Here, we represented catchments as networks of variables connected by cause-and-effect relationships. By comparing the performance of statistical and machine learning methods with and without incorporating causal information to predict runoff properties, we showed that causal information can enhance models' robustness by reducing the accuracy drop between the training and testing phases, improving the model's interpretability, and mitigating overfitting issues, especially with small training samples.

Precipitation (P) and soil moisture (SM) are critical components of the climate system but poorly understood in the Arctic. Using NASA's SMAP satellite, we analyzed SM and P patterns in Finland. SM and P correlate strongly in summer and fall but weakly in spring due to snowmelt. While the area of P can be estimated from SM, estimating its intensity is more challenging. Water bodies complicate SM retrieval. The promising results suggest this method could be applied across the Arctic.

This paper presents a new statistical method to estimate probable maximum precipitation (PMP), addressing flaws in traditional approaches. It accounts for uncertainty using the Pearson Type-I distribution. Applied to two Québec stations, it provides more objective PMP estimates. Challenges remain due to short-tailed models applied to heavy-tailed precipitation. Best practice recommendations are also provided for estimating PMP.

Satellite data reveal that woody plants in Texas’s Post Oak Savannah now return almost all rainfall to the atmosphere. In drier regions, once trees and shrubs blanket more than 80 % of the land, yearly water loss to the atmosphere even surpasses rainfall, shifting the region from a water surplus to a deficit and shrinking groundwater recharge. Without brush control, warming and further canopy growth could leave soils drier, streams weaker, and local water supplies increasingly strained.

In water-scarce regions, precipitation is a highly variable and essential resource for crop production. Developing countries like India have an uneven distribution of rain gauges, so reliance on satellite- and reanalysis-based precipitation products is critical for their prudent management. Hence, this study statistically evaluated different precipitation products against station data for water-scarce regions in Tamil Nadu and found that the European Centre for Medium-Range Weather Forecasts Reanalysis version 5 Land (ERA5-Land) performed the best, followed by the Multi-Source Weighted-Ensemble Precipitation (MSWEP).

Thawing permafrost changes how water is stored and moves across landscapes. We measured water inputs and outputs in a basin with thawing peatland complexes and three sub-basins. In addition to yearly changes in precipitation and evapotranspiration, we found that hydrological responses are shaped by thaw-driven landscape connectivity. These findings highlight the need for long-term monitoring of ecosystem service shifts.

The 2021 drought in Quebec showcased the province’s potential vulnerability. This study uses a storyline approach to explore the impacts of future extreme droughts under +2 °C and +3 °C global warming scenarios. Results show that future droughts similar to that of 2021 could worsen significantly, with water levels dropping for longer periods and impacting ecosystems and human activities.

We introduce the ratio of uncertainty to mutual information (RUMI), a new metric to improve rainfall-runoff simulations. RUMI better captures the link between observed and simulated stream flows by considering uncertainty at a core computation step. Tested on 99 catchments and with the GR4J model, it outperforms traditional metrics by providing more reliable and consistent results. RUMI paves the way for more accurate hydrological predictions.

This study compared global satellite and reanalysis precipitation datasets to assess diurnal variability. We found that all datasets capture key diurnal precipitation patterns, with maximum precipitation in the afternoon over land and early morning over the ocean. However, there are differences in the exact timing and amount of precipitation. This suggests that it is better to use a combination of datasets for potential applications rather than relying on a single dataset.

This study explores six methods to improve the ability of long short-term memory (LSTM) neural networks to predict peak streamflows, crucial for flood analysis. By enhancing data inputs and model techniques, the research shows that LSTM models can match or surpass traditional hydrological models in simulating peak flows. Tested on 88 catchments in Quebec, Canada, these methods offer promising strategies for better flood prediction.

This study presents a methodology for understanding heterogeneity in alluvial aquifers by combining borehole data, electrical resistivity tomography, and stochastic modeling. The approach uses hydrofacies to incorporate spatial variability into groundwater models. The hydrofacies distribution helps define hydraulic conductivity fields, essential for groundwater flow and contaminant transport simulations, providing a valuable tool for sustainable aquifer management in sedimentary environments.

Our paper introduces Saudi Rainfall (SaRa), a high-resolution, near-real-time rainfall product for the Arabian Peninsula. Using machine learning, SaRa combines multiple satellite and (re)analysis datasets with static predictors, outperforming existing products in the region. With the fast development and continuing growth in water demand over this region, SaRa could help to address water challenges and support resource management.

Three process-based and four data-driven hydrological models are compared using different training data. We found that process-based models perform better with small datasets but stop learning soon, while data-driven models learn longer. The study highlights the importance of memory in data and the impact of different data sampling methods on model performance. The direct comparison of these models is novel and provides a clear understanding of their performance under various data conditions.

This study describes a new methodology to identify regional structures in the dependence of extreme rainfall on global climate indices. The study area is North-Central Italy, but the methods are highly adaptable to other regions. We observe that while multiple climate indices show influence, the Western Mediterranean Oscillation Index and East Atlantic West Russia pattern appear to have the strongest effects, with regional structures that align with other studies. We also show that using these indices may improve estimation of extreme rainfall depth with a given probability.

The Shiyang River basin is a typical inland river basin in the arid zone, and, through this study, it was found that its soil salinization area has a general trend of increasing, and the degree of salinization gradually increases. External water transfers have alleviated water shortage and salinization to a certain extent, but soil salinization has already occurred in its periphery.

The objective of this study is to enhance the performance of 1D–2D flood models using satellite Earth observation data. The main factor influencing a 1D–2D flood model is the accuracy of the digital elevation model (DEM). Two workflows are introduced to improve the 1D–2D flood model: (1) in the DEM analysis workflow, 10 DEM products are evaluated using the ICESat-2 ATL08 benchmark, while (2) in the flood map analysis workflow, flood extent maps derived from multi-mission satellite datasets are compared with simulated flood maps.

Statkraft requires projections of future streamflow to plan hydropower investments. Setting up a hydrological model for new regions can be too time-consuming to meet the often short delivery deadlines. We have developed an interpretable machine learning method that links streamflow to precipitation and temperature, and can serve as a first screening approach. This method is then applied to climate model simulations of precipitation and temperature to obtain streamflow projections for Brazil.

We used a new method to measure how streamflow responds to precipitation across a network of watersheds in Iran. Our analysis shows that streamflow is more sensitive to precipitation when groundwater levels are shallower, climates are more humid, topography is steeper, and drainage basins are smaller. These results are a step toward more sustainable water resource management and more effective flood risk mitigation.

Our study compares neural network models for predicting discharge in ungauged basins. We evaluated convolutional neural networks (CNNs), long short-term memory (LSTM) and gated recurrent units (GRUs) using 28 years of weather data. CNNs showed the best accuracy, while GRUs were faster and nearly as accurate. Adding static features improved all models. The research enhances flood forecasting and water management in regions lacking direct measurements, offering efficient and accurate predictive tools.