Prediction of soil organic matter content in a litchi orchard of South China using spectral indices

摘要

An important drawback of the partial least squares regression (PLSR) method is the complexity of the transfer of spectral prediction models from one sensor to another. The performance of four visible and near infrared (VNIR) spectral indices in predicting the soil organic matter (SOM) content was compared to that of PLSR model using 30 soil samples collected from inside and outside the litchi canopy area of 15 different orchards in South China. The four types of spectral indices are the sum of the first derivative data at spectral region of high correlation (Sum), the maximum band depth magnitude (BDmax), total area (TA), and left area (LA) of characteristic absorption feature. The linear regression method was applied to correlate the spectral indices and SOM contents. The results showed that the left area below the profile of absorption spectrum at 2140-2240 nm (LA_2140-2240) was positively correlated with SOM contents (F value = 82.46), which presented the best performance in the examined spectral indices for the prediction of SOM with the highest coefficient of determination (R-cv(2)) and residual prediction deviation (RPD), and the lowest root mean square error of cross-validation (RMSECV) (R-cv(2) = 0.81, RPD = 2.11, and RMSECV = 0.27%). The accuracy of this LA_2140-2240 index-based model was comparable to that of the PLSR method (R-cv(2) = 0.81, RPD = 2.11, and RMSECV = 0.27%). We concluded that the absorption area index in near infrared spectral range can provide an effective way to estimate the SOM content in litchi orchard of South China. The SOM prediction model based on LA_2140-2240 spectral index can also be transferred from one sensor to another conveniently, which cannot be accomplished with the conventional PLSR method. The calibration method in this study was applied to the largest litchi plantation area of South China and even in the world. It has the potential to be used in other litchi orchards worldwide. (C) 2012 Elsevier B.V. All rights reserved.