Improved crop residue cover estimates obtained by coupling spectral indices for residue and moisture

Remote sensing assessment of crop residue cover (fR) and tillage intensity can improve predictions of the environmental impact of agricultural practices and promote sustainable management. Spectral indices for estimating fR are sensitive to soil and crop residue water contents, therefore the uncertainty of fR estimates increases when moisture conditions vary. Our goals were to evaluate the robustness of spectral residue indices based on the shortwave infrared region (SWIR) for estimating fR and to mitigate the uncertainty caused by variable moisture conditions on fR estimates. Ten fields with center pivot irrigation systems (eight partially irrigated and two uniformly dry fields) were identified in Worldview-3 satellite imagery acquired for a study site in Maryland (USA). The fields were mid-irrigation at the time of imagery acquisition, allowing comparison of residue cover under dry and wet conditions. Fields were subdivided into approximately equal-size wedges within the dry and wet portions of each field, and the SWIR bands were extracted for each pixel. Two crop residue indices (Normalized Difference Tillage Index (NDTI); Shortwave Infrared Normalized Difference Residue Index (SINDRI) and a water index (WI) were calculated. Reflectance in each band was moisture-adjusted based on the WI difference between wet and dry wedges, and updated NDTI and SINDRI were calculated. Finally, the probability density distributions of fR estimated from the residue indices were calculated for each field. SINDRI was more robust than NDTI for estimating fR. Moisture corrections of spectral bands reduced the root mean square error of NDTI fR estimates from 22.7% to 4.7%, and SINDRI fR estimates from 6.0% to 2.2%. The mean and variance of the probability density distribution of fR estimated from residue indices, before and after moisture correction, were greatly reduced in the partially irrigated fields, but only slightly in fields with uniform water distribution. The estimation of fR should be based on SINDRI if appropriate bands are available, but fR can be reliably estimated by combining NDTI with a water content index to mitigate the uncertainty caused by variable moisture conditions.