Spatial stochastic model for predicting soil organic matter using remote sensing data

Abstract

Accurate soil organic matter (SOM) estimation could provide critical information to understand soil organic carbon sequestration, soil fertility, and the global carbon cycle. A nearest-neighbourhood autoregressive moving average (NN-ARMA) modelling technique along with linear regression has been used to predict localized soil SOM variation based on topographical characteristics and vegetation indices in semi-arid region of Saudi Arabia. Seven topographic variables derived using DEM, and twelve vegetation indices obtained from Landsat 8 used in the model. The best NN-ARMA model showed seven significant variables explaining 96.4% of the total variation of SOM, whereas the best linear regression model could explain 78.8% of the total variation of SOM. The results showed that NN-ARMA model gave better results compared to the linear regression model. Our research gave a better understanding of the possibility of accurate estimation of SOM using the NN-ARMA approach using topographical characteristics and vegetation indices easily acquired by RS sensors.