Worldwide, strategies for adapting to climate change need to be developed through local participatory processes, particularly to minimize risks associated with changing water resources. To identify suitable adaptation measures, quantitative information about potentially hazardous future changes in local water resources is required. However, future changes in water resources can only be projected with large uncertainties (e.g., Tabari et al., 2021). For one specific future greenhouse gas emissions scenario, different climate models calculate rather different future changes in climatic variables, in particular in precipitation (Tarrín-Carrasco et al., 2025), and different hydrological models will translate the same time series of climatic input variables into rather different time series of, e.g., groundwater recharge (Reinecke et al., 2021). Therefore, decisions on climate change adaptation need to consider the uncertainty of future hazards of climate change (Döll and Romero-Lankao, 2017).

It is generally not possible to know whether a specific climate or hydrological model is superior to the other models in climate change studies. Even if a hydrological model has been shown to simulate historical streamflow or other hydrological variables well, it cannot be assumed to be superior to others when computing future climate change impacts. Most local or basin-scale hydrological models, for example, do not take into account the response of the vegetation to increased CO₂ and the changing climate when computing evapotranspiration. Therefore, it is state of the art to assess the potential future changes of hydrological variables for a specific greenhouse gas emissions scenario using the output of multi-model ensembles (MMEs) (Hagemann et al., 2013). These consist of the output of multiple hydrological models, each of which is forced by the bias-adjusted output of global climate models (GCMs) (Döll et al., 2015). The output of each model combination is considered to be equally likely, and the distribution of the changes calculated by each model combination of the MME approximates the uncertain range of future changes.

Most participatory climate change adaptation processes will not have access to multi-model local or regional studies on future changes in water resources due to financial and other constraints. Therefore, they are best served by the output of existing MMEs of global hydrological models (GHMs) that are driven by the bias-adjusted output of global climate models, as provided by ISIMIP (https://www.isimip.org, last access: 13 October 2026; Müller and Döll, 2024). Suppose a well-established local hydrological model can be applied to project future hydrological changes by forcing it with the bias-adjusted output of several global climate models. In that case, the local model ensemble changes should be contrasted with the changes of the GHM MME to embrace the uncertainty of translating climatic changes into hydrological changes. The spatial resolution of GHMs of 0.5° latitude by 0.5° longitude (55 km × 55 km at the equator) may seem, at first glance, unsuitably coarse for informing local climate change adaptation. However, given the even coarser spatial resolution of global climate models and their large uncertainty in translating greenhouse gas emissions scenarios into changes in climate variables, in particular precipitation, information on potential future changes in water resources in 0.5° grid cells is appropriate for informing on climate-driven changes in areas smaller than a 0.5° grid cell.

The output of the most recent ISIMIP3b GHM MME (https://www.isimip.org/protocol/3/, last access: 13 October 2026) is freely available for many model output variables as daily or monthly time series of 0.5° by 0.5° gridded values for the time period 1985–2100 in the form of NetCDF files, as described by Müller and Döll (2024) for the previous GHM MME developed under the ISIMIP2b protocol. As processing many large data files using, e.g., Python or R requires specialized skills and effort, we wanted to provide information on hydrological changes important for understanding climate change impacts on water resources and their uncertainty in an easily accessible format. To this end, we analyzed the available model runs and output variables of the ISIMIP3b GHM MME and processed the variables runoff, groundwater recharge, and actual evapotranspiration. We then provided and visualized the changes between a reference period and various future periods via a freely accessible map-based web application, the interactive data explorer (Neset et al., 2016) “Climate Change Impact on Water Resources Explorer” (https://ee-gwp.projects.earthengine.app/view/cciwr-explorer, last access: 13 October 2026), hereafter CCIWR Explorer. A link to the Explorer is provided on the ISIMIP website (https://www.isimip.org/outcomes/data-visualization/, last access: 13 October 2026).

The development of the CCIWR Explorer was guided primarily by the question: How can information on potential future changes in water resources, derived from a GHM MME, be optimally provided on a web portal, with a focus on supporting local climate change adaptation processes worldwide? The focus was on how to best communicate the uncertainty of the quantitative estimates of future changes, i.e., the range of potential future changes. In the Explorer, the deep epistemic uncertainty due to future greenhouse gas emissions (Döll and Romero-Lankao, 2017) is taken into account by the user being able to select between the MME output computed for three greenhouse gas emissions scenarios. The CCIWR Explorer is distinguished by the innovative way in which the low-to-medium epistemic uncertainty caused by (1) translating emissions scenarios into climate scenarios by global climate models and (2) translating climate scenarios into water resources scenarios by global hydrological models (Döll and Romero-Lankao, 2017), as quantified by the MMEs, is visualized. In the web app and documentation, we do not use the terms deep or low-to-medium uncertainty. MME results are always provided specifically for one of the three emissions scenarios, implicitly separating these two types of uncertainties. Statistical uncertainties are not mentioned at all; these are not quantified from the MME results for the CCIWR Explorer as they are beyond the scope of quantitative information input in the case of most participatory processes and rather difficult to grasp for most stakeholders (see visualization of changes in water resources in wet and dry years in Müller and Döll, 2024).

