Mapping and Yield Prediction of Castor Bean (Ricinus communis) Using Sentinel-2A Satellite Image in a Semi-Arid Region of India

Abstract

Abstract Castor bean (Ricinus communis) indigenous to the southeastern Mediterranean basin, eastern Africa and India is a crop having various industrial and medicinal applications. It is helpful in crop rotation and replenishing the soil nutrients due to less water consumption. The current study explores the utility of Sentinel-2A satellite image for mapping and yield prediction of castor beans. Several classification methods viz. migrating means clustering, maximum likelihood classifier, support vector machine and artificial neural network are used for the classification and mapping of different landscape categories. The overall classification accuracy was achieved to be highest for artificial neural network (85.81 %) subsequently support vector machine (80.12 %), maximum likelihood classifier (74.23 %) and migrating means clustering (73.03 %). The yield prediction is performed using Sentinel-2A-derived indices namely Normalized Difference Vegetation Index and Enhanced Vegetation Index-2. Further, the cumulative values of these two indices are investigated for castor bean yield prediction using linear regression from July 2017 to April 2018 in different seasons (pre-monsoon, post-monsoon, and winter). The regression model provided (adj R2=0.75) value using EVI-2 compared to (adj R2=0.55) using NDVI for yield prediction of Ricinus communis crop in the winter season. The methodology adopted in this study can serve as an effective tool to map and predict the productivity of Ricinus communis. The adopted methodology may also be extended to a wider spatial level and for other significant crops grown in semi-arid regions of world.