The stakes could hardly be higher. Groundwater quantity within Ibeator is already insufficient for its growing population, and inadequate investigation has produced a string of failed boreholes. Without a clear picture of how thick the protective layers above an aquifer are, how deep the water sits, and how readily contaminants could migrate downward, planners are essentially drilling blind. The study set out to close that knowledge gap by pairing a classic geophysical field technique with modern data-driven modeling, offering a template that could be transferred to regions with similar geology.

The fieldwork rested on twelve vertical electrical soundings conducted with an ABEM Tetrameter SAS 1000 resistivity meter using the Schlumberger configuration, with current electrode spacing reaching 900 meters. This method injects current into the ground and measures the resulting voltage differences, allowing researchers to reconstruct how electrical resistivity changes with depth. Because the resistivity of a layer depends on its lithology, porosity, water content, and salinity, these soundings act as a non-invasive X-ray of the subsurface. Several soundings were deliberately placed near existing boreholes so the interpreted geoelectric sections could be checked against observed lithology, a validation step that boosted confidence when the predicted depth to the aquifer of 90 and 81 meters compared favorably with the 110 meters recorded in the geological log at Okorobi.

Interpretation of the soundings revealed a seven-layer subsurface model across the study area, with resistivity values spanning from about 72 ohm-meters in conductive clays to more than 8,000 ohm-meters in resistive sands. At one representative station, the sequence progressed from lateritic topsoil through clay, silty sand, dry sand, wet sand, and finally a thick saturated sandstone aquifer extending to depths of around 170 meters before a basal clay was encountered. From these geoelectric parameters, the team derived the Dar-Zarrouk quantities, longitudinal conductance and transverse resistance, which link the electrical behavior of the subsurface to its hydraulic properties through Darcy-type relationships. Those derived parameters, including hydraulic conductivity, transmissivity, and storativity, became the raw material for the machine learning comparison.

The spatial patterns that emerged are striking. Aquifer thickness is greatest in the eastern part of the study area around Umudime, reaching 90 to 210 meters, while the western communities of Okorobi, Okahia, and Uhuala sit above much shallower aquifers. Overall aquifer depths ranged from about 45.6 meters to 408.85 meters, averaging 227.21 meters. Resistivity of the aquifer material was highest, between 10,000 and 28,000 ohm-meters, around Okorobi and Uhuala, and lowest, between 1 and 2,000 ohm-meters, toward the east and northeast. Crucially, hydraulic conductivity, with values between roughly 0.000174 and 0.000215 meters per second, showed an inverse relationship with resistivity: where the subsurface resists electrical current, it also tends to resist the flow of water.

Transmissivity and storativity maps echoed the same geography, with transmissivity values from about 0.0035 to 0.0455 square meters per second and storativity from roughly 0.0077 to 0.324 square meters per second, both declining from north to south and lowest at the western and eastern edges. These variations matter because they translate directly into vulnerability. A shallow aquifer with thin protective cover and high transmissivity offers contaminants a fast track to the water table, whereas deeper, better-shielded units are naturally more resilient. The lithologic shifts observed from sand and sandstone in the northeast toward shale-rich sequences in the northwest, near the contact between the Imo Shale and the Benin Formation, further modulate how water moves and how well each zone is protected.

With the dataset assembled, the researchers built two families of predictive models. The artificial neural network used a feedforward, supervised architecture with backpropagation learning, taking three inputs, aquifer depth, aquifer resistivity, and aquifer thickness, through a hidden layer of three nodes to a single output. Seventy percent of the samples trained the network and thirty percent were reserved for testing, with the architecture tuned experimentally by adjusting neuron counts and training proportions. The multivariate linear regression approach, by contrast, fit explicit equations relating the same three predictors to each hydraulic property, with predictor selection guided by Pearson correlation coefficients and standard assumption checks for linearity, multicollinearity, homoscedasticity, normality of residuals, and independence of errors.

The head-to-head results were unambiguous. For hydraulic conductivity, the ANN managed an R-squared of only 0.0869, with training errors of 4.037 times ten to the minus five in root mean squared error and 0.343 in mean absolute percentage error, while the MLR achieved an R-squared of 0.9775 with far lower training errors. For transmissivity, the ANN reached an R-squared of 0.9157 against 0.9958 for the regression model, and for storativity the MLR again led with 0.9868 versus 0.9157. The authors attribute the neural network’s underperformance to the small dataset: ANNs typically require large volumes of data to learn complex nonlinear patterns, and with only twelve soundings the network likely underfit the underlying relationships, whereas the regression equations captured the dominant linear structure with remarkable precision.

Sensitivity analysis of the trained models also revealed which inputs mattered most, with aquifer depth and resistivity ranking highest in different orders depending on the target property. The practical payoff is a set of explicit equations that water managers in Ibeator and geologically similar terrain can apply to new geoelectric data to estimate hydraulic conductivity, transmissivity, and storativity without expensive pumping tests. That, in turn, enables targeted protection: communities over shallow, highly transmissive aquifers can be prioritized for stricter sanitation controls and careful borehole siting, while deeper, better-protected zones can absorb more intensive development.

The broader lesson resonates well beyond southeastern Nigeria. In an era when deep learning dominates headlines, this study is a reminder that model sophistication must match data availability, and that transparent statistical models can deliver both accuracy and interpretability where data are scarce. The authors suggest that the hybrid strategy, using neural networks to probe nonlinear structure and regression to quantify the influence of each factor, offers a holistic picture of aquifer vulnerability and a practical path toward sustainable groundwater management, improved water security, and environmental conservation for the region’s growing population.

Subject of Research: Prediction of aquifer protective capacity and hydraulic parameters in southeastern Nigeria using artificial neural networks and multivariate linear regression applied to vertical electrical sounding data

Article Title: Enhancing aquifer protective capacity prediction over Ibeator and environ, Southeastern Nigeria using artificial neural networks and multivariate linear regression analysis

Article References: Usman, A. O., Akakuru, O. C., Azuoko, G.-B., Abraham, E. M., Chinwuko, A. I., & Chizoba, C. J. (2024). Enhancing aquifer protective capacity prediction over Ibeator and environ, Southeastern Nigeria using artificial neural networks and multivariate linear regression analysis. BMC Environmental Science, 1(1), Article 13. https://doi.org/10.1186/s44329-024-00013-3

Image Credits: AI Generated

DOI: 10.1186/s44329-024-00013-3

Keywords: groundwater, aquifer vulnerability, artificial neural network, multivariate linear regression, vertical electrical sounding, hydrogeophysics, hydraulic conductivity, transmissivity, storativity, Nigeria, geoelectric survey, water resource management

The conventional route to forward-looking characterisation factors runs through Integrated Assessment Models, or IAMs. These are vast coupled simulations of the economy, energy system, land use and climate that researchers use to sketch coherent scenarios of how the world might evolve. IAMs have powered most prospective life cycle assessments to date, including a well-known 2022 study that generated future water scarcity factors using output from the IMAGE model. But IAMs come with baggage. They are computationally expensive, they bundle similar technologies and regions into coarse aggregate categories, and their deterministic pathways can clash with the fine-grained spatial resolution that life cycle assessment demands. The characterisation factors they produce often do not line up neatly with the geographic units used in background inventory databases such as ecoinvent. Engberg and his collaborators hypothesised that a data-driven alternative, one that learns directly from historical records rather than from chains of socio-economic assumptions, could sidestep some of these frictions.

Their target variable was the AWARE factor, the consensus method adopted by the water-focused life cycle initiative WULCA, which expresses the relative availability of water in a region as the ratio between a consumption-weighted world average of available water and the water available in that specific basin. A high AWARE value means water is scarce; the factor can be read as the surface-time equivalent needed to generate one cubic meter of unused water in that place. The official AWARE dataset covers only a single year, 2010, which is far too thin to train a machine learning model. So the team rebuilt the factor from scratch, month by month, from 1960 to 2016, using the global freshwater model WaterGAP v2.2d. For each of roughly 9,700 watersheds they computed an availability-minus-demand quantity that subtracts human water consumption and environmental flow requirements from natural runoff, then normalised it against the world average. Environmental water requirements were weighted according to the scheme of Pastor and colleagues, protecting low-flow months more stringently than high-flow ones. The same boundary conditions as the original method, capping factors at 100 and flooring them at 0.01, were applied throughout.

