Douglas R. Morgenthaler and Timothy L. Haithcoat
Abstract
The objective of this research is the design of a
semi-automated classification mechanism for landforms based on
user-defined parameters set in a qualitative and quantitative manner
through a simplified user interface. The basis for the calculations
of land form will be a digital elevation model. Validity of the
modeling techniques will be assessed through comparison against
landforms delineated on 7.5 minute quadrangles by experts.
This project has created a potentially powerful tool that, coupled
with other spatial datasets, can more 'accurately' categorize and
classify landforms, serving as a tool for landslide evaluation, soil
movement, or watershed analysis. These user-defined parameters will
be placed within an ArcTool, providing the user with the flexibility
to control automation of the classification process.
Statement of Problem and Objectives
Landform accuracy is difficult to define across a continuous
surface. It is only when it is artificially categorized into a set of
features that an assessment can be made. The result is a set of
user-controlled breaks within and on a continuous surface. Attempts
to successfully classify landforms and their characteristics are
often subjective, largely based upon the interpreter's background,
experience, and current application. In recent years, due to the
staggering increase in elevation data types and surface modeling
tools, the need for raster-based landform classification models have
become more apparent. Most research concerning landform
classification has taken place at the conceptual level (Dikau 1990,
Evans 1980, Zevenbergen & Thorne 1987). While efforts toward more
standardized classifications are underway (Schmid-McGibbon 1993),
classification discrepancies still exist, illustrating the need for a
more'comprehensive' landform modeling protocol.
The objective of this research is the design of a semi-automated
classification mechanism for landforms based on user-defined
parameters set in a qualitative and quantitative manner through a
simplified user interface. Successful creation of landform breaks
will include the quantification and determination of the
effectiveness of slope, aspect, and profile and planar curvature to
classify geomorphologic units. A second objective of this research is
to examine expert landform delineations by application class to
determine similarity of landform process by application group. Groups
from which participation is sought include: hydrologists, geologists,
ecologists, and soil scientists.
Methods and Procedures
The initial step in model development will be the
determination of digital source data to be utilized as the basis for
model development. We will be using parameters evaluated on 1:24,000
scale datasets in order to determine the optimal output for the
classification model. It is the goal of this research to integrate
classification schemes advanced into the landform model, and through
further analysis expand upon them. The developed model will draw on
concepts set forth by several experts within geomorphologic modeling.
Dalrymple et al. (1968) developed the idea of using a nine-element
matrix incorporating all possible slope and curvature combinations to
describe landform. This matrix forms the basis for initial parameter
delineation to be implemented. The interface will be written in
Esri's Arc Macro Language (AML) to successfully migrate the developed
models into an ArcTool. The parameters to be evaluated and
implemented in the landform model include: slope, aspect, planform
curvature, and profile curvature. Quantitative values for each slope
classifier will initially be determined using suggested ranges of
degree slope provided by Dalrymple et al (1968). These quantitative
ranges, describing qualitative parameters, will be incorporated
within a flexible user-based matrix for user control of the
classification process.
Analysis, adjustment, and categorization of the model will be based
upon validation performed in several phases. First, a map of the
Lewis Hollow quadrangle in south central Missouri will be provided to
a number of experts in landform analysis, including the application
areas of ecology, geology, hydrology, and soil science. Each expert
will be asked to delineate landform features given certain mapping
criteria for a selected 9 square mile area. The study area was
selected because of terrain complexity and availability of digital
data. Some experts from each application area will be identified who
reside outside of the state and will be asked to participate in order
to minimize any local influences. A modified classification scheme
after Dalrymple et al. (1968) is to be implemented, comprised of six
classifiers of slope: crest, shoulder, fall face, midslope,
footslope, and toeslope (Figure 1). The
minimum area to be mapped is 10 acres, addressing both maintainance
of relief and limitations of the digital datasets to be used. To
reduce possible error introduced in synthesizing slope and curvature
information, delineation of these attributes will be performed
separately, resulting in separate maps illustrating profile and
planform curvature. These maps will be entered into a spatial
database and coregistered for further analysis.
Another initial phase will be implemented in order to quantify the
selected parameters to be used in classifying landform. This will
lead to the creation of a modified ArcInfo classification model.
