K-means clustering applied to vegetation indices for mapping cultivated areas using high-resolution Moroccan Mohammed VI satellite imagery

Abstract

This study presents a pixel-based unsupervised classification approach for mapping cultivated land using high-resolution imagery from the Moroccan Mohammed VI satellite. The proposed method integrates the K-means clustering algorithm with spectral features derived from vegetation indices, particularly the Normalized Difference Vegetation Index (NDVI) and the Modified Normalized Difference Water Index (MNDWI), together with the Near-Infrared (NIR) band. The output is a classified map composed of three classes: background, bare soil, and crop-dominated areas. The method was evaluated over a 175-hectare agricultural region in northern Morocco and achieved a relative error of 1.41%, significantly outperforming NIR threshold-based classification (7.2% error), NDVI-based classification (6.95%), and standard K-means classification using spectral bands only (5.47%). The results demonstrate the effectiveness of combining vegetation indices with unsupervised clustering and highlight the potential of the high-resolution satellite imagery for field-scale agricultural mapping, precision irrigation support, and sustainable land management.