Multiyear crop residue cover mapping using narrow-band vs. broad-band shortwave infrared satellite imagery

Abstract

Crop residue serves an important role in agricultural systems as high levels of fractional crop residue cover ( f R ) can reduce erosion, preserve soil moisture, and build soil organic carbon. However, the ability to accurately quantify f R at scale has been limited. In this study we produced annual maps of f R for farmland in Maryland, USA using WorldView-3 (WV3) imagery paired with on-farm photographs ( n = 895) classified to f R using SamplePoint software. Univariate linear regressions were used to compare photograph f R to WV3 crop residue indices including: 1) Shortwave Infrared Normalized Difference Residue Index (SINDRI), 2) Shortwave Infrared Difference Residue Index (SIDRI), 3) Normalized Difference Tillage Index (NDTI), and 4) Shortwave Infrared Angle Index (SWIRA). SINDRI and SIDRI are based on narrow bands capable of measuring lignocellulose absorption features. NDTI and SWIRA are based on Landsat-comparable broad bands. Our findings demonstrated that SINDRI outperformed other indices in f R estimation in terms of coefficient of determination ( R 2 = 0.869) and root mean square error (RMSE = 0.111), when R 2 and RMSE were averaged across six individual years. For a univariate analysis combining five years of high-quality WV3 imagery, SINDRI again exhibited the highest f R estimation performance ( R 2 = 0.795; RMSE = 0.141), suggesting that SINDRI can map f R accurately with a singular relationship, potentially reducing the need for labor-intensive ground data collection. For broad-band indices, a multiple linear regression analysis that included a Water Index (WI) and Normalized Difference Vegetation Index (NDVI) as additional predictors increased the accuracy of f R estimation significantly, particularly for SWIRA ( R 2 = 0.767; RMSE = 0.144), but also NDTI ( R 2 = 0.654; RMSE = 0.174). Our findings suggest that while indices computed from narrow-band imagery are most accurate for f R estimation, SWIRA has the potential to improve f R estimation compared to NDTI, especially when used in conjunction with WI and NDVI. An index suite of SWIRA, WI, and NDVI can be computed with Landsat 4–9 imagery, providing a more accurate record of global f R dating back to 1982.