ESTIMATING INFILTRATION PARAMETERS FROM BASIC SOIL PROPERTIES

Abstract

Infiltration data were collected on two rectangular grids with 25 sampling points each. Both experimental grids were located in tropical rain forest (Guyana), the first in an Arenosol area and the second in a Ferralsol field. Four different infiltration models were evaluated based on their performance in describing the infiltration data. The model parameters were estimated using non-linear optimization techniques. The infiltration behaviour in the Ferralsol was equally well described by the equations of Philip, Green–Ampt, Kostiakov and Horton. For the Arenosol, the equations of Philip, Green–Ampt and Horton were significantly better than the Kostiakov model. Basic soil properties such as textural composition (percentage sand, silt and clay), organic carbon content, dry bulk density, porosity, initial soil water content and root content were also determined for each sampling point of the two grids. The fitted infiltration parameters were then estimated based on other soil properties using multiple regression. Prior to the regression analysis, all predictor variables were transformed to normality. The regression analysis was performed using two information levels. The first information level contained only three texture fractions for the Ferralsol (sand, silt and clay) and four fractions for the Arenosol (coarse, medium and fine sand, and silt and clay). At the first information level the regression models explained up to 60% of the variability of some of the infiltration parameters for the Ferralsol field plot. At the second information level the complete textural analysis was used (nine fractions for the Ferralsol and six for the Arenosol). At the second information level a principal components analysis (PCA) was performed prior to the regression analysis to overcome the problem of multicollinearity among the predictor variables. Regression analysis was then carried out using the orthogonally transformed soil properties as the independent variables. Results for the Ferralsol data show that the parameters of the Green–Ampt and Kostiakov model were estimated relatively accurately (maximum R2 = 0.76). For the Arenosol, use of the second information level together with PCA produced regression models with an R2 value ranging from 0.38 to 0.68. For the Ferralsol, most of the variance was explained by the root content and organic matter content. In the Arenosol plot, the fractions medium and fine sand explained most of the observed variance.