Enhancing USDA NASS Cropland Data Layer with Segment Anything Model

Crop-specific land cover mapping is a vital application in agro-geoinformatics with the proliferation of remote sensing data and machine learning techniques. This paper presents a novel approach to enhance the well-known Cropland Data Layer (CDL) product by U.S. Department of Agriculture (USDA) National Agricultural Statistics Service (NASS) using Meta's Segment Anything Model (SAM). The study leverages SAM's zero-shot generalization capability to automatically delineate cropland fields from Sentinel-2 images. By voting for the major crop types within each delineated land unit, a substantial number of noisy pixels is CDL can be eliminated, leading to notable improvements in mapping accuracy. Preliminary experimental results across key agricultural regions in the U.S., such as California's Central Valley and Corn Belt, suggest that SAM can significantly enhance the quality of the original CDL data. This ability to refine crop-specific land cover data, like CDL, demonstrates SAM's practical applicability within agricultural monitoring systems. Moreover, the result showcases the promising potential of integrating SAM into existing crop type classification workflows to create high-quality early- and in-season crop type maps on a national scale with minimal effort.