Integrating NDVI, Zonal Statistics, and Random Forest for Precision Agricultural Monitoring in the Mahalaxmi Kheda Region

Abstract

The aim of this study was to monitor and map various crop types and their locations in the Mahalaxmi Kheda region using geospatial tools. Harmonized Landsat and Sentinel (HLS) datasets provided multitemporal imagery, which underwent preprocessing steps like atmospheric correction and layer stacking. The Normalized Difference Vegetation Index (NDVI) was calculated to evaluate plant health, growth patterns, and stress conditions. The Random Forest algorithm was employed to classify primary land cover and crop types in the study area. An accuracy assessment validated the classification results. Zonal statistics were used to analyze class distribution and spectral responses across different zones. The findings indicated that NDVI effectively distinguished between healthy and stressed plants, while zonal statistics quantified these differences at the regional level. The Random Forest classifier successfully identified crop types such as cotton, sugarcane, and sweet lime, achieving satisfactory accuracy levels. The overall classification accuracy was 81.17%, with a kappa coefficient of 0.7539. This study highlights the effectiveness of integrating NDVI analysis, regional statistics, and Random Forest methods for crop identification and agricultural monitoring, offering a reliable and cost-effective approach for precision agriculture.