Comparison of different satellite bands and vegetation indices for estimation of soil organic matter based on simulated spectral configuration

Soil organic matter content (SOM) is an important indicator of soil productivity that governs biological, chemical, and physical processes in the soil environment. Previous studies have shown that remote sensing data provide useful information for SOM estimation in different soil types. However, no studies have estimated SOM based on simulated spectral configurations of different satellite sensors. Further study is required to investigate whether SOM estimation accuracy can be improved by combining data from different satellite sensors and developing appropriate algorithms. Therefore, this study investigated new methods for SOM estimation with the following three objectives: (1) analyze the reflectance changes of simulated bands for different SOMs using the spectral response function of various satellite sensors; (2) develop optimal difference index (ODI), optimal ratio index (ORI), optimal normalized vegetation difference index (ONDVI), and optimal enhanced vegetation index (OEVI) algorithms for estimating SOM based on simulated band reflectance; (3) evaluate all bands, ODI, ORI, ONDVI, and OEVI for all simulated bands derived from the data of each satellite, and then combine the simulated data to estimate SOM using the particle swarm optimization (PSO)-support vector machine (SVM) algorithm. The OEVI analysis of simulated WorldView-2 data provided the best SOM estimation accuracy (R-2 = 0.43 and RMSE = 2.62%). The OEVI and ODI algorithms provided better estimation accuracy of SOM from the different simulated satellite data than the ORI and ONDVI algorithms. The best estimation accuracy of SOM was achieved using the PSO-SVM algorithm and simulated WorldView-2 data (R-2 = 0.77, RMSE = 1.66%, and AIC = 99.62). Combination of simulated bands 4-9 of ASTER data and all bands, ODI, ORI, ONDVI, and OEVI of WorldView-2 data provided optimum SOM estimation results (R-2 = 0.82, RMSE = 1.41%, AIC = 82.86). The results indicate that a combination of different satellite data and the PSO-SVM algorithm significantly improves the estimation accuracy of SOM.