Near-Real-Time Field-Level Monitoring of Soybean and Corn Harvested Fields Using Satellite Imagery

Efficient monitoring and mapping harvested area over cropland with remote sensing data could provide essential agro-geoinformation for more accurate assessment of crop productivity at regional or national scale. In this study, we developed a machine learning model for near-real-time detection of crop harvest area using multi-source satellite imagery data from Landsat 8/9 and Sentinel-2 A/B. By validating against ground truth data, the predicted harvested area map achieved 90.89% accuracy in Nebraska and 85% accuracy in Maryland. The results suggested our model could effectively provide state-wide weekly harvest progress for two major crops, corn and soybean, across the United States, demonstrating its effectiveness and reliability for regional agricultural monitoring.