Before any forecasting began, the team checked whether their reconstructed historical factors were trustworthy. Compared against the published 2010 AWARE values, their WaterGAP-based calculations showed a correlation of 0.81, high but not perfect, largely because the original method relied on an older, inaccessible version of the hydrological model. More telling was the comparison with the IMAGE-based factors of the earlier IAM study, which correlated at only 0.53 with the original values. Many basins that were genuinely water-rich appeared substantially scarcer in the IAM-derived dataset. This discrepancy matters: it suggests that the choice of underlying model and scenario machinery can shift characterisation factors dramatically, an uncomfortable truth for a field that treats these numbers as authoritative multipliers.

With the historical time series in hand, the researchers turned to model selection. They benchmarked several algorithms, including the gradient boosting methods XGBoost and LightGBM, alongside deep learning architectures designed for time series. The winner was N-Beats, a neural network architecture built from stacks of fully connected blocks that decompose a signal into interpretable trend and seasonality components. N-Beats is what forecasters call a global model: rather than fitting one model per basin, as classical approaches like ARIMA would require across thousands of series, it learns shared temporal patterns across all basins simultaneously. The deployed network used 24 stacks of two blocks each, with 136 units per layer, a learning rate of 0.001 and a batch size of 1024. Training was disciplined with early stopping, which halts the process when validation loss stalls for five consecutive epochs, and a learning rate scheduler that halves the rate whenever performance plateaus. The input window spanned 72 months, six full annual cycles, long enough to capture hydrological seasonality without dragging in outdated patterns that invite overfitting.

Performance was measured with the symmetric mean absolute percentage error, or sMAPE, a metric that penalises over- and under-forecasting equally and ranges from 0 to 200 percent. On the 1,463 largest basins, those spanning at least three half-degree grid cells, the N-Beats model achieved a median sMAPE of about 25 percent, and more than 60 percent of basins came in below 30 percent. Some basins were forecast with remarkable precision, errors near 3 percent, while the worst performers exceeded 50 percent and occasionally approached 78 percent. XGBoost and LightGBM fared worse, with maximum errors above 100 percent and a larger share of poorly predicted basins. On raw accuracy, the deep learning approach clearly outclassed both its statistical ancestors and its tree-based rivals.

But accuracy on monthly values tells only half the story, and it is the half that matters less. When the team smoothed the forecasts into 30-month moving averages to reveal long-term trajectories, cracks appeared. The models tracked seasonal oscillations faithfully, yet their multi-year trends often diverged substantially from the validation data. In the Volga basin, the gap between forecast and reality at the end of the validation period reached roughly 20 cubic meters of world-equivalent water per cubic meter consumed. Even some basins with seemingly excellent sMAPE scores showed poor alignment in their long-term behaviour, a reminder that a model can nail the seasonal rhythm while missing the underlying melody. Attempts to forecast annual values directly produced unsatisfactory results altogether.

The final deployment, trained on the full 1960 to 2016 record, projected AWARE factors out to 2032. According to the forecasts, more than 70 percent of basins will show lower water scarcity in 2030 than in 2010, while roughly 400 basins that already had factors above 60 were predicted to climb toward nearly 100, the ceiling of the scale. Comparing these projections with the IMAGE-based factors for 2030 revealed substantial disagreement in many regions, particularly near the equator, in eastern North America and in western Australia, while South America, Europe and Central Africa showed closer agreement between the two methods. The authors are careful to note that the comparison is imperfect, since the two approaches rest on different hydrological models and the IAM version embeds explicit socio-economic pathways, in this case a middle-of-the-road scenario, whereas the machine learning forecasts simply extrapolate past dynamics with no policy or demographic assumptions baked in.

That absence of assumptions is both the method’s defining feature and its Achilles heel. Time series forecasting, by construction, cannot anticipate structural breaks: a new water management regulation, a leap in irrigation efficiency, a wave of desalination plants, or an unprecedented drought all lie invisible to a model trained only on historical scarcity values. The authors are refreshingly blunt about the implications. Machine learning forecasts, they conclude, are not a standalone alternative to IAMs; they excel at capturing seasonal patterns but cannot reliably predict the long-term structural trends that matter most for prospective life cycle assessment. They also flag a subtle but significant finding for the field at large: the world-average availability term that anchors the AWARE method shifted by about 30 percent between 2010 and 2011, which alone moves every basin’s factor by the same magnitude. Even without any machine learning, that observation argues for treating single-year characterisation factors with caution and for embedding historical trends into impact assessment.

Still, the study opens doors that IAMs cannot reach. The approach could be extended to other characterisation factors where local historical time series exist, such as land use change, and the authors point to covariates from climate model ensembles like ISIMIP and CMIP, variables such as surface temperature, groundwater recharge and leaf area index, as a route to forecasts that respond to climate signals rather than merely repeating them. Transfer learning offers another avenue: a pretrained network could be adapted by practitioners with local data to generate factors for regions the global model never saw. The recently updated AWARE 2.0 method, with improved consumption and environmental flow calculations, may provide cleaner training data for follow-up work. For now, the message is one of calibrated optimism. Deep learning can render the seasonal pulse of the planet’s watersheds with impressive fidelity, and it does so at a spatial resolution that scenario models struggle to match. What it cannot yet do is tell us where the current of change is heading when the change itself has never happened before. Bridging that gap, between pattern recognition and genuine foresight, is the next challenge for anyone hoping to forecast the future state of the environment.

Subject of Research: Machine learning forecasting of water scarcity characterisation factors for prospective life cycle assessment

Article Title: Forecasting future states of the environment with machine learning: a case study on water scarcity

Article References: Engberg, N., Blanco, C. F., Barbarossa, V., Bakker, C., Balkenende, R., Lian, J. Z., & Sprecher, B. (2026). Forecasting future states of the environment with machine learning: a case study on water scarcity. Journal of Industrial Ecology, 30(4), 1285-1298. https://doi.org/10.1007/s44498-026-00010-6

Image Credits: AI Generated

DOI: 10.1007/s44498-026-00010-6

Keywords: machine learning, water scarcity, life cycle assessment, AWARE, time series forecasting, N-Beats, characterisation factors, integrated assessment models, WaterGAP, hydrology, prospective LCA, deep learning

The study’s setting could hardly be more consequential. The Zabol Basin sits at the downstream end of the transboundary Helmand River system, a landscape where decades of drought, upstream water diversion, and geopolitical tension have combined to drain the once-vast Hamun Lakes into salt flats. Communities in Iran’s Sistan and Baluchestan province depend on the timing and volume of Helmand flows for agriculture, drinking water, and protection from the region’s infamous dust storms. In such a basin, a reliable drought forecast is not an academic luxury; it is the difference between managed adaptation and humanitarian emergency. Yet forecasting here is uniquely hard because the river’s behavior is shaped by two countries, multiple dams, irrigation withdrawals, and a climate that is itself changing, all of which break the assumption that the past is a reliable guide to the future.

To quantify drought, the study relied on standardized indices computed from more than five decades of Helmand River streamflow data spanning 1961 to 2014. Three indices took center stage: the Standardized Runoff Index (SRI) and the Standardized Streamflow Index (SSI), both of which measure how far current water availability deviates from long-term norms, and a more ambitious third option, the Non-Stationary Standardized Streamflow Index (NSSI), which attempts to account for the fact that the statistical baseline itself shifts over time as human and climatic pressures reshape the river. The models were tasked with forecasting these indices at four time horizons: 1, 3, 6, and 12 months ahead. The inputs to the models were moving averages of streamflow and runoff, a technique that smooths out daily noise and lets the algorithms focus on the persistent signals that carry drought information across seasons.

The headline result is encouraging for seasonal water managers. For the stationary indices, SRI and SSI, both models performed remarkably well, with correlation coefficients between 0.95 and 1.00 and Nash-Sutcliffe Efficiency values, the standard hydrological yardstick that compares model predictions to a simple average-based baseline, ranging from 0.88 to 0.99. The sweet spot for both approaches was the 6- and 12-month scales, precisely the horizons at which seasonal drought monitoring is most useful for planning reservoir releases, crop choices, and emergency water allocations. In other words, when the underlying drought signal follows relatively stable statistical patterns, modern machine learning, whether built on support vector mathematics or on attention-based Transformer architectures, can capture it with near-perfect fidelity. PhysicsSolver edged out SVR-RSM in most comparisons, but the margins were modest rather than transformative.

