In 2022, southwestern France saw exceptional wildfires, burning an area about 14 times the regional average. Using fire records, weather data, and climate simulations with and without human influence, we show that human-caused climate change made the weather conditions linked to the 3 largest wildfires about 2 to 10 times more likely; such conditions could become roughly 10 to 100 times more probable by 2100 under moderate emissions, highlighting a growing need for prevention.

The Ucayali River is the major Andean conveyor of sediment to the Amazon. By combining field measurements, satellite data and modelling over decades, we show that floodplains trap 36% of this sediment while recycling 22% back into the river during flood recession. During high waters, flooding reduces the river's capacity to carry sand, capturing 14% at peak discharge. These findings show that floodplains act as dynamic regulators of sediment transport in large tropical rivers.

In France, the 2022 summer witnessed severe drought conditions, leading to very low flows in rivers and widespread wildfire occurrences. This article proposes a model to estimate the probability of occurrence of the 2022 event, viewed from various angles: intensity of the drought and fire weather conditions, duration and spatial extent of the event. The model can also be used to estimate how this probability will evolve in the future under global warming.

We present smash, an open-source framework for high-resolution hydrological modeling and data assimilation. It combines process-based models with neural networks for regionalization, enabling accurate simulations from the catchment scale to the country scale. With an efficient, differentiable solver, smash supports large-scale calibration and parallel computing. Tested on open datasets, it shows strong performance in river flow prediction, making it a valuable tool for research and operational use.

Understanding and modeling flash-flood-prone areas remains challenging due to limited data and scale-relevant hydrological theory. While machine learning shows promise, its integration with process-based models is difficult. We present an approach incorporating machine learning into a high-resolution hydrological model to correct internal fluxes and transfer parameters between watersheds. Results show improved accuracy, advancing the development of learnable and interpretable process-based models.