Wavelet incorporated spectral analysis for soil property determination

Quantitative analysis of soil spectroscopic reflectance has been used by many agricultural researchers for soil property determination and is considered a fundamental step towards the development of in-field optical sensors for soil properties. In this study, wavelet analysis was applied to soil spectroscopic reflectance data: (1) to examine its spectral smoothing capability, and (2) to incorporate wavelet analysis with the conventional regression methods for soil property determination. The results showed that, in terms of representing the original data curve, wavelet smoothing obtained an MSE (mean square error) of 2.83e-5, outperforming both Fourier filtering (MSE of 9.28e-5) and the Savitzky-Golay technique (MSE of 9.20e-5). The soil properties under investigation were Ca, K, Mg, Na, P, Zn, clay, and sand. Compared to the regression models developed by the conventional method, those developed with wavelet analysis yielded almost the same or slightly higher R 2 values yet contained fewer regressors for all soil properties. The multi-resolution capability of wavelet analysis was able to provide deeper insight into soil property models from the spectroscopic perspective. The regression models developed by wavelet analysis successfully predicted Ca, Mg, P, and Zn with reasonably high R 2 values (greater than 0.5) and low RMSEs (smaller than 20%) when the validation soil sample dataset was used. 2006 American Society of Agricultural and Biological Engineers.

Wavelet incorporated spectral analysis for soil property determination Academic Article