Crop type mapping using Sentinel-1 and Sentinel-2 Data in the smallholder cropping system of cotton belt in Mali

Details

Supervisors

Defourny, Pierre ; Guindo, Abdoulaye

Faculty

Faculté des bioingénieurs

Degree label

Master [120] en sciences agronomiques et industries du vivant, à finalité spécialisée: Geo-Information Science and Earth Observation for Environmental Modelling and Management

Abstract

Effective crop type mapping is crucial for farmland monitoring, assessment of food security, and sustainable land management, particularly in heterogeneous smallholder farming systems. This study investigated the prospect of Sentinel-1 SAR data, Sentinel-2 optical imagery, and fused Sentinel-1 and Sentinel-2 datasets for crop discrimination applying tile-based and temporal classification methods. Preprocessing of Sentinel datasets and Random Forest classification were employed using field survey data for training and validation on Google Earth Engine platform. Independent tile-based classification, combined tile-based classification, fused sensor classification, and temporal growth-stage analysis were compared in this study. Findings showed that Sentinel-2 optical data generally provided better performance than Sentinel-1 SAR data for the stronger spectral and vegetation-related information. However, the fusion of Sentinel-1 and Sentinel-2 increased classification steadily and crop separability by integrating spectral, structural, and moisture-related characteristics. The fused combined tile-based classification produced the most balanced overall performance (0.57) among all approaches. Temporal analysis exhibited that classification performance differed across crop growth stages. Lower crop separability showed in early growth and vegetative stages, while harvesting-stage classification derived the best performance (0.53) due to stronger phenological differences among crops. Furthermore, feature importance analysis stated that textural variables (37.3% - 40.3%) and spectral (26.3% -27.4%) contributed most significantly to classification accuracy, whereas SAR variables supported by providing complementary information. Overall, the study reveals that multi-source as well as multi-temporal data integration can remarkably improve crop type mapping in complex agricultural landscapes.