Land Cover Mapping
48-class land cover map of
Vegetation Classification System (NVCS): International Classification of Ecological
Terrestrial Vegetation of the
1998), but used alliance aggregations as final map units. The map units were derived after consideration of the availability of statewide ancillary data sets for modeling, success achieved by other states with similar vegetation types, and expert review by state professionals.
The map was created using various combinations of input data including:
1. a vegetation classification based on satellite images,
2. a statewide Digital Elevation Model (DEM),
3. the National
Wetlands Inventory (NWI) of
4. various other coverages including ecoregions, hydrology, soils, and urban boundaries
5. a computer model to integrate the data.
vegetation classification used 29 Landsat Thematic Mapper (TM) images and a hybrid of supervised and
unsupervised classification routines and pre-classification image stratification
to produce the primary modeling input. A DEM of Kentucky was created by mosaicking 721 USGS DEM quadrangles comprising
Final map accuracy on a pixel-by-pixel, per-class basis was 51% with User’s and Producer’s accuracy ranging from 8-100%. A ‘fuzzy’ accuracy assessment was also conducted. The accuracy using this method was 75% with User’s and Producer’s accuracy ranging between 25-100%.
Predicted Vertebrate Distributions
The distributions of 365 native terrestrial vertebrate species, including 52 amphibians, 52 reptiles, 63 mammals, 152 breeding birds, and 111 wintering birds (65 bird species occurred in both the breeding and wintering groups) were predicted. Several steps were required to complete the modeling process. Experts from around the state reviewed the products from each step before the next step was taken. First, we determined range limits for each species based on current information about species’ presence or absence within counties and quads, or the Environmental Protection Agency’s (EPA) hexagon grid system. Second, the association of each species with habitat features such as land cover, water, edge, and elevation was researched and compiled in a Wildlife-Habitat Relationships (WHR) database. Third, the necessary GIS layers to represent these habitat features were prepared. Fourth, a raster-based modeling approach was used to combine the species’ ranges and WHR databases into maps of predicted distributions for each species at a resolution of 30-m grid cells. The final step was to determine the accuracy of the predicted species distributions.
Accuracy assessment was conducted at three spatial scales. In the first level, records of species’ occurrences were obtained within specific areas of the state. The data were based on species' checklists from 51 validation areas, including 47 Breeding Bird Survey routes, one national park, one national forest, one national recreation area and one state park. Predicted species’ presence and absence were then compared with those indicated on the species’ checklists. In the second level, an accuracy assessment of physiographic provinces was conducted. A predicted species’ occurrence list was created for each province and compared with the occurrences documented in the species’ checklists. Finally, an accuracy assessment at the state level was conducted by comparing the predicted species’ occurrences with observational data compiled from databases around the state. The latter two assessment processes only evaluated species’ presence and not species’ absence because of our inability to distinguish between observational errors and true absence of species. Thus, these two assessments were not complete.
Geographic patterns of richness of terrestrial vertebrate species indicated that biodiversity
(i.e., predicted species richness) was generally higher in the western portion of the state.
mean biodiversity was associated with land cover of wet forested habitats
(i.e., riparian, bottomland, and floodplain forests) and drier
deciduous/coniferous forest habitats. Lowest biodiversity was predicted in the
Comparisons between predicted and observed species presence/absence at 51 validation areas indicated high agreement rates overall. Low rates of omission errors (failure to predict a species at a location in which it has been recorded) occurred, averaging 4% for all taxa combined. Commission error rates (prediction of a species in a location in which it has not been recorded) were also relatively low (< 10%) for all groups except breeding birds. High rates of commission errors for breeding birds (32%) showed that the models were more likely to overpredict bird distributions than to underpredict them. High commission errors are preferable to high omission errors in the context of management decisions. Failure to predict a species in an area in which it actually occurs (omission errors) can lead to management decisions that inadvertently harm the state’s biodiversity. In contrast, if a species is predicted to occur where it has never been recorded (commission errors), then that species can be targeted for future surveys and can be considered in land use decisions.
Gap Analysis Program (GAP) uses a scale of 1 through 4 to denote the relative degree
of management for biodiversity maintenance for each tract of land, with “1”
being the highest, most permanent and comprehensive level of maintenance, and “4”
being the lowest, or unknown status. Status codes were assigned to land parcels
by KDFWR staff based on conversations with managing entities and field staff
regarding management goals and practices. A flow chart adapted from the GAP
Analysis Handbook was used to make final status determinations. The gap
than 2% of
Of the 428 native terrestrial vertebrate species assessed, 27 (6.6%) had >10% of their predicted distributions within lands assigned management Status 1 or 2 while 57 (13.3%) species had <1%. Only 1 (0.2%) species had > 50% of predicted distribution in Status 1 and 2 lands. The distribution of this species (red-breasted nuthatch) was restricted to a single 7.5’ USGS quadrangle. Similarly to vegetation units, vertebrate species on the whole showed low-to-moderate levels of protection.
GAP analysis of