Tracy L. Emery, Timothy L. Haithcoat, and Ronald D. Drobney
CREATING AND EVALUATING A FRAMEWORK FOR LAND COVER
GENERALIZATION IN MISSOURI
National Gap Analysis protocol calls for land cover at a
scale of 1/100000 with a minimum mapping unit of 100 hectares. This
scale is not appropriate for the types of detailed studies occurring
at the sub-state level for which the databases will be utilized. As a
result, it was determined that Missouri land cover would be created
at a larger scale than that required for Gap Analysis reporting, and
that this land cover would be generalized for GAP reporting and use
at the national level.
The challenge this decision poses is the development of a methodology
for the generalization of land cover that maintains a level of detail
appropriate for the modeling of vertebrate species habitat
relationships (modeling that may occur at the national level).
Unfortunately, the problems of generalizing and maintaining detail
are not adequately addressed by existing methodologies. Current
algorithms rely on the combination and simplification of polygons and
the elimination of attribute information for those polygons
assimilated. This results in a database that has been spatially
generalized, and quality compromised. This research purposes to
evaluate and conduct spatial generalization with the preservation of
the attribute detail. Generalization will be approached as a method
for the compression of spatial data, the spatial presentation of the
data will be simplified, but the complexity of the attribute
information will not change.
Though the stated outcome of this land cover generalization process
is somewhat different than traditional generalization outcomes, the
previously established principles of generalization still apply.
These principles will provide the basis for this generalization
methodology.
There are three distinct steps to developing a generalization
methodology. The first step is the development of a conceptual
framework. This framework gives structure to the process, and assures
that the generalization procedure will be approached in a systematic
manner with some known goal as the endpoint. In the second step, the
actual generalization procedure will be developed. The procedure is
the specific approach that will be taken, based on the specific goal
or outcome, not the specific generalization rules. With a conceptual
framework and an approach to the problem in place, the final step is
pooling and then culling expert knowledge into a coded set of rules
that will drive the actual generalization.
The development of a conceptual framework
The Brassel and Wiebel model (1988) is a five step approach
to automated map generalization. The first step is structure
recognition. The objects in the dataset that need to be generalized
are identified. Their spatial relationships and relative importance
are also studied. The next step is process delineation. Before
generalization can occur the specific methodology that will yield the
desired output must be outlined. Questions necessary at this stage
include: Will the data be transformed? How will this occur? What
types of conflicts may occur? And what information will be part of
the final dataset? With these decisions made we enter the third step:
process modeling. Here, rules and procedures for the generalization
are compiled. These rules and procedures will be applied to the
methodology developed in the process delineation step. At this point
the information that needs to be generalized has been identified, a
methodology has been developed that yields a desired output, and the
rules necessary to implement the methodology are in place.
Generalization, the fourth step, now occurs. The final step is to
display the outcome of the process. It is important to note that the
final data display had been determined within step two, the process
delineation stage.
The development of generalization procedures
The specific procedures used in the application of
generalization to the GAP datasets have their roots in geographic
information abstraction, because the intent is to preserve the
highest possible level of detail during the generalization process.
Geographic information abstractions have been defined as
'"'constructs that use both definitions and geographical
relationships in order to develop the conceptual content of
databases'"' (Nyerges, 1991). Abstraction focuses on the suppression
of detail, instead of the elimination of detail. This makes
abstraction a more robust method of managing data. However, most work
in this field has been done in the realm of cartographic
generalization, both manual and digital. Though specific answers to
the problems of effective generalization cannot be gleaned from these
sources, they do provide a valuable foundation for the discussion of
generalization.
The pooling and culling of expert knowledge
Digital generalization requires that the artistic side of
generalization be turned into a science with clear definitions and
known implications of decisions. The difficulty lies in gathering the
information and making the rules that will create a replicable and
defensible process that can then be applied within real world
applications with a high level of confidence. There are three factors
to overcome in developing rules that will guide generalization.
First, the artistic nature of manual generalization does not lend
itself to rule formation. Because of the subjective and holistic
nature of manual generalization, the manual process never developed a
set of concrete rules (or even a step by step procedure) that could
be effectively replicated. This makes automating generalization very
difficult, as computer processes tend to rely on binary (either
yes/no, or true/false) questions. The first step towards digital
generalization, therefore, is the development of a systematic
approach to the creation of decision rules for the generalization
algorithm. The approach is a general process, whereas the decision
rules are specific to the problem being solved. This subjective
process has not, and perhaps can not, be dissolved into a set of
rules. Secondly, rules are specific to the purpose for which they are
written. It may be impossible to consistently apply a rule, even
within a database because of changes made as the data is generalized.
A corollary difficulty is knowing when a rule no longer applies.
Finally, very slight differences in the relationships among objects
may call for a different rule. The accuracy of the rule writing and
the proper application of the rules will determine the overall
effectiveness of the process and applicability of the resultant
map.
Levin (1992) stated that '"'the objective of the model should be to
ask how much detail can be ignored without producing results that
contradict specific sets of observations, on particular scales of
interest.'"' For this project, the focus will be on generalizing
(modeling) the land cover. Expert knowledge applies not only to the
processes that will occur in the generalization, but also to the
expert knowledge concerning the data to be generalized. Central to
the generalization of the land cover will be the development of
relationships between land cover types. This is based on many
factors, including the morphology, physiology, and function of the
plants as well as their interrelationships with wildlife habitat.
These decisions will effect not only the generalization, but the
results of any species/habitat modeling done at the national level
with these data. Regardless of how good the generalization
methodology is, if there are flaws in the data or models the results
of the entire project will be suspect.
References
Brassel, Kurt E. and R. Weibel. 1988. A Review and
Conceptual Framework for Automated Map Generalization.
International Journal of Geographic Information Systems,
2(3):229-244.
Levin, Simon A. 1992. The Problem of Pattern and Scale in Ecology.
Ecology, 73(6):1943-1967.
Nyerges, Timothy L. 1991. Representing Geographical Meaning. in
Buttenfield, B. P. and McMaster, R. B. (eds) Map Generalization:
Making Rules for Knowledge Presentation, Longman, London
59-85.
Author Information
Tracy L. Emery
Graduate Research Assistant
Department of Geography / Geographic Resources Center
University of Missouri
20 Stewart Hall
Columbia Missouri, 65211
phone: 573-882-1404
fax: 573-884-4239
c548975@showme.missouri.edu
Timothy L. Haithcoat
Senior Research Specialist
Department of Geography / Geographic Resources Center
University of Missouri
16 Stewart Hall
Columbia Missouri, 65211
phone: 573-882-1404
fax: 573-884-4239
grctlh@showme.missouri.edu
Ronald D. Drobney
Cooperative Fish and Wildlife Research Unit
University of Missouri
112 Stephens Hall
Columbia Missouri, 65211
phone: 573-882-3436