PSEUDOSPECTRAL ANALYSIS FOR MODELING DISCRETE SPATIAL DATA

Pseudospectral analysis is a new method for the classification, analysis, and modeling of discrete geospatial data. The method uses multispectral processing techniques to classify and model trends in discrete spatial data. Multiple point- and line-data sets are converted to coincident raster datasets using kernel density calculations. The raster images, each analogous to a band in a multispectral remotely sensed image, can then be stacked, possibly along with other coincident raster images, and analyzed using multispectral analysis and classification methods. Regions are classified based on unique combinations of density values from the individual raster layers. Within classified regions, the mean density value of each layer is calculated for each region yielding a matrix of values by regions and layers. Further statistical analyses, such as cluster or factor analyses, can be performed on the matrix and used to model the classified regions. Pseudospectral analysis of karst in Missouri illustrates the technique. The analysis revealed distinct areas that are statistically grouped into regions of upland recharge, valley recharge, transport, and discharge.