|
THE STATE OF THE ART OF LAND USE MODELING
I. Current State of the Art in Land Transformation Modeling
- Models are either stochastic (logit, markov, cellular automata, etc.) or
processed based (dynamic ecosystem model).
- Models are developed independently by various teams with essentially no
reuse or interoperability components.
- Models are linked to a heterogeneous group of external applications,
such as GIS, statistical packages, and visualization packages.
- A few models exist which utilize distributed computing, but are
application or model specific.
II. Projected State of the Art in Land Transformation Modeling
- Models are linked stochastic, process based, and other (fuzzy expert
system, individual based, etc.).
- Integrated Problem Solving Environment (PSE) will be widely used.
- The PSE will incorporate a uniform interface to external applications.
- Transparent distributed computing will be widely available.
- Data interoperability is widely supported.
- Uniform GUI - Learning environment and modeling interface.
COMPARATIVE MODELING
Land use transition models, while reflecting a varied heritage and
disciplinary background, share many commonalities. Common approaches are
the use of transition probabilities in a class transition matrix, the use
of multinominal logit methods, cellular modeling, and the use of the GIS
weighted overlay approach. Emerging models that show potential are object
oriented, expert system assisted, and use cells as their spatial unit
rather than parcels.
Model choice issues related to source data include the choice of data
model (features, cells or objects), and the comparability of
multitemporal land
use maps. Also important is the spatial aggregate structure of the
modeling unit, be it watershed, county, metropolitan area, etc. Of almost
unanimous concern in land transition modeling are the stability, estimation
of, and autocorrelation in the transition probabilities.
Research on comparisons between models, such as that conducted on global
climate models, is important to evaluate model performance, and to assist
in validation.
DATA ISSUES AND NEEDS
Land Use and Land cover information is essential for the calibration,
baseline, and testing of land use models. The information needs to be
continuously updated, at spatial and temporal resolutions that are
compatible with the models and have appropriate thematic content and
detail.
Issues:
Are discrete categories appropriate? Spatial boundaries are
occasionally
more appropriately represented as fuzzy gradations. However, land use
boundaries are typically fairly crisp because they are based on ownership.
However, probabilistic models might benefit by maintaining probabilities as
more continuous representations of land cover/use. Further, because aerial
and satellite image processing provides much of the data, probabilistic or
fuzzy approaches might be used. More continuous data might also be used
and represent land use intensity, and changes in intensity.
Time and space scales are co-dependent and critical. Higher spatial
resolution data are typically constrained to less frequent updates. For
example, land use/cover information from Landsat TM might be updated over a
large area every 5 years or decade (e.g. MRLC) whereas data at a resolution
of AVHRR might be updatable monthly. Generally, detail as well as quality
decreases with age of data.
Remote sensing provides raster data. Is that appropriate or sufficient?
What land use modeling approaches use grids vs. networks vs. objects as
data models?
These issues may be spatial and process-scale dependent. For example, very
detailed models often work with individual entities (plants, people),
whereas more aggregate models work better with a grid framework.
Use vs. Cover. Some of the models predict land use change based on
economics, some predict land cover based on biophysical processes. Yet,
the difference between use and cover are often not distinct for the
purposes of data collection. Anderson's classification attempts to combine
use and cover. Satellite remotely sensed data provide land cover
information well, but land use information requires more spatial detail and
interpretations.
Needs:
Rethink Anderson - Do we need a new classification system? In what way?
Can Anderson be improved? Do land use and land cover need separate
systems? Can they be amalgamated?
Landsat is critical - provide support for program.
There are more data choices coming online with new private and public
sector satellites. Modelers should be aware and flexible to make use of
the most appropriate resolution data.
Monitoring and Rapid Assessment Protocols. We need to develop methods to
quickly monitor and observe changes on large scales.
Is MRLC model appropriate?
APPLICATIONS
Applications are problem specific, tied to a geographic hierarchy. The
spectrum of analysis ranges from global to local issues and areas of
interests. Land use analysis models need to operate at the different
geographic levels. The resolution of the data should be transparent to the
model.
A land use model is applied to generate the results of a proposed
scenario.
A scenario usually has a focus which has either an urban or ecological
driver(s). A focus or driver is comprised of factors, parameters or
criteria that the user of the model can alter, prioritize, or weight to
serve alternative scenarios. The land use model should allow the user to
vary the time frame for each scenario being generated. The results of the
scenario or application can be used as output to launch another model, to
calibrate the land use model, and to assist in the decision making process.
POLICY/SOCIAL/ECONOMIC FACTORS
This is a very complex subject area so no single consensus can be
reached
in a short session. A list of topics/concerns emerged. This is a summary
of some of the elements under ten of these.
- Policy, social and economic (human dimension factors) are indispensible
to models of land use change and, therefore, land cover.
