A review of predator exclusion fencing to create mainland islands in Hawaiʻi

Fence design, purpose, and maintenance

We collected information on the fence itself (length, area enclosed, materials used, type, year completed, and location), on the project purpose (which species were protected), the habitat type protected, the funding source, and whether the project was for protection of existing resources or mitigation required under the United States Endangered Species Act (ESA), i.e., to compensate for negative impacts to protected species that occurred or were expected to occur elsewhere.

We compiled information about maintenance issues related to fence materials or construction for all completed fences. This included construction defects, materials defects, design problems, corrosion, environmental impacts, and events such as extreme weather or tree fall. We examined the cumulative information to search for trends related to material type, construction, environment, or fence component.

Predator removal and incursions

We collected information on the mammal species present at each site prior to fence construction, which species were targeted for removal, the methods used to remove them, and whether each species was successfully eradicated. We considered a species to have been eradicated if it was not detected with at least two detection methods for more than four months. While this is shorter than the two-year standard used in larger island-wide eradications (Parkes, Fisher & Forrester, 2011), the small size of many of these sites allowed a higher detection probability compared to large sites. We also gathered information on mammal incursions that occurred after a species had been eradicated, including whether they were isolated, infrequent events, or ongoing and chronic (i.e. happening multiple times without appearing to establish a breeding population), and whether an incursion resulted in re-establishment of a breeding population (i.e. a reinvasion). We used six months as the definition of chronic because it typically takes 1–2 months to identify that an incursion has occurred, the source of the incursion, and another 2–4 months to determine if breeding occurred and to re-eradicate anything that had re-invaded.

Monitoring of outcomes

Extensive biological monitoring was conducted within completed fences that protected existing populations of native species, primarily on the reproductive success of the target native bird species. We did not include two fences (Hiʻi and Kilauea Point National Wildlife Refuge; KPNWR) that were only completed in 2023, because outcome monitoring efforts were not yet complete.

Methods for monitoring wedge-tailed shearwaters (Ardenna pacifica), Laysan albatross (Phoebastria immutabilis), and other ecosystem parameters can be found in Young et al. (2009), VanderWerf et al. (2014), Young et al. (2012) and Young et al. (in press) and were used at both Kaʻena Point and Kuaokala. Briefly, this included monitoring of hatching, fledging, and reproductive success for Laysan albatross, and measuring overall reproductive success for wedge-tailed shearwaters. Methods used for monitoring waterbirds at Honouliuli can be found in Christensen et al. (2021).

Of the completed fences where the purpose was mitigation to create a new seabird colony through either social attraction and/or translocation (VanderWerf et al., 2023a; VanderWerf et al., 2023b), four of seven had been erected long enough for results to have been achieved: Nihoku on Kauaʻi, two sites at Makamakaole on Maui, and James Campbell National Wildlife Refuge on Oʻahu. At the other three sites on Kauaʻi, social attraction had been conducted for only two years, which is not enough time to determine if the projects were successful because many seabirds do not breed until they are 3–5 years old. We excluded them from analysis for this reason.

Fence purpose and species protected

All 18 existing or planned full predator exclusion fences in Hawaiʻi were built to protect native birds, including 15 to protect seabirds and three for wetland birds (Table 1). Ten of the fences for seabirds were designed to protect Newell’s shearwater (Puffinus newelli), Hawaiian petrel (Pterodroma sandwichensis), or band-rumped storm-petrel (Oceanodroma castro), which are the only three seabirds listed under the ESA in Hawaiʻi. Only one completed fence (Honouliuli) was built to protect endangered waterbirds, though two more planned fences are for waterbird protection (Kanaha Pond and James Campbell National Wildlife Refuge boundary fence). As of 2023, no fences had been built for the protection of endemic forest birds and only one fence, Kaʻena Point, had been built with recovery of native plants listed as a co-objective to protecting seabirds.

Twelve of the 18 fences (67%) were built to protect existing bird populations, and six fences 33%) were built for mitigation purposes to compensate for impacts that occurred or were expected to occur to species listed under the ESA. Of the six mitigation fences, all were for seabirds, of which five used social attraction and one used translocation; none of the mitigation fences protected existing bird populations.

Six of the 18 fences were peninsula-style with open ends and 12 of the fences were complete enclosures. The peninsula-style fences were either in wetland habitat or coastal shrub and all of them terminated at either a cliff face or steep, rocky shoreline that was intended to act as a natural, if semi-porous, barrier to predator ingress. There were no peninsula-style fences in montane environments. Of the 12 fully enclosed fences, two were in coastal shrub and the remaining 10 were in montane environments above 500 m in elevation.

Of the 18 fences, 11 (61%) were exclusively funded by private foundations or landowners, four (22%) were a combination of private and federal funding, and three (17%) were entirely federally funded. Nine of the fences were on state land (50%), five (28%) were on federal land (all national wildlife refuges), and four (22%) were on private land.

