Divergent spatial runoff trends in Western Patagonia projected by hybrid glacio-hydrological modelling

Western Patagonia's vast freshwater ecosystem, where glacial and non-glacial waters converge, is increasingly threatened by climate change, disrupting runoff patterns and endangering water resources. Here, we present estimates of past and future glacio-hydrological changes for 2,236 catchments in Western Patagonia, projecting impacts through the 21st century under contrasting climate change scenarios. Leveraging recent advances in the development of regional and global datasets, we applied a hybrid approach combining Long Short-Term Memory (LSTM) neural networks with process-based glacier modelling (Open Global Glacier Model). We evaluated the approach’s ability to predict streamflow in ungauged basins (PUB) and regions (PUR) through 10-fold cross-validation, and compared the results with those from a pure LSTM model and two process-based hydrological models. Additionally, we assessed how the different model predictions extrapolate spatially and project over time. The results show that the hybrid modelling approach outperformed all conventional approaches in more than 70% of the catchments considering PUB and PUR evaluations. Using this approach, we estimated a regional discharge of nearly 20,000 m³ s⁻¹ (2000-2019), with an average glacial contribution of 20%. By the end of the century, we project marked shifts in river seasonality under climate change scenarios. Under a high emissions scenario, the northern region (>46°S) is projected to experience the largest reductions in runoff, with dry season runoff decreasing by almost 50%. In contrast, glacierised basins in the southern regions are projected to show slight increases, with average relative changes of 20%. The results highlight the potential of hybrid modelling to provide new information for climate change adaptation in Western Patagonia.