University of Tehran Press Desert 2008-0875 30 2 2025 12 01 A Comparative Analysis of Feed-Forward and Long Short-Term Memory Networks for Solar Radiation Estimation. 388 416 106182 10.22059/jdesert.2025.106182 EN Soheila Mohtashami Department of Irrigation & Reclamation Engineering, University of Tehran, Karaj, Alborz Iran. Zahra Aghashariatmadari Irrigation & Reclamation Engrg. Dept. University of Tehran Karaj, Iran. Journal Article 2026 02 25 Accurate estimation of solar radiation is essential for numerous industrial applications, energy management, and agricultural planning. This study investigates the effectiveness of advanced machine learning models for solar radiation prediction in Kerman Province, Iran, utilizing a comprehensive set of meteorological variables. Following rigorous quality control procedures and correlation-based feature selection, the dataset was divided into training (80%) and testing (20%) subsets. Two Neural Networks, namely Long Short-Term Memory (LSTM) with the Adam optimizer and Feed-Forward Neural Network (FFNN), were developed and trained under six input scenarios, employing various learning algorithms including Levenberg–Marquardt (LM), Bayesian Regularization (BR), Gradient Descent (GD), and Resilient Propagation (RP) at both daily and monthly timescales. The results indicate that the FFNN-BR model under scenario 6, incorporating a wide range of meteorological inputs, yielded the highest accuracy for monthly radiation estimation (R² = 0.92, ARE = 4.5%). For daily radiation prediction, the LSTM model under scenario 4 provided superior performance (R² = 0.91, ARE = 1.35%). These findings underscore the importance of model selection and input configuration in enhancing solar radiation estimation accuracy, offering valuable insights for renewable energy resource assessment in arid regions.