The real story, and the scientifically provocative one, lies in what happened when the models confronted the non-stationary NSSI. Here the tidy agreement collapsed. At the 1-month horizon, PhysicsSolver demonstrated a clear advantage, achieving a Nash-Sutcliffe Efficiency of 0.98 compared with 0.88 for SVR-RSM, suggesting that the physics-informed Transformer’s ability to encode physical constraints and attend to long-range temporal dependencies gives it genuine power for very short-term prediction even when the data-generating process is shifting underfoot. But at 3- and 6-month horizons, both models failed outright, producing negative efficiency values, which in hydrological practice means the forecasts were worse than simply guessing the historical mean. The finding is a sobering reality check for a field that has grown accustomed to celebratory performance metrics.

Why does non-stationarity break medium-term forecasting so completely? The answer lies in what the NSSI is trying to represent. A stationary index assumes that the probability distribution of streamflow is fixed, so a drought is simply an unusually low draw from a known deck of cards. The non-stationary index acknowledges that the deck itself is being reshuffled by upstream dam operations, changing irrigation demand, land-use shifts, and evolving climate patterns. When a model trained on historical data tries to forecast several months ahead, it must implicitly extrapolate how those human and climatic drivers will evolve, and neither a support vector machine nor a physics-informed Transformer, however sophisticated, can conjure information about future dam releases or geopolitical water-sharing decisions that is not present in the training data. The physics constraints embedded in PhysicsSolver help it stay physically plausible, but they cannot substitute for knowledge of anthropogenic forcing.

The architecture behind PhysicsSolver deserves attention because it represents a broader movement in the geosciences. Physics-informed neural networks embed physical laws, such as mass conservation or flow equations, directly into the training objective, penalizing solutions that fit the data but violate known physics. The Transformer component, borrowed from the deep learning revolution in language modeling, uses attention mechanisms to weigh which parts of the historical record matter most for a given prediction, allowing the model to capture long-range temporal dependencies that older recurrent architectures struggle with. The concept was originally developed for solving and forecasting partial differential equations, and its adaptation to drought indices is part of a wave of hybrid approaches seeking to combine the flexibility of data-driven learning with the reliability of physical understanding, particularly valuable in data-scarce regions where pure machine learning risks learning spurious correlations.

For the Zabol Basin and the millions who depend on the Helmand, the practical implications are twofold. First, the study validates a workable toolkit for seasonal drought monitoring: agencies can use either model, at 6- to 12-month scales, to anticipate drought conditions with high confidence, providing lead time for water rationing, crop switching, and international coordination. Second, and more soberly, the study shows that medium-term forecasting of drought in human-dominated basins remains an open problem that no amount of architectural cleverness alone can solve. The authors’ conclusion is explicit: substantial methodological advances are needed before non-stationary drought can be forecast reliably at the 3- to 6-month horizons where early warning would matter most. That likely means incorporating covariates that explicitly represent anthropogenic pressures, such as upstream reservoir storage, irrigation withdrawals, and climate oscillation indices, rather than expecting streamflow history alone to carry the signal.

The broader lesson resonates far beyond the Iran-Afghanistan border. Transboundary basins cover nearly half of the world’s land surface and supply water to some two billion people, and many of them, like the Helmand, are experiencing exactly the kind of compound human-climate stress that renders historical statistics unreliable. As climate change accelerates and water infrastructure multiplies, the assumption of stationarity that underpins much of hydrology is eroding everywhere. Studies like this one perform a valuable service by mapping, with honest numbers, where the current generation of AI tools succeeds and where it hits a wall. PhysicsSolver’s near-perfect short-term performance on non-stationary indices hints that physics-informed architectures are the right direction of travel; its equally dramatic failure at medium horizons tells researchers precisely where the next breakthrough must come from. In the arid lands of Sistan, where the Hamun wetlands have already paid the price of unforecastable drought, that breakthrough cannot arrive soon enough.

Subject of Research: Physics-informed machine learning for hydrological drought forecasting in the transboundary Zabol Basin

Article Title: PhysicsSolver: A physics-informed transformer for hydrological drought forecasting in the transboundary Zabol Basin

Article References: Piri, J. (2026). PhysicsSolver: A physics-informed transformer for hydrological drought forecasting in the transboundary Zabol Basin. Earth Science Informatics, 19(10), Article 167. https://doi.org/10.1007/s12145-026-02214-7

Image Credits: AI Generated

DOI: 10.1007/s12145-026-02214-7

Keywords: hydrological drought, physics-informed neural network, Transformer, Zabol Basin, Helmand River, transboundary water, non-stationarity, drought indices, machine learning, streamflow forecasting, Sistan, water resource management

The stakes in this landscape could hardly be higher. Lake Urmia, once one of the largest hypersaline lakes on Earth, has shrunk dramatically over recent decades, and the aquifers surrounding it supply drinking water, irrigation, and ecosystems across a semi-arid basin. As the lake desiccates, saline water and changing recharge patterns push hydrochemical conditions in nearby groundwater into flux. Water managers need to know not just where quality stands today, but where it is heading, so that agricultural planning and environmental protection can respond before wells turn brackish. The study was designed to give them exactly that foresight.

The researchers assembled records from 54 monitoring stations spanning 2008 to 2021, drawing on the physical and chemical parameters measured at each site. From these they calculated an annual WQI for every station, a single composite number that condenses multiple chemical indicators into one interpretable score of overall water quality. To visualize how quality varied across the landscape and through time, they mapped the index values using inverse distance weighting, a geostatistical interpolation method that estimates values at unsampled locations based on the weighted influence of nearby measurements. The resulting maps exposed clear spatial and temporal gradients, with the northeastern region standing out as the area of greatest concern by the end of the study period.

With the historical picture established, the team turned to prediction. Three models were trained to forecast the WQI for the year following each observation. The baseline was a standard artificial neural network, or ANN, a flexible but relatively simple learner. The second was a hybrid convolutional neural network paired with a long short-term memory network, known as CNN-LSTM. The convolutional layer excels at extracting local patterns and features from input sequences, while the LSTM component is specifically designed to retain information over long time lags, making it well suited to the slow-moving, memory-laden dynamics of groundwater systems. This combination has proven powerful in other hydrological forecasting problems, and it outperformed the plain ANN here as well.

The third model added a crucial preprocessing step that proved decisive. Successive Variational Mode Decomposition, or SVMD, is a signal decomposition technique that breaks a complicated, noisy time series into a set of distinct intrinsic mode functions, each representing a cleaner, more regular component of the original signal. Groundwater quality data are notoriously non-stationary: they drift with drought cycles, pumping regimes, and salinity intrusion, all superimposed on measurement noise. By decomposing the input before it reaches the deep learning stage, SVMD reduces that non-stationarity and strips away noise, presenting the network with simpler, more learnable patterns. The idea echoes a broader trend in hydrological machine learning, where decomposition-based hybrids have improved forecasts of everything from dissolved oxygen to lake levels.

The performance numbers tell a striking story. The CNN-LSTM model already beat the ANN, achieving a root mean square error of 46.881, a coefficient of determination of 0.921, a Nash–Sutcliffe efficiency of 0.791, and a mean absolute percentage error of 0.097. Each of these metrics captures a different facet of accuracy: RMSE penalizes large mistakes, R-squared measures how much of the variance the model explains, NSE compares predictions against the simple benchmark of using the observed mean, and MAPE expresses typical error in relative terms. Then the SVMD-CNN-LSTM pushed every metric further, delivering an RMSE of 29.654, an R-squared of 0.967, an NSE of 0.946, and a MAPE of just 0.032. Relative to the ANN, the full hybrid reduced error by 58.28 percent, and relative to CNN-LSTM alone, by 36.75 percent.

Those gains matter because hydrochemical time series in this region are short and data-limited. Fourteen annual observations per station is a modest dataset by deep learning standards, and noisy, non-stationary signals make every data point harder to exploit. The study demonstrates that signal decomposition before deep learning substantially enhances predictive accuracy precisely in these challenging conditions. In practical terms, a water manager using the SVMD-CNN-LSTM model gets a forecast close enough to act on, whether that means adjusting irrigation allocations, flagging wells at risk of exceeding salinity thresholds, or prioritizing monitoring investment in the most vulnerable zones.

The spatial findings add urgency to the modeling. By 2021, WQI values in the northeastern part of the basin had fallen into the very poor category, a deterioration the authors link to the broader environmental stress afflicting the Lake Urmia system. Previous work in the region has documented how drought and land-use change degrade aquifer quality around the shrinking lake, and the new maps and forecasts provide a quantitative, station-by-station record of that decline. For a semi-arid, saline lake basin, the ability to anticipate which areas will cross critical quality thresholds gives planners a window for intervention that historical monitoring alone cannot offer.

