Mapping heterogeneous land use/land cover and crop types in Senegal using sentinel-2 data and machine learning algorithms

Abstract

ABSTRACT In rainfed and dryland agricultural areas with smallholder farms (less than 2 ha), crop diversity is high due to farmers' decisions and local climatic conditions, leading to a complex spatial–temporal distribution of crops. Monitoring and mapping crops is crucial for food security and implementing agricultural support programs. This study aims to map crop types across Senegal using Sentinel-2 satellite imagery and the limited ground reference data available, which has been increasing recently. The study compares conventional supervised classification algorithms to unsupervised classification algorithms using high-resolution satellite imagery. Crop type classification for 2020 in Senegal employed supervised machine learning algorithms, including Classification and Regression Trees (CART), Random Forest (RF), and Support Vector Machine (SVM) on the Google Earth Engine (GEE) cloud platform, and the unsupervised Iso-clustering classification algorithm with Spectral Matching Techniques (SMTs). Due to limited ground data, supervised classifiers achieved 45-55% accuracy, whereas the unsupervised semi-automatic approach achieved over 75% accuracy. The study indicates that supervised classifiers' performance depends on ground data quantity, while SMT shows good performance even with limited ground data. This SMT approach is valuable for classifying crop types in dryland areas with smallholder farms and diverse cropping patterns.