Discrete Wavelet Transform Approach for the Estimation of Crop Residue Mass From Spectral Reflectance

Abstract

Estimation of crop residue mass (CRM) using cellulose absorption index (CAI) from spectral reflectance data is a widely used approach in crop residue management. A specific limitation with the CAI approach is its inefficacy to predict CRM at high residue loadings and its failure to account for the overlapping of residue fragments on soil surface. In this study, we used a combination of discrete wavelet transform (DWT) and partial least square regression (PLSR) to estimate CRM of rice, wheat, maize, sugarcane and soybean. We followed a wavelet packet approach to select appropriate DWT coefficients by examining the variance (referred to as DWTv-PLSR) and correlation (referred to as DWTc-PLSR) structure of the multi-resolution DWT coefficients. Results showed that the DWTc-PLSR approach yielded excellent predictability regardless of crop residue types. An interesting observation of this study is that the wavelet-based approaches showed significant spectral features in the visible and NIR range in contrast to the commonly used SWIR (2100 nm) range representing the CAI. Spectral reflectance curves in our study and those reported in the literature clearly show that both the depth and width of cellulose absorption peaks generally do not vary much with the residue mass. Such lack of sensitivity may have been portrayed in the DWTc-PLSR approach and this method appears to overcome the limitations of using CAI for crop residue assessment.