Remote Sensing Based Soybean Crop Health Assessment Using Sentinel-2 Vegetation Indices and Machine Learning

Abstract

Remote sensing methodologies have received increased importance for the surveillance of the vitality of crops and the enablement of precision agriculture. Satellite based multispectral imagery provides critical information about the vegetation status using spectral reflectance measurements and vegetative indices. In the current investigation, imagery collected from Sentinel-2 satellite was used for assessing the health condition of soya bean crops in a 4-acre agricultural plot located in the Sillod region of Maharashtra, India. Two temporal observations, which were dated at 1st October, 2022 and 26th October, 2022 were analyzed in order to understand the temporal variations in crop vigor during the advanced growth stages of soybean cultivation. Sentinel 2 Level 2A surface reflectance data were pre-processed by cloud and spatial clipping to the delineated area of interest. The calculation of vegetation indices, i.e. Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), Normalized Difference Red Edge Index (NDRE), and Chlorophyll Index Green (CIgreen) was used to characterize the canopy vigor and the concentration of chlorophyll.Machine learning classifiers, namely, Random Forest (RF) and Support Vector Machine (SVM) were used to analyze the crop condition patterns using the extracted vegetation indices. Crop health classification using NDVI threshold values classified the plot into various zones of healthy, moderate and poor vegetation conditions. The data showed that at 1st October 2022, most of the plot had healthy crop conditions, with 42 locations having healthy conditions and two having moderate conditions. As of 26 October 2022, the number of healthy locations had fallen to 34 but moderate and poor conditions had risen to eight and two locations respectively. Mean vegetation index values also showed a decrease between the two dates with NDVI showing a decrease from 0.746 to 0.673 while CIgreen showed a decrease from 4.141 to 3.666. The average NDVI change of 0.073 shows a slow decline in canopy vigor as the soybean crop came closer to phenological maturity. The results show the value of Sentinel-2 derived vegetation indices in monitoring temporal variations in the health of crop in soybean fields.