Mapping Variations in Crop Growth and Irrigation in a Crop Field with Landsat-Derived Spectral Products

Abstract. The shift from conventional irrigation methods to sprinkler systems is intended to improve accuracy and efficiency; however, it requires thorough validation of the uniformity of water distribution. This research focuses on a particular agricultural issue related to possible coverage deficiencies in a 33-hectare field that has recently been fitted with a sprinkler system. The main goal was to detect spatial discrepancies in crop growth utilizing Landsat-derived data from the 2024-25 growing season. The analytical approach employed was Mean Shift Clustering (MSC), a non-parametric, unsupervised machine learning technique, to segment images of the Normalized Difference Vegetation Index (NDVI). In contrast to parametric techniques that necessitate predetermined cluster counts, MSC interprets the flattened 1D NDVI feature space as an empirical probability density function. By applying an adaptive bandwidth (calculated using a 0.1 quantile estimate), the algorithm iteratively adjusted data points towards high-density modes to autonomously ascertain the optimal number of growth zones. Concurrently with this machine learning-driven segmentation, a visual examination of Normalized Difference Moisture Index (NDMI) pixels was conducted to evaluate moisture levels. The MSC algorithm identified six distinct spectral clusters with the following average NDVI values: Cluster 0 (0.84), Cluster 1 (0.72), Cluster 2 (0.82), Cluster 3 (0.71), Cluster 4 (0.19), and Cluster 5 (0.68). Importantly, both the unsupervised NDVI clustering and the NDMI moisture assessment produced consistent findings: no significant spatial anomalies or indications of water stress were observed. The study verified that the sprinkler system delivered a uniform water supply, effectively addressing the farmer’s concerns regarding variations in irrigation.