Digital soil mapping of arable land in Sweden – Validation of performance at multiple scales

In this study, we produced a detailed digital soil map of topsoil texture and soil organic matter (SOM) content for 2.4 million ha of arable land in Sweden (DSMS). Three spatially exhaustive datasets (a laser-scanned digital elevation model, airborne gamma radiation scanning data and a legacy Quaternary deposit map) were calibrated against topsoil texture and SOM content in around 13,500 soil samples, using multivariate adaptive regression splines (MARSplines) modelling. We then deployed the MARSplines models to produce raster maps (50 m x 50 m) of clay, sand and SOM content. The modelling procedure was validated by an independent dataset of about 24,000 samples clustered on 544 farms (with a local sample density of one per 3 ha). The error in clay content was < 8% in 75% of the validation samples, while for sand content and SOM content it was 13% and 2%, respectively. Corresponding values for the farm-average level were 6%, 11% and 2%, respectively. Modelling efficiency values revealed that the clay content map was a considerable improvement over the mean of the reference values at national level, regional level and, in many cases, also farm level. However, SOM content predictions showed little or no improvement over the mean of the reference samples (at any scale), due to poor correlation with the exhaustive predictor variables at all three scales investigated. The DSMS soil geodatabase will continue to be improved and have more layers added, e.g. derived layers calculated from the primary clay, sand and SOM layers by use of pedotransfer functions. Practical use of DSMS is exemplified here in an internet application for deriving prescription files for precision agriculture.