Forecasting meteorological droughts in a hydrological basin using artificial neural networks

Artificial intelligence and machine learning methods have become valuable and effective tools across various research fields. However, despite their widespread adoption worldwide, a significant gap remains in Mexico regarding the application of these techniques for drought forecasting. This study presents a methodology based on artificial neural networks for predicting meteorological droughts, using the Conchos River Basin in Mexico as a case study. The approach involves combining climatic variables (precipitation and temperature) and meteorological drought indices – the Standardized Precipitation Index (SPI) and the Standardized Precipitation-Evapotranspiration Index (SPEI) – as input variables for the models. A multilayer perceptron with backpropagation was implemented to evaluate different architectures and hyperparameters, identifying the best configuration for predicting the three-month SPEI (SPEI-3). The resulting models achieved good performance (MSE < 0.1 and R2 > 0.90), particularly those using more complex architectures with two hidden layers (256–128) and optimizers such as Adam and Adamax, outperforming SGD and RMSprop. The results confirm the potential of artificial neural network (ANN) models to capture relevant drought patterns in hydrological sub-basins, thereby strengthening early warning systems and supporting decision-making in water resource management. This work highlights the use of machine learning in drought research and the need to expand its implementation in the Mexican context.