Predicting Sandstorms and Drought Using Artificial Intelligence Techniques Based on Climate Data: A Case Study of Jifarah Plain, Libya

Abstract

Sandstorms and drought are major climate hazards in North Africa, particularly across Libya's Jifarah Plain, where population, agriculture, and critical infrastructure are concentrated. This study develops an artificial-intelligence framework for predicting drought severity and sandstorm occurrence using monthly climate data for 1985–2022. The dataset combines observations from the Libyan National Meteorological Center (LNMC) at Tripoli International Airport with NASA POWER reanalysis data for Al-Aziziyah and Al-Asa. Three models were evaluated: Random Forest (RF) for drought-severity prediction, Long Short-Term Memory (LSTM) for temporal sandstorm prediction, and a weighted Ensemble model for joint spatio-temporal prediction. After correcting the preprocessing workflow to eliminate data leakage, all scaling parameters are fitted using the training subset only and then applied to the validation and test subsets. The embedded performance results show that the Ensemble model provided the strongest continuous-prediction performance, with R² = 0.955 and RMSE = 0.057, compared with R² = 0.888 and RMSE = 0.089 for RF and R² = 0.838 and RMSE = 0.115 for LSTM. A marginal warming trend of +0.014 °C/year (p = 0.054, R² = 0.10) was identified, while the hazard analysis indicates that the principal period of combined sandstorm and drought risk occurs during March–June. The spatial risk surfaces derived from only three stations are interpreted as illustrative rather than as operationally validated downscaling products. The analysis covered three meteorological stations: Tripoli Airport, Al-Aziziyah, and Al-Asa—over the 1985–2022 period. The proposed framework provides a reproducible approach that can be adapted to other arid regions of Libya and North Africa characterized by sparse ground-based monitoring networks. The Sandstorm Occurrence Probability Index (SOPI) and drought-risk layers provide a preliminary basis for supporting early-warning systems and climate-adaptation planning.