Field-scale crop yield prediction using multi-temporal WorldView-3 and PlanetScope satellite data and deep learning

Agricultural management at field-scale is critical for improving yield to address global food security, as providing enough the world’s growing population has become a wicked problem both scientists and policymakers. County- or regional-scale data do not provide meaningful information farmers who are interested in forecasting effective timely field management. No studies directly utilized raw satellite imagery prediction using deep learning. The objectives of this paper were twofold: (1) develop imagery-based learning approach prediction, (2) investigate contribution in-season multitemporal grain with hand-crafted features WorldView-3 (WV) PlanetScope (PS) direct input, respectively. Four WV-3 25 PS collected during season soybean utilized. Both 2-dimensional (2D) 3-dimensional (3D) convolution neural network (CNN) architectures developed that integrated spectral, spatial, temporal contained data. For comparison, hundreds carefully selected textural, optimal crop growth monitoring extracted fed into same model. Our results demonstrated was able predict extent comparable feature-fed approaches; 2D 3D CNN models explain nearly 90% variance yield; (3) limited number outperformed multi-temporal entire mainly attributed RedEdge SWIR bands available WV-3; (4) increased power compared due its ability digest from