Fence design and materials

Completed fences ranged in length from 304–3,200 m and enclosed 1.2–101 ha. Planned fences ranged in length from 482–4,877 m in length and will enclose 2.5–640 ha. The design of the first predator exclusion fence at Kaʻena Point Natural Area Reserve (NAR) is similar to the design used in New Zealand (Day & MacGibbon, 2007; Young et al., 2012). The two Maui fences also used a variant of this design, but more recent fences used a variety of designs that differed in several ways, including mesh material and design, post material, and hood shape (Table 2). The overall design used for nine of the completed fences is shown in Fig. 2.

Fences in Hawaiʻi have used two mesh types (welded wire panels or rolls, and mini chain link rolls) and three mesh materials (stainless steel, galvanized steel, and PVC-coated galvanized steel). Fences built of welded wire consisted of two sections; an upper section that comprised most of the vertical fence, and a lower section that formed the underground skirt, which were joined with a horizontal metal bar. In fences composed of two mesh sections, the mesh type and material were the same for the upper and lower sections, but at Honouliuli and JCNWR, repairs to the lower section were made with a powder-coated stainless steel mesh. Twelve fences (67%) used a mini chain link design that came in 10 m rolls and could be contoured to sloping terrain. All but one of the mini chain link fences used 304 stainless steel as the material. Welded wire mesh was typically less expensive than mini chain link and could be constructed in panels, but the inflexibility made it difficult to work with in areas with slopes and thus was used in only six (33%) of the fences. Of the welded mesh fences, one was stainless steel, two were galvanized steel, and the remaining three (all planned fences) used a PVC-coated galvanized steel.

The three earliest fences in Hawaiʻi used the hood design from New Zealand shown in Young et al. (2012), which extends diagonally downward from the top of the fence and has a curled lip on the outer edge. The 15 later fences (83%) used a modified design in which the hood sloped upward and then downward and had no curl at the outer edge (Fig. 2). Thirteen fences (72%) had a hood made of 304 stainless steel, four fences (22%) had a galvanized steel hood, and one fence had a powder-coated 304 stainless steel hood. The hood material was the same as the mesh material in most cases, but in the first fence at Kaʻena Point the mesh was 304 stainless steel and the hood was galvanized steel, and at three of the planned fences the mesh is PVC-coated galvanized steel and the hood is 304 stainless steel. The hood bracket material was always the same as that of the hood itself.

Post materials were either 13–16 cm diameter treated round wooden posts or five cm diameter stainless steel posts. Twelve fences (67%) used wooden posts exclusively, two (11%) used 304 stainless steel posts exclusively, three (17%) used a mix of metal and wooden posts depending on the specific terrain, and one used powder coated steel posts. Wooden posts were generally used in areas with soil substrates whereas metal posts were typically used in rocky areas.

Two of the completed fences (Kaʻena Point and Nihoku) each had a single culvert installed in ephemeral streams beds to allow for periodic water flow during periods of heavy rainfall. None of the existing fences crossed regularly flowing streams, and the fence lines were deliberately selected to avoid the use of culverts because they have been previously identified by several fence managers in New Zealand as the area with the highest potential for pest breaches. Culverts had a grate on the outside to prevent animals from entering, and at Nihoku, a grate was placed in the inside to prevent debris from entering the culvert since water flows from inside to outside the fence at that location. At two of the planned fences, Kanaha and JCNWR boundary, the fence will cross canals more than five m wide and up to several m deep and that have constant water flow, where it is not practical to use a culvert. At Kanaha, the fence is being built on existing bridges over the canals, and the steep, concrete banks of the canals are anticipated to limit entry by animals by swimming and climbing the banks. At the JCNWR boundary fence, the fence will extend a short distance into the canals, and intensive trapping will be done to intercept any animals that attempt to swim inside.

Maintenance

Maintenance issues were reported at 11 of the 12 completed fences. Most maintenance issues arose within the first year after fence completion and almost all had arisen by the end of the second year, with the exception of an extreme rainfall event. The fence components most commonly requiring maintenance were the skirt (N = 6/11 fences, 55%) and the hood (N = 4/11 fences, 36%; Table 3).

Several of these issues were isolated events (tree fall, extreme rainfall events, defective gate latch) that were easily fixed, but several were chronic issues that resulted in large and expensive repairs, such as mesh failures along the entire fence. The most common preventable issue was using a less corrosion-resistant material (such as galvanized steel) instead of 304 stainless steel, which created the fence-wide repairs needed at Makamakaole and Kuaokala. The Kuaokala fence, despite being over 500 m in elevation in a montane environment and almost 1 km from the ocean, was still subjected to salt spray because of strong prevailing winds and steep terrain that forced winds upward. The use of galvanized steel rather than stainless steel resulted in complete failure of the mesh within two years of construction along the side of the fence facing the prevailing wind direction.

