We used machine learning to explore what causes streamflow droughts across the U.S. We found that different regions are influenced by different factors like temperature, snow, and rainfall. Our new method can also predict droughts in areas without streamflow data, helping improve water resource planning.

An improved method is used to integrate satellite-derived terrestrial water storage and surface soil moisture observations into a hydrological model within the Brahmaputra River basin (South Asia). This integration leads to a more realistic representation of the water stored in the land, allowing us to better understand its changes in space and time, which is crucial in this basin due to its increasing water demand and vulnerability to extreme events due to climate change.

Towards the central question of "are more complex models better" this study proposes a novel framework to score the process complexity of large-scale hydrological models. By systematically assessing their linkage with snowmelt runoff performance, we find while more model complexity does not lead to better runoff magnitude, centroid timing performance is positively correlated with model complexity. The study highlights the unique gains from process improvements to guide model development.

Snowmelt runoff (SMR) is a critical freshwater resource. This study conducts the first comprehensive assessment of SMR, specifically its volume, peak, and timing, across tens of large-scale models and thousand of basins. We also test the models by assessing their performance drop as basins become more and more complex. Our results highlight the systematic biases and certain model categories with stronger robustness in stern conditions, paving the way for process diagnosis for SMR in models.

Traditional analysis of extremes often relies on complex tools when the statistics change over time. We developed a novel, efficient framework based on a maximum return period change over a specific timeframe. Simple formulas are derived to determine the variation in average frequency and magnitude of events. The approach has minor approximations compared to more complex methods, thus providing a reliable tool for practitioners to forecast risk assessment under changing environmental conditions.

We estimate the contribution of rainfall peaks and antecedent wetness to changes in small and large floods in Brazil. While changes in large events are usually rainfall-driven, changes in small floods mostly follow antecedent wetness. In regions with high storage capacity, antecedent wetness drives both small and large floods; in low-storage capacity regions, small floods follow antecedent wetness, while large floods are mainly rainfall-driven.

Our study in the Tarim River Basin tackles the tough balance between farming and nature in dry areas. Using a model, we found that boosting farm income often means less water for ecosystems and more pollution. The solution requires different water use strategies across the region. Importantly, farmland area and crop choices must change with the weather, shrinking in dry years. This approach provides a practical plan to manage water, ecology, and agriculture together.

Ecosystem water use efficiency exhibited a reduced correlation with drought and a delayed response timing from 1982 to 2018, a pattern primarily driven by CO₂ fertilization. Surface soil moisture emerged as the dominant hydroclimatic driver, exerting contrasting influences and exhibiting pronounced threshold effects. This study highlights the divergent vegetation adaptation strategies across arid and humid regions, underscoring the need to reassess the water use efficiency–drought relationship.

Large wildfires are reshaping California’s water-supply forests. Using 40 years of satellite and river data, we found that severe fires greatly reduce vegetation water use, sometimes increasing runoff during dry years. Recovery varies widely across basins and can take many years. The work shows that wildfire can disturb the entire water budget and highlights the need to include fire effects when planning for future water resources.

Distributed simulations and geochemical data suggest that groundwater hysteresis dampens the elasticity of flowing stream networks. This effect is expected to become more important as intense rainfall events replace gradual snowmelt in a warmer climate.

In mountain karst regions, seasonal snow is a key component of groundwater recharge, making them highly vulnerable to climate change. Using a rainfall-snow-discharge model in the southern French Alps, we show that warming shifts high-flow periods and causes a strong summer flow decline. Rapid karst flow paths lead to quick spring responses, limiting storage above the base level and increasing sensitivity to future hydrological change.

Vegetation affects surface runoff in several ways. Surface roughness is increased by stems and leaves, roots and their remnants enhance infiltration into the soil, and through evaporation, the soil water content changes at the beginning of a rainfall event. Our study investigates, how a simulation model is able to react to these effects. While roughness and infiltration are well covered, initial soil moisture is still an unsolved problem.