When developing the CCIWR Explorer, we envisaged that scientists and experts involved in local to national climate change adaptation processes around the world would use it and disseminate the information it offers. Additionally, multidisciplinary scientists from fields such as hydrology, ecology, agricultural sciences, and economics may wish to be informed about potential future changes in water resources in their study regions. Furthermore, the web application can serve as a valuable tool for teaching on climate change and sustainability.

The CCIWR Explorer was developed by bringing together expertise in global hydrological modeling, communication of climate change hazards, and geoscientific web application development. Two of the authors were involved in the development of a global-scale web application on freshwater-related hazards of climate change in the framework of the large research project CO-MICC (https://www.co-micc.eu, last access: 13 October 2026). The CO-MICC web application (data portal) was co-designed with meteorologists, hydrologists, and water resources managers from Spain, Morocco, Algeria, and Tunisia and involved more hazard indicators and visualization options than the CCIWR Explorer (Döll et al., 2021; https://www.co-micc.eu/data/data-portal/, last access: 13 October 2026). Unfortunately, due to technical and organizational issues, the data portal was only accessible for a few months. Different from the CO-MICC web application, which was technically developed by a large company funded within the research project, the CCIWR Explorer was developed without any third-party funding on a very low budget. The CCIWR Explorer was also inspired by more recent experiences of two co-authors in communicating uncertain climate change hazards in participatory processes concerning the adaptation to climate change in the water sector (Müller and Döll, 2024; Söller et al., 2026). We also reviewed existing web applications that present future climatic changes and their impacts.

In this paper, we present the CCIWR Explorer and show the results of its evaluation by pilot users. We then discuss (1) how the CCIWR Explorer represents uncertainty as compared to similar web apps, (2) the suggestions made by pilot users for improving it and their implementation in the revised version of the Explorer, and (3) how to best apply the Explorer in participatory processes on climate change adaptation in the water sector. This paper is of broad interest to researchers and consultants beyond the specific web application due to the following. Firstly, it shows how, in climate change adaptation processes, whether relating to water resources or other impacts such as crop yields, natural vegetation or wildfires, multi-model ensemble-based information on the uncertainty of climate change impacts can be most effectively communicated through web apps, and, consequently, in participatory climate change adaptation processes. Secondly, it informs researchers and consultants on how a web app that provides map-based information on climate change impacts can be implemented at low cost using basic programming skills. The code of the web application developed on the Google Earth Engine platform is made available to support the development of similar web applications.

When calling the URL of the CCIWR Explorer, the user sees the landing page depicted in Fig. 1. It encompasses two map windows on the right-hand side, the user interface on the upper part of the left-hand side, and, below it, the “Local Insights” percentile boxes showing changes between the reference period and three future time periods for a grid cell on which the user has clicked on one of the maps. By interacting with the computer via the user interface, the user selects the variable, the emissions scenario, the time horizon, and the season (or the annual aggregation of the monthly values) (Table 1). These specifications determine what is shown on the two maps and the “Local Insights” graph.

The upper map window shows the median percent change of the variable, i.e., the change that is not exceeded by half of the MME members. The bottom map shows the percentage of MME members simulating that the variable's change is either above or below a certain threshold that the user selects (default setting: below the threshold of -20 % change). By consecutively varying the threshold, the user can obtain a comprehensive impression of the spatially distributed hazards of climate change. For example, the spatial patterns of the number of models that exceed a certain decrease in groundwater recharge inform about the spatial pattern of (approximate) likelihood for exceeding problematic decreases. This is an informative addition to the median future change shown simultaneously in the top map window, which, without the bottom map, may hide a problematic change, as some MME members may project a problematic change even though the MME median indicates a non-problematic change. Both map windows permit zooming and panning, and the user can increase the transparency of the data layer to see the base layer with city names.

