Quantitative analysis of salinized soil reflectance spectra during microbial remediation processes based on PLSR

Abstract

The spectral reflectance characteristics of saline soils changing with soil salt content (SSC) were measured and analyzed during the microbial remediation process. To explore the influence of these pre-processing methods on improving performance for estimating the SSC, seven spectral transformation data sets were produced and used to find the optimal sensitive bands and spectral characteristics of SSC in this study. Partial least squares regression (PLSR) method was used to construct relational models between SSC and spectra to estimate soil salinity based on full bands (400–1650nm) and optimal sensitive bands, respectively. The prediction accuracies of these models were assessed by comparing determination coefficients (R2) and root mean squared error (RMSE). The results showed that the optimal spectral bands for eight spectral data sets, concentrated on 947.11nm-949.31nm, 1340.27nm, 1394.11nm, 1419nm, 1457.81nm–1461.31nm, 1537.68nm–1551.39nm and 1602.32nm. The full bands-based model using PLSR method obtained better prediction accuracies on the whole compared with the optimal bands-based model. Among all of the eight spectral data sets in full bands, the prediction accuracy of SGSD was the best, and its values of R2 and RMSEP for the predicted model were 0.67 and 1.26, respectively. For the PLSR predicting models of SSC based on optimal bands, the SGSD (LogR) obtained more robust calibration and prediction accuracies than other pre-processing methods. The optimal bands-based models were much simpler and reducing computation significantly than the full bands-based models.