Legacy soil maps as a covariate in digital soil mapping: A case study from Northern Iran

Abstract

Digital soil mapping (DSM) can be used for updating soil surveys. Legacy soil survey maps are often used as a covariate for updating soil surveys because such soil survey maps are logically assumed to contain significant information about the spatial distribution of soil classes. In the present study the usefulness of including conventional soil survey maps as a DSM covariate was investigated. Random forest and multinomial logistic regression models were built using two different covariate sets: covariate set 1 included the legacy soil survey, covariate set 2 excluded the soil survey. Soil Great Groups, Subgroups, and Series taxonomic classes were modeled using both models and covariate sets for an area of ~85,000ha in Golestan Province, northern Iran. Overall model accuracy, the Kappa statistic, and individual covariate importances were used to assess the influence of including the legacy soil survey.Including the conventional soil map as covariate generally increased model accuracy, but the improvement in model accuracy was surprisingly small at all taxonomic levels. This may be due to soil change or the mapping scale of the legacy soil survey. Random forests was found to be more accurate than multinomial logistic regression at all taxonomic levels. Multinomial logistic regression models at the soil Series level were less accurate than the legacy soil survey.