Description of GEOMOD

Text by Myrna Hall, April 2001

GEOMOD was developed by researchers at the SUNY College of Environmental Science and Forestry (Hall, et al. 1995a; 1995b; Pontius et al. 2001) with funding from the US Department of Energy, Carbon Dioxide Research Program, Atmospheric and Climatic Change Division. The model simulates land use change using what we refer to as geographic modeling (Hall, M. et al. 2000). It determines the rate of empirical land use conversion from forested to non-forest, extrapolates that rate into the future, and most importantly simulates the location of future land use change based on statistical analysis of the empirical pattern.

The rate of change is derived by comparing the area of forest found in a land cover/use map at one point in time to that found in another at a different (either earlier or later) point in time. The model can be run either forward or backward and simulated results can be tested against the actual landscape as derived from satellite or aerial photography imagery. Future rate trends can also be established from various independent variables that may influence deforestation such as population growth, economic activity, employment history, infrastructure establishment, etc., using regression analysis.

The spatially specific location of that estimated land use conversion, derives from a statistical deduction approach, which analyzes historical patterns of land use change against user-supplied map layers of bio-physical and cultural attributes. The change observed in any given landscape 'cell' is analyzed against a number of candidate drivers, which researchers (Hall et al. 1995, Flint and Richards 1991, 1994; Dale 1994), have determined as potentially the most important factors influencing human settlement patterns. GEOMOD2 essentially uses non-linear multiple-regression to weight each driver according to its assessed importance in determining the pattern and location of changed cells over time. The underlying philosophy is based on the “maximum power principle” (Odum 1983). It is assumed that those conditions which are most likely to significantly impact an individual's energy return on investment (EROI) (Hall, Cleveland and Kaufmann 1986) as he/she develop and work new ground include such variables as: topographic position (elevation and steepness of slope), distance from rivers, roads and already established settlements, climate variables, etc. Each of these variables can affect the energy cost or the expected return from a development decision. In other words we assume that developers are knowledgeable to some degree about where development would be most profitable in energy and hence economic terms. Using the assigned weights GEOMOD can then develop or undevelop land going forward or backward in time. Starting from seeds of earliest development it extends human activity across the landscape, creating a pattern of development from undisturbed to disturbed that closely mimics reality. As it 'nibbles' away at the landscape the model adheres to the three following principles: (1) adjacency, which is the tendency to develop land next to land already developed; (2) dispersion, which is the phenomenon to 'jump' from one place to another relatively favorable location; and (3) regional heterogeneity, which accounts for significant differences in the pattern and rates of land use change among subregions or countries because of the population density, economic and political factors particular to those places.

The variation in accuracy of model predictions depends on the time scale used, the number of land classes modeled and the accuracy of the initializing data. We have found that the drivers of land use change pattern are both scale- and terrain-dependent: specifically, topographic features are more important than climate variables for large-scale simulations where topography is rugged. We have achieved highest accuracy when predicting only two land use classes, forested and non-forested. Going to three classes, undisturbed, disturbed, and non-forest our accuracy declined to 84%. Including all 12 land use categories delineated in Brown et al. (1994) we achieved 78% accuracy. The amount of land use change is influenced greatly by population growth and land use policy in the region being modeled. The model, when linked with biomass estimates, can be used to estimate the magnitude of carbon released or sequestered in the vegetation (Brown et al. 1994).

Data Requirements

The model is FORTRAN-based and requires as inputs a spatially-referenced set of equally-dimensioned digital grid (raster) maps. The following inputs, at a minimum, are required.

