Estimating crop residue cover using SPOT 5 data

Abstract

Remote sensing of percentage residue cover (PRC) is required to manage crop residue and to provide a key parameter for several biogeochemistry models. However, the growth of the next crop, crop residue decomposition, and precipitation or irrigation can affect the spectral features of crop residue and decrease the spectral difference between crop residue and soil. These factors limit the time window of image acquisition for estimating PRC, especially for double cropping systems (two crops harvested in one year). Remote sensing sensors with a short revisit interval allow imagery to be captured for a specific period (e.g., soon after harvest) when these effects are minimal. SPOT imagery has a short revisit interval of two to three days because of its oblique viewing capacity, and provides balance between high resolution and wide-area coverage. The study aimed to investigate the potential of SPOT 5 imagery for predicting PRC in a typical area of China with a wheat (Triticum aestivum)–corn (Zea mays L.) double cropping system. A SPOT 5 image was acquired with three visible and near infrared bands and one shortwave infrared (SWIR) band, from which six normalized difference indices (NDI) were built. Wheat PRC values (n = 94) were measured immediately after wheat harvest using ground vertical photographs and divided into a calibration set (n = 62) and a validation set (n = 32). To estimate PRC, linear spectral analysis and three regression methods were used. Simple linear regression (SLR) was used for its simplicity, and stepwise regression (SR) and principal component regression (PCR) were used because they could exploit the spectral information from all available image bands. Besides, the bootstrap was used to assess the prediction uncertainty of the regression methods. The results demonstrated that the NDI41 index (derived from SWIR and green bands) performed the best among the four bands and the six indices for estimating PRC when using the SLR method, with a coefficient of determination (R²) of 0.588 and a root mean square error (RMSE) of 0.114 for the validation set. This estimation accuracy was similar to that observed for the SR and PCR models and much better than that for the linear spectral analysis (R² and RMSE were 0.326 and 0.208, respectively). However, NDI41 accompanied by SLR was the optimal choice for estimating PRC because of its simplicity. These results indicate that SPOT 5 imagery could be used to estimate PRC. Therefore, this study should contribute to crop residue management.