Modeling crop yield using NDVI-derived VGM metrics across different climatic regions in the USA.

Abstract

An accurate, cost-efficient, and early crop yield projection is a national and global necessity. This study is aimed to achieve the national demands by proposing crop yield estimation models on the Google Earth Engine (GEE) platform. This study directly used dynamic crop phenology metrics to model soybean crop yield by considering climatic regions in the USA (e.g., Central, East, Northeast, South, Southeast, and West North Central regions). We modeled soybean yields with vegetative growth metrics (VGMs) of NDVI narrated as VGM70 (avg. NDVI of 70days from the emergence), VGM85 (avg. NDVI of 85days from the emergence), VGM98T (total NDVI of 98days from the emergence), VGM120 (avg. NDVI of 120days from the emergence), VGMmean (avg. NDVI of the growth season), VGMmax (maximum NDVI of the growth season), and climatic factors (i.e., daytime surface temperature: DST, night-time surface temperature: NST, and precipitation) from 2000 to 2019. This study further explored individual predictors and a combination of predictors in modeling crop yield for diverse climatic regions. Therefore, we proposed six crop yield linear models for each of the climatic divisions, and these models are then compared with support vector machine (SVM) models. All models showed reliable predictability based on adjusted R-square, normalized root mean square error (NRMSE), normalized mean prediction error (NMPE) parameters, and a p-value of less than 0.001.The impact of the independent predictor in the best crop yield models is discussed based on the regression weights (beta weight: β), and, the VGMmax is identified as the significant predictor in crop yield modeling for different climatic regions. Overall, this study will help the national agricultural management system for better monitoring and forecasting of soybean yield to support and manage soybean production.