COMPARISON OF LINEAR AND NONLINEAR MODELS TO ESTIMATE THE RISK OF SOIL CONTAMINATION.

Abstract

The study of pollution in geographical areas includes spatial dependence, non-normal distribution, and heteroscedasticity. However, the modelling of edaphological data has not taken these features into consideration. Therefore, this study included the analysis and comparison of the behavior of the estimators of generalized linear regression (GLM), generalized linear mixed (GLMM), generalized additive (GAM), and generalized additive mixed (GAMM) models, through the simulation of a response variable generated with different statistical distributions, with five weighing matrixes (W, B, C, U, and S) and several autocorrelation levels. The results showed a strong U-adjacency matrix for all spatial autocorrelation levels. As was expected, GAMs and GAMMs were higher than GLMs and GLMMs, as a consequence of their flexibility which is represented by smoothing splines and the incorporation of mixed effects. The concentration of heavy metals and the risk probability of surpassing permissible limits in the Mezquital Valley, Hidalgo, were subject to prediction mapping.