Vegetation mapping inventory project: Hawaiʻi Volcanoes National Park

Green, Kass; Hall, Mark; Lopez, Chad; Ainsworth, Alison; Selvig, Megan; Akamine, Kathryn; Fugate, Steve; Schulz, Keith; Benitez, David; Wasser, Mark; Kudray, Greg

This report details the methods used and results of the vegetation inventory for Hawai‘i Volcanoes National Park (HAVO). HAVO’s unique vegetation spans 130,888 ha (323,431 ac) on the Hawaiian Island of Hawai‘i and includes two active volcanoes, Kīlauea and Mauna Loa. The park’s dynamic landscape, huge expanse, and extreme range of elevation and climate combine to create a distinctive and exciting national park. The vegetation inventory was a team effort managed by the National Park Service (NPS), Pacific Island Network Inventory and Monitoring Program (PACN), and staffed by personnel from the NPS, University of Hawaii, Cogan Technology, Inc., NatureServe, and Kass Green & Associates (KGA). The NPS collected 590 vegetation classification plots and observation sites. NatureServe used these data to classify 77 vegetation associations at HAVO. A dichotomous key included 68 of these associations and detailed descriptions were developed for 49 of these associations. To ensure data compatibility throughout the NPS and other agencies the National Vegetation Classification Standard, Version 2 was used to characterize HAVO’s vegetation. Digital Globe Quickbird and Worldview2 imagery of HAVO were obtained from the US Department of Agriculture – National Resource Conservation Service. The imagery was segmented in eCognition software and both the imagery and the segments were taken to the field to support the calibration trip. During the calibration trip to HAVO by KGA, NatureServe, and PACN staff, four additional vegetation associations were discovered and added to the association key and HAVO staff provided a model for assignment of labels to the coastal grasslands. Following the calibration trip, KGA labeled the segments using the imagery, ancillary data sets, and classification and regression tree analysis to create a preliminary map. The preliminary map was reviewed by PACN staff, carefully compared to the 1992 vegetation map of HAVO, and assessed in the field with KGA during the validation trip. Manual edits based on the comparison, comments and results from the validation trip were made to the preliminary map to create HAVO’s draft vegetation map. PACN personnel collected 964 accuracy assessment (AA) samples of HAVO’s vegetation. The sample reference labels were compared to the draft map labels at each sample location to create an error matrix. PACN and KGA personnel reviewed the AA data and decided which associations would be collapsed into map classes for creation of HAVO’s final vegetation map, resulting in 80 vegetated and four unvegetated map classes. Following the AA, a short second validation trip was conducted with KGA, PACN, and HAVO staff. During that trip the coastal grasslands model was updated and some of the AA reference labels were clarified. KGA then edited the areas of error in the draft map and applied the new coastal grasslands model to produce the final map.