp\u003eTo ensure HydroBOT meets management needs and is trusted by management users, a team was created at project inception consisting of primary HydroBOT developers, ecologists, hydrologic modelers, and managers. Collaboration among this team ensured decisions about HydroBOT construction reflected the needs of the end-users. As implementation proceeded, the collaboration provided a mechanism by which goals could be adjusted or the implications of decisions discussed and understood. By having this window into HydroBOT development, the managers obtained a much more granular view of how it operates, along with its capabilities and limitations. This view into HydroBOT development and reasoning was crucial to avoid becoming a \u0026lsquo;black box\u0026rsquo;, and enhances trust, interpretability and usability of the HydroBOT outputs.\u003c/p\u003e \u003cp\u003eKey to the collaborative development was identification of the response models to include. These models needed to provide information about values relevant to water management decisions. Early in the development process, a wide scan was taken to identify candidate response models. This scan considered models being used or developed within the MDBA itself, other government agencies (both federal and state), and externally. A number of response models were found to be available, but many were outdated or highly manual, and their use was patchy both within and across management agencies (Holt et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Ultimately, only one was suitably modern and well-developed to include in HydroBOT, the Environmental Water Requirements (EWR) Tool (MDBA, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Other in-development modules were identified as candidates to include during the development of HydroBOT, including economic and social models. Further, this scan identified values that are mandated management targets for which there were no available or in-development response models, highlighting areas needing future work.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Advantages and tradeoffs\u003c/h2\u003e \u003cp\u003eIntegrating new modules and shifting workflows into a new package always comes at a cost, and using HydroBOT is no exception. Repeatedly doing the same operations suggests that those operations should be done in a software package, not a script (Wickham \u0026amp; Bryan, \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). This is particularly true if many of those operations are checks or data manipulations to meet certain conditions, e.g. to handle multi-dimensional aggregation or compare scenarios in multiple ways (Wilson et al., \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2014\u003c/span\u003e), as they are here. Thus, if a response model already produces outputs that are essentially finished, or if an analysis is a one-off that is unlikely to be repeated or adjusted, HydroBOT may not be needed. However, in our experience, if a response model produces outputs that provide useful management information, the running and onward processing are likely to be desired across a range of projects, as they have been for the EWR tool discussed here.\u003c/p\u003e \u003cp\u003eIt is not unusual for a single project to require hundreds to thousands of runs, particularly when small tweaks and changes to the analysis are considered. These adjustments are a frequent source of error, particularly when multi-dimensional aggregation is involved, with quality control rapidly becoming unwieldy or impossible (Wilson et al., \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Indeed, it is just these issues, compounded across different sets of scenarios and different management questions, that led to creation of HydroBOT as an end-to-end package. Allowing HydroBOT to handle that processing in tested, documented functions while keeping the analysis script short and clear provides advantages for readability, time, consistency, and correctness. Identical processing of a set of scenarios, as well as internal parallelization, provides similar advantages in terms of consistency and speed.\u003c/p\u003e \u003cp\u003eIntegrating new response models into HydroBOT requires wrapper functions that send relevant arguments to the response model, and potentially (as is the case with the EWR tool), clean up the output of those models for onward processing by the Aggregator. These wrappers should typically be lightweight, performing only simple data formatting. However, it is also possible to forego complete integration and use the Aggregator and Comparer on the outputs of such response models. This approach would lose some advantages (e.g. scripting, automated metadata, parallelization) but can be useful if the input integration is difficult, as might be the case with proprietary models or those with only GUI (non-scriptable) interfaces. The outputs would still need to be cleaned into an appropriate format for the Aggregator, though this process is rarely more onerous than would need to happen for analysis outside HydroBOT.\u003c/p\u003e \u003cp\u003eThe Aggregator simply needs dataframes with columns of output values, any required grouping variables (e.g. scenarios), and those that link the values to the relevant spatial, temporal, or value dimension. This same information would be needed by most analyses. Similarly, while a new response model is likely to need its own causal network this would be needed for any analysis that aggregated along the value dimension. Spatial units are likely to be shared across models unless HydroBOT is being used in wholly-different locations. If they are needed, that would again be true of any spatial analyses with or without HydroBOT. Both causal networks and spatial data can be passed as arguments rather than needing to be incorporated into HydroBOT itself.\u003c/p\u003e \u003cp\u003eHydroBOT is not the answer for all projects and there are certainly tradeoffs. Our belief and experience is that the marginal cost of its implementation, combined with the advantages HydroBOT brings in terms of safety, reproducibility, and speed, make those tradeoffs fall in favor of HydroBOT quite quickly, when compared with conducting the same analyses outside HydroBOT (e.g. with a set of bespoke scripts). This investment is particularly worthwhile for informative response models with clear management importance such as the EWR tool demonstrated here. Once HydroBOT is set up, using it for new scenarios and new questions is straightforward and robust, enabling rapid exploration of new avenues of analysis not otherwise possible.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. HydroBOT demonstration","content":"\u003cp\u003eHere, we present example results produced using HydroBOT, a set of demonstration hydrological scenarios, and the EWR tool for the response model (see Section \u003cspan class=\"InternalRef\"\u003e3.2\u003c/span\u003e for a description of the EWR module and its associated causal network). These results demonstrate the flexibility and robustness of HydroBOT, which give it the capability to produce rigorous but interpretable and management-relevant results in a reproducible way. The scenarios used here are developed to illustrate HydroBOT functionality with perturbed hydrographs that do not represent real flow scenarios. The plots shown here demonstrate several of the plot types designed to illustrate some potential visualizations relevant to these scenarios. Many more visualizations are possible due to the flexibility of the plotting functions, giving users the ability to generate plots relevant to different projects and questions.\u003c/p\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003e3.1 Demonstration scenarios\u003c/h2\u003e\n \u003cp\u003eOur input data consists of hypothetical flow timeseries generated from historical hydrographs at 43 gauges. These gauges fall within the Lachlan, Macquarie\u0026ndash;Castlereagh, and Namoi Sustainable Diversion Limit (SDL) units of the Murray-Darling Basin, Australia. These SDL units have detailed Long Term Watering Plans, and so well-specified environmental water requirements and causal networks (described in Section \u003cspan class=\"InternalRef\"\u003e3.2\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eWe develop a set of simple scenarios that capture two sorts of changes that may be commonly represented in management analyses. First, we consider scaled flow throughout the period of the hydrograph, representing overall increases or decreases in flow as might occur from changes in large-scale climate patterns. Second, we consider short-duration additions to flow, representing periodic pulses of change as might happen from targeted interventions. Each scenario characterizes all water in the system including natural inflows, extraction, and release of environmental water, yielding a complete hydrograph (SI Section 9.4 shows a selected subset). These do not reflect realistic future scenarios but provide an avenue to test and illustrate the capabilities of HydroBOT.