Etudiant : JABBARI MOHAMMED

Filière : Master Big Data Analytics & Smart Systems (BDSaS)

Encadrant : Pr. EL BEQQALI OMAR

Annèe : 2025

Résumé : This study presents a comprehensive approach to estimating Surface Soil Moisture (SSM) in semi-arid Moroccan agricultural regions using machine learning (ML) techniques integrated with multi-source remote sensing data. Conducted as part of an internship at the Center for Remote Sensing Applications (CRSA), Mohammed VI Polytechnic University (UM6P), the research leverages Landsat 8/9, MODIS, ERA5/ERA5-Land datasets, and in-situ measurements from four study sites to develop spatially and temporally continuous SSM estimates. The methodology employs the ROBOT spatiotemporal fusion algorithm to combine Landsat's high spatial resolution (30 meters) with MODIS's high temporal frequency (near-daily), enhancing the dataset's ability to capture rapid moisture dynamics driven by precipitation and evapotranspiration. Three ML models Random Forest (RF), XGBoost, and LightGBM were trained and evaluated using Python, with preprocessing conducted on Google Earth Engine and validation using in-situ measurements from the Moroccan study sites. Without data fusion, RF achieved a validation Root Mean Square Error (RMSE) of 0.051012 m³/m³, R² of 0.599934, and bias of -0.002 m³/m³ under random split conditions. The advanced spatio-temporal split yielded reduced performance (RMSE = 0.051970 m³/m³, R² = 0.541654, bias = -0.003 m³/m³), highlighting generalization challenges across different temporal and spatial contexts. With fusion implementation, RF significantly improved performance (RMSE = 0.033767 m³/m³, R² of 0.829991, Pearson correlation of 0.914682, bias = -0.001 m³/m³), effectively capturing SSM changes and outperforming both random and advanced spatio-temporal split approaches. Challenges include cloud cover gaps in optical data, spatial resolution mismatches in ERA5 interpolation, and potential synthetic artifacts from data fusion processes. The study demonstrates the substantial potential of ML-based SSM estimation for precision agriculture and water resource management applications, suggesting future enhancements through refined fusion algorithms, advanced feature engineering, and broader geographic validation.