Extending the Landscape Pattern Metric (LPM) model analysis in Smith et al. (2001) into a LPM decision model, decadal scale prediction of fecal coliform compromised South Carolina watersheds is developed. The model’s parameter variability identifies the greatest contributors to a compromised watershed’s prediction. The complete set of model parameters include Land Cover Land Use (LCLU) & slope, along stream proportion, Fourier Metric of Fragmentation (FMF), Fourier Metric of Proportion (FMP), and Least Squares Fourier Transform Fractal Dimension (LsFT). The 1992 National Land Cover Data (NLCD) Land Cover Land Use (LCLU) within fecal coliform compromised watersheds is used to train the model parameters, and the 2001 NLCD LCLU is used to test the LPM model. The most significant model parameters are along stream bare rock LsFT, FMF between urban/recreational grasses and evergreen forests, and FMF between deciduous forests and high density residential areas. These metrics contribute significantly more than the best proportion descriptor: proportion of urban/recreational grasses. In training, the proposed model correctly identified 92 % of the compromised watersheds; while the Smith et al. (2001) model 94 % of the compromised watersheds were correctly identified. This study reveals the ability of Fourier metrics to interpret ecological processes, and the need for more appropriate landscape level models.
