Session Information
| Session | Poster Session | | Date | Monday (2008-04-07) | | Time | 5:00 PM - 7:00 PM | | Room | Grand Terrace |
Presentation Information
| Presenter | James Millington | | Title | The inferential nature of constructing landscape-level simulation models: An example from a managed forest landscape | | Affiliation | Michigan State University | | Authors | James Millington, Joseph LeBouton, Michael Walters, Kimberly Hall, Megan Matonis, Edward Laurent, Jianguo Liu | | Keywords | CHANS, Deer herbivory, Simulation model, Species distributions, Timber harvest | | Presentation Type | Poster | Abstract:
Landscape-level simulation models of coupled human and natural systems are often constructed because data and understanding about interactions within the systems are limited. In such cases, inferences are needed to produce generalizations about the interactions. Here we demonstrate why such inferences are needed in the construction of an integrated ecological-economic simulation model of the spatial interactions between deer browse and forest regeneration following timber harvest. We present the process of constructing a spatial sub-model of stand-level browse intensity via an examination of factors influencing deer density.
We use stand-level deer fecal pellet survey data as a proxy for white-tailed deer density in a ~4,000 km2 study area in Michigan’s central Upper Peninsula (UP). For each deer density estimate, land cover compositions and pattern metrics derived from Landsat imagery were calculated for surrounding local landscapes across a range of areas (0.125 – 8.0 km2). We found few significant statistical relationships between deer density and land cover compositions and pattern metrics. Furthermore, these relationships were insensitive to changing local landscape area. The best stepwise linear regression model explained 20% of the variation in deer density. This model used land cover compositions (for a 2.0 km2 local landscape) and distance to nearest lowland conifer stand as independent variables. A predictive model of deer density for our study area, using distance to lowland conifer alone, reproduced the weak spatial relationships evident in the empirical deer pellet data.
This ongoing model construction process has demonstrated the challenges of using fine-scale empirical data to represent phenomena over large extents. Previous studies in the UP have used deer-density estimates aggregated at more coarse spatial grains (e.g. county level) than we considered. These studies found significant statistical relationships between deer density and land cover compositions and spatial pattern metrics. For stand-level deer-density estimates we found no significant relationships with land cover pattern metrics and weak relationships with land cover compositions. Consequently, to represent browse intensity at the stand level in our integrated model we must take an alternative modeling approach that makes inferences based on multiple information sources. We will combine understanding gleaned from our data (e.g. deer density is influenced by distance to lowland conifer) with results from previous research and theoretical understanding. This modeling process highlights that inferential processes are often needed when constructing models of poorly understood spatial interactions. Such inferential approaches will continue to be useful when integrating sub-models in our landscape-level ecological-economic simulation model. |
|