Quantification and simulation of errors in categorical data for uncertainty analysis of soil acidification modelling

Abstract

Simulation studies that use maps to generate georeferenced model input may be prone to errors in the definition and delineation of the map units. Our study aims at the estimation of errors in categorical data, i.e., a generalized soil and vegetation class map of the EU vs. a highly detailed soil/vegetation map of the Netherlands. From this, an error model evolves containing (i) an index of map purity and misclassified area fractions and (ii) indicator variograms describing the spatial autocorrelation structure of the degree of error at individual locations. Furthermore, we describe a method to evaluate the effect of these errors on the uncertainty of the outcome of the soil acidification model Simulation Model for Acidification's Regional Trends, version 2 (SMART2). This method involves the application of joint sequential simulation to produce equiprobable realisations of soil/vegetation maps. Results show that the errors in the EU-soil/vegetation map are considerable, because 69% of the area is misclassified when compared to highly detailed maps from the Netherlands. Simulated maps reproduced the error model for the dominant soil/vegetation map units well. Results of the uncertainty analyses show that errors in categorical data do have a pronounced influence on the uncertainty of SMART2 results. This influence was between 20% of the total variance for Al 3+concentrations and exceedance probabilities, and 40%–50% of the total variance for NO 3 − concentrations and exceedance probabilities.