Incorporating Probability Density Functions of Environmental Covariates Related to Soil Class Predictions

Abstract

The distribution of continuous soil properties and their environmental covariates within soil classes are often times unknown or not evaluated. Understanding and defining the distribution of environmental covariates within soil classes is fundamental to the fuzzy logic inference mapping processes. Under knowledge-based applications of fuzzy logic mapping, the user typically utilizes predetermined distribution functions to define a representative relationship between soils and their covariates. If the predetermined distributions do not adequately describe the soil–covariate relationship, the misrepresentation can lead to inadequate prediction of soil properties while requiring a high level of user input. To move away from knowledge-based “guesses” of distributions, we present a new and innovative method of modeling the distribution of environmental covariates, specifically terrain attributes (TAs), within the landform-based soil classes. This eliminates the need for manually manipulated, user-defined curves and works to more accurately represent the distribution of TAs within soil classes. The fully automated method fits a variety of probability distribution functions (PDFs) to TA values within algorithm-derived landform classes. We compared the Pearson’s correlation coefficient for goodness of fit to determine which PDF best models the distribution of TAs within soil classes. This fully automated method works to improve our understanding of how terrain attributes vary within soil classes, allowing for more accurate and reliable model predictions.