The mapping efforts at PUHE focused on the benthic habitats of the waters offshore of PUHE (fig. 2) and excluded inside the Kawaihae Harbor. PUHE boundaries do not officially extend into the marine environment; however, effects downslope of any activity in PUHE are of concern to NPS management (personal communication). The area of Kawaihae Bay mapped by USGS extends from the north edge of Kawaihae Harbor approximately 3.5 kilometers south to the north edge of the Mauna Kea Golf Course and Beach Resort at Waikoloa, and from the shoreline to depths of approximately 40 m below local mean sea level (LMSL) where the fore reef drops off to the sandy shelf.

PUHE is located within the Kawaihae watershed, which contributes approximately 75 percent of the drainage in the northern portion of the study area. The Waikoloa and Waiulaula watersheds contribute approximately 25 percent of the drainage in the southern portion of the study area (Cochran and others, 2007b). Drainages from these watersheds are responsible for the increased input of sediment and urban runoff relative to other unpopulated areas of the island (Gove and others, 2023).

The only park on the Kona Coast that includes submerged lands and marine resources within its official boundaries is KAHO (fig. 3). KAHO was established in 1978 and is 1,160 acres in size, including 596 acres of marine area. As of 2025, only a portion of the park has been acquired by the NPS, with the remaining acreage either held privately or by the State of Hawaii Department of Land and Natural Resources Division of Aquatic Resources. KAHO is bordered on the south by the Honokōhau small-boat harbor and on the north by residential homes, a resort, and a golf course.

Although the zonation and community structure transitions described above are evident in KAHO, the seafloor morphology in the park is more complex than along much of the Kona Coast. The mid shelf cliffs, the semi-protected environments provided by the pinnacles off Kaloko Fishpond, and the embayed coastline close to the harbor support localized coral ecosystems that are unique to KAHO. The highest coral covers are generally off the harbor mouth and Kaloko Fishpond (Gibbs and others, 2007).

The area mapped for PUHO extends from Mīana Point around Puʻuhonua Point to Kiʻilae Bay and past the southern park boundary (fig. 4). PUHO park boundaries do not officially extend into the marine environment; however, effects downslope of any activity in PUHO are of concern to management (personal communication). PUHO lies within the nearly 850-square-kilometer Kiʻilae watershed, which begins at the crest of Mauna Loa (Cochran and others, 2007a). The Kiʻilae Watercourse runs through the southern area of the park and empties into Kiʻilae Bay, but only during periods of extreme rainfall.

PUHO experiences the lowest land-based effects of all three of the parks (Gove and others, 2023), though groundwater flow may be of concern (Cochran and others, 2007a).

Baseline underwater imagery for this study were acquired during three field activities conducted by the USGS in 2003 and 2004 (activity IDs A-803-HW, A-204-HW, and A-604-HW), as part of a benthic habitat mapping project done in collaboration with NPS (Cochran and others, 2007a, 2007b; Gibbs and others, 2007).

Georeferenced underwater video was collected along single pass transects distributed throughout the offshore portions of each of the three Kona Coast parks using an underwater standard definition (640 by 480 resolution) video camera illuminated with a light-emitting diode ring designed by SeaViewer Underwater Video Systems. The camera was mounted on a small aluminum hand-deployed frame with a rear-mounted plastic fin. Time, date, location, and ship speed were overlaid on the video using the Sea-Trak global positioning system (GPS) method developed by SeaViewer Cameras, Incorporated. Geographic positioning was obtained using a wide area augmentation system (WAAS) enabled Garmin GPS76 receiver. The stated horizontal accuracy of the Garmin GSP76 is less than 3 m when receiving WAAS corrections.

Video imagery was acquired either by towing the camera along transects while underway (towed video) or by lowering the camera near the seafloor at a given location while drifting slowly (drop camera). Still images were acquired from the videos by sampling one frame every 10 seconds using an automated routine in the Media Mapper software package. The software recorded the GPS coordinates for the resulting images in the image Exif headers. The resulting set of georeferenced images was used for the baseline classifications. The original image set contains 10,229 640- by 480-pixel resolution images of the seafloor from varying heights above the seafloor and from various orientations ranging from oblique to downward looking driven by currents and drag on the camera while the boat was underway.

