Accuracy Assessment: Russell Cave National Monument

Smart, Lindsey; Jones, Erin

Brief Description: This report presents an accuracy assessment for the digital vegetation map of Russell Cave National Monument (RUCA). Full Description: The map accuracy was assessed by comparing mapped vegetation types to field verified vegetation types at randomized evaluation points. The evaluation points were chosen prior to field work using statistical methods to ensure full representation of the range of map classes in the park. Accuracy was calculated for each individual map class and for all map classes combined. Remotely-sensed imagery is limited in its ability to differentiate between certain forest types and even the most experienced mappers can’t differentiate between certain species of oaks or pines in a remotely sensed image. Sources of error for the mapping project are varied and include more than solely “remote sensing error” but also included “ecologist error” caused by poor interpretation of the vegetation community concept, “field worker error” caused by mistakes made by fieldworkers while collecting the data (including misreading of the key), and temporal error when conditions on the ground change between the mapping and assessment processes. It is difficult to isolate a single error that is causing accuracy issues without more research. The accuracy assessment, therefore, should be used more as a tool to discern usability of map classes rather than a way to judge the performance of the mapmakers.The University of Georgia (UGA) Team focused on generating the highest level of detail possible during park vegetation mapping to provide the most accurate information for the National Park Service. As a consequence, assessment of the finished project requires a two step approach: (1) assessing the overall accuracy of the finest-scale map produced, and (2) combining the most “confused” map classes to determine the accuracy measures at coarser scales. The report provides the best approximation of individual map class accuracy and also suggests combinations of map classes to produce a more reliable map at a coarser scale. For RUCA, the overall accuracy of the final map, which includes one grouped map class, is 89%, with a kappa statistic of 0.85 (85%). This version of the map is the most appropriate for use by the standard user; what it misses in fine-scale detail, it makes up for in the relatively high level of accuracy. Only two vegetation associations were combined for the purpose of increasing the overall accuracy of the final map: White Oak - Mixed Oak Dry-Mesic Alkaline Forest (CEGL002070) and Rich Low-Elevation Appalachian Oak Forest (CEGL007233).The accuracy assessment for this combined version of the map considered points as a match if the vegetation observed on the ground matched any of the dominant, secondary, or tertiary vegetation types attributed to the map by the mapmaking team. The strictest analysis of the data (before combining map classes/ NVC associations and only considering a point a match if the vegetation observed on the ground matched the dominant vegetation type attributed by the mappers) showed an overall map accuracy of 62% with a kappa statistic of 0.55 (55%). This lower accuracy reflects the difficulty in differentiating the vegetation associations that were combined in the final analysis and that these classes most likely share similarities in composition on the ground and/or in appearance on aerial photography. Key findings: For users interested in preserving the full detail of the map for highly detailed studies or management of the landscape, we recommend use of the fine-scale map as published by UGA. For all other users, we recommend combining map classes as specified above to allow for an overall map accuracy well above 80%. In this way, the vegetation maps are useful for the widest audience possible and retain their potentially important fine-scale detail.

Accuracy Assessment: Russell Cave National Monument

Collections