Graduation date: 2008
Conventional soil maps represent a valuable source of information about soil
characteristics, however they are subjective, very expensive, and time-consuming to
prepare. Also, they do not include explicit information about the conceptual mental
model used in developing them nor information about their accuracy, in addition to the
error associated with them.
Decision tree analysis (DTA) was successfully used in retrieving the expert
knowledge embedded in old soil survey data. This knowledge was efficiently used in
developing predictive soil maps for the study areas in Benton and Malheur Counties,
Oregon and accessing their consistency. A retrieved soil-landscape model from a
reference area in Harney County was extrapolated to develop a preliminary soil map for
the neighboring unmapped part of Malheur County. The developed map had a low
prediction accuracy and only a few soil map units (SMUs) were predicted with
significant accuracy, mostly those shallow SMUs that have either a lithic contact with
the bedrock or developed on a duripan. On the other hand, the developed soil map based
on field data was predicted with very high accuracy (overall was about 97%).
Salt-affected areas of the Malheur County study area are indicated by their high
spectral reflectance and they are easily discriminated from the remote sensing data.
However, remote sensing data fails to distinguish between the different classes of soil
salinity. Using the DTA method, five classes of soil salinity were successfully predicted
with an overall accuracy of about 99%. Moreover, the calculated area of salt-affected
soil was overestimated when mapped using remote sensing data compared to that
predicted by using DTA. Hence, DTA could be a very helpful approach in developing
soil survey and soil salinity maps in more objective, effective, less-expensive and
quicker ways based on field data.