Beyond its immediate regional value, the framework is designed to travel. The authors describe it as a transferable approach that can be extended to other basins, integrated with additional hydrogeological drivers such as recharge estimates or aquifer properties, and coupled with uncertainty quantification to support decision-making under changing environmental conditions. As climate change intensifies drought pressure on aquifers worldwide, tools that convert sparse monitoring data into reliable one-year-ahead quality forecasts are likely to move from research novelty to operational necessity. For the communities around Lake Urmia, the new model offers something rare: a quantitative glimpse of their groundwater’s future, early enough to change it.

Subject of Research: Hybrid deep learning prediction of spatiotemporal groundwater quality index in the Eastern Lake Urmia region

Article Title: A novel SVMD-CNN-LSTM hybrid model for spatiotemporal groundwater quality index prediction in the ecologically sensitive Eastern Lake Urmia Region

Article References: Abdi, E., Asadi, E., Zarrintan, N., & Ibrahim, O. R. (2026). A novel SVMD-CNN-LSTM hybrid model for spatiotemporal groundwater quality index prediction in the ecologically sensitive Eastern Lake Urmia Region. Earth Science Informatics, 19(10), Article 172. https://doi.org/10.1007/s12145-026-02228-1

Image Credits: AI Generated

DOI: 10.1007/s12145-026-02228-1

Keywords: groundwater quality, Water Quality Index, Lake Urmia, SVMD, CNN-LSTM, machine learning, signal decomposition, hydrochemistry, spatiotemporal prediction, semi-arid basin, deep learning, water resource management

The core problem the researchers tackle is that conventional WHPA delineation relies on deterministic models built from fixed parameter values. Hydraulic conductivity, aquifer thickness, and effective porosity are each estimated from sparse field data, and a single ‘best estimate’ of each is fed into a model that then produces one crisp boundary. Studies dating back to the 1990s have shown that such deterministic approaches can produce overly optimistic delineations that mask the true range of plausible capture zones. Because WHPA boundaries carry real consequences, including land-use restrictions and potential compensation for affected landowners, an artificially narrow zone can leave a drinking water source exposed while an inflated one can impose unnecessary burdens. Recent legislative shifts, notably in Sweden, now demand risk-based approaches that explicitly account for uncertainty, but the advanced numerical tools capable of doing so, such as MODFLOW coupled with groundwater modeling systems, demand data, expertise, and computational resources that many rural utilities simply do not have.

The workflow developed by the Swedish team combines two established techniques in a novel, accessible package. The first is the Analytic Element Method, or AEM, pioneered by Otto Strack, which represents groundwater flow by superimposing elementary analytical solutions for wells, rivers, lakes, and other hydrogeological features. Because AEM requires no spatial grid, it avoids grid-related numerical artifacts and lets modelers specify inputs directly in terms of real features and boundary conditions, striking a balance between the simplicity of closed-form analytical methods and the flexibility of full numerical models. The second ingredient is Monte Carlo simulation, a decades-old statistical technique that propagates uncertainty by repeatedly sampling input parameters from probability distributions and running the model anew for each sample. Coupling the two produces not a single capture zone but an entire ensemble of equally plausible ones, each reflecting a different combination of aquifer properties.

Technically, the researchers built their workflow around TimML, an open-source Python and Fortran AEM package created by Mark Bakker, whose open code made it possible to add a dedicated Monte Carlo component. The implementation is delivered as three Jupyter notebooks covering the full pipeline: a pre-processor that defines uncertain inputs as probability distributions, a sampling engine that runs TimML repeatedly with different parameter sets, and a post-processor that visualizes the results in a GIS environment. Three parameters were treated as uncertain because they are typically poorly constrained yet strongly influence travel-time-based capture zones: hydraulic conductivity, saturated aquifer thickness, and effective porosity. Hydraulic conductivity was represented as lognormally distributed, a standard choice in hydrogeology because the parameter is strictly positive and often spans orders of magnitude; the authors verified this assumption with quantile-quantile plots and a Shapiro-Wilk test on the log-transformed data from 70 Hazen-based estimates. Thickness and porosity were assigned truncated normal distributions bounded by physically plausible limits, with the bounds set at roughly the mean plus or minus three standard deviations.

The treatment of uncertainty is conceptually careful. The authors distinguish aleatory uncertainty, arising from natural spatial variability in the aquifer, from epistemic uncertainty, which stems from limited measurements and imperfect knowledge of site conditions. In the present implementation, the aquifer is represented as a single hydrogeological unit with effective properties, so the Monte Carlo ensemble primarily captures epistemic uncertainty in the effective parameterization. Where conservative estimates were needed, the team computed upper confidence limits of the mean using t-distributions, and for lognormal data applied a Cox-modified method in log space. For each Monte Carlo realization, a unique parameter set is sampled and assigned to the TimML model, reverse particle tracking generates pathlines for the chosen travel times, and the procedure repeats until an ensemble of plausible WHPA realizations accumulates. Latin Hypercube Sampling could reduce the number of runs, but ordinary runtimes proved manageable on available hardware.

The output is summarized through percentile-based envelopes rather than a single boundary. Particle locations along the simulated pathlines are pooled across all realizations, and percentiles of their distances from the pumping well define nested WHPA polygons. Lower percentiles yield larger, more precautionary zones because only points beyond the percentile distance are excluded before polygon construction, while higher percentiles yield tighter, more central delineations. The researchers computed the 50th, 75th, 95th, and 99th percentile envelopes using a convex hull algorithm, exported them as georeferenced GIS layers in the Swedish SWEREF 99 coordinate system, and thereby produced maps that water managers can overlay directly on land-use data. This percentile framework turns an abstract statistical ensemble into something a municipal planner can read, compare, and discuss with stakeholders.

The case study applied the workflow to the Varnum aquifer near Borås in southwest Sweden, an unconfined glaciofluvial delta deposit forming part of the Rångedala esker. The aquifer spans roughly three square kilometers in a valley at about 170 to 175 meters above sea level, with deposit thicknesses between 20 and 55 meters, and consists of fine sand overlying medium to coarse sand on bedrock. Ten years of historical head measurements show a stable water table, justifying a steady-state model under the Dupuit-Forchheimer approximation. Crucially, the team benchmarked their probabilistic results against an existing deterministic numerical model of the site that had been developed, calibrated, and officially approved by the Municipality of Borås. The digitized 100-day pathlines from that numerical model fell squarely within the central portion of the probabilistic ensemble, most consistent with the 50th to 75th percentile envelopes, exactly what one would expect if a deterministic parameter set represents one plausible draw near the center of the assumed distributions.

The magnitude of the uncertainty effect is striking. At the 50th percentile, the 100-day WHPA covered 32,260 square meters, or about 3.23 hectares, concentrated around the well. At the 75th percentile it ballooned to roughly 15.6 hectares, and at the 95th and 99th percentiles it reached 61.7 and 102.1 hectares respectively, a more than thirtyfold expansion across the range. As the envelope widens, it progressively intersects agricultural land, residential areas, cemeteries, and peat bogs, each bringing new potential contaminant sources into scope, from septic systems and household chemicals to leaching from burial grounds and elevated dissolved organic carbon in peatland environments. The authors emphasize that this exposes a genuine trade-off: the safest decision under uncertainty may differ from the statistically optimal one, and choosing a very conservative percentile pulls farmland, forest, and peatland into the protection zone with corresponding management implications, such as fertilizer restrictions. They also caution that final zoning in practice is often adjusted to follow roads and parcel boundaries, which can amplify the effect of the percentile choice on the ultimate protected area.

The researchers are candid about the limitations. Results depend on the probability distributions assigned to the inputs, which at data-poor sites should be read as plausible representations of uncertainty rather than definitive aquifer characterizations, and the parameters were sampled independently, a deliberate simplification that correlated sampling could later relax. The single-unit aquifer representation matches common practice for small supplies, where the realistic alternative is usually a deterministic model under the same homogeneous assumption rather than a heterogeneity-resolving model. Computational cost grows with ensemble size, and location-specific capture probabilities would require additional post-processing. Yet the value proposition is clear: the workflow is a screening-level first step for sites with limited but not absent data, requiring only basic hydrogeological and GIS skills, and integrating preprocessing, modeling, and GIS-ready output in one Python environment reduces the software fragmentation that undermines transparency and reproducibility. As the authors note, risk communication research has long recognized that getting the numbers right is only the beginning; making uncertainty visible, as these percentile envelopes do, allows decision-makers and communities to debate precaution versus land-use burden concretely. For the thousands of small and medium-sized water supplies worldwide that currently rely on a single deterministic boundary despite substantial subsurface uncertainty, this open-source workflow offers a defensible, feasible path toward risk-based groundwater protection.

