An error budget for different sources of error in digital soil mapping

Abstract

Digital soil mapping is gathering momentum across the globe, driven by a need for soil information and made possible by the increasingly widespread availability of spatial data that can be used to represent Jenny's (1941) soil forming factors. Much of the focus is on the predicted values of soil properties but it is equally important to quantify associated prediction errors.Previous studies have considered individual sources of error in a digital soil map but none have considered the combined effect of all sources. In this study, we develop an error budget procedure to quantify the relative contributions that positional, analytical, covariate and model error make to the prediction error of a digital soil map of clay content. We consider four scenarios corresponding to typical levels of data quality: (i) good, (ii) legacy, (iii) spectroscopic and (iv) poor quality data.The error budget procedure uses both geostatistical and Monte‐Carlo simulation to produce numerous realizations of the data and their underlying errors. Linear mixed models are used to construct the digital soil map. In this implementation we consider the error associated with the measurement of the clay content; the location of the survey sites; the interpolation of the covariate (ECa); and the estimation of the fixed effects, random effects and interpolation from the linear mixed model.For all data quality scenarios, the error from the model dominated the prediction error (mean square error of 67.24–72.41% Clay2). Where the analytical error was small, the error attributed to the covariate was greater than the analytical error. This relationship was reversed in the data quality scenarios where the analytical error was large. Under all data quality scenarios, the effect of positional error was negligible.