Neural Network Approach for Modeling Future Natural River Flows: Assessing Climate Change Impacts on the Tagus River

Abstract

[Abstract]: Study region: Tagus River basin (Iberian Peninsula). Study focus: An innovative methodology is developed to analyze the impact of climate change on the hydrological cycle. Initially, natural river flow is reconstructed to address the challenge posed by river regulation, which complicates accurate hydrological modeling and can obscure the true impact of climate change. The Iber+ hydrodynamic model is applied to account for downstream reservoir contributions, which allows reversing their influence. Then, neural networks of varying configurations, with specific requirements such as data bucketing, are trained to replicate river flow utilizing recorded precipitation and temperature datasets, subjected to validation procedures. A multi-model ensemble is constructed to address uncertainties inherent in modeling future hydrological climate scenarios. This ensemble, supplied with climate model data, derives historical and projected river flows, allowing analysis of their temporal evolution. New hydrological insights for the region: The findings affirm the efficacy of the proposed methodology and reveal, for the considered high-risk SSP5–8.5 scenario, the intensification of the Tagus hydrological cycle. Within the inherent uncertainty of climate models, average ensemble outputs indicate a reduction of about −20 % in available water at the end of the century, especially critical during summer, with an almost 600 % rise in dry months. Average ensemble results also indicate an increase in flooding events, with extreme floods that currently have five-year frequency, projected to double by the century’s end.