Subject of Research: Probabilistic delineation of groundwater wellhead protection areas using Monte Carlo simulation and the Analytic Element Method

Article Title: Accessible probabilistic modeling of wellhead protection areas for small and medium-sized water supplies

Article References: Gärtner, N., Zamzami, M., & Lindhe, A. (2026). Accessible probabilistic modeling of wellhead protection areas for small and medium-sized water supplies. Discover Geoscience, 4(1), Article 378. https://doi.org/10.1007/s44288-026-00741-w

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00741-w

Keywords: groundwater, wellhead protection areas, Monte Carlo simulation, analytic element method, hydrogeology, drinking water, uncertainty analysis, open-source software, TimML, water policy, risk assessment, GIS

The research team, led by Hyeong-Yun So, Hyeon-Cheol Yoon, and Tae-Gyun Kim of the National Integrated Drought Center at the National Disaster Management Research Institute, together with Se-Jeong Lee of the Korea Institute of Hydrogical Survey, focused on the domestic and industrial water supply sector of South Korea’s drought warning system. Unlike the meteorological sector, which relies on indices such as the Standardized Precipitation Index, or the agricultural sector, which monitors reservoir storage and soil moisture, the domestic and industrial sector bases its four warning stages—attention, caution, alert, and severe—on dam-specific water supply adjustment criteria tied to storage volumes and storage ratios. Predicting those storage levels quantitatively, months in advance, therefore translates directly into earlier and better-calibrated drought warnings for cities and industry.

The study area was Juam Dam itself, a multipurpose dam in the Boseonggang River basin, a tributary of the Seomjingang River, with a total storage capacity of 457 million cubic meters. During the prolonged meteorological drought that stretched from the second half of 2022 into the first half of 2023, Juam Dam’s storage fell to the severe stage, disrupting major industrial complexes at Yeosu and Gwangyang and prompting consideration of domestic water supply restrictions in Gwangju. The researchers assembled 32 years of daily hydrological and meteorological observations, spanning 1991 to 2022, from the National Water Resources Management Information System and the Korea Meteorological Administration’s Weather Data Portal, and reserved the drought-stricken first half of 2023 as an independent test period.

A central challenge emerged immediately from correlation analysis. Using the Pearson correlation coefficient after min–max normalization, the team found that even the most strongly related input variable, dam outflow, showed a correlation with storage of only r = 0.253, followed by inflow, humidity, wind speed (negatively correlated), temperature, and rainfall. In other words, the linear relationships between the available observations and the target variable were weak across the board. This is precisely the regime in which deep learning models, capable of capturing temporal lags and complex nonlinear interactions, outperform traditional regression and conventional machine learning approaches—but only if the input representation is designed well. The team therefore applied feature engineering, generating lagged variables, arithmetic transformations, and moving averages from the raw observations, and built three distinct training datasets. Dataset A contained only the original variables; Dataset B incorporated all expanded features; Dataset C retained only the strongly correlated expanded features derived from inflow, outflow, and antecedent storage, deliberately excluding meteorological parameters.

Each dataset was paired with four deep learning architectures, producing twelve candidate models. The Long Short-Term Memory network, a recurrent architecture with forget, input, and output gates and a cell state that preserves long-term information, served as the field’s standard baseline. The Bidirectional LSTM added a backward pass over the sequence, combining past and future context. The hybrid CNN-LSTM stacked one-dimensional convolutional feature extraction in front of recurrent sequence learning. The fourth architecture, the Temporal Convolutional Network, replaced recurrence entirely with causal one-dimensional convolutions—ensuring that predictions depend only on current and past inputs—and dilated convolutions, which insert gaps between filter elements to expand the receptive field exponentially across time. That dilation mechanism allows the TCN to learn patterns over very long horizons while remaining fully parallelizable during training, a structural advantage over step-by-step recurrent processing.

The comparative results were decisive. Models trained on Dataset B, with its full set of engineered features, reduced test RMSE by roughly 14 percent relative to Dataset A, while Dataset C—stripped of meteorological variables—performed dramatically worse, with test RMSE increasing by about 34 percent over Dataset A. The lesson was counterintuitive but important: although temperature, humidity, and wind speed are physically coupled to inflow and outflow through evapotranspiration, removing them deprived the models of nonlinear context about long-term storage depletion. Deep learning architectures, the authors conclude, thrive on multi-dimensional interactions among many weakly correlated variables rather than on a few strongly correlated linear inputs. Across architectures, the TCN dominated, achieving average RMSE values of 12.150 in training, 8.823 in validation, and 21.337 in testing, while the LSTM fared worst at 44.818, 17.296, and 57.950 respectively—reductions of roughly 73, 49, and 63 percent in the TCN’s favor. Qualitatively, however, the LSTM reproduced the temporal shape of storage variation most faithfully, and the TCN delivered the closest numerical agreement, a tension the team resolved by optimizing the TCN further.

Optimization proceeded through a two-stage grid search over the TCN’s hyperparameters, which are known to strongly influence its behavior: activation function, dilation rate, filter size, kernel size, learning rate, batch size, number of epochs, and the lookback period of historical input. The first stage, generating 256 model configurations, converged on the ReLU activation, a dilation rate sequence of 1, 2, 4, and 8, and 200 training epochs. The second stage revealed a striking finding: the optimal lookback window was 1,460 days—approximately four years. The authors attribute this to the operational rhythm of South Korean dam management, in which the standardized Dam Water Supply Adjustment Criteria are revised every two to four years. A four-year memory allows the model to internalize these cyclical regulatory shifts and the human-controlled discharge patterns they produce. Notably, the configuration with the lowest training and validation error showed unstable oscillatory predictions and poor generalization, so the final hyperparameters were selected from the first-stage results to guarantee robustness over raw precision.

The final optimized model was evaluated four times across 2023, each with a different forecast origin, and its predictions were compared against Juam Dam’s operational warning thresholds, where consecutive drought stages are separated by roughly 4.5 to 5 percent of total storage. The model captured the severe drought of early 2023, reproduced the storage recovery toward the end of June, tracked spring and flood-season dynamics, and matched the stable conditions of late year. Its one clear weakness was the abrupt storage surge in mid-July driven by an extreme rainfall event totaling 306 millimeters in five days, which the model underestimated. Quantitatively, the four evaluations yielded an average Mean Absolute Error of 28.491, an average RMSE of 35.440, and an average error rate of 6.23 percent of total storage, with the best quarter achieving an error rate of just 3.68 percent. In earlier comparative testing, the TCN model also achieved a 78.9 percent hit rate within a 5 percent error margin of observed storage, with a correlation coefficient exceeding 0.6 in the Taylor diagram analysis.

The authors are candid about what a 6.23 percent error rate means operationally. Because warning stages are densely spaced, a forecast of this precision could misclassify a specific stage if used for short-term, deterministic actions such as triggering water restrictions, which demand errors below roughly 3 percent. But the purpose of a six-month outlook is different: it functions as an early-stage monitoring tool for macro-level storage trends and precursory signs of impending shortage. Within that framing, capturing a downward storage trajectory within about 6 percent provides water managers an extended proactive window to secure alternative supplies and establish conservative safety margins well before crisis points arrive—the very lead time that was missing during the Juam and Obong crises.

The research team outlines an ambitious path forward. They plan to extend the framework to all 33 dams covered by South Korea’s domestic and industrial drought forecasting system, developing dam-specific datasets and models that reflect each reservoir’s unique hydrology, and to investigate stretching predictions to a twelve-month horizon. An ensemble framework is planned that would combine the TCN’s error-suppressing precision, the LSTM’s skill at reproducing temporal trends, and the Transformer’s capacity for long-range context, with performance assessed using additional metrics such as the Nash-Sutcliffe Efficiency and the Kling-Gupta Efficiency. To address the models’ struggle with the non-stationary extremes of a changing climate—intense short-duration rainfall and prolonged drought alike—the team intends to accumulate training data on unprecedented events and generate synthetic, physics-informed virtual datasets. If those efforts succeed, the six-month dam storage forecast demonstrated here could evolve from a research result into a routine instrument of national drought preparedness, giving water managers the one resource drought victims never have enough of: time.

