Detecting winter cover crops and crop residues in the midwest us using machine learning classification of thermal and optical imagery

Abstract

Cover crops are an increasingly popular practice to improve agroecosystem resilience to climate change, pests, and other stressors. Despite their importance for climate mitigation and soil health, there remains an urgent need for methods that link winter cover crops with regional-scale climate mitigation and adaptation potential. Remote sensing is ideally suited to provide these linkages, yet, cover cropping has not been analyzed extensively in remote sensing research. Methods used for remote sensing of crops from satellites traditionally leverage the difference between visible and near-infrared reflectance to isolate the signal of photosynthetically active vegetation. However, using traditional greenness indices like the Normalized Difference Vegetation Index (NDVI) for remotely sensing winter vegetation, such as winter cover crops, is challenging because vegetation reflectance signals are often confounded with reflectance of bare soil and crop residues. Here, we present new and established methods of detecting winter cover crops using remote sensing observations. We find that remote sensing methods that incorporate thermal data in addition to traditional reflectance metrics are best able to distinguish between winter farm management practices. We conclude by addressing the potential of existing and upcoming hyperspectral and thermal missions to further assess agroecosystem function in the context of global change.