Assessing spatial uncertainty in mapping soil erodibility factor using geostatistical stochastic simulation

Soil erosion is one of most widespread processof degradation. The erodibility of a soil is a measure of itssusceptibility to erosion and depends on many soil properties.Soil erodibility factor varies greatly over space andis commonly estimated using the revised universal soil lossequation. Neglecting information about estimation uncertaintymay lead to improper decision-making. One geostatisticalapproach to spatial analysis is sequentialGaussian simulation, which draws alternative, equallyprobable, joint realizations of a regionalised variable.Differences between the realizations provide a measure ofspatial uncertainty and allow us to carry out an erroranalysis. The objective of this paper was to assess themodel output error of soil erodibility resulting from theuncertainties in the input attributes (texture and organicmatter). The study area covers about 30 km2 (Calabria,southern Italy). Topsoil samples were collected at 175locations within the study area in 2006 and the mainchemical and physical soil properties were determined. Assoil textural size fractions are compositional data, theadditive-logratio (alr) transformation was used to removethe non-negativity and constant-sum constraints on compositionalvariables. A Monte Carlo analysis wasperformed, which consisted of drawing a large number(500) of identically distributed input attributes from themultivariable joint probability distribution function. Weincorporated spatial cross-correlation information throughjoint sequential Gaussian simulation, because model inputswere spatially correlated. The erodibility model was thenestimated for each set of the 500 joint realisations of theinput variables and the ensemble of the model outputs wasused to infer the erodibility probability distribution function.This approach has also allowed for delineating theareas characterised by greater uncertainty and then tosuggest efficient supplementary sampling strategies forfurther improving the precision of K value predictions.