Positional accuracy and uncertainty varied for baseline imagery collection depending on the method used. The uncertainty of towed video was tens of meters, especially in deeper water where the angle of the cable to the camera was greater than in shallow water. Drop camera imagery had estimated uncertainties of around 5 m from the position of the GPS antenna. The final dataset used for this study was predominantly composed of imagery obtained from towed video (1,512 images from towed video, 112 images from drop camera imagery).

Resurvey underwater imagery was acquired during a single field activity conducted by the USGS in 2022 (USGS activity ID 2022-615-FA) to investigate changes to the nearshore benthic habitat relative to baseline surveys conducted in 2003 and 2004. A total of 146 sites were resurveyed across all three parks. Sites for resurvey were selected based on where coral cover was observed in the baseline surveys. Imagery was acquired at each site while drifting by lowering the camera to 1 to 2 m above the seafloor at each site. The length of the site transects were similar in length to the baseline survey transect lengths because the goal was to “resurvey” the baseline sites.

The underwater image collection system was comprised of three major components. First, a pole-mounted, downward facing digital camera clamped to the side of the surface vessel which extended approximately 1 m below and above the water surface. A global navigation satellite system (GNSS) antenna was fixed at the top of the pole above the downward facing camera and remained approximately 1 m above the water surface to provide accurate image capture locations. Second, a tethered drop-camera system was lowered below the vessel and held roughly 1–3 m above the seabed (fig. 5). Third, a data acquisition and control system was located in the weather protected space of the surface vessel and operated in real-time to maintain image quality by adjusting camera collection rate, exposure, and gain settings of both cameras. The data acquisition system provided real-time position information via a local area network to a separate laptop located on the flying bridge to assist with navigation in the context of the previous survey and planned resurvey locations.

The surface camera was a 12.3-megapixel machine vision camera with a global shutter and a 2.8-centimeter (cm) sensor. The camera was made by Teledyne FLIR (model #BFS-PGE-122S6C-C). The lens used was a manual focus computer (model #V12280MPY) 12-millimeter (mm) focal length, which is approximately equivalent to a 30-mm focal length full frame (35-mm) camera lens. The lens was fixed at f-stop 5.6. The purpose of the surface camera was to collect larger area imagery coincident with the drop camera to facilitate estimating the extent of and continuity of features imaged at close range with the drop camera. The surface camera was designed with a slightly longer focal length lens and higher pixel count than the drop camera so the ground sample distance (amount of seabed imaged per pixel) would be more closely comparable to the drop camera.

The drop camera was a 5.0 MP machine vision camera with a global shutter and a 1.7-cm sensor. The camera was made by Teledyne FLIR (model #BFS-PGE-50S5C-C). The lens used was a manual focus Fujinon (model #HF6XA-5M) 6-mm focal length, which is approximately equivalent to a 23-mm focal length full frame (35-mm) camera lens. The lens was fixed at f-stop 5.6. The drop camera package was tethered to the surface with a sea-cable that provided unlimited power and enabled real-time camera control and image data transfer. In addition to the camera, the subsurface package included dual Ikelite DS160 light-emitting diode strobes modified to accept external power and were included to improve color rendering and image exposure with the limited natural light available at depth. A Tritech PA500 precision altimeter was mounted on the drop camera to measure and display the height above the seafloor to assist with our attempt to maintain a relatively consistent height above the reef.

The drop- and surface-camera housings were designed with identical 50-mm diameter dome windows made from 5-mm-thick optical quality BK7 glass. In both cases, the cameras were rigidly mounted longitudinally in alignment with the long axis of the housings and adjusted fore and aft so that the entrance pupil of the lens aligned with the dome window radius of curvature. This provided the maximum depth of focus and minimized distortion away from image center.

