Knowledge-based soil type classification using terrain segmentation

Abstract

Soil information covering regional, continental, or even global scales is needed for modelling, prediction, or estimation of environmental risks, crop yield estimation, carbon stock estimation, or research on climate change. This study aims to evaluate the extent to which geographic object-based image analysis and expert-knowledge, using digital maps of climate, topography, vegetation, and geology as soil covariates (GEOBIA approach), might model and reproduce a conventional soil map at a scale 1:1000000 in the south-west of Romania. The environmental variables were segmented with a region-growing algorithm, the resulting objects being subsequently classified into soil types using expert-knowledge fuzzy classification rules. To assess the geographical support of classification for the modelling of a conventional soil map, we quantitatively evaluated a pixel-based soil map produced using the same expert-knowledge classification rules, as an alternative to an object-based approach. To evaluate the source of soil information, we quantitatively assessed the map of the World Reference Base soil groups produced by the data-driven global soil information system, SoilGrids, as an alternative to expert-knowledge rules. The digital soil maps were quantitatively compared with the conventional soil map. Evidence was provided that the similarity of soil types with the conventional soil map was higher when the modelling was conducted through GEOBIA approach (general similarity of 65% and fuzzy kappa index of 0.58) than the pixel-based approach and SoilGrids. Furthermore, the results showed that the SoilGrids map achieved higher similarity to conventional soil map than the pixel-based soil map. When tested in another area, without modification to the knowledge-based methodologies, the same conclusions could be drawn, although the two maps recorded lower similarity values. The overall reduction in similarity values is explained by a high variability of some soil types under different environmental conditions.