Identifying Longleaf Ecosystems Across the Southeast (Hogland 2005)     

ABSTRACT | HIERARCHICAL CLASSIFICATION SCHEME | RESULTS | CONCLUSION | DOWNLOAD

Abstract

Longleaf ecosystems have declined to a mere 5% of their original range since European settlement. These dramatic losses, in what was once the dominant pine ecosystem across the southeastern U.S., are the principle reasons for the listing of many plants and animals as threatened and endangered, and have been the driving factor for recent longleaf ecosystem restoration efforts. While studies have documented the regional decline of longleaf ecosystems, they provide little information on fine scale fragmentation patterns and current ecosystem locations. This lack of information often limits the efficacy of restoration efforts.  

To aid longleaf restoration efforts we have developed a hierarchically organized classification scheme that produces a series of fine grain (30 m) ecosystem probability distributions using multitemporal Landsat enhanced thematic mapper plus imagery, digital elevation models, field data, ancillary data sets, polytomous logistic regression, and a hierarchical classification scheme. Using our ecosystem probability distributions, resource managers can identify the most probable locations for longleaf ecosystems, locate potential restoration sites, prioritize restoration efforts, and estimate ecosystem area.

Link to Hogland (2005) Thesis

Back to top

Hierarchical Classification Scheme

Our hierarchical classification scheme is a 2 level, multi-stage, classification that uses the conditional probabilities of a preceding PLR model (stage) to constrain the conditional probabilities of subsequent PLR models (stages). Using this approach, we were able to reduce the impact of temporal features in multitemporal imagery and were able to hierarchically organize our classification without losing spatially explicit class information. The benefits of our hierarchical classification scheme include fewer field samples, preserving modeling and classification errors, and the ability to account for confounding temporal features.

Conceptual Model of our Hierarchical Classification

LUC: Land Use Change
S, F, and W correspond to the seasonality of image acquisition dates (Spring, Fall, and Winter)

Back to top

Results (Download ESRI Grid Files)

Generalized Land Cover Classification (Level One)

Ecosystem Probabilities (Level Two)

Back to top

CONCLUSION

We successfully mapped longleaf ecosystems at a fine spatial resolution (30 m grain), across a large portion of the Southeast. These probabilistic ecosystem classifications provide resource managers with a level of detail that is statistically accurate and precise and flexible enough to begin addressing fine scale longleaf ecosystem restoration questions. In addition, model and classification errors have been maintained in a spatially explicit manner across our study area, thereby allowing other researchers to incorporate our model errors into their work. Future studies could potentially improve upon our results by incorporating ETM+ imagery from a spring leaf-on season, adding a textual component to the analysis, and/or directly incorporating the spatial locations of FIA data.

DOWNLOAD PROBABILITY GRIDS

Level 2 Ecosystem Probability of Occurrence grids can be downloaded from our Data Products Page. These grids have been compressed to a zip file format to reduce download time. To extract these zip files, you can use a program like WinZIP. Link to WinZIP