Digital Soil Mapping (DSM) aims at the creation of reliable, reproducible and dynamic spatial soil information according to specific users' requests and demands. Positional and temporal inaccuracies as well as the question of an optimal resolution of digital elevation models (DEMs) indicate scale-related issues which represent typical challenges for DSM.

In this study, the effective map scale (EMS) approach is presented which enables the detection of operational scales where soil-related processes take place, the localization of corresponding process domains, as well as the statistical and spatial visualization of their scale-specific inaccuracies. The underlying algorithm can be considered as a test procedure for predictive efficiency where measurements, characterizing a soil-related process as well as a proxy variable and its scale-specific variation, are optimized. In doing so, positional and semantic inaccuracies of legacy data can be detected.

The EMS approach is applied to the example of an agricultural parcel where soil erosion by tillage is assumed. Auger samples have been taken in order to quantify the amount of soil loss and accumulation during the last 80 years in a German landscape with a complex topography and dominating loess parent material. The measurements have been related to the terrain attribute Mass Balance Index (MBI), which acts as an indicator for tillage erosion and has been varied according to both scale and soil surface complexity. The indicator MBI is derived from a high resolution digital elevation model and combines the basic terrain attributes slope, curvature and vertical distance to channel network due to their importance for tillage erosion processes. Different scale levels have been created by a region-growing segmentation algorithm. Each scale level contains discrete soil-terrain objects, represented by polygons.

The scale-related analysis of MBI variations and measurements has revealed a range of EMSs where process domains are visible. Their accuracies are characterized from various perspectives: (1.) The analysis of single MBI variants of scale and complexity by linear regression expresses the spatial and statistical variance of EMSs. (2.) The application of the data mining algorithm random forest on all the MBI variants of complexity per scale level leads to a spatial and statistical suppression of uncertain process domains and an emphasis of process domains, which could be predicted with a higher reliability. In this study, the process domains of accumulation could be identified on a range of operational scale levels. Due to the positional inaccuracies of auger samples and temporal inaccuracies based on overlaying long-term soil transport processes, the process domains of soil loss could not be sufficiently located.