The surface acquisition system included a Trimble R7 GNSS, an electronics interface box that provided power and network communication to the drop camera, a hand reel which incorporated a slip ring for the drop camera tether, and a computer system on which two instances of custom acquisition software were run concurrently to control and acquire images from the two cameras and record the associated GNSS position and precise Coordinated Universal Time (UTC) time at the instant of image capture from either system. Image acquisition frequency was typically set on both cameras to record at a rate of one image every 2 seconds.

The GNSS data for the imagery were post-processed using nearby GNSS base stations from the National Park Service real-time GNSS network. The North American Datum of 1983 (NAD83) ellipsoid heights for the GNSS positions were transformed to orthometric heights referenced to LMSL using the National Geodetic Survey GEOID12B. Precise positions were derived for each image using the GNSS-synchronized image acquisition times and were written into the image Exif headers. Orthometric heights for the imagery from the boat-mounted camera were calculated using the 2.005-m fixed vertical offset between the GNSS antenna and the camera lens. Vertical positions were not recorded for the towed camera imagery; instead, the vertical measurements to the seabed from the camera-mounted altimeter were recorded in the image Exif tags.

Images from the baseline and resurvey datasets were subsampled to facilitate comparable classification of the images. For the baseline data, still frames from the underwater video were created by sampling one frame every 10 seconds using an automated routine in the Media Mapper software package. These baseline images were further subsampled by targeting selected sites where coral cover was previously observed (n=1,621; Cochran and others, 2007a, 2007b; Gibbs and others, 2007) but does not fully exclude other predominant benthic habitat types such as sand and volcanic pavement. For the resurvey data, to ensure a more consistent comparison with the baseline imagery, images from the surface camera were excluded from classifications and only the images from the resurvey drop camera system were used. The images were subsampled to select one image every 10 seconds as an initial step to reduce the large number of images from the same location. The resurvey images were further sub-sampled by only keeping images with GPS positions within a 5-m radius of a baseline image location. Finally, if multiple resurvey images fell within a 2-m grid space, only a single image was kept for classification (n=3,332).

Percent live coral cover (referred to hence forth as “coral cover”) and dominant bottom type classifications of selected baseline and resurvey imagery were performed manually by a single specialist with a background in ecological studies. The open-source software, Label Studio version 1.4 (Tkachenko and others, 2020), was used to enable efficient classification of the baseline and resurvey images using a custom, two-variable classification scheme for live coral cover and dominant bottom type (described in more detail in the “Image Classification of Live Coral Cover” section). The software provided the functionality to sequentially display each image to the specialist and allowed for the use of keyboard hot keys to rapidly assign classification labels to each image. At the end of each classification session, a tabulated comma delimited text file was generated containing the image filenames, geographic coordinates, and classification labels, which were then imported to Python for data analysis.

Coral cover is a commonly measured attribute of coral community composition and was the primary metric used to assess change in coral abundance at the Kona Coast parks. Coral cover was determined visually by assessing the live coral cover as a fraction of total coral (live and dead) cover in each image relative to reference diagrams of percent cover after Terry and Chilingar (1955). Cover estimates were made in multiples of 10 from 0 to 100 percent. A total of 4,282 baseline and resurvey images were visually classified for percent cover. A total of 783, 2,798, and 701 images were classified for PUHE, KAHO, and PUHO, respectively. The classified baseline and resurvey images and associated classifications are available in a USGS data repository (Logan and others, 2025).

The baseline and resurvey imagery differed in resolution, field of view, and camera orientations at each site. This unavoidably resulted in differing image footprints and scales between the surveys, meaning we were unable to ensure the same amount of reef area was covered and analyzed between sampling periods. To mitigate this, we averaged the coral cover by site before determining change in coral cover. We also relied on park-wide density distributions of coral to assess larger scale change.

