LAUSR

Abstract A soil map (1:50,000 scale) was recently produced in Sardinia (Italy) using a cost-effective GIS approach. In this study we aimed to verify, in two pilot areas and by means of statistical analysis, the effectiveness of the adopted methodology in representing and predicting the spatial distribution of soil types and properties. We focused on evaluation of 1) the influence of landforms and parent materials on soil types (WRB Reference Soil Groups) and selected soil properties and 2) the suitability of the adopted methodology for calibrating a model to predict land unit composition in terms of different soil types. Leptosols, Regosols and Cambisols were prevalent on slopes, with Leptosols being more frequent on convex slopes and Regosols and Cambisols on concave slopes. In flat areas, soil types mainly depended on the type and age of parent material, with Regosols and Cambisols prevailing on Holocene deposits and highly developed soils (mainly Luvisols) largely prevailing on Pleistocene deposits. On hard rock, Leptosols were very frequent on terrigenous metamorphic rock and frequent on granite. Besides Leptosols, Regosols occurred more frequently than Cambisols on both parent materials. Landforms strongly influenced soil depth and available water capacity. Soils on plains were deeper than those on slopes, where convex forms had shallower soils than concave forms. A similar trend applied to the available water capacity. The parent material had a significant effect on topsoil properties (thickness, texture, pH and organic carbon content) of soils belonging to the same WRB Reference Soil Group (analysis done on the most relevant WRB Reference Soil Groups, i.e. Leptosols, Regosols and Cambisols). We calibrated and tested stepwise multiple linear regressions (MLR) and general linear models (GLM) to predict the composition of map units in terms of different WRB Reference Soil Groups. The two models gave very similar results, with distinct distribution patterns that were coherent with the relationships observed between soil groups and specific combination of terrain attributes and parent materials. Results showed that both models were more reliable in predicting the absence rather than presence of a given soil type.