Mapping topsoil physical properties at European scale using the LUCAS database

- The LUCAS harmonised soil survey comprising 20,000 observations was used in this study.
- Soil texture and coarse fragments were mapped over the extent of Europe using MARS.
- MARS modelled soil texture with good accuracy whilst constraining their values.
- AWC, soil bulk density and USDA textural classes were derived from soil texture maps.
- These maps constitute a first approximation of the GlobalSoilMap products for Europe.

The Land Use and Cover Area frame Statistical survey (LUCAS) aimed at the collecting harmonised data about the state of land use/cover over the extent of European Union (EU). Among these 2 · 105 land use/cover observations selected for validation, a topsoil survey was conducted at about 10% of these sites. Topsoil sampling locations were selected as to be representative of European landscape using a Latin hypercube stratified random sampling, taking into account CORINE land cover 2000, the Shuttle Radar Topography Mission (SRTM) DEM and its derived slope, aspect and curvature.In this study we will discuss how the LUCAS topsoil database can be used to map soil properties at continental scale over the geographical extent of Europe. Several soil properties were predicted using hybrid approaches like regression kriging. In this paper we describe the prediction of topsoil texture and related derived physical properties. Regression models were fitted using, along other variables, remotely sensed data coming from the MODIS sensor. The high temporal resolution of MODIS allowed detecting changes in the vegetative response due to soil properties, which can then be used to map soil features distribution. We will also discuss the prediction of intrinsically collinear variables like soil texture which required the use of models capable of dealing with multivariate constrained dependent variables like Multivariate Adaptive Regression Splines (MARS).Cross validation of the fitted models proved that the LUCAS dataset constitutes a good sample for mapping purposes leading to cross-validation R2 between 0.47 and 0.50 for soil texture and normalized errors between 4 and 10%.