A total of eight coral and non-coral categories were used to classify the dominant bottom type of the imagery. Each image was only assigned one dominant bottom type class (fig. 6). Coral dominant bottom type categories were selected based on the most common coral species occurring along the Kona Coast and included Montipora capitata, Pocillopora meandrina, Porites compressa, and Porites lobata. We distinctly classified P. compressa (finger coral) and P. lobata (lobe coral), but because of morphological genus it is possible that corals classified as either may have similarities across the different Porites species.

The non-coral dominant bottom type categories were macroalgae, coral rubble, sand, and volcanic pavement. The non-coral categories were selected based on 1) what common benthic habitat classes were previously classified at the three parks (for instance, sand and volcanic pavement; Cochran and others, 2007a, 2007b; Gibbs and others, 2007), and 2) the hypothesis that reef composition shifts from coral to non-coral cover after coral bleaching events. We hypothesized that macroalgae occurrence would increase between the two survey periods due to a coral-macroalgal phase shift that often occurs following bleaching events (Fukunaga and others, 2022). Furthermore, we hypothesized that coral rubble would increase in occurrence as bleaching events produce dead and bleached coral that can then break down and produce rubble.

A total of 2,047 images were assigned one of the 8 above dominant bottom type classifications and 2,235 images were assigned an “Unknown” dominant type classification (table 2).

Change in live coral cover for each park and site was determined by:

where

Sites that did not have at least two images classified for the baseline and resurvey were excluded from the coral change analysis (PUHE: 2 sites; KAHO: 6 sites; PUHO: 4 sites).

Because of the uneven distribution of the number of images within each dominant bottom type category (table 2), all eight dominant types were aggregated by park into either coral or non-coral prior to calculating percent change of these two groups by:

where

Unknown classifications were not included in the total number of classifications.

Percent change of individual dominant bottom types was also calculated, but M. capitata, macroalgae and volcanic pavement were excluded from the results because of low sample size (less than n=5) in either one or both survey periods (table 2).

The mean coral cover of PUHE was 60.4 plus or minus (±) 25.2 percent and 23.2±24.5 percent in the baseline and resurvey periods, respectively (table 4; fig. 7). There were no obvious spatial patterns of live coral cover or coral cover change between the baseline and resurvey (fig. 8). There was widespread decrease in live coral cover (park mean= −61.6 percent; table 3; figs. 7, 8), although there were three sites where positive changes in live coral cover were observed (fig. 8C).

The occurrence of coral cover between 0 and 40 percent increased, whereas the occurrence of coral cover greater than 40 percent decreased at PUHE between the baseline and resurvey periods (figs. 9A, 9B). Most of the site-specific coral cover change was negative, with a strong bias towards 100-percent loss in the distribution curve of percent change (fig. 9C).

Coral types were classified as the dominant type in 90.4 percent of the baseline imagery (table 2; fig. 9). P. compressa was the most common coral type, comprising 55.7 percent of the known dominant type classifications. P. lobata was also relatively common, making up 34.8-percent of the known baseline classifications. The most common non-coral type was sand. In the resurvey, coral types were classified as the dominant type in 80.4 percent and non-coral types were classified in 19.6 percent of the images. Again, P. compressa and P. lobata were the most common coral types in the resurvey at 53.6 percent and 26.0 percent, respectively. Sand continued to be the most common non-coral type, whereas rubble increased between the baseline (0 percent) and resurvey (7.0 percent).

PUHE experienced the greatest increase in non-coral of the three parks (105.2 percent; table 3; fig. 9F), and the greatest decrease of coral (−1.1 percent). This pattern was driven by an increase in the proportion of sand and rubble classification and a decrease in the proportion of P. lobata.

The mean percent live coral cover at KAHO was 28.7±24.9 percent and 25.3±24.7 percent for the baseline and resurvey periods, respectively (table 4; figs. 7, 10A, 10B). The variability of coral cover was higher at KAHO than PUHE and PUHO. In both the baseline and resurvey, the highest percent coral cover was found adjacent to Kaloko Fishpond and near the northern end of the park (fig. 10).