  1. A digital elevation model, or better yet, a digital coverage of elevation contours and maximum/minimum elevation points from which a DEM can be prepared using ARC/INFO's (ESRI) Tin generator. Slope and aspect, potentially important drivers, are derived from this.
  2. A digital hydrography coverage (streams, lakes) also used to generate the DEM.
  3. A digital coverage of roads
  4. A coverage of any other transportation routes (rail, air, boat) that give people access to the interior.
  5. Classified and geo-referenced land use maps derived from either aerial photography or satellite imagery for preferably two points in time, and preferably at the same scale and no smaller than 1:24,000 with a grid cell resolution no larger than 30 x 30 meters. Existing settlements should be one of the identified land use classes. Also any land guaranteed as 'set aside' (i.e. protected) should be indicated (see summary below).
  6. Population data over the same period of time.
  7. Climate data, particularly if there is a considerable elevation gradient. These would include both mean annual temperature, mean minimum and mean maximum and precipitation measurements from one or more nearby weather monitoring stations, if available. (The elevation and geographic coordinates of that station are required as well.)

Additional useful data:

  1. Economic data such as crop production, investments, or exports, employment, average income, building permits, local government budgets for roads, bridges, sewage treatment, etc.
  2. Social data such as average highest level of education, land use preference. (This needs Valerie Luzadis' input.)
  3. A digital soil map and accompanying information on such soil characteristics as soil thickness, drainage characteristics, annual flooding, infiltration rate, % silt, clay, loam.
  4. Any other climatic data, PET, daily insolation, cloud cover, etc. could be useful.
  5. Biomass estimates for digitized vegetation polygons, or, if the land use map is classified at a level that supports biomass estimates, then an accompanying table would be useful.

Methods

  • Reclassification of Land Use Types
    • Geomod can evaluate change in two land use types at a time. Therefore, each map of land use change must be reclassified similarly. The most common reclassification is to classify all undisturbed forest as type 1, and all other land use types, which can be characterized as having undergone some human intervention, such as urban and agricultural areas, as type 2. In addition, land that will not be evaluated, such as land that is unlikely to ever be used by humans for productive activity, e.g. deserts, can be excluded from the analysis. Normally in the original land use map these areas are assigned a unique value. Furthermore, land that sits within the grid map, but lies outside the area being modeled, also must have a unique identifying number, indicating that it will not be included in the evaluation. This could include land in another country or region, oceans, or areas for which no data exists. A normal land use map, ready for input to GEOMOD, therefore, will usually consist of four possible values as follows:
      1. forested
      2. agriculture, cities, pasture, degraded land
      3. deserts, water (lakes, oceans, rivers)
      4. outside region of interest
  • Rate Estimation
    • If two maps of land use from two points in time are available they must be reclassed as shown above. Using an area calculation we can determine how many cells of forest existed in each period. Subtraction of the second from the first tells us how many cells were deforested in the interim period. This number times the area per cell yields total area deforested. When we divide the area by the number of years we have our rate per year.
    • We can also calculate the rate of deforestation per year as a function of population or economic growth. Linear analysis of deforestation versus population, or multi-linear regression of deforestation versus population, gross regional product, exports, agricultural production, etc., allows us to project different rates of growth according to these indicators and hence simulate the land use change that can be expected as a function of changes in these independent variables. This gives us more flexibility to test different future scenarios.

Products

GEOMOD will produce a time series of land use maps, at a time interval to be selected, over the 40-year life of the project. Each digital map will be produced as a color print, and the area in each land-use type will be reported in an accompanying table. Additionally, the output and accompanying predicted data, can be displayed in a time-series display module called ECOPLOT (RPA 1997). ECOPLOT displays the changing landscape, over time, as a central image, and graphs important driving variables, and program output as graphs surrounding this image (e.g. population growth, gdp, linear miles of roads built, biomass, carbon stored in the vegetation, etc.) (See http://nrmsc.usgs.gov/research/glacier_model.htm (Hall, M. 1995, Hall, M. and Fagre 2000) for a 'virtual' example of ECOPLOT displayed on the USGS web site – minus the graphs described above.) This tool is particularly intuitive for illustrating rapid and significant landscape change and is a useful informational aide when presenting land use alternatives and decision-making options, and their environmental consequences, to politicians, resource managers, developers, financial backers, and the general public.