Subject of Research: Midterm drought forecasting using deep learning prediction of dam reservoir storage dynamics

Article Title: Midterm drought forecasting based on dam storage prediction using deep learning algorithms

Article References: So, H.-Y., Yoon, H.-C., Kim, T.-G., & Lee, S.-J. (2026). Midterm drought forecasting based on dam storage prediction using deep learning algorithms. Environmental Earth Sciences, 85(16), Article 407. https://doi.org/10.1007/s12665-026-13138-2

Image Credits: AI Generated

DOI: 10.1007/s12665-026-13138-2

Keywords: drought forecasting, deep learning, temporal convolutional network, LSTM, dam storage prediction, feature engineering, grid search, hyperparameter optimization, water resource management, South Korea, Juam Dam, hydrological modeling

The research, led by Mohammadali Alijanian, Narjes Salmani-Dehaghi and Hamed Yazdian of the University of Isfahan, addresses a question that sounds almost trivially simple until you realize how much rides on the answer. The water balance of a landscape can be expressed in several mathematically defensible ways. You can treat storage change as the sum of surface water and groundwater changes. You can compute it as precipitation minus evapotranspiration and runoff. Or you can take that three-variable formulation and adjust it to account for groundwater withdrawals, the water pumped out of aquifers that may never show up as streamflow. Each version is internally consistent, yet each can yield dramatically different estimates of how fast a region is draining or refilling its water reserves.

Disentangling this ambiguity required serious data engineering. The team first confronted GRACE’s coarse resolution, roughly 150,000 to 200,000 square kilometers per pixel, far too broad for local water management. They downscaled the satellite observations to a much sharper 0.25-degree grid, approximately 25 to 30 kilometers, using a Spatially Promoted Support Vector Machine, or SP-SVM, model. This machine learning approach, previously developed by the same group, fuses ground-based and satellite datasets to sharpen the gravity signal without drowning it in noise. The downscaled estimates were then compared against independent in-situ observations across four sub-basins, giving the researchers a rigorous test bed spanning the years 2005 to 2020.

Against these data, the team pitted three candidate formulations of water balance change, which they abbreviated WB-SG, WB-3V and WB-4V. WB-SG simply adds up changes in surface water and groundwater storage. WB-3V calculates storage change as precipitation minus evapotranspiration and runoff, the classic flux-based approach. WB-4V extends that framework by adjusting for groundwater withdrawal, acknowledging that in heavily pumped basins, extraction itself is a significant term in the ledger. The trio then evaluated all three formulations at both monthly and annual timescales, deploying two of hydrology’s workhorse statistical tools: the Mann-Kendall trend test and Sen’s slope estimator, applied to both original and detrended series to separate long-term signals from seasonal cycles.

The verdict was clear. The three-variable formulation, WB-3V, proved the most accurate match to GRACE-derived water balance change. In the monthly analysis using the original, untrended data, WB-3V achieved coefficients of determination ranging from 0.56 to 0.63, with root mean square errors between 5.12 and 11.08 centimeters of equivalent water height. Its rival WB-SG performed dismally by comparison, explaining almost none of the variance with R-squared values of just 0.03 to 0.08 and errors that ballooned to nearly 20 centimeters in some sub-basins. The contrast matters because the flux-based approach inherently captures the full hydrological cycle, whereas a simple sum of surface and groundwater changes omits soil moisture and snowpack, two reservoirs that dominate storage variability in semi-arid regions.

One of the study’s most methodologically interesting findings concerns detrending. When the researchers stripped out long-term trends from the time series before analysis, the root mean square error dropped significantly for every formulation tested. This makes physical sense: persistent trends, such as steady aquifer depletion driven by years of over-pumping, can mask the seasonal and interannual fluctuations that GRACE and ground observations share. By isolating the variability around the trend, the agreement between satellite and in-situ estimates sharpened, suggesting that trend contamination has been quietly degrading water balance comparisons in previous studies. For analysts auditing drought-prone basins, detrending may be a cheap and powerful preprocessing step.

The practical payoff goes beyond picking a winner among three equations. The authors demonstrate that GRACE can more effectively estimate unrecorded terrestrial water balance changes by applying adjustment coefficients derived from the statistical relationship between the GRACE-based water balance and the WB-3V formulation. In regions where hydrological records are sparse, politically fragmented or simply never collected, this offers a way to reconstruct the missing ledger from orbit. That is a tantalizing prospect for arid and semi-arid basins in the Middle East, Central Asia and beyond, where unregistered groundwater extraction runs into billions of cubic meters per year and confounds every conventional accounting method.

The study is also a reminder of how much the GRACE enterprise has matured since the satellites launched in 2002. Early applications treated the gravity data as a blunt instrument, good for continent-scale assessments of ice loss and major aquifer decline. Today, downscaled products can interrogate sub-basin dynamics, and machine learning frameworks like SP-SVM have made the transition from research curiosity to operational tool. The Isfahan-based team, working in one of the world’s most water-stressed countries, embodies that shift: their analyses lean on decades of accumulated ground truth, refined satellite retrievals and careful statistical hygiene to turn a noisy planetary scale reading into something a water manager can act on.

For the broader hydrology community, the message is that formulation choice is not a formality. Researchers combining GRACE data with precipitation, evapotranspiration and runoff products must consciously choose how they define storage change, and the wrong choice can silently undermine their conclusions. The four-variable version, adjusted for groundwater withdrawal, did not win the accuracy contest here, but the study’s framework shows how such adjustments could be tuned regionally through calibration coefficients. As GRACE Follow-On extends the gravity record and downscaling techniques push effective resolution ever finer, identifying the right water balance formulation becomes a foundational question for anyone trying to close the water budget in a warming, increasingly thirsty world.

Subject of Research: Identifying the most accurate terrestrial water balance formulation using downscaled GRACE satellite gravity data

Article Title: Identifying the Terrestrial Water Balance Formulation Using Downscaled GRACE Data

Article References: Alijanian, M., Salmani-Dehaghi, N., & Yazdian, H. (2026). Identifying the Terrestrial Water Balance Formulation Using Downscaled GRACE Data. Water Resources Management, 40(11), Article 523. https://doi.org/10.1007/s11269-026-04664-6

Image Credits: AI Generated

DOI: 10.1007/s11269-026-04664-6

Keywords: GRACE, terrestrial water storage, water balance, satellite gravimetry, SP-SVM downscaling, groundwater, evapotranspiration, runoff, Mann-Kendall test, Sen's slope, machine learning, hydrology

The research, published in the journal Water Resources Management, addresses a persistent weakness in conventional reservoir allocation. Traditional approaches, such as Brazil’s national methodology formalized by the National Water and Basic Sanitation Agency (ANA), rely on fixed operating rules, predefined hydrological states, and rigid planning horizons. These simplifications, the authors argue, fail to capture the enormous diversity of reservoir sizes and shapes found across semiarid landscapes—where storage capacities in their dataset ranged from less than 10 cubic hectometers to more than 700—and can obscure the real risks that decision-makers and water users face.

At the heart of the new framework are “allocation abaci”—graphical calculation charts that synthesize how a reservoir behaves under many combinations of initial storage, demand levels, and planning horizons. Rather than producing a single yes-or-no answer based on a fixed set of conditions, each abacus maps a continuous landscape of possibilities, allowing managers to see at a glance whether a proposed withdrawal is sustainable, how much water would remain at the end of a planning period, and how close the system would come to collapse. In total, the team generated 650 such abaci across five simulation horizons for each reservoir analyzed.

The study area is the state of Paraíba, a territory of roughly 56,000 square kilometers, about 70 percent of which lies within Brazil’s semiarid region—one of the most drought-vulnerable zones in the country. Rainfall there is extraordinarily variable, ranging from around 1,400 millimeters per year in the humid coastal Mata region to as little as 400 millimeters in the interior Borborema highlands. Meanwhile, potential evaporation remains relentlessly high, between 1,200 and 1,800 millimeters annually, and is relatively stable across the seasons. This climatic asymmetry—rain arriving in pulses, evaporation draining steadily year-round—is what makes storage-dominated water systems in the region so fragile, and why the researchers placed evaporative losses at the center of their methodology.