\u003c/p\u003e\n \u003cp\u003eTo scale flow, we apply nine flow multipliers, ranging from 0.5 to 2.0, to the historical hydrographs (Table 1). We refer to these as \u0026lsquo;climate\u0026rsquo; scenarios, reflecting a common representation where entire hydrographs might shift to represent climate change. Though these scenarios are not derived from climate models and do not represent hypothesized climate change, they do fall within the range of predictions from GCMs (Devanand et al., \u003cspan class=\"CitationRef\"\u003e2025\u003c/span\u003e). To achieve pulsed change within each of the \u0026lsquo;climate\u0026rsquo; scenarios, four flow additions were applied including 1) no addition, 2) addition of 250 ML/d, 3) addition of 6500 ML/d, and 4) addition of 12000 ML/d (Table\u0026nbsp;1). These additional flows were added throughout the period of September to December to target the characteristics of specific environmental water requirements, as might be done by water managers. For example, there are 202 EWRs in the three catchments used here with flow thresholds above 250 ML/d starting in September, which the smallest addition would ensure are met. We refer to these scenarios as \u0026lsquo;climate adaptations\u0026rsquo; because management options are often available in the form of altering water availability for short time periods through mechanisms like water releases, though the options here do not represent proposed actions. These scenarios should not be interpreted as potential climate impacts or adaptations, but instead as different ways flows might change (multiplicative or additive) and different magnitudes of change. The \u0026lsquo;E1\u0026rsquo; scenario (Table\u0026nbsp;1) is the observed historical data with no climate change or adaptation options applied, and so is a natural baseline for comparisons.\u003c/p\u003e\n \u003cp\u003eTable 1: Demonstration scenarios are a factorial combination of \u0026lsquo;climate\u0026rsquo; (scaled flow) and \u0026lsquo;adaptation\u0026rsquo; (pulsed additions). \u0026lsquo;Climate\u0026rsquo; scenarios included in this demonstration were produced by applying a flow multiplier to historical flows. \u0026lsquo;Adaptation\u0026rsquo; options were applied to each climate scenario with additional flows added throughout the period of September to December. For ease of display in some figures, we provide alpha codes for the \u0026lsquo;climate\u0026rsquo; changes and numeric codes for the \u0026lsquo;adaptations\u0026rsquo;. In most plots, we present the \u0026lsquo;A\u0026rsquo; (0.5 x historical flow), \u0026lsquo;E\u0026rsquo; (1.0 x historical flow and \u0026lsquo;I\u0026rsquo; (2.0 x historical flow) scenarios, with adaptions 1 (no flow addition), 2 (addition of 250 ML/d) and 3 (addition of 6500 ML/d), with \u0026lsquo;adaptation\u0026rsquo; scenario 4 and the intermediate \u0026lsquo;climate\u0026rsquo; multipliers only appearing in heatmaps and smoothed fits.\u003c/p\u003e\n \u003cdiv align=\"char\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\u0026nbsp;\u003ctable id=\"Tabb\" border=\"1\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eClimate code\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFlow multiplier\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAdaptation code\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFlow addition (ML/d)\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e250\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6,500\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e12,000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n \u003ch2\u003e3.2 Environmental Water Requirements (EWR) Tool\u003c/h2\u003e\n \u003cp\u003eThe EWR tool, forming the core of this demonstration, is a response model for hydrologic indicators of ecological outcomes. The EWR tool is written in Python and is used internally by the MDBA, the states of New South Wales and Queensland for water planning, and other interested stakeholders (Sheldon et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e; MDBA, \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). It models the response of hydrologic indicators established in Long Term Watering Plans (LTWPs) and Environmental Water Management Plans (e.g. DEW, \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e; DPIE, \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e; DRDMW, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e; DELWP, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eThe EWR tool holds databases of the environmental water requirements (EWRs; the indicators), characterized by the discharge, frequency, timing and duration of flows (SI Table S3). These indicators were developed based on hypothesized relationships to the ecological objectives of the MDB, which protect or enhance environmental assets and ecosystem functions that are valued based on ecological significance (DPIE, \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e; Sheldon et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). The EWR tool assesses hydrologic indicators, determining whether spatially explicit flow timeseries meet each EWR (indicator) at each gauge (illustrated in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). The precise definitions for each EWR differ at each gauge, due to the unique hydrology and channel morphology. For example, a small fresh for 10 days is required every year between October and April to meet indicator EWR SF1 (small fresh 1), but the discharge defined as a \u0026lsquo;small fresh\u0026rsquo; differs between gauges. Moreover, the expected ecological outcome for achieving SF1 might vary between gauges or catchments due to different impacts of small freshes on local species or ecological processes, as defined in the LTWPs.\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003cp\u003eThe EWR tool returns binary responses of hydrologic indicators. However, the LTWPs describe not only the indicator requirements assessed by the EWR tool, but also how those hydrologic indicators are expected to influence both proximate and larger-scale ecological outcomes according to the best available information from water managers, ecologists, scientific publications, and analysis of gauged and modeled flows (e.g. DPIE, \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e; Lobegeiger \u0026amp; Prior, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). We extracted these expected links from tables within the LTWPs for each planning unit, producing causal networks defining the expected relationships between the EWRs and several higher levels of ecological values. These clean causal networks are provided in HydroBOT (all links for a single gauge are shown in SI Figure S2). While extracting these causal networks was a significant undertaking, it may not be for other response models that capture ecological values or if the hypothesized links are developed from the outset to be conceptualized causally and are machine-readable.\u003c/p\u003e\n \u003cp\u003eCausal networks link the hydrologic indicators produced by the EWR tool to the values for which the system is being managed. Thus, these networks greatly expand the utility of the EWR tool by allowing modeling of a range of ecological values dependent on those indicators. These values are defined at several levels, from components of the life cycle of single species to broad ecological groupings to whole-community outcomes at 10- or 20-year target dates (see SI Section 9.1 and Figure S2). This increases the transparency in the causal relationships that underpin HydroBOT results and builds understanding and trust in the outcomes. For example, the SF1 indicator described above might contribute to multiple ecological objectives pertaining to native fish (ecological objectives NF1-9; e.g. NF1\u0026thinsp;=\u0026thinsp;No loss of native fish species), native vegetation (NV1), and ecosystem functions (EF1-5). These ecological objectives are defined to support the completion of all elements of a life cycle of an organism or group of organisms (taxonomic or spatial) to achieve a \u0026ldquo;defined goal for a state, condition or characteristic of an ecological asset or function\u0026rdquo; (DPIE, \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e glossary). Outcomes for ecological objectives are then linked in the causal network to five large-scale ecological groups (native fish, native vegetation, waterbirds, other species, and ecosystem functions, MDBA, \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e) and are associated with long-term targets (5-, 10-, and 20-year). This structure not only provides a visual definition of the links in the LTWP, it also enables assessment of outcomes that are directly equivalent to the LTWP management strategies.