Consequences of uncertainty in field observations of soil depth for digital soil mapping of soil functions

Many large soil profile databases are compilations of multiple surveys and include a variation in observation uncertainty. This uncertainty is often unknown and therefore ignored in studies. Question remains how the uncertainty can be derived and incorporated in the subsequent analysis? Over the past 15 years a database of over 1800 georeferenced soil profile observations has been compiled during the WU-SGL Field-training Geosciences course for a 340 km2 study area in Southern Spain. In this database of site and soil profile observations, effective soil depth was one of the quantitative variables observed by students. However, in what way can we actually assess best the uncertainty of these observations? Secondly, how can we produce the best prediction for soil depth in this study area, using the available explanatory variables and taking observation uncertainty into account? Soil depth in the study area depends a.o. on parent material, slope angle and landscape position. For both the aforementioned questions the knowledge on spatial heterogeneity of the explanatory variables between and within discrete mapping units is crucial. In this study we elaborated on observation uncertainty and recommend which digital soil mapping model yields the best prediction of soil depth in the study area, using the calibrating and validating potential of such a large database. Finally, the impact of soil depth uncertainty on the available water capacity is evaluated for different soil mapping units.