Hybrid Deep Learning Models for Predicting Meteorological Variables Associated with Santa Ana Wind Conditions in the Guadalupe Basin

Santa Ana winds are extreme meteorological events that strongly affect the U.S.–Mexico border region, often associated with droughts, high fire risk, and hydrological imbalance. Understanding the temporal behavior of key atmospheric variables during these events is crucial for integrated water resource management in semi-arid regions such as the Guadalupe Basin in northern Baja California. In this study, we explored the predictive capability of several hybrid deep learning architectures—Long Short-Term Memory (LSTM), Convolutional Neural Network combined with LSTM (CNN–LSTM), and Bidirectional LSTM with Attention (BiLSTM–Attention)—to model the temporal evolution of wind speed, wind direction, temperature, relative humidity, and atmospheric pressure using Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) reanalysis data from 1980 to 2020. Model performance was evaluated using RMSE, MAE, and R2 metrics and compared against persistence and climatology baselines. The BiLSTM–Attention model achieved the best overall performance, showing particularly high accuracy for temperature (R2 = 0.95) and relative humidity (R2 = 0.76), while maintaining angular errors below 35° for wind direction. The results demonstrate the potential of hybrid deep learning models to capture nonlinear temporal dependencies in meteorological time series and provide a methodological framework to enhance hydrometeorological understanding and water resource management in the Guadalupe Basin under Santa Ana wind conditions.