Reservoirs change how rivers behave, making them harder to predict. We tested whether machine learning models can learn these effects by training on river sites with and without upstream reservoirs across Great Britain. Models without reservoir training overestimated regulated flows, while reservoir-only models performed poorly on natural rivers. Training on both performed well everywhere. All models failed at heavily managed reservoirs where human decisions dominate river flow.

Many flood-prone regions lack the river observations needed for flood planning. We compared two widely used global datasets by testing how well they reproduced observed river flow at more than 5,000 gauging stations worldwide. One dataset better captured daily river flow changes, while the other provided more reliable estimates of long-term change. The results show that the most suitable dataset depends on whether the goal is flood monitoring or long-term risk assessment.

Global floods pose serious risks, but existing models are too slow for large-scale prediction. We redesigned the CaMa-Flood model for GPUs by reformulating key components such as irregular river networks, flux updates, and floodplain dynamics into highly parallel, GPU-native algorithms. The resulting CaMa-Flood-GPU runs global simulations in hours instead of days with the same accuracy, enabling larger ensembles, improved flood-risk analysis, and better preparedness worldwide.

This study addressed the essential need for detailed spatial economic data to accurately assess global economic losses from natural disasters by developing high-resolution global maps of sector-specific (Service, Industry, and Agriculture) Gross Domestic Product (GDP) for 2010, 2015, and 2020. The new maps showed improved accuracy in reflecting local economies, revealing that sectoral GDP distribution differs by country and affects disaster impact estimates.

The Global Lakes and Wetlands Database (GLWD) version 2 distinguishes a total of 33 non-overlapping wetland classes, providing a static map of the world’s inland surface waters. It contains cell fractions of wetland extents per class at a grid cell resolution of ~500 m. The total combined extent of all classes including all inland and coastal waterbodies and wetlands of all inundation frequencies – that is, the maximum extent – covers 18.2 × 10⁶ km², equivalent to 13.4 % of total global land area.