A Hybrid Machine Learning–Deep Learning Framework for Early Wheat Crop Stress Detection in Northwest Tunisia Under Climate Change Using Multi-Temporal Satellite Data

Artificial Intelligence (AI) is increasingly transforming the agricultural sector by improving crop monitoring, reducing production costs, and enhancing yield. However, in many developing regions, agriculture still faces major challenges related to climate variability, crop diseases, and the limited adoption of advanced technologies. To address these challenges, this paper proposes a hybrid machine learning and deep learning framework for agricultural parcel condition monitoring based on multi-temporal Sentinel-2 and Landsat-8 satellite imagery combined with meteorological data. Vegetation indices and biophysical parameters are used to characterize crop dynamics. Random Forest, Support Vector Machine, and ConvLSTM models are integrated within a two-stage strategy for parcel classification and short-term condition prediction. Preliminary results highlight the consistency of the multi-source dataset and demonstrate the potential of the proposed approach for early crop stress detection and precision agriculture applications under Mediterranean climate conditions.