To characterize that evaporation, the team needed to know precisely how each reservoir’s geometry changes as water levels fluctuate. They compiled elevation–area–volume (EAV) curves for 130 monitored reservoirs, describing how each basin’s surface area expands or contracts as storage rises and falls. This matters because evaporation acts on the water surface: a shallow, sprawling reservoir that spreads rapidly as it fills loses proportionally far more water to the atmosphere than a deep, steep-walled one holding the same volume. The researchers then applied hierarchical cluster analysis—validated with standard statistical criteria including the Calinski–Harabasz index and silhouette coefficients—to group the reservoirs into five morphometrically distinct classes based on the shapes of their standardized elevation–area curves.

The clustering revealed striking differences in vulnerability. Reservoirs in Cluster 5, characterized by irregular geometries with limited surface area at low storage but rapid expansion near full capacity, proved the most susceptible to evaporative losses, particularly when starting from high storage levels. At the other extreme, Cluster 4 reservoirs, with more regular, roughly cylindrical shapes, showed the lowest sensitivity, losing only about 1.25 percent of their total volume per 10 percent decrease in initial storage, compared with roughly 2 percent for the most sensitive classes. Because reservoirs in Clusters 2, 3, and 1 dominate the regional network—comprising 66, 42, and 11 reservoirs respectively—the classification offers water agencies a practical way to stratify their infrastructure by inherent risk profile.

The simulation engine behind the abaci is deliberately conservative. Rather than attempting to forecast river inflows, which in intermittent semiarid streams are highly unreliable, the model assumes zero inflow and tracks only the “water bank” already in storage, depleted by evaporation and withdrawals. The researchers describe this as a precautionary screening baseline: it ensures that allocation decisions rest on tangible stored volumes rather than on uncertain recharge, providing a robust safety margin during the critical months when evaporation and consumption are the only predictable drivers of depletion. In more humid regions, they note, the framework could be adapted by adding stochastic inflow scenarios.

The team demonstrated the tool’s power by comparing it directly with the conventional ANA approach at the Acauã reservoir, one of Paraíba’s strategic storage sites. Under the traditional method, hydrological states such as the Green, Yellow, and Red categories are defined by guide curves tied to a fixed planning horizon—in this demonstration, corresponding to demands of 2,800, 2,000, and 1,200 liters per second. The new abaci reproduced those classifications but expanded the analysis across a continuum of horizons. For an 18-month planning period, the reservoir would need initial volumes of approximately 169, 128, and 86 cubic hectometers to satisfy the Green, Yellow, and Red state demands respectively. Stretch the horizon to 30 months, and even a completely full reservoir could no longer guarantee the Green state demand, while the Yellow and Red demands would require roughly 210 and 139 cubic hectometers. Read another way, at maximum initial storage, the Green demand could be sustained for only about 27 months, the Yellow for around 36, and the Red for more than 48.

These numbers carry real consequences for the region’s negotiation-based allocation system, in which government agencies, water users, and civil society jointly decide how to ration water during droughts. Because the abaci display not only whether a demand can be met but also the final storage percentage—the safety margin—for every scenario, stakeholders can explore both conservative and permissive options with a shared, transparent evidence base. The framework also distinguishes sharply between reservoirs: large systems such as Coremas and Acauã display high resilience, sustaining demands of 1,000 liters per second over long horizons with only moderate depletion, while Acauã collapses if demands exceed roughly 2,400 liters per second over 30 months. Medium reservoirs such as Araçagi and Capoeira are far more sensitive to both demand and horizon length, and small systems like Marés deplete rapidly due to limited storage and proportionally larger evaporative losses.

The study builds on a growing body of research documenting how evaporation from reservoirs represents a substantial and often overlooked drain on stored water, particularly in water-stressed regions. Recent work by Lorenzo-Lacruz and colleagues in 2025, along with studies by Zhao and Gao and by Nevermann and colleagues, has shown that intensive damming and regulation can intensify evaporative losses and erode usable storage. What the Brazilian team contributes is the translation of that concern into an operational allocation instrument—one that quantifies how evaporation interacts with reservoir geometry, demand, and time, and flags precisely when storage becomes vulnerable.

The authors are candid about the framework’s data requirements. Applying it elsewhere demands three core inputs: reliable and regularly updated EAV curves, which are especially important in reservoirs affected by sedimentation; representative and recent climatic data, particularly precipitation and evaporation, given that climate change is pushing evaporation rates upward; and realistic operational demand estimates reflecting local water use. The methodology is also tailored primarily to arid and semiarid settings, where the zero-inflow assumption is most defensible, though the authors emphasize that in humid regions it can be extended with additional water-balance components.

Beyond its technical contributions, the study carries a message about governance. Evidence from Brazil suggests that allocation processes grounded in transparent technical criteria, participatory mechanisms, and explicit strategic objectives achieve greater legitimacy among water users—improving the odds that hard decisions during droughts are actually followed. By replacing opaque rule charts and single-scenario assessments with intuitive graphics that any stakeholder can read, the framework aims to make negotiation itself more evidence-based. The researchers describe the tool as automated, replicable, and user-friendly, with immediate potential for adoption by Paraíba’s Executive Agency for Water Management and by other agencies confronting similar conditions.

Looking ahead, the authors call for curated data repositories and open, adaptable computational tools to broaden the framework’s applicability, and for validation across additional reservoir systems in different climatic, morphological, and governance contexts. As climate change intensifies hydrological uncertainty worldwide, the need for flexible, scenario-based alternatives to fixed operating rules is only expected to grow. In showing that reservoir shape—not just size or inflow—strongly controls both the magnitude and the pace of evaporative losses, this study offers water managers everywhere a deceptively simple but powerful idea: look the reservoir in its geometry, chart its fate graphically, and let the trade-offs speak for themselves.

Karst aquifers form when slightly acidic rainwater dissolves carbonate rock over millions of years, carving conduits, caverns and fissure networks into what was once solid bedrock. The result is an aquifer fundamentally unlike a sandstone or alluvial system. Water moves through discrete channels rather than through the pore space of a homogeneous matrix, so two wells drilled a few dozen meters apart can yield dramatically different results—one flowing freely, the other dry. This extreme heterogeneity has made groundwater exploration in karst regions a notoriously high-risk endeavor, with failure rates for conventionally sited wells that can exceed half of all attempts.

The research team, led by Zhijie Zheng of the Institute of Karst Geology at the Chinese Academy of Geological Sciences, working with colleagues from Hefei University of Technology and the SinoProbe Laboratory, argues that this risk can be dramatically reduced by treating geomorphology as the master variable. In their framework, the distribution of groundwater potential is closely tied to landform type, and each of the major landform units of southwest China’s karst imposes a characteristic structure on the aquifer beneath it.

The first setting is the plateau valley, a landscape of narrow valley floors, steep flanks and deeply incised river channels. Here the water table lies far below the surface, and the subsurface plumbing is governed by relatively impermeable boundaries that confine the aquifers both laterally and vertically. Exploration in this terrain is therefore a search for those confining boundaries rather than for the water itself. The authors note that geophysical methods capable of resolving sharp lateral contrasts—electromagnetic techniques and direct-current resistivity soundings—are the standard instruments of choice. By mapping where resistant limestone gives way to conductive shale or fault gouge, geophysicists can bracket the compartments where water is stored under pressure, and drilling targets are placed against these hydrogeological walls.

The second setting, the plateau slope area, presents an almost inverse problem. Slopes are comparatively gentle, but the unsaturated zone—the dry rock between the surface and the water table—can be extraordinarily thick. Water exists at depth, but only where low-resistivity impermeable boundaries trap it above poorly permeable strata. Because the depth to the target is the dominant uncertainty, exploration here prioritizes depth-determination methods, most often electromagnetic soundings that quantify how conductivity changes with depth. Getting the depth wrong by even tens of meters can mean drilling straight past a water-bearing fracture zone into barren rock, a failure mode the authors identify as common in earlier campaigns across the Yunnan-Guizhou Plateau slopes.

The third and most productive setting is the peak-cluster depression, the iconic terrain of cone-shaped hills surrounding enclosed depressions. These areas combine low elevation, shallow groundwater levels, large catchment areas and comparatively high-yielding aquifers, making them the most attractive targets for water supply development. The exploration problem shifts accordingly: instead of finding boundaries, geophysicists must identify well-connected, fissured karst zones within the soluble rock. In open terrain, electrical resistivity tomography, or ERT, is highly effective, imaging the subsurface by injecting current through electrode arrays and mapping the resulting resistivity contrasts that highlight water-filled fissures. Where terrain or cultural noise constrains ERT, the magnetotelluric method—which measures natural variations in the Earth’s electromagnetic fields to probe to greater depths—can be combined with ERT for joint interpretation, marrying shallow resolution with deep penetration.