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n \u003ch2\u003e3.3 Demonstration aggregations\u003c/h2\u003e\n \u003cp\u003eThe aggregation step is critically important for outputs and should be tailored to the values being aggregated and the questions being asked, as described in Section \u003cspan class=\"InternalRef\"\u003e2.1.4\u003c/span\u003e. The examples used here are designed for a simple demonstration; many other aggregations are possible, with the capacity explored in-depth on the documentation website. For any analysis, aggregation choices should be considered carefully (see SI Section 9.3). Since our purpose here is a simple demonstration, we use arithmetic means for nearly all aggregations. The exception is the first value dimension step (following temporal aggregation) due to the structure of the EWRs. In the EWR tool, the returned outcomes (and so finest scale on value dimension) gives achievement of multiple versions of EWR indicators, typically representing different leniency in flow conditions required to pass. Here, we consider that if any of the versions of an EWR is achieved, that EWR is achieved. We thus use a maximum for that aggregation step; first we obtain a time-average of each EWR version, and then we take the maximum of each of those versions (typically 1\u0026ndash;4 values) to obtain the most-lenient outcome for that EWR. Specifying a different aggregation sequence would yield a different result and have different meaning (e.g. a minimum would require the most-strict version to pass, or we might skip this step entirely and treat the versions as fundamentally distinct indicators).\u003c/p\u003e\n \u003cp\u003eThough we use means for other steps, their interpretation changes depending on the level of aggregation. For simplicity, we assume the value of each outcome represents some \u0026lsquo;condition\u0026rsquo;, and the \u0026lsquo;condition\u0026rsquo; at a particular level can be assessed as the mean of the conditions contributing to it. For example, the condition of detailed ecological values such as fish spawning or bird fledging might simply be the proportion of EWRs that contribute to each value that are achieved. Then, the condition of ecological groups, e.g. native fish, might be the mean condition over all the NF1\u0026hellip;n values, despite those depending on different numbers and sets of EWRs. This captures the idea that native fish condition improves when the life-cycle components captured by constituent lower-level values are met, whether it takes 1 or 10 EWRs.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\n \u003ch2\u003e3.4 Demonstration outputs\u003c/h2\u003e\n \u003cp\u003eHere, we provide example illustrations of HydroBOT outputs for the demonstration scenarios and aggregations described above. Just as the aggregation steps demonstrate a small part of the capacity of HydroBOT, tailored for this example, so too do these figures demonstrate example capacity for a small subset of the potential outputs. Many more figures might be generated to address other questions or needs, with further detail available on the documentation website for HydroBOT. Moreover, because the outputs of the plotting functions are \u003cem\u003eggplot\u003c/em\u003e objects, additional bespoke tailoring is straightforward using the most common plotting package in R.\u003c/p\u003e\n \u003cdiv id=\"Sec17\" class=\"Section3\"\u003e\n \u003ch2\u003e3.4.1 Spatial relationships\u003c/h2\u003e\n \u003cp\u003eComparison of the direct outputs from the response model (in this example, pass/fail of EWRs) without any aggregation along the value dimension gives an overview of ecologically-relevant hydrologic outcomes at different locations (e.g. gauges) or areas (e.g. SDL units) to different scenarios (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e; noting that these scenarios do not represent real climate simulations or proposed adaptations). Three focal \u0026lsquo;climate\u0026rsquo; scenarios (A, E and I; 0.5x, no change, and 2x the historical base level, respectively) and three adaptation options (1, 2 and 3; additions of 0, 250 and 6500 ML/d, respectively) affect each SDL unit differently, with each catchment having some scenarios in which they have higher EWR achievement than the others. In all situations, the \u0026lsquo;climate\u0026rsquo; (A, E, I) scenarios have less of an impact than the \u0026lsquo;adaptation\u0026rsquo; (1, 2, 3) scenarios, though neither is reflective of expected change in these dimensions. The point here is simply that different types of changes to the hydrograph can affect locations differently. These outcomes are not linked to ecological values here (e.g. ecosystem functions or environmental assets) and so each hydrologic indicator (EWR) at each gauge is represented with equal importance, whether it is required for all ecological values or just one.\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec18\" class=\"Section3\"\u003e\n \u003ch2\u003e3.4.2 High-level scenario comparisons\u003c/h2\u003e\n \u003cp\u003eWe use the causal networks to link response model outputs to ecological outcomes and investigate how changes in flow affect ecological values at several levels of the \u0026lsquo;value\u0026rsquo; dimension. Thus, we can compare how native fish, native vegetation, waterbirds, other species, and ecosystem function targets are likely to respond to our hypothetical set of multiplicative and additive changes to the hydrograph (\u0026lsquo;climate\u0026rsquo; and \u0026lsquo;adaptation\u0026rsquo; scenarios; Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e). Indeed, one of the primary purposes of HydroBOT is to provide this assessment of how high-level values respond to different scenarios. Such a summary collapses thousands of datapoints (EWR outcomes or proximate ecological responses at every gauge) across many scenarios into digestible yet scientifically-robust visualizations. At a glance, we can see in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e panel a that Ecosystem function outperforms Native vegetation in the halving (A) and no-change (E; historical hydrograph) scenarios, but is less sensitive to changes between scenarios. Halving water (moving from E to A) reduces Native fish outcomes more than doubling water (E to I) improves them, while the opposite is true for Waterbirds.\u003c/p\u003e\n \u003cp\u003eThe choice of aggregation sequence can be interrogated with the Comparer to better understand their implications. For example, aggregating the \u0026lsquo;versions\u0026rsquo; of the EWRs using the minimum (i.e. the strictest requirement for a given EWR) reduces the condition of all outcomes (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e panel b) compared to the case with the maximum in panel a. Using the median for aggregations at every step following an initial time average (panel c) yields much lower outcomes for the historical (E) and halving (A) climate scenarios, but idiosyncratic patterns in the doubling (I) scenario, with Waterbirds and Native Vegetation much lower than when using the means, while Other species is higher. These are relatively crude changes for the purpose of illustration; more complex aggregation functions or different functions used at different stages are likely appropriate in many situations. These choices should reflect the underlying biology (e.g. how do fish outcomes arise from the different set of flow conditions) and analysis questions (e.g. risk \u003cem\u003evs.\u003c/em\u003e resilience) and the consequences of these choices can be investigated as illustrated in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e or with other Comparer visualizations (see SI Section 9.3).\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec19\" class=\"Section3\"\u003e\n \u003ch2\u003e3.4.3 Values at multiple levels of detail\u003c/h2\u003e\n \u003cp\u003eHigh-level outcomes are essential, but managers need to drill down to understand how they arise, including comparing lower-level values, spatial areas, and the scenarios themselves. While there are many ways to display this information (See SI Section 9.6.1 for spatial analysis of each step), Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e consolidates a large amount of information across scenarios, planning units, and multiple value scales. Each set of colors (and row of panels) represents outcomes at the scale of large ecological value groups. Within each bar, the different shades represent single smaller-scale ecological values (e.g. WB1; Maintain the number and type of waterbird species) contributing to the larger value group, with the heights of these shades being the proportion of the constituent hydrologic requirements that are met (EWRs). Because the overall bar heights are a sum of these proportions, we provide a dashed line as reference that shows the situation where all relevant EWRs are met. This reference is necessary because the number of EWRs contributing to each small-scale ecological value varies, as does the number of small-scale ecological values contributing to the larger ecological value groups and spatial units. Dividing total bar heights by this line would thus give the proportional \u0026lsquo;condition\u0026rsquo; score for the larger groupings.