Remote estimation of soil organic matter content in the Sanjiang Plain, Northest China: The optimal band algorithm versus the GRA-ANN model

Abstract

Soil organic matter content (SOMC) is an important factor that reflects soil fertility, land production capacity, and the degree of soil degradation. The objectives of this study were to (i) test various regression models for estimating SOMC based on published spectral parameters in the Sanjiang Plain, (ii) develop optimal band difference, ratio, and normalized difference algorithms for assessing SOMC using spectral data, and (iii) compare the performance of the proposed models using grey relational analysis-artificial neural networks (GRA-ANN) and the band difference algorithm. The SOMC data and concurrent spectral parameters were acquired in the Sanjiang Plain of Northest China in 2006. For the GRA-ANN model, GRA was used to select the sensitive spectral parameters and ANN was established to estimate SOMC. The results showed that reflectance (R) gradually decreased with increasing SOMC and the regression equations based on the spectral parameter 1/R588, Diff (1/R835), R610, R654, R550, and R520 could be used to estimate SOMC, respectively. The SOMC model based on the optimal difference index (ODI; R2=0.63 and RMSE=1.43%) outperformed those based on the optimal ratio vegetation index (ORVI; R2=0.48 and RMSE=1.82%) and normalized difference vegetation index (ONDVI, R2=0.57 and RMSE=1.56%). The GRA-ANN model presented better SOMC estimation results (R2=0.90 and RMSE=0.88%). Thus, the GRA-ANN model has great potential for SOMC estimations; however, the ODI also has merit, especially when taking into consideration the simplicity of its application. Combining different algorithms may improve SOMC estimations on a regional scale.