Crop Yield Prediction Using Remotely Sensed Derived Primary Production and Deep Learning

Crop yield prediction before harvest is crucial for policy-making and decision-making related to imports and exports of grains. Yield prediction on large scale and before harvest is possible due to the availability of a high volume of remote sensing data and suitable tools. Remote sensing has been used recently to characterize different aspects of crops. Rapid developments in machine learning have made it easier to apply these techniques to high volumes of data. This study aims to investigate the potential of remote sensing derived gross primary production for crop yield prediction using deep learning. Moderate Resolution imaging spectroradiometer (MODIS) has a synthesized product known as Gross Primary Productivity (GPP) which is available as MOD17A2 (Terra), and MYD17A2 (Aqua). Corn yield data for 15 years was acquired from the United States Department of Agriculture (USDA) which is available on the county-level scale. GPP data obtained from MODIS were aggregated to the county level and time-series of GPP for crop duration were generated. Potential of several machine learning and deep learning algorithms to predict corn yield were analyzed in this study.