\u003c/p\u003e\n \u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e provides a useful overview of the impact of different scenarios on large ecological value groups across space, while also retaining the ability to assess more proximate constituent ecological values. Each shade represents a unique smaller-scale ecological value, and so this view shows how those smaller values contribute to the overall condition at the larger scale without undue complexity. If an analysis needed to examine these contributions more explicitly, a different figure might be used to accentuate the lower-level environmental objectives more clearly (e.g. Figure S14, Figure S15).\u003c/p\u003e\n \u003cp\u003eThough the scenarios here are only demonstrations, they illustrate the sorts of comparisons that are possible. For example, WB1 (Maintain the number and type of waterbird species) and WB2 (Increase total waterbird abundance across all functional groups) are only achieved to any appreciable extent in the Macquarie\u0026ndash;Castlereagh in the high water addition (3) scenarios, while they are met more generally in the Namoi, and show an interaction with the \u0026lsquo;climate\u0026rsquo; scenarios in the Lachlan (met in \u0026lsquo;adaptation\u0026rsquo; scenarios 1 and 2 only in the doubled water \u0026lsquo;climate\u0026rsquo; scenario). At the larger ecological value group scale, we can see that the effect of halving water (moving from the baseline E1 scenario to the A1 scenario) has a disproportionately greater impact on Native fish than doubling water (the I1 scenario) in the Macquarie\u0026ndash;Castlereagh, while the Lachlan and Namoi show slightly larger impacts of doubling than halving.\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec20\" class=\"Section3\"\u003e\n \u003ch2\u003e3.4.4 Quantitative scenario descriptions\u003c/h2\u003e\n \u003cp\u003eScenarios can often be described quantitatively, most obviously here in the flow multipliers used for \u0026lsquo;climate\u0026rsquo; and flow additions used for \u0026lsquo;adaptation\u0026rsquo;, though these can also be combined to yield total changes in daily flow volumes. This quantification makes additional visualizations possible, as well as quantitative analyses to explore the impact changing water has on changing ecological values. The next three figures provide examples of different ways this might be done.\u003c/p\u003e\n \u003cp\u003eCalculating comparisons from a baseline clarifies disproportionate impacts and thresholds where management actions could be most effective. Here, we represent both our \u0026lsquo;climate\u0026rsquo; (multiplicative) and \u0026lsquo;adaptation\u0026rsquo; (additive) scenarios on a single axis by quantifying the difference in mean flow from historical hydrograph (E1) using the baselining functionality provided by HydroBOT (see Section 9.5). Plotting the proportion of EWRs achieved for each of the Native fish ecological values shows how that EWR achievement is related to changes in the overall levels of flow. One striking feature of Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e is that the lines do not smoothly increase; EWR achievement is not simply a direct relationship to discharge. Instead, the timing matters. There are particularly steep slopes between the \u0026lsquo;1\u0026rsquo; (no-addition baseline) and \u0026lsquo;2\u0026rsquo; (+\u0026thinsp;250 ML/d) \u0026lsquo;adaptations\u0026rsquo;. The application of water at those particular times is thus disproportionately impactful in these example scenarios. Indeed, the \u0026lsquo;A2\u0026rsquo; scenarios yield better outcomes than the \u0026lsquo;E1\u0026rsquo; for nearly all objectives even though they have less water. The same is true for \u0026lsquo;E2\u0026rsquo; compared to \u0026lsquo;I1\u0026rsquo; in the Macquarie\u0026ndash;Castlereagh and Namoi. Though these scenarios are not developed from real adaptation proposals, they identify real impacts of the timing of water additions. Such analyses can identify highly efficacious points to add water to the system, provided they represent real responses rather than model artefacts.\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003cp\u003eWith a large number of scenarios, we can characterize how values respond along axes relevant to how those scenarios change. Using smoothed fits, we show trends in the performance of ecological values (and groups thereof) in response to changes in flow from both \u0026lsquo;climate\u0026rsquo; and \u0026lsquo;adaptation\u0026rsquo; scenarios (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e, SI Figure S12). We use baselining (see Section 9.5) to show the relative change on both the x (flow) and y (response) axes. Considering these relative changes on a log-log axis then illuminates whether doubling flow has the same impact as halving, and whether doubling (or halving) flow doubles (or halves) response. Thus, any slope other than 1:1 on the log-log axes represents a disproportionate impact of increases \u003cem\u003eversus\u003c/em\u003e decreases in flow, and whether ecological condition shifts proportionately with flow.\u003c/p\u003e\n \u003cp\u003eThe responses seen in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e vary between SDL unit (columns) and ecological values (rows), indicating different sensitivity to both \u0026lsquo;climate\u0026rsquo; and \u0026lsquo;adaptation\u0026rsquo; flow conditions across space and among ecological groupings. For example, Ecosystem function responds disproportionately positively on a multiplicative scale to flow across the range in the Lachlan (and so doubling flow more than doubles ecological condition). Ecosystem function responds approximately proportionately to flow changes in the Namoi. Ecosystem function and Native fish respond more strongly to decreased than increased flows in the Macquarie\u0026ndash;Castlereagh, while all other groups are both more sensitive to increases. By using smoothed fits, we can identify thresholds or areas of flow change that are disproportionately more important (i.e. where small changes in flow can yield large shifts in condition). Though the scenarios here do not represent real changes, these results do show disproportionate change across a range of flows; whether that is also true for realistic ranges would require a similar analysis of meaningful scenarios.\u003c/p\u003e\n \u003cp\u003eThe response to \u0026lsquo;adaptation\u0026rsquo; options (simulated seasonal water additions; difference between the lines in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e panel b) tends to be much stronger than the response to \u0026lsquo;climate\u0026rsquo; (multiplicative changes to the whole hydrograph; slopes), even though the shift in total water may be lower (particularly for the baseline \u0026lsquo;1\u0026rsquo; and +\u0026thinsp;250 ML/d \u0026lsquo;2\u0026rsquo; adaptations; see x-axis of Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e). Moreover, the adaptation options are not equally sensitive to the climate scenarios; adaptation options \u0026lsquo;2\u0026rsquo; and \u0026lsquo;3\u0026rsquo; show less change in response to climate than \u0026lsquo;1\u0026rsquo; (no adaptation) (Figure S12). Thus, these \u0026lsquo;adaptations\u0026rsquo; are increasing resilience to holistic \u0026lsquo;climate\u0026rsquo; flow changes. Although these scenarios do not represent true adaptation options or climate scenarios, this shows that such changes to resilience are possible with targeted interventions, and HydroBOT provides the capacity to investigate them.\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003cp\u003eScenarios will often be defined along more than one axis. In this demonstration, we define a flow multiplication axis as a proxy for \u0026lsquo;climate\u0026rsquo; shifts, and a flow addition axis as a proxy for \u0026lsquo;adaptations\u0026rsquo;. Similarly, values often respond to several different aspects of the flow regime, for example the mean flow and the flow variance or the return interval of floods or low-flow periods. Heatmaps or contour surfaces provide a powerful tool for visualizing these interacting responses (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e, SI Figure S13). By examining multiple driver axes, such plots can highlight important interactions and identify where thresholds occur and where change along one axis (e.g. management actions, illustrated with our \u0026lsquo;adaptation\u0026rsquo; options) can mitigate change along the other (e.g. \u0026lsquo;climate\u0026rsquo; shifts). This ability to identify the axes that provide resilience or sensitivity to changes in others will be critically important for management uses of HydroBOT targeting climate change adaptation or efficient use of limited water. Such assessments must also take spatial and value differences into account; the different ecological values and geographical SDL units show very different patterns in how they respond to the