The percentile boxes in the “Local Insights” plots are designed like those introduced by Müller and Döll (2024). This publication compares percentile boxes to four alternative methods for visualizing cumulative distributions in the form of percentiles, letter-value plot, violin plot, box plot, and the bar chart as used by GERICS, the Climate Service Center Germany. GERICS bar charts and percentile boxes are easier to understand than the other three alternatives. Both indicate just five percentiles, i.e., they show five values of change of the variable of interest, where, e.g., P90 refers to the change that is not exceeded by 90 % of the MME members. A percentiles box shows P10, P30, P50 (median), P70, and P90, while a GERICS bar chart shows the minimum change, P20, P50, P80, and the maximum change (e.g., https://www.gerics.de/imperia/md/content/csc/projekte/klimasignalkarten/gerics_klimaausblick_germany_version1.2_deutsch.pdf, last access: 13 October 2026). Showing the minimum and maximum may reduce the readability of the GERICS bar chart, as outliers may increase the box size strongly. Not including all MME members in the percentile box, but only 80 %, increases the precision with which the three central percentile values can be identified from the visual. In addition, showing P10 and P90 in percentile boxes often fits the notion of both scientists and stakeholders of excluding extreme values from consideration. When clicking on the “Local Insights” plot, the plot opens in a separate page, from where either the plot can be downloaded as png file, or the data behind it as csv file. The gridded data shown on the maps cannot be downloaded.

Before engaging with the user interface, the user can easily detect, at the top left of the page, the word “documentation”. By clicking on this word, the documentation of the CCIWR Explorer opens. The original documentation available to the participants of the pilot user survey is included as Supplement file 3. The current version was updated in response to the user survey and is included as Supplement file 4 (see Sect. 5.2). The documentation introduces the purpose of the web application, and then answers the following questions: What type of information is provided by the CCIWR Explorer? How is the information provided? How is the provided information best utilized for informing local climate change adaptation processes? After having read this short documentation of about three pages, the user understands the purpose, applicability, and content of the web app and knows how to operate the interface in a meaningful way. The revised version also includes a tutorial.

To utilize the CCIWR Explorer to inform stakeholders about potential future changes in water resources, even if there is no internet access during the stakeholder engagement, a user may print parts of the screen to pdf files in advance and then combine them. As an example, Fig. 2 shows potential future changes in renewable water resources in Southern Africa. While the two top panels are the same and show the median changes, the lower panels enable them to grasp the uncertainty of the projections by showing, on the left side, the share of MME members that project a general decrease in total renewable water resources, and on the right side, the share that projects a decrease of at least 25 %. For many but not all grid cells with a projected median increase in water resources (blue cells in the top maps of Fig. 2), more than 20 % of the MME members project a decrease, but less than 20 % a decrease of more than 25 %. In some grid cells with a median increase, e.g., in southern Tanzania, more than 20 % of the MME members predict a decrease of more than 25 %.

After showing the larger regional pattern of future changes in water resources using maps, local participatory processes for climate change adaptation are best informed by “Local Insights” plots. These plots show the range of potential changes in a selected 0.5° grid cell using percentile boxes. Section 5.3 provides information on how the interpretation of percentile plots can be explained to stakeholders, and the revised documentation also gives some guidance. The “Local Insights” plots offer, for instance, concise information on projected seasonal changes in the variable of interest. As an example, Fig. 3 shows seasonal and annual changes of diffuse groundwater recharge between the reference period and the near, intermediate, and far future for a grid cell in the mountains north of the city of Frankfurt, Germany. While most MME members project an increase in annual groundwater recharge, i.e., renewable groundwater resources, 10 % project a decrease of more than 20 % until the far future (lower end of box), which may motivate risk-averse stakeholders to undertake adaptive water management measures. The predominantly projected increases in annual groundwater recharge until the far future are due to rather stable values in winter (December to February) and strongly increasing values in spring (March to May), the two seasons that dominate annual groundwater recharge in the region. These projections are suitable for countering fears of a lack of drinking water supply from groundwater if waterworks have access to large groundwater reservoirs. However, if waterworks rely on groundwater bodies with a small storage capacity, they may be negatively affected by the projected decrease in groundwater recharge in the summer (June to August) and fall (September to November) seasons (Fig. 3). 30 % of the MME members project a decrease in summer groundwater recharge of more than 40 % until the end of the century (lower end of dark yellow box), while 10 % of the MME members project declines of more than 70 %. Figure 3 shows a clear decreasing tendency in summer groundwater recharge during the 21st century.

The first author has already integrated the CCIWR Explorer into her hydrogeography class. In this class, students successfully used the CCIWR Explorer to characterize the impacts of climate change on water in selected large river basins. They took several screenshots after zooming in on the basin and then juxtaposed them on a PowerPoint slide. They also used “Local Insights” plots to show their classmates the potential future temporal development of water resources at a specific location within the basin.

