Wet Aggregate Stability Predicting of Soil in Multiple Land-Uses Based on Support Vector Machine

The stability of soil aggregates plays a vital role in soil quality and represents the ability of soil-aggregates to resist destruction under disturbance. The direct determination of wet aggregate stability (WAS) is time-consuming and expensive. In this paper, we attempted to estimate WAS including by using different methods, including multiple linear regression (MLR), artificial neural network (ANN), and support vector machine (SVM). We chose sand, silt, clay, organic carbon (OC) and particle density (DP) as input variables. The 134 soil samples from different land-uses (crop, grass, and bare) were utilized to evaluate the utility of these techniques and confirm the effective variables. 107 samples were selected to calibrate the predictive model and the rest was utilized for testing. The result show that SVM is superior to ANN and MLR where R²=0.685 and the root mean squared error (RMSE) = 9.54. it is clear that OC > silt > clay > DP > sand is the order of sensitive variables predicted by WAS.

I. Introduction

Soil resource, as a vital component of earth ecosystems, is one of the most significant and fundamental factors for human development [1]. Soil degradation which is greatly controlled by aggregate stability, has captured much attentions [2–3]. Low aggregate stability means that soil structure can be readily destroyed, which will lead to soil productivity deterioration, soil erosion aggravation, ecosystem degradation, and surface crusting [4].