- The indicators that are used to record or measure the factors to be
entered influence very much how many converging (confounded) factors are or
are not included.
- The united variables that are selected often shapes (or reshapes) the
way the system is understood.
- The assumptions about human behavior, human choice, and system
stability and predictability determine the degree to which the model is
simple and manageable but removed from reality or complex and unwieldy but
approximating reality. The assumptions must be stated clearly and
challenged regularly.
- The incorporation of human dimension factors means necessarily that
other disciplines are involved. They bring both different assumptions
(different indicators) and different methodologies. These complicate
interaction but open doors for enhancing model performance (for example
should multinominal logit be used where probit analysis can be used?)
- In the system we wish to model, experimentation is not possible and
much of the bases of reductionist science is challenged.
- Scales of space and time: The processes of human activities operate at
very different time scales from ecological processes with which they are
linked. The local spatial scale of landscape units are affected by social,
economic and policy factors at smaller scales (larger area).
- Surprise in the system is inevitable and all systems are ultimately
idiosyncratic. This raises the question of the value of models that
purport to be general. At one scale, individual decisions are determined
by very personal circumstances and quality of life considerations that are
hard to measure. At another scale, things like earthquakes, fires, war,
economics cycles, are unpredictable but leave indelible marks in land use
processes.
- Therefore, for all of these reasons it seems as though it might be
futile to attempt to model. But such models are necessary because they (1)
consolidate data, (2) make assumptions explicit, (3) permit testing of
hypotheses, and (4) allow a basis for prediction about change.
- But to the extent that the resultant information becomes a part of the
decision making process, the model is a part of the process it is modeling
and so the role of modeling as a means of shaping the future (rather than
just predicting it) needs to be considered.
RESEARCH NEEDS
What are the big science, blue sky questions (requiring the expenditure
of tens of millions of dollars) regarding land use and land cover
modeling that can be answered through an organized research effort?
- Are there a common (and limited) set of factors (variables)
which explain (across a sample of the largest urban areas):
- the extent and rate of urbanization/land cover change
- composition of urbanization/land cover change.
Can we identify them?
Does data and modeling scale matter to the identification of these
factors?
How do we validate these factors (predictive validation,
cross-validation)?
- Which dimensions of scalability matter the most?
- grid size
- polygon grain
- grid shape
- temporal scale (noise vs. signal)
- Across a sample of representative urban regions, does the extent
and composition of
urban land use forms really affect the differential production of:
- ozone and greenhouse gases (vs. meteorological conditions)
- particulate (vs. Greenhouse gases)
- watershed nitrogen and nutrient loadings (vs. hydrology)
Are there critical periods that deviate from the typical?
- What land use/land cover/landscape ecology primitives/objects
would be most appropriate for a next-generation object-oriented land
use/land cover modeling system. How do these differ from current
conceptualizations of time, space, and scale in GIS?
- What are the key public policy issues that land use/land cover
modeling can usefully inform (i.e., shape appropriate interventions).
At what scale? Can we develop some useful heuristics/screening criteria
for answering this question?
FUTURE COLLABORATION AND INTERACTION
Multiple pathways should be used to collaborate and interact within and
without the group represented here.
INTERNAL
Internal communication can be sustained most easily and consistently
using
a common web site. The site begun for this meeting could be continued for
the purpose, perhaps using "forum" software so that all participants can
post messages/information directly and immediately to the site. The
contents of this user-maintained web site could then periodically be edited
or distilled.
A follow-up to this meeting could be convened in a year or two. Timing
could capitalize on reconvening of this group under current sponsorship,
other joint funding opportunities, and the International GIS/Modeling
Conference.
EXTERNAL
The above web page will be useful outside the group as well but any
directly interactive function will probably be less important. Links to
related pages will perhaps be more important.
A paper collection could be completed either as an NCGIA in-house
report or
a special issue of a journal. The first would not be peer-reviewed.
Papers from this meeting could be used in either case with moderate
adaptations.
A review paper could be done by joint authors as an article, a USGS open
file report, or the front piece to a book which would give a systematic
treatment, perhaps with case studies. A book could indicate the state of
the art most comprehensively, particularly at this "moment of
reinvigoration" and expansion.
SPECIAL
A "working group on land use modeling" could be formed as a virtual
entity
with the advantage that the burden of participation could be limited by
individual desires yet its existence can be cited to endorse or sanction
collective actions.
NOTES
Gray literature and intensive web page development are methods which do
not have strong incentives attaching.
A writing project was begun at Santa Fe by Pijanowski, Parks, and Maxwell.
A potential framework has been discussed and participation is invited.
A fourth GIS/Modeling Conference (following Santa Fe) is under
consideration using a modified approach. Time, place and form are
undecided.
OPTIONS
(previously listed in charge to group):
Web page
Review papers
Paper collection
Special issues
NCGIA Report
Open file report
|