Integrating Remote Sensing and Machine Learning for Accurate Detection of Agricultural Zones in El Jadida, Morocco

This article presents an in-depth study on the detection and location of agricultural zones in El Jadida, Morocco. Accurate identification of these zones is crucial for sustainable land management and effective regional planning. To achieve this objective, a methodology integrating remote sensing techniques, geographic information systems (GIS), and machine learning algorithms was implemented. The results demonstrate the accuracy and reliability of this approach, providing essential information to decision-makers and agricultural stakeholders in the region. This study highlights the potential of remote sensing technologies to improve the mapping of agricultural areas, thus contributing to sustainable agricultural practices and optimal territorial planning. The methodological approach adopted involves the pre-processing of imagery data from Landsat 7 and 8 satellites, covering the city of El Jadida. We extracted the relevant features, divided them into training and test sets, and then applied three supervised learning algorithms: random forest (RF), support vector machine (SVM), and gradient boost tree (GTB). Through several experiments, we evaluated the performance of each machine learning method in terms of accuracy and Kappa coefficient for the years 2000 and 2020. We also analyzed changes in agricultural areas between these two periods. The results show that random forest is the best performing algorithm, with an accuracy of 98.14% in 2000 and 98% in 2020, and Kappa coefficients of 0.96 in 2000 and 0.95 in 2020.