Land-Use Changes in Southern Appalachian Landscapes: Spatial Analysis and Forecast Evaluation
- Cite this article as:
- Wear, D. & Bolstad, P. Ecosystems (1998) 1: 575. doi:10.1007/s100219900052
- 598 Views
Understanding human disturbance regimes is crucial for developing effective conservation and ecosystem management plans and for targeting ecological research to areas that define scarce ecosystem services. We evaluate and develop a forecasting model for land-use change in the Southern Appalachians. We extend previous efforts by (a) addressing the spatial diffusion of human populations, approximated by building density, (b) examining a long time period (40 years, which is epochal in economic terms), and (c) explicitly testing the forecasting power of the models. The resulting model, defined by linking a negative binomial regression model of building density with a logit model of land cover, was fit using spatially referenced data from four study sites in the Southern Appalachians. All fitted equations were significant, and coefficient estimates indicated that topographic features as well as location significantly shape population diffusion and land use across these landscapes. This is especially evident in the study sites that have experienced development pressure over the last 40 years. Model estimates also indicate significant spatial autocorrelation in land-use observations. Forecast performance of the models was evaluated by using a separate validation data set for each study area. Depending on the land-use classification scheme, the models correctly predicted between 68% and 89% of observed land uses. Tests based on information theory reject the hypothesis that the models have no explanatory power, and measures of entropy and information gain indicate that the estimated models explain between 47% and 66% of uncertainty regarding land-use classification. Overall, these results indicate that modeling land-cover change alone may not be useful over the long run, because changing land cover reflects the outcomes of more than one human process (for example, agricultural decline and population growth). Here, additional information was gained by addressing the spatial spread of human populations. Furthermore, coarse-scale measures of the human drivers of landscape change (for example, population growth measured at the county level) appear to be poor predictors of changes realized at finer scales. Simulations demonstrate how this type of approach might be used to target scarce resources for conservation and research efforts into ecosystem effects.