Coupled Hydrological-Deep Learning Models Reveal Nonlinear ENSO Effects on Run-of-River Hydropower and Downstream Navigation

Abstract This study analyzes the influence of the El Niño-Southern Oscillation (ENSO) on energy generation at the Jirau and Santo Antônio Hydroelectric Plants and on downstream navigation within the 1.4-million km 2 Madeira River basin, a strategic sub-basin of the Amazon River. Using the PHYSITEL/HYDROTEL semi-distributed hydrological modeling platform, the basin’s streamflow dynamics were simulated with KGE values ranging from 0.80 to 0.89 and NSE_Log values ranging from 0.78 to 0.82. Results indicate that the integrated distributed hydrological and LSTM models significantly outperform traditional approaches, achieving test R 2 values > 0.88 in simulating water levels at ungauged stations. Furthermore, Maximal Information Coefficient (MIC) analysis revealed a strong non-linear dependence (MIC = 0.61) between ENSO-induced climatic signals and the operational constraints of run-of-river hydropower plants and downstream navigation depth. The results demonstrate that high-intensity ENSO events have induced pronounced flow reductions, which have directly impacted hydroelectric output, reducing hydropower generation to below 20% of its full production potential and constrained river navigability, especially during severe drought conditions. These findings demonstrate the model’s robustness in capturing complex, non-linear teleconnections that linear models typically fail to resolve. This study contributes to advancing the understanding of ENSO’s effects on the hydrological variability and highlights the need for resilient infrastructure and strategic planning to safeguard sustainable energy production and downstream waterway transport.