Airborne hyperspectral imaging of cover crop outcomes and tillage intensity in croplands by machine learning and radiative transfer modeling

As major conservation practices, cover crop and reduced tillage can strongly influence cropland nutrient and carbon dynamics. Beyond traditional remote sensing to quantify adoption of these management practices, the spatially and temporally resolved information of cover crop growth outcomes (e.g. nutrient and carbon uptakes) and tillage intensity (e.g. crop residue fraction) is important for evaluating nutrient and carbon outcomes of management practices. With hundreds of narrow wavelengths and high spatial resolution, airborne imaging spectroscopy provides great opportunities to quantify cover crop growth outcomes and crop residue fraction accurately and efficiently. Here, we used visible to shortwave infrared (4002400 nm) airborne hyperspectral imaging with high spatial (0.5 m) and spectral (35 nm) resolution to acquire surface reflectance over cropland in Champaign County, IL from December 2020 to April 2021. In cover crop fields, we collected leaf area index, aboveground biomass and nutrients. We used a hybrid approach to integrate machine learning and soil-canopy radiative transfer models to retrieve cover crop leaf area index, aboveground biomass and nutrients. Extensive photos of ground cover were used to derive ground-truth crop residue fraction through image segmentation and active learning. Furthermore, we used long-short term memory networks to upscale the ground-derived crop residue fraction to airborne hyperspectral imagery. Results show that the integration of high-resolution airborne hyperspectral imagery with machine learning can provide highly accurate estimates of cover crop growth outcomes and crop residue fraction. These high-resolution estimates can provide valuable information for stakeholders to take adaptive strategies for sustainable agriculture.