interaction between the \u0026lsquo;climate\u0026rsquo; and \u0026lsquo;adaptation\u0026rsquo; scenarios (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003ePerhaps most usefully, if scenarios are carefully developed to explore the range of potential change in the system, then the heatmap would represent the response surface of values over this range. Developing the response surface can be iterative, with subsequent scenarios developed to increase resolution near areas of rapid change in outcome. This approach differs from the usual approach of assessing only a small set of scenarios targeting specific proposed actions or environmental conditions (Marchau et al., \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e; Helgeson, \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e). Instead, by covering the range to generate a response surface, it can provide hypotheses about how proposed scenarios might perform and assess how sensitive the outcomes are to uncertainty in single scenarios. Specific proposed scenarios could, of course, also be assessed and mapped onto the surface. For example, we might think of the flow multipliers here as defining the entire range of plausible \u0026lsquo;climate\u0026rsquo; changes, and the additive changes as the entire potential range of \u0026lsquo;adaptation\u0026rsquo; options. Then, specific proposed adaptations or climate sequences could be mapped onto the heatmaps in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e at particular points. The reverse is also possible; identification of areas of high sensitivity in the heatmaps could be used to choose a set of potential management actions designed to increase climate resilience. This idea extends to additional dimensions, which can then be collapsed in various ways (e.g. via principal component analysis, or by identifying dominant hydrometrics) to visualize with 2d heatmaps even when the response surface itself is best defined and studied in more dimensions.\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec21\" class=\"Section3\"\u003e\n \u003ch2\u003e3.4.5 Outcomes on the causal network\u003c/h2\u003e\n \u003cp\u003ePlotting the condition of each value directly on the network shows how condition changes across values at different scales (illustrated here for the historical [E1] scenario, Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e10\u003c/span\u003e). Perhaps most usefully, tracing through the causal network shows how higher-level outcomes result from the complex sets of dependencies on lower levels. For example, in the E1 scenario, the EWRs themselves (leftmost nodes) either nearly always fail (LF1 and above), or nearly always pass (SF2 and below). Many detailed ecological values (middle column) depend only on those EWRs that frequently fail, and so have very low condition (EF4 and above). In contrast, ecological values that depend on a range of EWRs (EF5 and below) show variable condition, depending not only on the values of those EWRs, but the mix of EWRs contributing to each. For example, EF5 has low condition because only two of its eight dependencies had high condition, while EF1 has high condition because only one of its six dependencies had low condition. These outcomes then combine to yield larger-scale ecological value groups (rightmost column). Here, we can see that no detailed values contributing to Waterbirds has good condition, and so neither does the Waterbird group. In contrast, the other groups show a range of conditions reflective of the number and condition of their constituent lower-level values.\u003c/p\u003e\n \u003cp\u003eThis network view provides a clear visualization of the importance of the aggregation functions. By calculating the \u0026lsquo;condition\u0026rsquo; at each step as the mean of the constituent values at earlier steps, the network connections act to smooth out impacts. While the EWRs themselves tend to have extreme values (leftmost nodes), they combine to yield more-similar values in the ecological groupings as they incorporate multiple contributing lower-scale values (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e10\u003c/span\u003e). This smoothing is a feature of averaging and occurs with spatial and temporal aggregation as well; it is simply more evident here because the networks show the steps in the process (see SI Section 9.6.1 for a spatial example). Whether the mean is appropriate is an empirical question relevant for each particular use, and may vary within and between sets of analyses. Using a mean implies that adding more constituent nodes would increase resilience, as good condition for some would compensate for poor condition of others. Using minima or maxima, for example, would have different biological meaning and management interpretation and would yield more persistent differences through the network (see SI Section 9.3). These decisions are critical for extracting meaning and interpretation and should be interrogated thoroughly to ensure they are appropriate for the particular analysis. HydroBOT encourages comparison of different aggregation choices to assess their impact, as we illustrate above in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003cp\u003eVisualizing the causal network can also show the holistic consequences of the different scenarios across the complex and interrelated sets of values, as well as identify nodes that are unusually resilient or sensitive to change using comparisons with a baseline scenario (SI Section 9.6.5). While the relative changes are a useful way of understanding change in condition for each node, they should not themselves be propagated through the network because they are nonlinear transformations and do not capture the meaning of the causal relationships. Instead, the condition scores (Figure S16) capture the causal relationships between conditions at nodes as a consequence of conditions at constituent nodes within a scenario. Then, these condition scores for each node can be compared across scenarios to visualize differences (Figure S17).\u003c/p\u003e\n \u003cp\u003eAnalysis of network structure, combined with aggregated values at each node, could identify critical nodes with outsized importance for outcomes at other nodes in the network. Any such nodes should be investigated to determine whether they capture key aspects of the system or are instead an artifact of model structure (possibly resulting from the way the network was developed) that may bias results. If their sensitivity does in fact capture the system, these nodes may be ideal targets for management intervention. The edges also identify the dependence between values, where we can see that different environmental outcomes depend on not only different hydrologic indicators (EWRs), but also different numbers or sets of those indicators. HydroBOT provides functionality to extract parts of the causal network relevant to a node or nodes, allowing more targeted visualization and investigation of these effects.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e"},{"header":"4. Implications","content":"\u003cp\u003eHydroBOT provides consistent, scientifically robust, and repeatable capacity to model responses to flow, producing management-relevant outcomes at multiple dimensions. This capacity provides a step change for management agencies such as the Murray-Darling Basin Authority to move beyond assessment of hydrologic outcomes and towards assessing the values that depend on that hydrology. Water managers must make decisions with the tools and information currently available, while also seeking to identify knowledge gaps to address for future decisions. Even where HydroBOT is lacking (e.g. additional response models, better knowledge of aggregation choices), it serves to accentuate this missing information, leading to improvements and acknowledgment of limitations. Attempting these models and identifying weaknesses is a far more robust basis for decision making than being unaware of or ignoring these issues.\u003c/p\u003e \u003cp\u003eRequirements such as those for a \u0026ldquo;healthy working basin\u0026rdquo; (MDBA, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2011\u003c/span\u003e) that demand attention across socio-economic, cultural, and environmental values are commonly included in management plans for large basins (e.g. Reclamation, \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; DSC, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Connor, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Ziolkowska \u0026amp; Ziolkowski, \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2016\u003c/span\u003e), and reflect the interconnections of a range of values provided by these systems. Key to assessing this range of values is modeling their dependence on hydrology and other management levers, and requires the capacity to make decisions relating to multiple spatial, temporal, and value dimensions. Management is most successful when multiple scales within those dimensions are considered (Moore, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The HydroBOT aggregation and scaling framework can navigate those dimensions even when the underlying response models are highly detailed. Combining such disparate information in a standard and comparable manner can illuminate synergies and trade-offs, which could be critical for decision making. Moreover, the scaling provided by HydroBOT condenses and synthesizes large numbers of complex outcomes to results tailored for interpretation for management questions, and provides consistency, rigour, and reproducibility. Our demonstration concerns the ecological values of the Murray-Darling Basin; however, the framework is equally applicable to social, economic, and cultural values and to other locations. HydroBOT itself has a modular design to incorporate disparate response models and spatial units.