CLIMATE AND CATCHMENT SIGNATURES OF THE BRAZILIAN STREAMFLOW REGIME

Código

XXVI-SBRH1114

Autores

Abderraman Róger Amorim Brandão, Admin Husic, Andre Almagro, Dimaghi Schwamback, Paulo Tarso Sanches de Oliveira

Tema

F - Processos Hidrológicos e Monitoramento Integrado de Recursos Hídricos

Resumo

South America holds immense freshwater reserves, yet the controls on streamflow in its tropical basins remain poorly resolved. We applied explainable artificial intelligence to disentangle how climate and catchment attributes govern three discharge (Q5, Qmean and Q95) across 735 Brazilian watersheds. Random-Forest ensembles trained with sixteen key predictors explained over seventy percent of variance for most metrics. SHapley Additive exPlanations highlighted the aridity index (PET/P) as the dominant national driver, with streamflows declining sharply once values exceed 1.30. Regional patterns underscored heterogeneity: in central Brazil, land-cover change and topography most strongly reduced low flows; in the Caatinga and Pantanal, wetter seasons and low aridity magnified floods; while in the Pampa, high silt combined with increasing aridity produced the largest extremes. Amazon and Cerrado basins showed that intact soils and geology temper extreme discharges, whereas aridification offsets this buffering. Nationwide, climate drives average conditions, but catchment properties modulate extremes, demonstrating the necessity of considering both when forecasting hydrological responses to change. The study illustrates how data-driven models coupled with XAI can reveal causal mechanisms, critical breakpoints and spatial variability that traditional correlations overlook, offering a transferable framework for tropical regions.