Tillage and biomass detection for estimating winter-time cropland management practices with satellite remote sensing

Abstract

Supportive policies to promote sustainable agriculture have been implemented across countries and regions. For example, continuous vegetative groundcover and reduced tillage have been advocated for sustainable post-harvest biomass management. Accurate and timely information on cropland management practices is needed for agricultural policy evaluations, evidence-based planning, and agri-environmental assessments. We show that a satellite-based approach can yield off-season cropland management information on preferred spatial and temporal scales from a narrow window of opportunity in early spring after snow melt and before seedbed preparation. Agricultural parcel geometries from an administrative registry were used to extract information on Sentinel-1 backscatter and coherence, and Sentinel-2 spectral reflectance. Based on a large survey-based dataset of 6,623 fields, we show that the highest impact on model performance comes from the spectral regions of near-infrared and upper red edge of the Sentinel-2 mission, whereas Sentinel-1–based features had a relatively small contribution to classification performance. Our proposed method for tillage and biomass detection generalises well in the study area of boreal environmental zone with dominantly mineral soils, as confirmed by the high test set classification accuracy of 85%. The supporting dataset and codes are stored in a publicly accessible repository.