Integration of tillage indices and textural features of Sentinel-2A multispectral images for maize residue cover estimation

Abstract

In this work, Sentinel-2A images were used to estimate maize residue cover (MRC) using tillage indices (TIs) and textural features. A MRC estimation model based on partial least squares regression (PLSR) was constructed using different features and methods (TIs method, textural feature method, and combination method). Results showed that the normalized difference tillage index (NDTI; r:0.854; R2: 0.729) and the simple tillage index (STI;r:0.853; R2: 0.728) had good linear relationships with MRC. Several models involving different combinations of TI and textural feature indicators were examined in regard to their accuracy. The best performing model (r:0.885; R2: 0.783; and RMSE = 10.63%), a combination of seven TIs and eight textural feature indicators, was selected to estimate MRC in the Songnen Plain, Northeast China. Results showed that MRC ranged from 0.05% to 95% over the study area (mean: 47%) and was generally higher in the northern and southern areas than in the central portion of the Songen Plain. From these results, the distribution of conservation tillage practices across the region was determined. These results demonstrate the spatial distribution of MRC can be monitored in a systematic way using remote sensing technologies, and this information can help guide research and outreach efforts for the emergence of more sustainable agricultural production systems.