The technical foundation of the Explorer is the ISIMIP3b multi-model ensemble, in which four global hydrological models, WaterGAP 2.2e, CWatM, MIROC-INTEG-LAND and JULES-ES-VN6P3, were each driven by the bias-adjusted output of five CMIP6 global climate models under three greenhouse gas emissions scenarios. That combination yields up to 20 equally likely ensemble members for total runoff and actual evapotranspiration, and 15 for diffuse groundwater recharge, since one hydrological model did not provide that variable. Monthly time series of these three variables, gridded at 0.5 degrees latitude by 0.5 degrees longitude, roughly 55 by 55 kilometers at the equator, cover the period 1985 to 2100. The reference period 1985 to 2015 is compared against three future windows, 2016 to 2045, 2041 to 2070 and 2071 to 2100, and because direct human impacts on water use and land cover are held constant at 2015 levels in the future runs, the computed changes reflect climate change alone.

What makes the Explorer genuinely innovative is how it handles uncertainty. The top map displays the median percent change of the selected variable, the change not exceeded by half of the ensemble members. The bottom map shows the percentage of ensemble members projecting a change above or below a user-selected threshold, set by default at a decrease of 20 percent. By varying that threshold, a user can trace the spatial pattern of approximate likelihood for exceeding any problematic level of change. This matters because a median that looks benign can conceal a real hazard: even when half the models suggest little change, a substantial fraction may project a decrease that stakeholders would consider dangerous. The Explorer makes that hidden risk visible at a glance, something that maps showing only a median or mean cannot do.

Clicking on any grid cell opens a panel called Local Insights, which renders percentile boxes showing the P10, P30, P50, P70 and P90 values of future change for that cell. The P10, for example, is the change not exceeded by 10 percent of the ensemble members. By deliberately excluding the most extreme 20 percent of the ensemble, the boxes sharpen the readability of the central percentiles while still fitting the common intuition of scientists and stakeholders that extreme outliers deserve caution rather than inclusion. The design was chosen after a systematic comparison with alternatives such as letter-value plots, violin plots, conventional box plots and the bar charts used by the German Climate Service Center GERICS, which were found to be harder to interpret. Users can download the Local Insights plots as images or the underlying data as spreadsheets, and the entire application runs on the Google Earth Engine cloud platform, which made development possible without any dedicated server infrastructure and at a very low budget.

The team evaluated the tool with an online survey of 21 pilot users, framed by the international software quality standard ISO/IEC 25019:2023, which measures quality in use through effectiveness, efficiency, operability, user engagement and beneficialness. Participants first completed two realistic tasks, exploring renewable groundwater resources in Central America and examining changes in total runoff around Ulaanbaatar, before rating the application. On average, users answered five of eight task questions correctly, and the time needed ranged from 4 to 55 minutes with a median of 21 minutes. User satisfaction was high overall, and satisfaction correlated rather strongly with user expertise, with an R-squared of 0.53. Only three of the 21 participants said they did not intend to consult the Explorer again, and the bottom map, which visualizes the distribution of model results, received somewhat more critical ratings than the more familiar median map, likely because most users had never seen ensemble distributions presented this way before.

The survey also produced concrete improvements. Sixteen pilot users offered suggestions, and the most frequent request, from seven users, was a step-by-step tutorial and clearer guidance on interpreting the displays. The revised documentation now includes that tutorial, and the survey tasks themselves were folded into the guidance. Some interface suggestions could not be implemented because of constraints in the Google Earth Engine user interface library, which limits custom styling and layout flexibility, but the developers have released the application code freely so that others can build similar tools.

Perhaps the most compelling evidence for the Explorer’s value comes from three real participatory adaptation processes in which the underlying approach was tested. In the KlimaRhön process in the biosphere reserve Rhön in central Germany, 31 stakeholders, almost none of whom had worked with climate models before, were shown percentile boxes of projected changes in total water resources and groundwater resources after a careful explanation of emissions scenarios and multi-model ensembles. The researchers stressed that the percentiles represent shares of models, not probabilities, and that the median is not the most likely future. At a World Café discussion, all but one of the stakeholders advocated preparing for the worst-case decrease in water resources, guided by the precautionary principle, and the presentation of uncertain summer declines alongside stakeholders’ own experiences of recent dry summers steered the whole process toward adapting to potential summer water scarcity. By the end of the first workshop, 60 percent of participants said the ensemble information improved their understanding of future water changes, and 75 percent said they better understood the uncertainties.

Similar presentations of ensemble-based groundwater recharge projections in a region south of the Harz in Germany and on the Croatian island of Krk enabled stakeholders to develop management strategies that are robust across the entire range of plausible futures. In the German case, participants explicitly geared a groundwater management concept toward dealing with uncertainty, and on Krk, projections of year-round declining water resources supported a large set of adaptation measures spanning both water supply optimization and reductions in tourism-related water demand. In all three regions there were no local projections of groundwater recharge and hardly any streamflow observations, so the global model ensemble represented genuinely new information for the communities involved.

The authors are careful about who should use the tool directly. Because satisfaction and correct interpretation correlated with expertise, they recommend that experts such as water engineers, hydrogeologists or local climate managers extract the relevant information from the Explorer and communicate it to stakeholders, explaining how the ensemble quantifies change and how to read the maps and percentile boxes. They also advise clearly verbalizing the findings, for example by stating that 10 percent of the models project a decrease in average groundwater recharge of more than 20 percent. When stakeholders question the coarse 0.5-degree resolution, which covers roughly 1500 to 3000 square kilometers, the expert should explain that the uncertainties of climate projections for smaller areas are so large that finer-gridded information would not be meaningful for adaptation decisions. The Explorer is, however, well suited to direct use by university students and researchers, and the first author has already integrated it into her hydrogeography class, where students successfully characterized climate impacts on large river basins.

Beyond water, the paper carries a broader message for the entire climate adaptation community. It demonstrates how multi-model ensemble information on the uncertainty of climate change impacts, whether for crop yields, natural vegetation or wildfires, can be communicated effectively through low-cost map-based web applications, and it shows that such tools can be built with basic programming skills on free cloud platforms. The authors argue that the ISIMIP ensemble output presented in the Explorer could become a primary source of global climate impact information for a future integrated IPCC Atlas supporting the seventh assessment cycle, and they note, strikingly, that even the prominent IPCC Interactive Atlas has never been formally evaluated by its users. With climate impacts accelerating and adaptation budgets tightening worldwide, a free tool that lets any region on Earth see the plausible range of its water future, and decide how much risk it is willing to live with, may prove to be one of the quiet breakthroughs of the decade.

Subject of Research: climate change adaptation, water resources, uncertainty visualization, multi-model ensembles, participatory processes

Article Title: Supporting climate change adaptation worldwide: a web application for exploring uncertain future changes in water resources

Article References: Döll, P., Attard, G., Kneier, F., & Müller, L. (2026). Supporting climate change adaptation worldwide: a web application for exploring uncertain future changes in water resources. Geoscience Communication, 9(4), 473-488. https://doi.org/10.5194/gc-9-473-2026

Image Credits: AI Generated

DOI: 10.5194/gc-9-473-2026

Keywords: climate change, water resources, hydrology, uncertainty, web application, ISIMIP, adaptation, Google Earth Engine, Supporting, climate, change, worldwide