Hyperspectral estimation of soil organic matter content based on partial least squares regression

Abstract: Soil organic matter (SOM) plays an important role in soil fertility and carbon (C) cycle. Soil spectral reflectance provides an alternative method to soil's classical physical and chemical analysis in laboratory for the estimation of a large range of key soil properties. In order to achieve rapid measurement of soil organic matter content (SOMC) based on hyperspectral analysis, in this paper, 46 soil samples at 0-20 cm depth were collected as research objects from Gong'an County in Jianghan Plain, and these samples were highly representative for the SOM. The raw hyperspectral reflectance of soil samples was measured by the standard procedure with an ASD FieldSpec3 instrument equipped with a high intensity contact probe under the laboratory conditions. Meanwhile, physical and chemical properties of these soil samples were analyzed. Twenty-eight of 46 samples were used for building hyperspectral estimation models of SOMC and the other 18 samples were used for model prediction. In the next, the raw spectral reflectance (R) was transformed to 3 spectral indices, i.e. logarithm of reciprocal reflectance (LR), first-order differential reflectance (FDR) and continuum removal reflectance (CR) to analyze the correlation coefficients between the 4 spectral indices and their SOMC. Then, the correlation coefficients of the 4 spectral indices by F significant test were got (P<0.01), which could be used to extract significant bands. At last, we used partial least squares regression (PLSR) method to build quantitative inversion model of SOMC based on full bands (400-2 400 nm) and significant bands for this study area, respectively. The prediction accuracies of these optimal models were assessed by comparing determination coefficients (R2), root mean squared error (RMSE) and relative percent deviation (RPD) between the estimated and measured SOMC. The results showed that, after conducting the CR transformation on raw soil spectral data, there were prominent differences among the absorption peaks of spectral curves in different soil samples, and the heterogeneity of different spectral curves was decreased to a certain extent, at the same time, their correlations were also improved by about 0.2 in the range of visible bands. Compared to the significant bands, the full bands using PLSR method could obtain more robust prediction accuracies. Among all of the 4 spectral indices based on processing inversion models in full bands, the prediction accuracy of CR was the best, and its values of R2, RMSE and RPD between the estimated and measured SOMC for the predicted model were 0.84, 3.86 and 2.58, respectively, which were better than those in significant bands. For the PLSR models based on significant bands, although there was a slight gap in the prediction accuracy with that based on full bands, they also had their own unique advantages: these models were much simpler and thus the model computation was reduced significantly, and they could play an important role under the circumstances in which increasing modeling speed and reducing model computation were more important than improving prediction accuracy. At last, it could be concluded that the CR-PLSR model for SOMC was better than R-PLSR, LR-PLSR, FDR-PLSR models not only in full bands but also in significant bands. In the future, the CR-PLSR hyperspectral inversion model can be used as a reference for aerospace hyperspectral remote sensing of soil fertility information in this region, and can realize the timely monitoring of SOMC.