Improving digital soil mapping in Bogor, Indonesia using parent material information

Abstract

Digital soil mapping depends on the accurate representation of soil forming factors in the form of spatial layers called scorpan. Within the scorpan factors, digital elevation model (DEM) and its derived data are the most commonly used covariates, while parent materials and soil age received the least attention. This may be due to the coarse resolution or complex interpretation of parent material maps. This study examines the role of parent material derived from a semi-detailed soil map as a covariate for improving the accuracy of digital soil maps. The study was conducted in Bogor, West Java (2663.81 km2). Observations of 56 soil map units consisting of 19 soil subgroups of the USDA Soil Taxonomy were made on an existing 1:50,000 soil map. The study evaluated the contribution of covariates representing soil, organisms, relief and parent material in predicting soil classes using random forest models. The results show that the inclusion of parent material covariate can increase the total accuracy of predicting soil classes from 44.60 to 66.19% to 59.91–73.89%. The combination of 13 covariates representing topography, soil, organisms and parent materials achieved the highest accuracy above 70%. On the other hand, the most parsimonious set of covariates that can achieve 66.02% accuracy includes DEM, NDVI, NDWI, and parent material. The use of parent material covariate derived from detailed soil maps can improve the accuracy of digital soil mapping of soil classes in regions derived from complex volcanic materials.