\u003c/p\u003e \u003cp\u003eWater management faces an increasing need and expectation for scientifically-sound and interpretable modeling from a range of stakeholders (Ryder et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Gawne et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). These issues are magnified when facing major long-range decision making to understand the impacts of climate change and potential policy or operational adaptations management agencies might undertake (Neave et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Tonkin et al., \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Co-design of HydroBOT with a diverse team within the Murray-Darling Basin Authority yielded distinct benefits to achieving these goals, with developers and researchers better understanding and building in the needs of the management agency, while the relevant managers better understand the limitations, uses, and potential of the tool. The focus on communication both within management agencies and to external partners, including model provenance, production of tailored synthesis figures, and causal networks, is a key outcome to build trust and understanding in HydroBOT as a tool and the results it produces. HydroBOT provides new capacity to assess a wide range of target values across scenarios and an exciting opportunity for continued advances and improvement with the development of new response models and climate scenarios.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch3\u003e5. CRediT authorship contribution statement\u003c/h3\u003e\n\u003cp\u003eGalen Holt*: Writing \u0026ndash; original draft, review \u0026amp; editing, Visualization, Validation, Software, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Georgia Dwyer*: Writing \u0026ndash; original draft, Visualization, Methodology, Investigation, Formal analysis, Conceptualization. David Robertson: Writing \u0026ndash; review \u0026amp; editing, Supervision, Conceptualization, Funding acquisition. Martin Job: Writing \u0026ndash; review \u0026amp; editing, Supervision, Conceptualization. Rebecca E Lester Writing \u0026ndash; review \u0026amp; editing, Supervision, Investigation, Conceptualization, Funding acquisition.\u003c/p\u003e\n\u003cp\u003e*Holt and Dwyer co-first authors.\u003c/p\u003e\n\u003ch3\u003e6. Declaration of competing interest\u003c/h3\u003e\n\u003cp\u003eThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e\n\u003ch3\u003e7. Acknowledgements\u003c/h3\u003e\n\u003cp\u003eThis research was supported through funding from the Australian Government Murray\u0026ndash;Darling Water and Environment Research Program (MD-WERP). We thank Murray-Darling Basin Authority collaborators including Lara Palmer, Elisha Freedman, Tumi Bjornsson, Matt Coleman, and Joel Bailey for their assistance and support.\u003c/p\u003e\n\u003ch3\u003e8. Data availability\u003c/h3\u003e\n\u003cp\u003eAll data and their sources have been explained in the manuscript. HydroBOT code is available at github.com/galenholt/HydroBOT, and code to create this manuscript is available at github.com/galenholt/toolkit-demo-paper. Documentation with examples is available at github.com/MDBAuth/HydroBOT_website. All are archived at 10.6084/m9.figshare.27605157. \u003cem\u003eLinks are public. HydroBOT and paper code at submission are archived at https://figshare.com/s/699c18f781df7ed2f39d.\u003c/em\u003e\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAllaire JJ, Teague C, Scheidegger C, Xie Y, Dervieux C (2022) \u003cem\u003eQuarto\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArthington AH, Kennen JG, Stein ED, Webb JA (2018) Recent advances in environmental flows science and water management \u0026ndash; Innovation in the Anthropocene. Freshw Biol 63:1022\u0026ndash;1034. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/fwb.13108\u003c/span\u003e\u003cspan address=\"10.1111/fwb.13108\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAscough JC, Maier HR, Ravalico JK, Strudley MW (2008) Future research challenges for incorporation of uncertainty in environmental and ecological decision-making. Ecol Model 219:383\u0026ndash;399. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ecolmodel.2008.07.015\u003c/span\u003e\u003cspan address=\"10.1016/j.ecolmodel.2008.07.015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBengtsson H (2021) A unifying framework for parallel and distributed processing in R using futures. R J 13:273\u0026ndash;291. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.32614/RJ-2021-048\u003c/span\u003e\u003cspan address=\"10.32614/RJ-2021-048\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBond NR, Brooks S, Capon S, Hale J, Kennard M, McGinness H (2021) Chapter 4 - Water-based assets of the Murray\u0026ndash;Darling Basin and their ecological condition. In: \u003cem\u003eMurray-Darling Basin, Australia: Its future management\u003c/em\u003e. Ecohydrology from Catchment to Coast, (Eds B.T. Hart, N.R. Bond, N. Byron, C.A. Pollino \u0026amp; M.J. Stewardson), pp. 75\u0026ndash;93. Elsevier\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrooks S (2021) \u003cem\u003eANAE Classification of the Murray-Darling Basin Technical Report, Revision 3.0\u003c/em\u003e. Commonwealth Environmental Water Office, Department of Agriculture, Water; the Environment, Australia\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrown RR, Keath N, Wong THF (2009) Urban water management in cities: Historical, current and future regimes. Water Sci Technol 59:847\u0026ndash;855. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2166/wst.2009.029\u003c/span\u003e\u003cspan address=\"10.2166/wst.2009.029\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCampbell IC (2016) Integrated management of large rivers and their basins. Ecohydrol Hydrobiol 16:203\u0026ndash;214. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ecohyd.2016.09.006\u003c/span\u003e\u003cspan address=\"10.1016/j.ecohyd.2016.09.006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eConnor R (2015) The United Nations world water development report 2015: Water for a sustainable world. UNESCO, France\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCosgrove B, Gochis D, Flowers T, Dugger A, Ogden F, Graziano T et al (2024) NOAA\u0026rsquo;s National Water Model: Advancing operational hydrology through continental-scale modeling. JAWRA J Am Water Resour Association 60:247\u0026ndash;272. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/1752-1688.13184\u003c/span\u003e\u003cspan address=\"10.1111/1752-1688.13184\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDELWP (2022) Environmental water management plan guidelines for rivers and wetlands. Department of Environment, Land, Water, and Planning, State of Victoria, Melbourne\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDevanand A, Khan Z, Chiew F, Zheng H, Potter N, Robertson D et al (2025) Hydroclimate projections for the murray-darling basin\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDEW (2020) Long-term environmental watering plan for the South Australian River Murray water resource plan area. Department for Environment and Water, South Australian Government, Adelaide\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDPIE (2020) Macquarie-Castlereagh long term water plan part A: Macquarie-Castlereagh catchment. NSW Department of Planning, Industry; Environment, Parramatta\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDRDMW (2022) Warrego Paroo Bulloo and Nebine long-term watering plan. Department of Regional Development, Manufacturing; Water, Queensland Government, Brisbane\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDSC (2013) The Delta Plan. Delta Stewardship Council, California\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFrota RL, Souza Filho F, de Barros A, Silva LS, Porto SMO V.C., Rocha RV (2021) Network socio-hydrology: A case study of causal factors that shape the Jaguaribe River Basin, Cear\u0026aacute;-Brazil. Hydrol Sci J 66:935\u0026ndash;950. