Hazard Rating Model This image depicts the probability of observing Armilaria root disease on randomly located sample plots in the Black Hills National Forest of South Dakota. Important varibles used in describing the large-scale spatial variability included elevation, site index, and slope. The small-scale spatial variability was modeled using kriging. The final model accounted for 93% of the spatial variability in the sample data.

Forest Inentory Analysis The four images depict some results of a forest inventory of a management unit in the Ejidi el Largo, which is located in the western part of the state of Chihuahua, Mexico. In the summer of 1998, the management unit was sampled to obtain stand level estimates (top figure in each image). The same information was used in developing a spatial model to describe the variable of interest across the management unit at a 60m resolution (bottom figure in each image). Elevation and various Landsat MSS bands (60m resolution) were used in modeling the large-scale spatial variability, and kriging was used to describe the small-scale spatial variability.

Micro-ecological Units This image demonstates the application of a multivariate spatial clustering algorithm for identifying micro-ecological units using forest inventory data from northern Mexico. In this example, areas with the same color have similar spatial characteristics with respect to habitat suitability for the wild turkey, standing wood volume, stand density index, and 10 year diameter growth.

Soil Types This image depicts the results of a model developed to predict soil types in the Rocky Mountain National Park, Colorado. Discriminant analysis was used to develop a set of discriminant functions for 15 soil types identified from 229 soil pits, strategically located throughout the park. Variables identified as being important in discriminating among soil types include such variables as, individual Landsat TM bands, pH, depth to bedrock, percent lamella, and percent rockiness. A spatial autoregessive model was used to spatially predict pH, depth to bedrock, percent lamella, and percent rockiness, as function of percent slope, elevation, and various Landsat TM bands. The spatially predicted surfces of soil characteristics, along with the Landsat TM bands were passed through the discriminant functions, one pixel at a time, to produce the final map.

Species Richness These two images depict the spatial distribution of native and non-native species on 9,500 ha of the Rocky Mountain National Park, Colorado. The large-scale spatial variability in the number of non-native species was modeled as a function of various Landsat TM bands, percent sand, percent silt, carbon, elevation, and the number of native species. The model as fit using a spatial autoregressie model. There was no small-scale spatial variability associated with the spatial distribution of non-native species. The large-scale spatial varibility in the number of native species, percent sand, percent silt, and carbon was modeled using various Landsat TM bands and elevation. Kriging was used to model the small-scale spatial variability. When the maps were developed no data was available for the high elevation tundra and so the models over estimate both the number of native and non-native species in this area.

Soil Texture A trend surface model was used to spatially predict percent clay as a function of various Landsat TM bands, radar imagery, and elevation. Kriging as used to describe the small-scale spatial varibaility in percent clay. Percent sand was modeled as a function of percent clay, elevation, aspect, and various Landsat TM bands. The prediction of percent clay and sand were constrained to have the same probability density function as the field data. Percent silt was estimated by subtracting percent clay and sand from 100. Finally, the soil texture map was generated by classifying each pixel of the image into one of 10 soil types based on the amount of sand, silt, and clay.

Rock Outcropping A linear discriminant function was developed to spatially predict the location of rock outcropping as a function of various Landsat TM bands and elevation (middle figure). The model falsely predicted the presence of rock outcroppings on 4% of the area (bottom figure), and falsely predicted the absence on 10% of the area, for an overall accuracy of 85.5% (middle figure).

Back to Robin's Homepage