I am an Assistant Professor in the Department of Geography at Michigan State University. After obtaining a B.A. in Geoenvironmental Studies at Shippensburg University, in Pennsylvania, I obtained two graduate degrees (M.A. and Ph.D.) in Geography from the University of North Carolina at Chapel Hill. My dissertation research involved developing a logistic regression model to predict the patterns of vegetation at alpine treeline in Glacier National Park, Montana, using topographical, bioclimatic, and disturbance gradients. I teach courses in geographic information systems (GIS), integration of GIS and remote sensing, digital terrain analysis, and people and the environment. My research focuses in three broadly defined areas: digital terrain analysis, ecological mapping, and land use/cover change. Each of these areas involves the application of remote sensing, GIS, spatial analysis, and ecological modeling tools. I have not developed a land use change model, but share my thoughts on the application of one below.
A research group at MSU (of which I am PI) has recently been awarded
a three-year grant under NASA's Land Cover and Land Use Change
(LCLUC) program to study and model the relationships between socioeconomic
and demographic changes and changes in the types and patterns
(i.e., fragmentation) of land use and land cover across the Upper
Midwest between 1970 and 1990. This project builds on two on-going
projects. One is a collobration with the USDA Forest Service that
is examining increased fragmentation of land ownership throughout
the same region. The other, headed by Dr. Bryan Pijanowski (another
workshop attendee) among others, involves the devlopment of a
GIS-based spatial model of land use change. My interests are not
in describing the model, but rather some of the issues involved
in the interplay between the model and a data-rich empirical investigation
on a regional scale.
The two previous projects:
The objectives of the LCLUC project are as follows:
Figure 1. Location of study region (not shaded), sample counties
(in blue), sample sites (in red), and MSS scenes (outlined).
I am also working on a project recently funded by the National Science Foundation to compare observed spatial patterns of vegetation at treeline with the spatial patterns that result from a plant growth model (in the JABOWA-FORET tradition, but with some modifications to incorporate spatial processes). In this study we will attempt to predict the spatial pattern of productivity, using the leaf area index (LAI) as one measure. Both of these projects are just now getting started. What follows are thoughts that have arisen from my involvement in the planning stages of these two projects.
"Nobody believes a model save its developer, everybody believes a data set except its collector"
-- Anonymous
I recently read this quote in the proposed framework for the National Environmental Monitoring Initiative by
the National Science and Technology Council (NSTC) Committee on
Environment and Natural Resources. It suggests something about
the relative roles of data and models in science. In reality modeling
and observation are two completely different endeavors. Yet they
are dependent on one another in completing the cycle of scientific
inquiry.
Reconciling models and data can be difficult. One illustration
in the context of land use modeling can help make the point. Multi-spectral
satellite sensors are typically proposed as fairly consistent
sources of data with which land use change models can be compared.
Most remote sensing specialists know that obtaining land use
information from satellite data with a resolution the Landsat
sensors (TM or MSS) is nigh to impossible (see Figure 2 below).
Satellite data are good at providing information on land cover,
not land use. However, modeling efforts tend to focus on land
use because it is more directly a product of the social value
placed on the land. Land cover is a consequence of how the land
is used.
Figure 2. Color Infrared photo from Crawford County, Michigan
(resolution = 2 m). Residential areas bounded by the road network
in the subdivision on the left is spectrally very similar to the
forest on the right. A spectral classification would most likely
identify some of this area of residential USE, with the exception
of the roads, as forest COVER, especially when aggregated to 30
or 60 meters.
Although difficult, reconciliation of data and models is necessary.
Observational data are necessary if one is to test the performance
of a model or to calibrate any unknown parameters. Also, a t0
map is typically needed to set a base line for the model. The
source of the baseline map is typically remote sensing. Furthermore,
the observation of changes without the structure of a model can
be very difficult to generalize.
How can mapped land data be reconciled with land use models and
vice versa? I can think of four different modes for such reconciliation,
most of which have already been used to some extent. Although
these are not new approaches, I think it useful to structure our
thinking about how data supports modeling and vice versa. First,
one can simplify the problem of change to one that involves only
land use types that are expressed on the landscape as easily distinguishable
land cover types. This is the approach taken in models that address
urbanization (i.e., the conversion of land from rural to urban
uses) as the only change (Clarke et al., 1996; Pijanowski et al.,
In Review). Unfortunately, all change is not urbanization, nor
does urbanization always look the same (see Figure 2). Secondly,
models can be compared and constructed with maps of land use (Geoghegan
et al., 1997). Land use mapping is reasonable over a small area,
but on a regional or global scale can be an overwhelming task
because (a) data at the resolution of Landsat or coarser are insufficient
for mapping most land uses and (b) automated multi-spectral algorithms
are ineffective for identifying land use. Third, an attempt
can be made to model land cover changes (Turner, 1990). This is
usually done through the use of empirically based approaches (i.e.,
Markov transitions) which tend to be limited by the low level
of explanation in the model. Therefore, these models have more
serious limitations on how far they can be extrapolated in space
or time. Finally, a model might attempt to explicate the
linkage between land use, which is a product of many social driving
variables, and land cover, which is a product of those driving
variables and the resulting land use.
I am interested in exploring the degree to which different land uses result in certain types, proportions, and patterns of land cover. This requires information on both land use and land cover. In our study, we will map land use from aerial photography and parcel boundaries within our 136 sample sites, and land cover region-wide from MSS data. A modeling question becomes how to model land cover change regionwide as a consequence of land use change, for which only a sample of sites is available. Linkages might also be developed to predict changes in other ecosystem properties (e.g., green leaf biomass, net primary production, and leaf area index) as a consequence of land use change. These properties can provide information about the biogeochemical cycle implications of land use change and can also be monitored.
Clarke, K.C., Hoppen, S., and Gaydos, L. J., 1996, Methods And Techniques for Rigorous Calibration of a Cellular Automaton Model of Urban Growth. Third International Conference/ Workshop on Integrating GIS and Environmental Modeling, Santa Fe, New Mexico, January 21-25, 1996. Santa Barbara: National Center for Geographic Information and Analysis. WWW and CD.
Geoghegan, J., Wainger, L., and Bockstael, N., 1997, Spatial landscape indices in a hedonic framework: an ecological economic analysis using GIS. Ecological Economics, In Press.
Pijanowski, B.C., Long, D.T., Gage, S.H., and Machemer, P.L., In Review, A spatially explicit land transformation model: integrating policy, socioeconomic, and environmental factors. Ecological Modelling.
Turner, M. G. 1990. Landscape changes in nine rural counties in Georgia. Photogrammetric Engineering and Remote Sensing, 56(4): 379-386.