You are here:
Predictive Soil Modeling in Soil SurveyEPA Grant Number: U915632
Title: Predictive Soil Modeling in Soil Survey
Investigators: Scull, Peter R.
Institution: San Diego State University
EPA Project Officer: Boddie, Georgette
Project Period: August 1, 1999 through August 1, 2002
Project Amount: $76,117
RFA: STAR Graduate Fellowships (1999) RFA Text | Recipients Lists
Research Category: Fellowship - Geography Earth Sciences , Ecological Indicators/Assessment/Restoration , Academic Fellowships
The objective of this research project is to integrate Geographic Information Systems and Remote Sensing technology into standard soil surveys in order to make large-scale soil mapping more cost-effective, and to produce more robust data products.
Decision-tree analysis (DTA) is a model framework that can be used to create predictive models of soil occurrence. DTA was selected because of its capability to integratingintegrate a wide range of data sets (remote sensing and DEM products, as well as a variety of ancillary data), and because it makes intuitive sense, allowing easy communication with soil experts. DTA involves successively partitioning the dependant variable into increasingly homogeneous subsets. Splits, or rules defining how to partition the data, are selected based on information statistics that define how well the split decreases impurity within the data set. Once the tree has been constructed (or grown), it has encoded a set of decision rules that describe the data -partitioning process. These rules can be applied to a geographic database to predict the value of a response variable in an area where the predictor variables are known, but the response variable is not. Binary decision -tree models have been developed in areas previously mapped, and will be applied in similar areas that have not yet been mapped. This method can be used to provide the soil mapper with a set of maps characterizing the probability of mapping unit occurrence in an unmapped area.