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/02626667.2021.1913282\u003c/span\u003e\u003cspan address=\"10.1080/02626667.2021.1913282\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGawne B, Capon SJ, Hale J, Brooks SS, Campbell C, Stewardson MJ et al (2018) Different conceptualizations of river basins to inform management of environmental flows. Front Environ Sci 6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.3389/fenvs.2018.00111\u003c/span\u003e\u003cspan address=\"10.3389/fenvs.2018.00111\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHarrison PA, Beauchamp K, Cooper J, Dickie I, Fitch A, Gooday R et al (2023) An adaptable integrated modelling platform to support rapidly evolving agricultural and environmental policy. Environ Model Softw 169:105821. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.envsoft.2023.105821\u003c/span\u003e\u003cspan address=\"10.1016/j.envsoft.2023.105821\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHart BT, Bond NR, Byron N, Pollino CA, Stewardson MJ (2021) Chapter 1 - Introduction to the Murray\u0026ndash;Darling Basin system, Australia. In: \u003cem\u003eMurray-Darling Basin, Australia\u003c/em\u003e. Ecohydrology from Catchment to Coast, (Eds B.T. Hart, N.R. Bond, N. Byron, C.A. Pollino \u0026amp; M.J. Stewardson), pp. 1\u0026ndash;17. Elsevier\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHelgeson C (2020) Structuring decisions under deep uncertainty. Topoi 39:257\u0026ndash;269. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11245-018-9584-y\u003c/span\u003e\u003cspan address=\"10.1007/s11245-018-9584-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHolling CS (2001) Understanding the complexity of economic, ecological, and social systems. Ecosystems 4:390\u0026ndash;405. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10021-001-0101-5\u003c/span\u003e\u003cspan address=\"10.1007/s10021-001-0101-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHolt G, Chesson P (2016) Scale-dependent community theory for streams and other linear habitats. Am Nat 188:E59\u0026ndash;E73. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1086/687525\u003c/span\u003e\u003cspan address=\"10.1086/687525\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHolt G, Dwyer GK, Robertson D, Bailey J, Job M, Palmer L et al (2022) Architecture design for the Climate Adaptation Toolkit workflow. Murray-Darling Basin Authority, Canberra\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHolt G, Macqueen A, Lester RE (2024) A flexible consistent framework for modelling multiple interacting environmental responses to management in space and time. J Environ Manage 367:122054. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jenvman.2024.122054\u003c/span\u003e\u003cspan address=\"10.1016/j.jenvman.2024.122054\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJones NA, Shaw S, Ross H, Witt K, Pinner B (2016) The study of human values in understanding and managing social-ecological systems. Ecol Soc 21. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/http://dx.doi.org/10.5751/ES-07977-210115\u003c/span\u003e\u003cspan address=\"10.5751/ES-07977-210115\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKaye-Blake B, Schilling C, Nixon C, Destremau K (2014) Water management in New Zealand: A road map for understanding water value. New Zealand Institute of Economic Research\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKing JM, Tharme RE, De Villiers M (2000) Environmental flow assessments for rivers: Manual for the building block methodology. Water Research Commission Pretoria\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLester RE (2019) Wise use: Using ecological models to understand and manage aquatic ecosystems. Mar Freshw Res 71:46\u0026ndash;55. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1071/MF18402\u003c/span\u003e\u003cspan address=\"10.1071/MF18402\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLevin SA (1992) The problem of pattern and scale in ecology: The Robert H. MacArthur Award Lecture. \u003cem\u003eEcology\u003c/em\u003e 73, 1943\u0026ndash;1967. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi-org/10.2307/1941447\u003c/span\u003e\u003cspan address=\"https://doi-10.2307/1941447\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLobegeiger J, Prior A (2022) Border Rivers and Moonie long-term watering plan. Queensland Department of Regional Development, Manufacturing; Water, Brisbane\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLoire R, Pi\u0026eacute;gay H, Malavoi J-R, Kondolf GM, B\u0026ecirc;che LA (2021) From flushing flows to (eco)morphogenic releases: Evolving terminology, practice, and integration into river management. Earth Sci Rev 213:103475. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.earscirev.2020.103475\u003c/span\u003e\u003cspan address=\"10.1016/j.earscirev.2020.103475\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarchau VAWJ, Walker WE, Bloemen PJTM, Popper SW (eds) (2019) Decision Making under Deep Uncertainty: From Theory to Practice. Springer International Publishing, Cham\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMart\u0026iacute;nez LF, Toro J, Le\u0026oacute;n J C (2019) A complex network approach to environmental impact assessment. Impact Assess Project Appraisal 37:407\u0026ndash;420. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/14615517.2018.1552442\u003c/span\u003e\u003cspan address=\"10.1080/14615517.2018.1552442\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarwick B, Boettiger C, Mullen L (2018) Packaging data analytical work reproducibly using R (and friends). Am Stat 72:80\u0026ndash;88. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/00031305.2017.1375986\u003c/span\u003e\u003cspan address=\"10.1080/00031305.2017.1375986\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMDBA (2023) A guide to Traditional Owner. groups for Water Resource Plan areas\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMDBA (2019) Basin-wide environmental watering strategy. Murray-Darling Basin Authority, Canberra\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMDBA (2011) Delivering a healthy working basin. Murray-Darling Basin Authority, Canberra\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMDBA (2024) EWR tool\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMedema W, McIntosh BS, Jeffrey PJ (2008) From premise to practice: A critical assessment of integrated water resources management and adaptive management approaches in the water sector. Ecol Soc 13\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMentzafou A, Katsafados P, Papadopoulos A, Dimitriou E (2023) Hydrological regime alteration assessment in the context of WFD 2000/60: A European and global review. Sustainability 15:15704. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/su152215704\u003c/span\u003e\u003cspan address=\"10.3390/su152215704\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoore S (2021) Toward effective river basin management (RBM): The politics of cooperation, sustainability, and collaboration in the Delaware River basin. J Environ Manage 298:113421. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jenvman.2021.113421\u003c/span\u003e\u003cspan address=\"10.1016/j.jenvman.2021.113421\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNeave I, McLeod A, Raisin G, Swirepik J (2015) Managing water in the Murray-Darling Basin under a variable and changing climate. Water: J Australian Water Association 42:102\u0026ndash;107. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3316/informit.269194666182946\u003c/span\u003e\u003cspan address=\"10.3316/informit.269194666182946\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evon der Ohe PC, Apitz SE, Arbačiauskas K, Beketov MA, Borchardt D et al (2014) In: Brils J, Brack W, M\u0026uuml;ller-Grabherr D, N\u0026eacute;grel P, Vermaat JE (eds) Status and Causal Pathway Assessments Supporting River Basin Management. Springer, Berlin, Heidelberg, pp 53\u0026ndash;149. de\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOlden JD, Konrad CP, Melis TS, Kennard MJ, Freeman MC, Mims MC et al (2014) Are large-scale flow experiments informing the science and management of freshwater ecosystems? Front Ecol Environ 12:176\u0026ndash;185. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1890/130076\u003c/span\u003e\u003cspan address=\"10.1890/130076\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePang J, Chen S, Men B, Wang H (2024) Causal networks and spatiotemporal patterns of different droughts from the perspective of hydrological cycle - A case study of the Nenjiang River Basin, China. Ecol Ind 166:112326. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ecolind.2024.112326\u003c/span\u003e\u003cspan address=\"10.1016/j.ecolind.2024.112326\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePebesma E, Bivand R (2023) Spatial Data Science: With applications in R. Chapman and Hall/CRC\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePeeters LJM, Holland KL, Huddlestone-Holmes C, Boulton AJ (2022) A spatial causal network approach for multi-stressor risk analysis and mapping for environmental impact assessments. Sci Total Environ 802:149845. