Sugarcane Mapping and Harvest Monitoring using ML and Google Earth Engine

The increasing global population intensifies the pressure on food production, necessitating optimized agricultural practices. This study explores the potential of remote sensing and machine learning to enhance sugarcane identification, mapping, and harvest monitoring in Dera Ismail Khan, Pakistan. Employing a two-step approach, we leverage Sentinel-2 satellite imagery and NDVI analysis for sample selection followed by a comprehensive field survey using Kobo Toolbox. Of 467 collected samples, 80% served for model training, identifying a total sugarcane area of ~21,875 hectare. We implemented RF, SVM, CART, KNN, and K- Means classification models, with SVM achieving the highest overall accuracy of 90%. Furthermore, we propose a novel sugarcane harvest monitoring approach using an NDVI threshold of 0.35 and analyzing the decrease in NDVI value of sugarcane areas to delineate harvested areas, offering real-time insights into cropping patterns. This research aligns with two key objectives: (1) Enhancing precision in sugarcane identification and mapping, and (2) Automating and improving growth and harvest tracking. The findings demonstrate the potential of this methodology for sustainable crop monitoring, contributing to informed decision-making in agriculture and potentially mitigating the global food demand-production gap.

Перевод пока недоступен