Exploiting Machine Learning and Remote Sensing for Precision Crop Mapping in Africa

Abstract

As the global population continues to grow, the demand for food intensifies, making precision agriculture essential for enhancing productivity and resource efficiency. A critical component of precision agriculture is accurate crop mapping, which enables optimized resource allocation and better yield prediction. This study focuses on developing a machine learning-based predictive model for crop mapping using multispectral remote sensing data from the Senegalese groundnut basin. The dataset, collected via Sentinel-2 satellites, includes optical bands and vegetation indices over four time periods in 2023. After preprocessing and addressing class imbalance through data augmentation and SMOTE, several machine learning models were evaluated, with Random Forest emerging as the most effective. The results show a significant improvement in classification accuracy, particularly for underrepresented crop types, with the Random Forest model achieving 93.57% accuracy after SMOTE, compared to 57.43% before SMOTE. The outcomes of this research extend beyond technological advancements, contributing to environmental sustainability by enabling more precise resource use, reducing waste, and minimizing the environmental impact of agricultural practices. Furthermore, this type of crop mapping supports food security by helping farmers optimize yields and manage crops more effectively, directly benefiting communities reliant on agriculture. The integration of machine learning and remote sensing in precision agriculture has the potential to foster a more resilient and sustainable agricultural system globally.