The next most common maintenance issue was related to soil type and chemical interactions with the skirt, which happened at JCNWR and Honouliuli NWR. At JCNWR, the southern boundary of the fence was built onto a limestone substrate, which dissolved in rainwater to produce carbonic acid that greatly accelerated corrosion of the stainless steel mesh wherever it touched the limestone, and this was exacerbated by rainwater running down the fence. The limestone substrate also contained numerous small holes and natural tunnels, some of which could have extended under the fence and allowed rodents to enter. On the northern boundary of the same fence, the soil consisted of sand with a variable calcareous component, which also produced carbonic acid, but at a slower rate. This issue was largely corrected by adding cement over mesh sections that touched limestone or calcareous sand, though occasional small areas of corrosion still occurred on top of the cement. At Honouliuli on O’ahu, part of the fence was located on fine, salty silt that when flooded became anaerobic mud, resulting in the production of hydrogen sulfide by anaerobic bacteria, which was converted to sulfuric acid at ground level through an interaction of sulfur, salt, and air. The corrosion at Honouliuli was particularly rapid and occurred within the first year of fence construction and required replacement of mesh in the sections with anaerobic mud.

At two fences, digging by animals resulted in fence breaches. At Kahuama’a, pig rooting along the edge of the skirt resulted in gaps that allowed rodents to enter. This was fixed by adding cement or sod over the edge of the skirt to fill the gaps and discourage rooting. At JCNWR, rodents occasionally dug tunnels under the skirt in the sandy soil. This was corrected by adding an extension to the skirt in the sandy area so it was 80 cm wide and extended 30–40 cm underground (instead of 30 cm wide and 10 cm underground), which required uncovering the skirt and then re-burying it after the extension was added.

Culverts resulted in maintenance issues at two sites. At Nihoku, an extreme rain event in 2018 resulted in accumulation of debris over the grate, which caused the ephemeral stream to flood over the culvert, causing erosion and gaps between the culvert and mesh. This event required excavating some debris and repositioning the culvert. At Kaʻena Point, the culvert across an ephemeral stream caused no issues, but there was a pre-existing older culvert from the 1940s that during fence construction was thought to be blocked, and closer inspection after the fence was completed showed it to have an opening that had allowed rodents to enter the fence. This issue was detected using tracking cards that showed a “hotspot” around the culvert, and it was corrected by installing a grate over that culvert too.

Another maintenance issue was mixing of metal types that caused accelerated corrosion by electrolysis. At Kaʻena Point, the use of a galvanized hood with stainless steel mesh and screws resulted in electrolysis that severely corroded the hood within two years. Several other interactions occurred when metal fasteners (i.e. nuts, bolts, rivets, and screws) of a different metal type were used. While the Mokio fence has not yet been completed, the fence has already experienced corrosion, particularly at contact points between the hood (which is 304 stainless steel) and cut sections of the PVC-coated steel (B. Haase, pers. comm., 2023). At Kaʻena Point, another issue related to metal type was widespread fatigue and failure of aluminum rivets used to secure two sections of mesh to each other. Repeated thermal expansion and contraction associated with daily heating and cooling caused the rivets to fatigue and become brittle, and similar daily expansion and contraction of the mesh sections they connected put strain on the rivets that they eventually could not withstand, causing the rivet heads to pop off.

Predator eradication and incursions

Eleven of the 12 completed fences had completed eradications at the time of this publication; the KPNWR eradication was still underway and is not included in the results below. Cats were present at all 11 sites and were successfully eradicated from all of them. Cats removed themselves from nine (82%) of the sites by climbing over the fence, which is possible because the hood is directed outward, and if they remained, they were removed with a combination of live trapping and shooting. The largest fence where cats self-exported was 32 ha at Hiʻi on Lanaʻi.

Rats and mice were present in all 11 sites. All three species of rats were present in varying densities at each site with R. rattus being the most commonly detected species. Rats were targeted for removal at all sites and were successfully eradicated from eight of them (72%); successful projects used diphacinone bait stations placed 25–50 m apart and kill traps (Young et al., 2013; Young et al., 2018; Christensen et al., 2021, this article). Mice were targeted for removal in six of the 11 sites and were eradicated from five of those sites (83%) using the same methods as for rats. In 2023, the first aerial broadcast of rodenticide for rat eradication in Hawaiʻi took place within the Hiʻi fence on Lanaʻi and appears to have been successful for both rats and mice despite only rats being targeted.