This will be accomplished in part by utilizing ArcInfo's CURVATURE
function. This, in combination with slope and aspect, will provide
the basis for the initial classification output
(Figure 2). As a result of the user's
inability to adjust the parameter breaks within CURVATURE, several
modified classifications will be derived from the original surface
grid (Figure 2). A number of landform
modifiers will be performed on the original surface, including the
use of high and low pass filters, the inclusion/exclusion of ridges
and stream network information, using alternate methods for obtaining
slope and aspect measurements, and hydrologic iteration adjustments
in TOPOGRID, creating a series of potential surfaces on which to
obtain measures. These modified classifications will provide the
alternatives for categorizing a particular application group to
attain enhanced classification correlation in successive phases of
the research (Figure 3).
For each iteration within this phase of the classification process, a
summary classification from within each application group will be
developed. Expert class delineations from similar application areas
will be examined to determine areas of fuzziness and degree of
agreement within the application area. If a significant correlation
is made between experts within a application group, a summary
classification will be used to assess the appropriateness of the
model against the series of modified input surfaces. If there is poor
correlation between the experts, each will be assessed individually
against the classification model. If these conditions are met, each
groups' collective classification will be analyzed in relation to the
classification resulting from the ArcInfo process.
A walkthrough of the process to be implemented follows. In the first
iteration (Figure 3a) each of the aggregated
application maps delineated by experts will be compared to the
original ArcInfo classification. If there proves to be sufficient
correlation between these classifications, no changes will be made in
the ArcInfo model. However, if significant correlation does not
exist between the classifications, the application group
classification will be tested against each of the modified ArcInfo
model classifications to determine which modifiers or combination of
modifiers produce acceptable agreement. Successive iterations
(Figure 3b) will first compare the summary
classification of each group against previous groups' classification.
If there is significant correlation between the two groups, each
group will utilize the same model classification. If correlation does
not exist, the summary classification will be run against each of the
modified ArcInfo models to determine its proper set of
modifiers.
Finally an expert classification will be aggregated, resulting in a
summary classification across application area and discipline
(Figure 4). This summary classification will
be compared against the original ArcInfo model to determine its
applicability as a default classification model. If sufficient
agreement cannot be reached, the summary classification is compared
against each of the modified ArcInfo models to determine the
greatest degree of agreement.
A comparison between classification methods will be made in order to
determine the error distribution in space of the model in classifying
landform features. Statistical measures will be made to determine
significance and correlation between each of the delineated maps.
Based upon the validation results, further adjustments in the model
will be made to reflect greater validity for specific
application-based landform modeling if possible.
Bibliography
Dalrymple, J., Long, R., and Conacher,A. (1968) A
hypothetical nine-unit land- surface model. In Zeitschrift fur
Geomorphologie 12: 60-76.
Dikau, Richard. (1990) Digital relief models in landform analysis. In
GIS: Three Dimensional Applications in Geographic Information
Systems ed. J. Raper, 51-77.
Dikau, Richard. (1990) Geomorphic landform modeling based on
hierarchy theory. In4th International Symposium on Spatial Data
Handling, July 23-27, Zurich, 230-239.
Evans, Ian. (1980) An integrated system of terrain analysis and slope
mapping. In Zeitschrift fur Geomorphologie 36:
274-295.
Jenson, Susan, and Domingue, J. (1988) Extracting topographic
structure from digital elevation data for geographic information
system analysis. In Photogrammetric Engineering and Remote
Sensing 54: 1593-1600.
Schmid-McGibbon, Geshe. (1993) Landform mapping, analysis, and
classification using digital terrain models. Unpublished PhD thesis.
University of Alberta.
Zeverbergen, Lyle, and Thorne, C. (1987) Quantitative analysis of
land surface topography. In Earth Surface Processes and
Landforms 12: 47-56.
Douglas Morgenthaler
Head Research Assistant
20 Stewart Hall
University of Missouri-Columbia
Phone: (573) 882-1404
Fax: (573)884=4239
Email:
c572853@showme.missouri.edu
Timothy Haithcoat
Sr. Research Specialist
18 Stewart Hall
University of Missouri-Columbia
Phone: (573) 882-2324
Fax: (573)884=4239
Email:
grctlh@showme.missouri.edu