Beyond these three landform types, the study delves into the fine structure of hilly plain areas, where karst connectivity is more developed than in peak-cluster depressions and where the carbonate massif is organized into three distinct vertical layers. The shallow layer, extending from the surface to about 30 meters depth, is well connected hydrologically but is frequently choked with mud and sand deposits that clog the fissure network, limiting yields even where cavities are abundant. The middle layer, between roughly 30 and 100 meters, is the prize: here the karst is well connected, fully saturated and high-yielding, making it the primary target for production well drilling. Below 100 meters, karst development becomes weak and isolated, with cavities and conduits largely disconnected from one another, so deep drilling rarely pays off. This three-layer model gives drillers a quantitative depth window—and an equally important warning about where not to spend money on a borehole.

The practical significance of the framework lies in its integration of geomorphological reasoning with geophysical survey design. Rather than applying a generic survey template, practitioners can now select instruments and interpret anomalies according to the terrain unit they are working in. In a plateau valley, a conductive anomaly might represent the confining boundary that defines an aquifer’s edge; on a plateau slope, the same anomaly signals a target depth; in a peak-cluster depression, it may mark the fissure zone that will feed a high-yielding well. The same geophysical signature carries three different meanings depending on landscape context, and the authors show that recognizing this context-dependence is what separates successful exploration programs from expensive drilling failures.

The work draws on a substantial body of case studies across southwest China, including electromagnetic surveys in arid southern Sichuan, resistivity and induced-polarization soundings in mountainous terrain, and integrated campaigns in Yunnan villages such as Shuangnuo and Laohutiangou, where the same research group tested combined geophysical approaches under low-resistivity background conditions. It also builds on long-standing Chinese research into karst water systems dating back to foundational studies of the region’s underground rivers and peak-forest plains.

The broader stakes extend well beyond China’s borders. Karst terrains cover roughly 10 to 15 percent of the planet’s land surface and supply drinking water to perhaps a quarter of the global population. In many karst regions, climate change is intensifying both droughts and floods, and the fast drainage characteristic of conduit-dominated aquifers means that water arrives in pulses rather than as a steady supply. Reliable, science-based well siting is therefore a matter of water security for hundreds of millions of people, from the Mediterranean to Southeast Asia and the Caribbean. The southwest China karst is one of the largest continuous carbonate provinces on Earth, and a validated, terrain-specific exploration methodology developed there offers a transferable template for other karstic regions facing similar challenges.

The study also carries implications for engineering and environmental management. The same three-layer structure that governs water yield also controls contaminant pathways: the well-connected middle layer that makes an excellent aquifer can also transmit pollutants rapidly from the surface, while the mud-filled shallow layer acts as a partial but unreliable buffer. Understanding the vertical zonation of karst connectivity thus informs not only where to drill, but how to protect what is found. The authors’ research program was supported by regional geological survey projects in Guangxi, and their data are available from the corresponding author upon reasonable request.

For a region where rural communities have historically depended on cisterns and seasonal springs, the ability to translate the shape of the land into the geometry of the aquifer beneath it represents a quiet but consequential advance. The message of the new work is deceptively simple: in karst, the landscape is the logbook of the water. Read it correctly, match the geophysical method to the terrain, and the odds of a productive well transform from a gamble into an engineering calculation.

The award targets a major blind spot in American snow science. Western states have long benefited from extensive snow-monitoring programs, but northern New England lacks a standardized, automated system capable of tracking how snow accumulates, changes and disappears across mountains and watersheds. That gap matters because snow in the region is not simply a seasonal feature. It acts as a temporary water reservoir, stores precipitation in frozen form, influences spring runoff, affects flooding and supports a multibillion-dollar winter recreation sector. Yet snow depth and snow-water equivalent—the amount of liquid water contained in a snowpack—can vary dramatically over short distances because of elevation, wind exposure, vegetation, slope orientation and local temperature patterns.

Snow STORE will address that complexity by linking monitoring sites that already operate in different scientific and recreational environments. Researchers plan to enhance observations at the Mount Washington Observatory, the Sleepers River Research Watershed and Hubbard Brook Experimental Forest, while also working with winter recreation areas such as Stowe Mountain Resort, Wildcat Mountain Resort and Sugarloaf Mountain. Additional stations will be supported at the University of Vermont’s Jericho Research Forest and Mount Mansfield. Together, these locations span a wide range of elevations, landscapes and microclimates, allowing scientists to compare snow behavior from managed ski slopes to remote mountain ecosystems and long-term hydrological research sites.

The planned infrastructure is designed to replace isolated, occasional measurements with continuous information. Snow depth has traditionally been measured by hand, a method that can be accurate at a particular location but cannot capture rapid changes across complex terrain. Automated instruments can record snow depth and weather conditions at much shorter intervals, revealing how snowfall, compaction, melting and refreezing alter the snowpack over time. Measurements of temperature, precipitation, wind and humidity can be combined with snow observations to determine whether a storm produces light, dry powder or dense, water-rich snow. Such distinctions are critical for calculating water availability, predicting runoff and understanding why two nearby mountains can experience very different snow conditions during the same storm.

The project’s scientific challenge extends beyond collecting more data. Snow models must represent physical processes that occur at scales ranging from individual snow grains to entire watersheds. As snow settles, its crystals change shape, its density increases and liquid water may move through or refreeze within the pack. Wind can redistribute snow from exposed ridges into sheltered hollows, while forests intercept snowfall and alter the timing of melt. Warmer winter temperatures can produce more rain-on-snow events, in which rainfall and rapid melting combine to generate sudden runoff and flooding. By unifying field observations with computer simulations, the researchers hope to improve predictions of snowpack evolution, mountain weather and water movement across northern New England.

The initiative will also create a shared system for turning scientific observations into practical decisions. Its partners envision a real-time, one-stop data resource that aggregates information from weather stations, ski areas, research sites and forecasting networks. Rather than leaving measurements in separate databases, Snow STORE will synthesize them into dashboards and decision-support tools. Ski-area operators could use the information to refine snowmaking strategies, which depend on temperature, humidity, wind and the amount of natural snow already present. Forecasting agencies could use improved observations to assess winter hazards, while communities could gain better information about changing flood risks and the timing of spring runoff.

For the outdoor recreation industry, the value of more precise snow intelligence may become increasingly important as climate conditions shift. Ski areas and other winter businesses must make operational decisions under highly variable weather, often with only limited information about conditions a few kilometers away or at a different elevation. A regional monitoring network could help operators distinguish between temporary fluctuations and longer-term changes in snow reliability. Better data may support more efficient snowmaking, improve visitor advisories and help businesses plan staffing, transportation and infrastructure use. The project’s economic research will examine how snow conditions influence recreation patterns and regional development, connecting atmospheric measurements to the decisions made by businesses, workers, visitors and public agencies.

Snow STORE is organized around four linked goals: advancing snow and runoff modeling, improving awareness of mountain weather, delivering actionable information to decision-makers and strengthening the outdoor recreation economy. Achieving those goals will require more than instruments and software. The collaboration includes Ski Vermont, Ski NH and the Ski Maine Association, state outdoor recreation offices, the National Weather Service’s Burlington Weather Forecast Office and Northeast River Forecast Center, the Appalachian Mountain Club and the Consortium of Universities for the Advancement of Hydrologic Science. These partners bring operational experience that can help researchers determine which measurements are most useful in real-world settings and how technical information can be communicated clearly during rapidly changing conditions.

The project also has a workforce and education mission. By connecting universities, federal agencies, research forests, observatories and recreation companies, it will give students and early-career scientists experience with field instrumentation, snow physics, hydrological modeling, weather forecasting, cyberinfrastructure, economics and social science. The resulting network could provide a durable scientific foundation for northern New England, where snow influences ecosystems, reservoirs, rivers, forests, transportation and local economies at the same time. Researchers say the ultimate objective is to understand how the region’s snow is changing and to make that knowledge useful before hazards develop or opportunities are lost. With a coordinated observing system in place, a storm on a mountain ridge, a shift in snowpack density or an approaching rain-on-snow event could become part of a shared regional picture—one capable of linking atmospheric processes to water security, public safety and the future of winter recreation.