Summary

We have been very successful in the past in predicting land use change, for example in Costa Rica in 1983 from a 1940 map (Hall et al. 1995b; Pontius et al. 2000). Comparison of our predicted results with an actual land use map of 1983 shows an 87% accuracy, with a Kappa factor of 40-50) (Pontius, 2000). Cornell (2000) has improved the accuracy to 92% by setting aside those lands with national park status, i.e. protected areas. We feel, therefore, that given good data, GEOMOD should give highly accurate estimates of future potential land use change in the northeastern USA.

Literature Cited

Brown, S., L. R. Iverson, and A. E. Lugo, 1994. Use of GIS for estimating potential and actual forest biomass for continental South and Southeast Asia, pp. 67-116 in V. H. Dale (Ed.), The effect of land-use change on atmospheric CO2 concentrations, Springer-Verlag, New York.

Cornell, J. (2000). Assessing the role of parks for protecting forest resources using GIS and spatial modeling. Chapter. 19, in C. A. S. Hall, (Ed.) Quantifying Sustainable Development: the Future of Tropical Economies. Academic Press, San Diego, CA.

Dale, V. H. (Ed.), 1994. The effect of land-use change on atmospheric CO2 concentrations, Springer-Verlag, New York.

Flint, E. F. and J. Richards, 1991. Historical analysis of changes in land use and carbon stock of vegetation in south and southeast Asia. Can. J. For. Res. 21, 91-110.

Flint, E. F. and J. Richards, 1994. Trends in carbon content of vegetation in South and Southeast Asia associated with changes in land use, pp. 201-300 in V. H. Dale (Ed.), The effect of land-use change on atmospheric CO2 concentrations, Springer-Verlag, New York.

Hall, C. A. S., C. J. Cleveland and R. Kaufmann, 1986. Energy and resource quality: The ecology of the economic process. John Wiley and Sons, New York.

Hall, C. A. S., H. Tian, Y. Qi, G. Pontius, J. Cornell and J. Uhlig, 1995a. Spatially explicit models of land use change and their application to the tropics. DOE Research Summary, No. 31. (Ed. By CDIAC, Oak Ridge National Lab).

Hall, C. A. S., H. Tian, Y. Qi, G. Pontius, J. Cornell and J. Uhlig, 1995b. Modeling spatial and temporal patterns of tropical land use change. J. of Biogeography, 22, 753-757.

Hall, M. H. P., 1995. Predicting the Impact of Climate Change on Glacier and Vegetation Distribution in Glacier National Park to the Year 2100. State University of New York: Syracuse, New York. Master's Thesis, 194pp.

Hall, M. H. P., 1999. A dynamic time series display of vegetation and glacier change in Glacier National Park as a function of changing climate. USGS web site: http://www.nrmsc.usgs.gov.

Hall, M. H. P., C. A. S. Hall, and M. R. Taylor, (2000) Geographical Modeling: The Synthesis of GIS and Simulation Modeling, Chapter. 7, in C. A. S. Hall, (Ed.) Quantifying Sustainable Development: the Future of Tropical Economies. Academic Press, San Diego, CA.

Hall, M., and D. Fagre (submitted to Bioscience). Where have all the Glaciers gone?

Odum, H. T., 1983. Systems Ecology. Wiley Interscience, New York.

R G Pontius Jr, J Cornell and C Hall. 2001. "Modeling the spatial pattern of land-use change with GEOMOD2: application and validation for Costa Rica". Agriculture, Ecosystems & Environment.

R G Pontius Jr. 2000. "Quantification error versus location error in comparison of categorical maps". Photogrammetric Engineering & Remote Sensing 66(8) pp. 1011-1016.

RPA, 1997. ECOPLOT software, Resources Planning Associates, Inc., Ithaca, NY. (contact .(JavaScript must be enabled to view this email address))