The effectiveness of the CCIWR Explorer is quantified by the correct number of answers to questions related to the two tasks that the pilot users had to perform to become acquainted with the web application. On average. i.e., over all 21 pilot users, 5 out of the 8 questions were answered correctly. Only three users answered all the questions correctly. Relating the number of correct answers to the time required to do the tasks, including potentially reading the documentation, quantifies the efficiency of use. The reported times ranged from 4 to 55 min, with a median of 21 min. Assuming that the reported “4 min” represents a typo, as three questions were answered correctly, the efficiency ranges between 0.9 and 7.5 correct answers per 10 min, with a median of 2.4 correct answers per 10 min. The number of correct answers is positively correlated with user expertise, but only weakly (Fig. 4b). The level of user expertise was quantified by averaging, for each user, the Likert-scale (1–5) answers to four questions regarding the user's climate change and water resources knowledge, and their experience with hydrological modeling and the quantification of climate change impacts (Supplement file 2). We did not include, in the level of user expertise, the previous user knowledge about percentiles and the application of MMEs in climate change studies, as all survey participants except one indicated that they had this knowledge before taking the survey.

Evaluating the sub-criterion of user satisfaction operability (statements 1 and 2 in Fig. 5) separately, more than 60 % of the users agreed or strongly agreed with the statements. Statements 4, 5, and 7 were included to evaluate user engagement, specifically the clear and informative presentation of median changes (top map), MME member agreement (bottom map), and the “Local Insights” plots. The presentation of the top map obtained a better evaluation than the other two, which may be explained by the fact that this is the most typical and simplest presentation of climate change impacts, while the bottom map and the “Local Insights” plot visualize the distribution of MME results, which is something that most users might not have seen before. In terms of beneficialness (statements 3, 6, 8, and 9), only 10 %–20 % of the participants did not see a benefit in the CCIWR Explorer. Many survey participants evaluated the bottom map more negatively than the top map and the “Local Insights” plot, both in terms of user engagement and beneficialness. The level of user satisfaction correlates positively with the number of correct answers (Fig. 4a), but only weakly. However, with an R2 of 0.53, it correlates rather strongly with the level of expertise (Fig. 4c). Only 3 out of the 21 survey participants did not intend to consult the CCIWR Explorer again in the future (statement 10).

The survey participants are more knowledgeable about evaluating uncertain climate change impacts than the typical stakeholders in climate change adaptation processes. Based also on the positive correlation between user satisfaction and expertise, we would expect the participation of typical stakeholders in the survey to lead to less positive evaluations of the CCIWR Explorer. Furthermore, the survey results suggest that typical stakeholders should not be encouraged to peruse the Explorer without guidance. Instead, experts, such as researchers and consultants, should use the CCIWR Explorer to access information that they can then communicate to non-expert stakeholders in an appropriate manner.

The CCIWR Explorer is a low-budget web application for supporting climate change adaptation in the water sector under uncertainty. The development process focused on facilitating effective communication regarding the plausible range of potential future changes in water resources. The objective was to enable users to understand and embrace the uncertainty information provided by MMEs, and to use this information to identify adaptation measures. With the provided knowledge about the proportion of the ensemble members that project a change they consider hazardous, stakeholders can make informed adaptation decisions based on their risk aversion.

While the Explorer was evaluated very favorably by most pilot users, the evaluation, as well as our own experience with presenting multi-model ensemble projections in participatory processes for climate change adaptation, indicates that some hydrological and climate change expertise is required for a meaningful interpretation of the provided information. Therefore, we recommend including experts such as water engineers or climate managers in participatory climate change adaptation processes dealing with water resources. These experts can use the CCIWR Explorer to compile information for communication with stakeholders, using the maps and plots provided by the web application. We also believe that the CCIWR Explorer is suitable for direct use by university students and researchers, as well as for informing the public about the impact of climate change on water resources. The CCIWR Explorer may serve as an example for low-cost map-based web applications for showing uncertain climate change impacts for other impact sectors. It may inspire a future integrated, interactive cross-Working Group IPCC Atlas, which, according to Alegría et al. (2024), should be developed to support comprehensive AR7 climate change risk assessments. Alegría et al. (2024) propose that ISIMIP MME output, like the one presented in the CCIWR Explorer, becomes a primary source for global climate impacts in key WGII-relevant sectors such as health, water, forests, agriculture, and fisheries. Unfortunately, even very prominent web applications on climate change have not been formally evaluated by their users. To the best of our knowledge, this is not even the case of the IPCC Interactive Atlas (Cimadevilla et al., 2025). In line with the recommendations of VanderMolen et al. (2019), we advise the implementation of comprehensive user surveys, especially for current and future IPCC interactive atlases.