Prediction of Soil Enzymes Activity by Digital Terrain Analysis: Comparing Artificial Neural Network and Multiple Linear Regression Models

Abstract

This study was conducted to use soil and topographic attributes to predict the activity of three soil enzymes: L-asparaginase, L-glutaminase, and urease by artificial neural networks (ANNs) and multiple linear regression (MLR) approaches in a hilly region of central Iran. A total of surface (0–10 cm depth) soil samples were collected from the site under pasture. Sampling points were chosen in a stratified random manner from geomorphic surfaces, including summit, shoulder, backslope, footslope, and toeslope at the site. MLR and feed-forward back-propagation of ANNs were employed to develop models to predict soil enzymes activity (SEA). Results of the study showed that MLR models explained 37%–61%, and ANN models explained 96%–98% of the variability in the three SEA at the site studied. Overall, the results indicated that the ANN performed better in predicting the SEA than did MLR. Sensitivity analysis showed that topographic parameters as the easily accessible auxiliary variables were the most important factors for predicting the SEA prediction. It was concluded that digital terrain models (DTMs) can be applied to predict spatial distribution of the SEA at the hillslope scale.