Conventional Soil Maps (CSM) like the USDA-NRCS Soil Survey Geographic (SSURGO) database have provided baseline soil information for land use planning since their inception in the early 20^th century. Although CSM have been widely used, modern demands for high resolution soil information at a field scale are not suitable applications for CSM in many cases. CSM lack a realistic representation of soils at that scale because they were created with polygonal vector mapping format that uses crisp map boundaries and often aggregates multiple soils types within one mapping unit. These spatial issues with SSURGO create added work for natural resource professionals trying to implement conservation planning strategies that utilize soil survey data. We present a repeatable method to disaggregate SSURGO data into a one arc-second (~30-meter pixels) rasterized soil class map that also provides continuous representation of probabilistic map uncertainty, and ability to utilize fuzzy membership of classes if soil intergrades are desirable for end users. Methods included training set creation for each original SSURGO component soil class from soil-landscape descriptive language within the original survey database. Training sets were then used to build a random forest predictive model that utilized 54 environmental covariate maps derived from ASTER satellite imagery and the one arc-second USGS National Elevation Dataset. Results showed agreement at 70% of independent field validation sites and equivalent accuracy between original SSURGO map units and the finer resolution disaggregated map. Uncertainty was mapped by empirically relating prediction frequencies of the underlying trees of the random forest model and the validation sites success rates.