Sub-field crop yield prediction using satellite remote sensing and machine learning

Goals / Objectives
Crop yield is of great interest to economics and food security. Remote sensing is a tool that may be used to estimate crop yield from space, and could be especially useful over regions where ground data is not available. The current main limitation is the lack of publicly available ground truth data for calibrating estimates. The Hydrology and Remote Sensing Lab recently published an open, sub-field scale crop yield dataset at at the Beltsville Agricultural Research Center (BARC). The dataset extends from 2014 to 2023 and covers about 20 fields per year on average at 5 meter grid scale. ARS scientists will work with the Texas State University to conduct research on crop yield estimation using satellite data. The primary objective of this work is to assess the capability of machine learning in making accurate sub-field scale crop yield estimates.

Project Methods
The cooperative work proposed here will assess the accuracy of yield mapping at sub field scales that may be obtained by machine learning approaches. The work will focus on staple crops for which data is most plentiful (e.g., corn, soybeans). The work will investigate two different spatial scales: (1) performance at BARC fields, (2) and correspondence to yield statistics provided by the National Agricultural Statistic Service at the Maryland county or state level (whichever is available).