Tillage Detection with Sentinel-1 Radar Data.

Estimates of zero-till (ZT) adoption in the Indo-Gangetic-Plain (IGP) vary according to survey because these data are sparsely reported in agricultural statistics and ZT machinery is sometimes used for sowing after conventional till (CT) land preparation. The objective of this study is to assess the feasibility of using radar (Sentinel-1, SAR-C dual-pol) and optical (Sentinel-2) data to detect ZT planting of wheat in Punjab and Haryana, India. SAR data have the advantage that they are not affected by clouds and haze, but they have a low signal to noise ratio. For this reason, we used a stratified object-based approach. We delineated the segments based on optical images. The classification parameters include the γ⁰_VV, γ⁰_VH, and ratio backscattering coefficients from the SAR data. The training of the random forest classifier is based on a fields’ practices map collected through an extensive ground truth data collection campaign. We observed a general increase in the differences in backscattering intensities between VV and VH starting in late October. This coincides with the main land preparation and sowing period. The increase in backscattering was distinctively lower for no-tillage fields. This method detects differences in tillage practice with average accuracies ranging from 78 to 90 %, depending on the stratum. The resulting map shows large spatial differences in ZT adoption rates. Patches of practices tend to reflect previous efforts to promote zero tillage, with limited adoption outside of those areas. In the sampled regions, adoption rates range between 5 and 20%.