Live coral cover in KAHO decreased by 11.7 percent between the baseline and resurvey (table 3). Despite this loss, there were sites where an increase in coral cover was observed (fig. 10C), including many of the sites adjacent to Kaloko Fishpond and around the harbor. This positive change in coral cover was also reflected in the change of coral and non-coral types, where non-coral types decreased by 16.8 percent and coral types increased 22.3 percent (table 3).

Though there were similar distributions of coral cover between the baseline and resurvey periods at KAHO (fig. 11A), the lowest coral cover bin (0 to 10 percent) showed the greatest increase in occurrences, whereas coral cover bins with percentages greater than 10 percent varied in gains or losses (fig. 11B). Most of the site-specific coral cover change was negative, with a strong bias towards 100-percent loss in the distribution curve of percent change (fig. 11C). The distribution of coral change at KAHO was more weighted towards positive change than PUHE and PUHO with the tail of the distribution extending to gains of 400-percent (fig. 11C). These large gains in cover were relatively rare (n=2) and may be driven by high uncertainties in the locational data of the baseline imagery or the heterogeneity of coral cover of a particular site.

KAHO was more characteristic of lower coral cover in the baseline and resurvey periods as evidenced by a higher proportion of non-coral type classifications (table 2; figs. 11D–F). The most common dominant bottom type in the baseline survey was sand (37 percent), but rubble was significantly more common at KAHO than the other two parks across both surveys. In the resurvey, the occurrence of coral and non-coral types was nearly equal with corals comprising 52.5 percent and non-coral types comprising 47.4 percent of the images. P. lobata was the most common coral type in both survey periods. KAHO was the only site where positive change of coral types and negative change of non-coral types was observed (table 3; fig. 11F).

The mean percent live coral cover at PUHO was 52.3±31.5 percent and 25.0±23.0 percent in the baseline and resurvey periods, respectively (table 4; figs. 7, 12). Live coral cover decreased by 52.2 percent between the baseline and resurvey (table 3; figs. 7, 12). Generally, live coral cover decreased across the entire park (fig. 12C).

Like PUHE, the occurrence of coral cover between 0 and 40 percent increased, whereas the occurrence of coral cover more than 50 percent decreased at PUHO (figs. 13A, 13B). Most of the site-specific coral cover change was negative, with a strong bias towards 100 percent loss in the distribution curve of percent change (fig. 13C). Despite overwhelming negative coral change across the park, there were some sites that experienced positive change in live coral cover.

Coral types were classified as the dominant type in 91.2 percent of the baseline images at PUHO (table 2; fig. 13). For the baseline survey, P. lobata was the most common bottom type (52.3 percent), followed by P. compressa (32.0 percent). Non-coral types were classified as the dominant type in 6.6 percent of the baseline images. In the resurvey, coral types were classified in 89.2 percent of the images and non-coral in 10.8 percent. P. compressa was the most common dominant bottom type in the resurvey (47.9 percent); however, P. lobata was still relatively common (40.7 percent). Rubble increased from 1.0 percent of all classified dominant bottom types to 8.7 percent (a nearly 800-percent change) between the baseline and resurvey.

The depth distribution of coral cover bins demonstrated high variability between the baseline and resurvey periods across all three parks (fig. 14). However, high percent cover bins became more concentrated at mid-depths (20 to 15 m) in the resurvey, especially at PUHE and PUHO.

Water depth had moderate or little influence on changes in live coral cover across all three parks (fig. 14). The relation between live coral cover change and water depth was significant at all three parks (probability value less than 0.005), but the relations were corroborated by low correlation values (R2<0.2), especially at KAHO and PUHO. At KAHO, increases in coral cover up to 400 percent were observed at certain sites. PUHE exhibited the strongest relation between coral change and depth (correlation value=0.48), where live coral cover loss was pronounced at shallower depths. However, the opposite was observed at KAHO, where coral loss was pronounced at deeper depths and large gains occurred at shallow depths.