Several US federal agencies (e.g., USGS, NOAA, USDA, and NSF) collect information for river basins to support their water-related missions. Data is published online in named collection  that each have their own attributes and objectives.  HARBOR harmonizes and brings together all these datasets, just as many large cargo ships can be moored in one harbor. It also classifies basins based on hydrologic similarity, helping researchers find the best model for predicting their hydrologic response.

This article examines hydrological spells, ranging from successive floods to sub-seasonal alternations between high and low flows. To study these rare successions, we developed an original detection method using baseflow as a catchment recovery indicator. Applying this to 643 catchments in France, we characterize these spells, identifying distinct spatial variability. We found that rapid transitions from low to high flows are concentrated in the Rhone-Mediterranean and Rhine-Meuse basins.

To better predict mountain hazards like landslides, we studied how long soil retains rain moisture. Using 20 years of satellite data from China, we found a control shift at about five years. Short-term memory is governed by weather and plants, while long-term persistence is locked in by soil and terrain. This creates a lasting "background" wetness, especially in humid forests, pre-conditioning slopes for years.

Jinci Springs exemplify balancing water use and protection. The springs’ journey—harmony, dry-up, recovery—shows people’s awakening to water protection in crises, offering valuable lessons for today’s sustainable water use.

We assessed the U.S. National Hydrologic Model's ability to simulate several components of the water cycle using multiple datasets of environmental variables. We find that the model's accuracy in streamflow simulation is positively and negatively impacted by the additional constraints, and more model parameters are identified as important. Our results inform operational hydrologic modeling by illuminating the complexities of using the continually expanding suite of data products.

We developed a cost-effective way to monitor toxic petroleum contaminants in groundwater using machine learning and water quality measurements such as pH, dissolved oxygen, redox potential, and electrical conductivity. By simulating real-world conditions, we showed that machine learning models can predict contamination concentration due to reactions being triggered during contamination events. The presented framework complements existing monitoring strategies for better groundwater management.

River flow behaviour varies across Europe and is shaped by both climate and landscape. Using streamflow data from more than 7000 European catchments, we identified 10 distinct hydrological response types based on flow magnitude, timing, and seasonality. Climate is the main control at the continental scale, but landscape features are equally important or more influential in several regions. These results show that river behaviour emerges from the combined effects of climate and landscape.

We developed an open-source tool to combine weather radar and commercial microwave link data, tested on large, public datasets. Merging these measurements produced more accurate rainfall estimates than radar alone, and we show which methods work best in different situations to support improved rainfall monitoring and applications.

Understanding how carbon moves through rivers is vital for managing water quality and climate change. This study developed a new diagnostic framework that combines computer modeling and data analysis to explore why current river carbon simulations are often inaccurate. The results show that parts of the model describing plant growth and carbon cycling cause most errors, and improving these areas can make future environmental predictions more reliable.

Accurately evaluating land surface models is crucial for reliable climate forecast and water resource management. We proposed a new evaluation metric that avoids some traditional metrics' flaws by focusing on accuracy, variability, and pattern similarity. This work offers a more reliable alternative to evaluate land surface models, supporting better decisions in land surface model development.

We compare shallow landslides in shrubland and woodland on the Chinese Loess Plateau. Field monitoring, dye tracing and root tests show that woodland has deeper roots, stable preferential flow and stronger soil reinforcement, which enhance infiltration and delay shallow saturation. Shrubland hillslopes fail more frequently and at shallower depths. These contrasts imply lower landslide susceptibility and sediment yield in woodland.

This study examines the risk of possible water shortages in the Huaihe River Basin and explores whether dry periods can be predicted more reliably. Using decades of daily environmental records, we built a model that separates complex signals into simpler parts and learns their patterns. It forecasts dry conditions more accurately than earlier methods and performs well across regions and time spans. The findings can help the basin prepare for potential water shortages and support better planning.