Using Sentinel-1, Sentinel-2, and Planet Satellite Data to Map Field-Level Tillage Practices in Smallholder Systems

Abstract

Zero tillage has become more popular among smallholder farmers and understanding patterns of adoption is crucial for evaluating the financial, agricultural and environmental impact of tillage practices on agroecosystems. However, detecting tillage practices is still challenging in smallholder fields (<2 ha) because historically-available satellite data are too coarse in spatial resolution to map individual smallholder fields. In this study, we used newer, higher-resolution satellite data from Sentinel-1, Sentinel-2, and Planet to map tillage practices in northeast India. We specifically tested the classification performance of single sensor and multiple sensor random forest models, and the impact of spatial, temporal, or spectral resolution on classification accuracy. We found when considering a single sensor, Planet imagery (3 m) had the highest classification accuracy (86.55%) and radar Sentinel-1 data (10 m) did little to improve classification accuracy (62.28%). When considering sensor combinations, combining three sensors achieved the highest classification accuracy (87.71%), though this was only marginally better than the Planet only model. We also found that high levels of accuracy could be achieved by using imagery only available during the sowing period. Considering the impact of improved spatial, temporal, and spectral resolution, we found that improved spatial resolution from Planet contributed the most to improved classification accuracy. Overall, it is possible to use readily-available, high spatial resolution satellite data to map tillage practices of smallholder farms, even in heterogeneous systems with small field sizes.