Geoscience Reference
In-Depth Information
distance between cart passes about 10 m, resulting in about 6900 EC measurement points for the
studied area. Two sets of EC measurements were collected corresponding to depths of approxi-
mately 0 to 30 cm (shallow EC) and 0 to 90 cm (deep EC) (Sudduth et al., 1998).
Survey grid GPS (Leica 500 RTK) was used to measure elevation. About 1500 elevation mea-
surements were taken on a semiregular grid with a mean distance between measurements of approx-
imately 10 m. The map of field elevation ranges is shown in Figure 17.1. ArcView Spatial Analyst
(Environmental Systems Research Institute, 1996) was used to analyze elevation data and to derive
the topographical land features—namely, slope, curvature, and flow accumulation. ArcView was
also used to calculate the shortest distance from each of the sampling sites to the drainageway that
ran in the middle of the study area (Figure 17.1). This distance was used as one of the variables with
potential effect on soil drainage class.
17.2.2 d i s c R i M i in a in t a n a l y s i s a n d g e o s t a t i s t i c a l P R o c e d u R e s
In this study, discriminant analysis was used to find a decision rule for separating sites with differ-
ent drainage classes based on the values of the topographical and EC variables measured at each
site. The decision rule was first defined based on the measured data, and then the decision rule was
applied to predict drainage classes at unsampled sites where measurements of topographical and
electrical conductivity variables were available.
Two combinations of the variables were considered. The first combination included topographi-
cal variables, such as elevation, slope, curvature, flow accumulation, and distance to a drainageway.
The purpose of this combination was to determine how accurate drainage class prediction can be
if it is based on topographical data only. The second combination of the variables included all the
topographical and electrical conductivity variables to study the possibility of using soil electrical
conductivity data with topographical information for predicting soil drainage classes.
Stepwise discriminant procedure, STEPDISC (SAS Institute Inc., 1999) was applied to both
combinations of variables to select the variables that had significant influence on the soil drainage
class. Only variables significant at 0.15 significance level (Bell et al., 1992) were used in further dis-
criminant analysis. Discriminant analysis was conducted using DISCRIM procedure (SAS Institute
Inc., 1999).
Cross-validation was applied to evaluate accuracy of drainage data prediction and classification
by discriminant analysis. For cross-validation, each value from the data set was eliminated in turn
and, then, estimated using information from the rest of the data (Khattree and Naik, 2000). Poste-
rior probabilities of the three drainage classes obtained for each data point were compared, and the
site was assigned a drainage class with the highest posterior probability. Percent of correct drainage
class estimates was used to compare different combinations of variables and their effectiveness in
predicting drainage classes.
For geostatistical analysis, soil drainage class was treated as a categorical variable. At each data
location this variable assumed only one of the three mutually exclusive possible states, correspond-
ing to either WD, MWD, or SWPD/PD drainage classes. Indicator transformation was applied to
each drainage class data resulting in three indicator variables. Kriging procedures were used to
obtain indicator variable estimates at unsampled locations. The kriging estimate was equivalent
to the probability of finding a certain drainage class at this location and assumed a value between
0 and 1. For each drainage class, kriging produced a map of probabilities of finding this drainage
class at each particular location, and as in the discriminant analysis, the location was assigned to a
drainage class with the highest probability.
Two geostatistical procedures compared in the study included ordinary indicator kriging and
soft indicator cokriging (Goovaerts, 1997). Indicator kriging uses the drainage data only; cokriging
allows for combining primary drainage data with any available secondary data related to drainage.
One of the disadvantages of cokriging is that it becomes extremely cumbersome and time consum-
ing with a large number of secondary variables. In this study, we conducted cokriging with one
Search WWH ::




Custom Search