Remote sensing estimation of cover crop biomass using machine learning algorithms in Denmark

Knowledge of total aboveground biomass (AGB) at harvest time, from site to national scales, is critical to understanding the potential for C sequestration and residue utilization in agricultural ecosystems. Cover crop biomass estimation models using an integration of remote sensing and field data represent a useful inversion technique. Here we present different approaches to estimating cover crop biomass in Denmark, using a variety of different datasets and methods, such as observational data of Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Ratio Vegetation Index (RVI), Precipitation (mm), Land surface temperature (oC), as well as crop grain production (statistical data) for the years 2013 – 2023, supplemented with data from national soil mapping (e.g. texture, SOC, pH) and a Digital Elevation Model (DEM for estimating slope, aspect, etc.). These data were used to develop a spatially explicit peak biomass model for wheat. Our results show that the cover crop biomass in Denmark showed significant spatial and interannual variation from 2013 to 2023, with the spatial and temporal distribution characteristics closely related to NDVI. The linear regression between peak biomass and NDVI yielded a R2 of 0.78, which could be further improved by also considering other information on soils, fertilization, and landscape characteristics. Our results indicate a high potential to estimate biomass for the cover crops based on combined information by remote sensing, soil mapping and field management practices. This information, in combination with biogeochemical modeling, will allow assessment of the potential of agricultural landscapes to provide estimates of regional availability of residues for the bioeconomy while maintaining or improving soil organic C stocks.