Uncertainty quantification of soil property maps with statistical expert elicitation

Abstract

Accuracy assessment and uncertainty analyses are key to the quality of data and data analysis in a wide array of scientific disciplines. For soil science, it is important to quantify the accuracy of soil maps that are used in environmental and agro-ecological studies and decision making. Many soil maps, however, do not provide information about the associated accuracy. For these maps, the accuracy can be assessed by independent validation, but this requires additional fieldwork to collect validation data and does not provide a full probabilistic description of the map uncertainty such as required by uncertainty propagation analyses. Instead, in this study we aimed at quantifying the accuracy and spatial uncertainty in a soil property map using knowledge from multiple experts. The soil property map in question is a map of the volumetric soil water content at field capacity (SWFC) of the East Anglian Chalk area, United Kingdom. A formal statistical expert elicitation procedure was used to extract multiple experts' knowledge about the probabilistic model description of the error of the legacy soil map. A web-based tool for variogram elicitation was applied to elicit from experts the marginal probability distribution and spatial correlation of the error of this map. The elicited outcomes show that the spatial error of the SWFC over the study area is characterised by a multivariate normal random function. The variogram of the random function is the Matérn model with nugget=0.45%2, partial sill=54.6%2, range=25,400m, and kappa=0.40. The results show that experts, and the knowledge they provide, can be valuable in quantifying spatial uncertainty and hence can fill the gap of lacking information about the accuracy of soil property maps.