PARENT SESSION
Poster Session: Spatial Analysis

Cohen's Kappa, Classification Tables and Multi-criteria Model Selection Process: an ArcView 3x Extension to Verify and Select Spatially Explicit Predictive Models. ^*JENNESS, JEFF ¹ and ^*WYNNE, J. JUDSON ², ¹ USDA Forest Service, Flagstaff, AZ, USA² USGS-Southwest Biological Science Center, Flagstaff, AZ

ABSTRACT- Spatially explicit models are employed routinely in guiding land management decisions. We developed an ArcView 3x extension to 1) verify the accuracy of spatially explicit predictive models and 2) provide the metrics necessary for selecting the best model within a multi-criteria selection process. Input data required are an ArcView grid or shapefile of model predictions, and a point data file of independent field observations. Our extension calculates a classification table of correctly and incorrectly "predicted" and "observed" cases, whereby a Cohen's Kappa statistic, overall model accuracy, model specificity and sensitivity, and errors of commission and omission are derived. These metrics and the associated explanations within the user manual will enable end-users to understand and identify the highest quality models. When comparing multiple models, we recommend the use of this extension within the context of a multi-criteria selection process. We suggest weighting the highest Kappa value, highest overall accuracy, highest specificity and sensitivity, lowest omission and commission errors to select the highest quality model. Regardless of the multi-criteria model selection procedure employed, it must be clearly defined, objective, consistent and repeatable.

KEY WORDS: Cohen's Kappa, accuracy assessment, multi-criteria model selection, spatially explicit models, ArcView