Detection of Lime Crops Using Sentinel-2 Satellite Images and Machine Learning

Abstract

Lime is a widely consumed agricultural product in Mexico and is predominantly produced in the state of Colima, making it a key component of the region’s economy. Lime trees are commonly affected by Huanglongbing, a disease present in Colima since 2010. Its spread, combined with the lack of control measures, has significantly reduced cultivated areas and production, resulting in substantial economic losses and negatively impacting local growers. To address this issue, mapping cultivated areas is essential for monitoring plant health and conducting accurate, up-to-date, and reliable lime crop censuses. Recent studies have shown that remote sensing and machine learning are effective tools for crop mapping and monitoring in specific regions. Many of these studies leverage optical satellite imagery and phenological analysis to assess plant health. This study proposes a strategy for lime crop detection and mapping using the Google Earth Engine (GEE) code editor, Sentinel-2 imagery, and machine learning algorithms: Random Forest (RF), Support Vector Machine (SVM), Classification and Regression Trees (CART), and Gradient Tree Boosting (GTB). The study area covers the municipalities of Tecoman and Armeria, Colima. The models were trained using 3,932 sampling points. Model performance evaluation indicates that GTB achieved the highest accuracy in lime crop detection, supporting informed decision-making, efficient management of agricultural resources, and the implementation of strategic policies