Comparison of field-observed and simulated map output from a dynamic floodplain vegetation model using remote sensing and GIS techniques

Three different forms of the Kappa statistic were used to compare simulated and observed maps to assess the predictive capability of a dynamic vegetation model. Kappas were used to compare vegetation model performance considering both individual vegetation classes and merged vegetation phases. The overall Kappa increased slightly from 0.18 to 0.21 and 0.23, when different vegetation phases were resampled from the original 10 m cell size to 30 m and 50 m cell sizes, respectively. Similarly, overall Kappa improved from 0.18 to 0.55 and 0.71 when the neighborhood cells within a buffer distance of 10 m and 20 m, respectively, were considered. Different forms of the Kappa statistic (K, K_loc, K_qty) are valuable indicators to study the performance of vegetation models and identify parameters (e.g., quantity or location) for improvement in the model results.