Poster Number 557
Predictive Soil Mapping (PSM) has recently become an attractive method for soil scientists wishing to develop a more objective and efficient approach to mapping soils. Due to the potential PSM has for reducing the effort to produce soil maps, as well as its ability to improve the classification accuracy of soils, this method has peaked the interest of agriculturalists and land managers desiring quantitative information about soils and the landscape. In this study, PSM is applied in the Fremont National Forest (NF) of south-central Oregon, where an updated soils map and National Hierarchy Landtype Association (LTA) map are needed.
Decision-tree analysis (DTA) is a PSM technique and was used to derive the landscape model to produce both maps needed for the study area. DTA aided in the process of identifying errors in the original Soil Resource Inventory (SRI) map of the Fremont NF and in developing a corrected predictive soils map of the forest. Once a predicted soils map of the forest was made, an LTA map using the newly predicted soil boundaries and landforms was produced and designed for display at a scale of 1:100,000.
Both the accuracy of DTA prediction rulesets and a discrete multivariate technique, the kappa analysis, were employed to assess the accuracy of the predictive soils maps of the Fremont NF. Maps produced with training data from the original SRI of the forest yielded results between the lower sixtieth to mid eightieth percentiles, and maps produced with corrected training data yielded results in the lower ninetieth to upper ninetieth percentiles. Kappa for predictive maps using modified training data showed strong agreement between ground-truth maps and predicted maps.