Mapping Cover Crops in Southeastern Michigan with Sentinel-2 Remote Sensing data

Abstract

The River Raisin watershed, which spans southeastern Michigan and northwestern Ohio, is a hotspot for negative environmental impacts caused by industrialized and conventional agricultural practices, particularly excess nitrogen leaching and phosphorus runoff polluting the Great Lakes. Planting cover crops is one way for farmers to reduce nutrient losses, and has been adopted by some farmers in this area. To understand the extent of cover crop adoption in the region, in our study we used optical remote sensing data from Sentinel-2 to determine the spatial distribution of cover crops in the River Raisin watershed. The random forest classification algorithm achieved 86.37 % overall accuracy, and for the cover crops in the region, 75.33% (producer’s accuracy - PA) and 80.55% (user’s accuracy -UA) for cereal rye, and 85.90% (PA) and 83.98% (UA) for red clover. In particular, the red edge wavelengths of Sentinel-2 were the most important bands for classifying cover crops. Our study shows that we can use readily-available satellite data to map cover crops with high accuracies in the US Midwest. This implication will better the assessment process of the adoption and impacts of conservation practices on farms.