Evolución de los modelos de Deep Learning para el pronóstico de caudales en cuencas, 2026: Evolution of Deep Learning models for forecasting flow rates in basins, 2026

Evolución de los modelos de Deep Learning para el pronóstico de caudales en cuencas, 2026

Abstract

Flow forecasting is a crucial tool for water resource management, flood risk reduction, and hydraulic system planning. In recent decades, forecasting techniques have advanced from physical and conceptual hydrological models to data-driven methods, with a particularly strong focus on Machine Learning (ML) and Deep Learning (DL) techniques. This review explores the evolution of deep learning models applied to the estimation, calculation, and prediction of river flows in watersheds to date. Numerous methodological advances, model structures, data requirements, and evaluation criteria used in recent studies have been presented.

Special emphasis is placed on hybrid and deep architectures, such as Deep Belief Networks (DBN), Artificial Neural Networks (ANN), Support Vector Regression (SVR), and Wavelet Artificial Neural Networks (WANN). The results of this research indicate that traditional machine learning methods are very effective for short-term forecasts, while hybrid models are more efficient at capturing non-linearity over longer time horizons. Finally, current challenges are addressed, such as data scarcity, model interpretability, and uncertainty assessment, as well as new trends including physics-guided neural networks, graph neural networks, and mutable architectures.