Accuracy of landsat-TM and GIS rule-based methods for forest wetland classification in Maine

An investigation was undertaken to compare satellite image classification techniques to delineate forest wetlands in Maine. Four classification techniques were compared, including a GIS rule-based model. Accuracy assessments of the four methods on two study sites, Orono and Acadia, revealed very similar results. Overall accuracy for four super groups (forest wetland, other wetland, forest upland, other upland) ranged from 72% to 81% at Orono and 74% to 82% at Acadia. Pairwise significance tests indicated that the GIS model was significantly better than unsupervised classification at both study sites, and significantly better than tasseled cap (Acadia) in classifying the four super groups. Although Kappa coefficients were slightly higher for the GIS model compared to hybrid classification, there was no significant difference between the two methods at either study site. Forest wetland userʹs and producerʹs accuracy was in the 80% range for the highest accuracy achieved either by the GIS model or hybrid classification. Hydric soils, National Wetland Inventory data, and slope percentage were the most important variables in the GIS model. From this study, it appears that a combination of hybrid and GIS rule-based classification methods are the most promising for further investigations of forest wetland delineation.