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.scitotenv.2021.149845\u003c/span\u003e\u003cspan address=\"10.1016/j.scitotenv.2021.149845\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReclamation B (2012) of \u003cem\u003eColorado River basin water supply and demand study technical report D \u0026mdash; System reliability metrics\u003c/em\u003e. US Department of the Interior, Bureau of Reclamation\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRegan RS, Juracek KE, Hay LE, Markstrom SL, Viger RJ, Driscoll JM et al (2019) The U.S. Geological Survey National Hydrologic Model infrastructure: Rationale, description, and application of a watershed-scale model for the conterminous United States. Environ Model Softw 111:192\u0026ndash;203. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.envsoft.2018.09.023\u003c/span\u003e\u003cspan address=\"10.1016/j.envsoft.2018.09.023\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRobson BJ (2014) When do aquatic systems models provide useful predictions, what is changing, and what is next? Environ Model Softw 61:287\u0026ndash;296. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.envsoft.2014.01.009\u003c/span\u003e\u003cspan address=\"10.1016/j.envsoft.2014.01.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRuel JJ, Ayres MP (1999) Jensen\u0026rsquo;s inequality predicts effects of environmental variation. Trends Ecol Evol 14:361\u0026ndash;366. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S0169-5347(99)01664-X\u003c/span\u003e\u003cspan address=\"10.1016/S0169-5347(99)01664-X\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRyder DS, Tomlinson M, Gawne B, Likens GE (2010) Defining and using \u0026lsquo;best available science\u0026rsquo;: a policy conundrum for the management of aquatic ecosystems. Mar Freshw Res 61:821\u0026ndash;828. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1071/MF10113\u003c/span\u003e\u003cspan address=\"10.1071/MF10113\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eScheuerell MD, Ruff CP, Anderson JH, Beamer EM (2021) An integrated population model for estimating the relative effects of natural and anthropogenic factors on a threatened population of steelhead trout. J Appl Ecol 58:114\u0026ndash;124. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/1365-2664.13789\u003c/span\u003e\u003cspan address=\"10.1111/1365-2664.13789\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSheldon F, Rocheta E, Steinfeld C, Colloff MJ, Moggridge B, Carmody E et al (2024) Are environmental water requirements being met in the Murray-Darling Basin, Australia? Mar Freshw Res 75. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1071/MF23172\u003c/span\u003e\u003cspan address=\"10.1071/MF23172\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShenton W, Bond NR, Yen JDL, MacNally R (2012) Putting the ecology into environmental flows: Ecological dynamics and demographic modelling. Environ Manage 50:1\u0026ndash;10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00267-012-9864-z\u003c/span\u003e\u003cspan address=\"10.1007/s00267-012-9864-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSinger MB (2007) The influence of major dams on hydrology through the drainage network of the Sacramento River basin, California. River Res Appl 23:55\u0026ndash;72. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/rra.968\u003c/span\u003e\u003cspan address=\"10.1002/rra.968\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStern CV, Sheikh PA (2019) Management of the Colorado River: Water Allocations, Drought, and the Federal Role. Congressional Research Service, Washington, D.C.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTharme RE (2003) A global perspective on environmental flow assessment: emerging trends in the development and application of environmental flow methodologies for rivers. River Res Appl 19:397\u0026ndash;441. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/rra.736\u003c/span\u003e\u003cspan address=\"10.1002/rra.736\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTonkin JD, Poff NL, Bond NR, Horne A, Reynolds MDM L.V., et al (2019) Prepare river ecosystems for an uncertain future. Nature 570:301\u0026ndash;303. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/d41586-019-01877-1\u003c/span\u003e\u003cspan address=\"10.1038/d41586-019-01877-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUshey K, Allaire JJ, Tang Y (2023) Reticulate: Interface to \u0026rsquo;python\u0026rsquo;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWater Act (2007) (2007)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWelsh WD, Vaze J, Dutta D, Rassam D, Rahman JM, Jolly ID et al (2013) An integrated modelling framework for regulated river systems. Environ Model Softw 39:81\u0026ndash;102. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.envsoft.2012.02.022\u003c/span\u003e\u003cspan address=\"10.1016/j.envsoft.2012.02.022\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWickham H (2016) ggplot2: Elegant graphics for data analysis. Springer-, New York\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWickham H, Bryan J (2025) R packages, 2nd edn. O\u0026rsquo;Reilly\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWilson G, Aruliah DA, Brown CT, Hong NPC, Davis M, Guy RT et al (2014) Best practices for scientific computing. PLoS Biol 12:e1001745. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pbio.1001745\u003c/span\u003e\u003cspan address=\"10.1371/journal.pbio.1001745\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang A, Dutta D, Vaze J, Kim S, Podger G (2017) An integrated modelling framework for building a daily river system model for the Murray\u0026ndash;Darling Basin, Australia. Int J River Basin Manage 15:373\u0026ndash;384. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/15715124.2017.1315814\u003c/span\u003e\u003cspan address=\"10.1080/15715124.2017.1315814\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZiolkowska JR, Ziolkowski B (2016) Effectiveness of water management in Europe in the 21st century. Water Resour Manage 30:2261\u0026ndash;2274. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11269-016-1287-9\u003c/span\u003e\u003cspan address=\"10.1007/s11269-016-1287-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"Deakin University","isAcceptedByJournal":true,"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":"1) Environmental flows, 2) Water management, 3) Hydrology, 4) Integrated management, 5) R package, 6) Synthesis","lastPublishedDoi":"10.21203/rs.3.rs-6793584/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6793584/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eWater management rarely focuses only on water; instead, management targets values across many water-dependent responses. To assess past performance and plan future actions, water managers need to understand how changes to hydrology (over which they typically have most control) affect water-dependent values. Models relating values to hydrology can be difficult to integrate into management processes; they are often developed for other uses and target subsets of values. We describe and demonstrate HydroBOT (Hydrology-dependent Basin Outcomes Toolkit), a modeling toolkit co-designed with the primary federal water management agency in the Murray-Darling Basin, Australia. HydroBOT can integrate disparate response models and scale and synthesize those results across space, time, and groups of values. Outputs target water management needs and can be tailored to a range of questions, from local, short-term evaluation to basin-scale climate assessment. HydroBOT provides new capacity to move beyond hydrology to assess outcomes across diverse target values.\u003c/p\u003e","manuscriptTitle":"HydroBOT: an integrated toolkit for assessment of hydrology-dependent outcomes","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-03 12:41:40","doi":"10.21203/rs.3.rs-6793584/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":"26fa7c5a-d07c-4989-809e-d1449701d075","owner":[],"postedDate":"June 3rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":49344702,"name":"Ecological Modeling"},{"id":49344703,"name":"Marine and Freshwater Ecology"},{"id":49344704,"name":"Environmental Policy"}],"tags":[],"updatedAt":"2025-06-23T14:19:02+00:00","versionOfRecord":{"articleIdentity":"rs-6793584","link":"https://doi.org/10.1016/j.envsoft.2025.106579","journal":{"identity":"environmental-modelling-and-software","isVorOnly":true,"title":"Environmental Modelling \u0026 Software"},"publishedOn":"2025-06-19 00:00:00","publishedOnDateReadable":"June 19th, 2025"},"versionCreatedAt":"2025-06-03 12:41:40","video":"","vorDoi":"10.1016/j.envsoft.2025.106579","vorDoiUrl":"https://doi.org/10.1016/j.envsoft.2025.106579","workflowStages":[]},"version":"v1","identity":"rs-6793584","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6793584","identity":"rs-6793584","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}