Mongooses were present in six of the 11 fence locations (mongoose are absent from the islands of Kauaʻi and Lanaʻi); they were targeted and successfully eradicated at all sites with a combination of diphacinone bait stations, DoC 250 kill traps, and live trapping, depending on the site.

Pigs and axis deer (Axis axis) were each present at a single site, and they both were targeted for removal. Deer were removed from the Hi’i fence by targeted hunting. Pigs at KPNWR also are being removed by targeted hunting, and the eradication effort is still underway. It is too early to determine incursion rates for these species post-eradication.

Predator incursions occurred at all sites, but there was variation in the rate of incursions among predator species and among sites (Table 4). Two of the recently completed fences (Hiʻi on Lanaʻi and KPWNR on Kauaʻi) were excluded from incursion rate analysis because their eradications had just been completed or were still underway. There were no incursions by feral dogs at any site. There were a few cat incursions at peninsula fences, but none of the enclosed fences experienced a cat incursion and no sites have experienced a cat re-invasion, indicating that the fences are highly effective against cats.

Chronic mongoose incursions occurred at two sites, Honouliuli and Kaʻena Point. Mongoose incursion rates were low at Kaʻena Point, once every 217 days or about two per year, but persistent over time (Young et al., 2013). At Honouliuli, mongoose incursions were high at first, every 4–17 days (Christensen et al., 2021), when there was a corrosion problem with the fence skirt (see maintenance section above), but mongoose incursions stopped entirely after the problem was corrected by replacing the skirt. JCNWR experienced a single mongoose incursion of two individuals in 2019 during a period in which the skirt had holes due to corrosion; none have been detected inside the fence since the holes were patched. At Kuaokala, large scale failure of the mesh allowed at least three mongooses to enter the fence. At all sites that experienced mongoose incursions, the mongooses that entered the fence were trapped quickly, preventing reinvasion.

Rat incursions occurred in all fences. At fully-enclosed fences, incursion rates ranged from one rat in seven years (Nihoku) to 3.75 rats per year at the remaining four sites without reinvading. For the two peninsula-style fences with completed eradications, where rat incursions are expected, incursion rates ranged from one rat every 12 days to one every 56 days. Mouse incursions occurred at all sites and were most frequent. At some sites it was not clear if mice were ever eradicated, and their nearly continuous presence made it impossible to measure incursion rates. Four of the fully enclosed fences experienced full reinvasions.

At the most recently completed fence at KPNWR, seven motion detection cameras (Ridgetek branded cameras using animl detection software) were deployed at fence ends, gates, and cliff tie ins in places where the biosecurity risk was deemed to be high. The images were automatically sorted by artificial intelligence to delete images without animals in them, and to draw boxes around animals detected in the photographs. Images with animals flagged in them were then emailed to staff in real time to facilitate a timeline response.

Monitoring outcomes

In the completed fences that protected existing populations of native species (Kaʻena Point, Honouliuli, and Kuaokala), the response of native species to the removal of predators has been dramatic. At Kaʻena Point, reproductive success of wedge-tailed shearwaters rose from 0.29 young per nesting attempt before the fence to 0.50 after the fence, and the colony has grown dramatically in size (VanderWerf et al., 2014).

At Kaʻena Point and Kuaokala, hatching, fledging, and reproductive success of Laysan albatross were measured before and after fencing and subsequent predator removal. While hatching rates were comparable at both sites before and after predator exclusion fences and not significantly different, fledging rates increased by 25% from 0.60 ± 0.09 to 0.75 ± 0.04 chicks per pair per year (df = 1, N = 1,179; X² = 31.03, p < .00001) suggesting that chicks are more vulnerable to predation than eggs in this species. This resulted in an overall increase in reproductive success from 0.37 ± 0.05 to 0.43 ± 0.03 (df = 1, N = 1,984, X² = 9.21, p = .0024).

At Honouliuli NWR, Christensen et al. (2021) found that hatching success in Hawaiian stilts (Himantopus mexicanus knudseni) was significantly higher for pairs nesting inside the fence (0.83 hatching rate) than in pairs nesting in a nearby wetland with predator control but no predator exclusion fence (0.35 hatching rate). Additionally, a significantly higher number of eggs were laid per nest at the fenced site compared with the unfenced site. Together, this resulted in nearly three times the number of eggs hatched per nest inside the mammal-exclusion fence, compared with nests at the site with trapping alone.

Of the seven sites that were constructed for social attraction or translocation, sufficient time has elapsed since construction at four sites to evaluate whether they were successful. It should be noted that the success of social attraction is not related to the fence itself since birds are attracted to the sounds being played and decoys rather than the fence, but it is reported if it was listed as the purpose of building the fence. Of those four (two translocation and two social attraction) three have been successful in achieving their objectives of attracting birds to successfully breed at the site (VanderWerf et al., 2019; Learned et al., 2023; Young et al., 2023).