Thirteenth International Conference on Climate Change: Impacts & Responses

Abstract

This research explores rice yield prediction models using satellite remote sensing-based vegetation indices at the optimum harvesting time before flash flooding. Five relevant vegetation indices (VIs) were used to develop several empirical yield prediction models for rice production. From multiple regression analysis, it was observed that the datasets derived from the satellite imagery in the form of either spectral bands (Red, NIR, SWIR, Blue) or vegetation indices (NDVI, RGVI, and MSI) were the most effective parameters for the rice yield prediction models. The generated models were validated using both parametric (simple and multiple) and nonparametric (artificial neural network, ANN) regression analyses. The crop yield models developed using regression analysis showed very significant agreement with ground reference yield information. The best estimated performances for the RGVI (R^2 = 0.44), NDVI 〖(R〗^2=0.63), NDVI (R^2 = 0.55), and NDVI (R^2 = 0.67) in the simple regression analysis were observed for 2017, 2018, and 2019 and the average seasons during 2017-2019, respectively. On the other hand, the composite NDVI-RGVI 〖(R〗^2 = 0.49), NDVI-NDWI 〖(R〗^2 = 0.65), NDVI-NDWI 〖(R〗^2 = 0.56), and NDVI-MSI 〖(R〗^2 =0.69) indices were the best-performing vegetation indices in developing boro rice yield prediction models using multiple regression. Nevertheless, in the ANN-based machine-learning results, NDVI had a higher accuracy for the average boro rice season (2017-2019) using a simple regression approach (R2 = 0.84) and (R2 = 0.91) for the multiple regression analysis of the average NDVI-MSI composite index.