Shifting Limitations to Restoration across Dryland Ecosystems in Hawai ʻ i

: Hawaiian dryland ecosystems are important for global biodiversity conservation and contain numerous species threatened with extinction. Over the past century, wildfre frequency and size have increased dramatically because of invasion by fre-promoting non-native invasive species, greatly threatening these ecosystems. Native species restoration is a tool that can disrupt the cycle of increased fre and invasion in lowland dry forest communities, but restoration prescriptions have not been studied systematically in other dryland plant communities. We examined the restoration of three Hawaiian dryland plant communities (a high-productivity Diospyros sandwicensis and Metrosideros polymorpha lowland dry forest (HP), a moderate-productivity Myoporum sandwicense and Sophora chrysophylla dry forest/woodland (MP), and a low-productivity Dodonaea viscosa shrubland (LP)), using a community-assembly framework to understand the abiotic and biotic constraints to species establishment and growth in each community. Because active restoration methods are often needed, at both high and low levels of productivity, we also examined restoration treatments and outcomes across the three sites, which spanned a gradient of rainfall and substrate age. At each site, we used the same factorial feld experiment with three factors: habitat quality (high or low), weed control (yes or no), and species addition (none, seeding, or outplanting). Outplants (cohort 1) and seeds were added in the winter of 2009–2010, and outplants were added again in March 2011 (cohort 2). Dispersal limitation was apparent at the LP and HP sites, but was not observed in the MP site, which had, overall, greater native diversity and abundance. Outplant survival was greater in high-quality habitats at the HP site, likely due to reduced abiotic stress. Invasive species were found in greater abundance in certain types of microsites at the LP and MP sites, suggesting that shade or topography can be used to plan restoration and weed-control activities. Overall, active restoration methods improved restoration outcomes at the high- and low-productivity sites, and less so at the moderately productive site. Weed removal and outplanting were effective restoration prescriptions at the LP and HP sites, and habitat quality could also be used to increase survival at the HP site. Active restoration could be a lower priority for moderately invaded, moderate-productivity communities, which have the capability to maintain a native ecosystem state. Author Contributions: Conceptualization, S.C. and E.J.Q.; methodology, S.B., A.U., E.J.Q. and S.C.; investigation, S.B., A.U. and E.J.Q.; data curation, S.B., A.U. and E.J.Q.; writing—original draft preparation, E.J.Q.; writing—review and editing, S.B., A.U., E.J.Q. and S.C.; funding acquisition, S.C. E.J.Q.


Introduction
Dryland ecosystems in Hawai ʻi are critically important for the conservation of biodiversity and species at risk of extinction. There are more endangered species per square mile on these islands than any other place in the US, and most of these species, and the ecosystems in which they live, are found nowhere else in the world. Hawai ʻi is home to nearly one-third of all federally listed threatened and endangered species and almost half of all listed plants. The total number of listed plant species in Hawai ʻi has increased by prescriptions that are effective in lowland dry forests are also effective across the three dryland communities, or whether restoration prescriptions should vary across different dryland ecosystems. We established the same restoration feld experiment simultaneously across three dryland communities, which applied treatments to address the major barriers to restoration (i.e., invasion of non-native grasses; lack of native species seed and/or propagules, and absence of suitable microhabitat for native species). We hypothesized that active restoration in the form of invasive plant removal would have its greatest impact at the highest-productivity site, active restoration in the form of planting seedlings of native species and fnding suitable microhabitats would have its greatest impact in the lowestproductivity site, and the moderate-productivity site could be restored more passively.

Study Sites
We identifed native dryland communities that represent a large proportion of dryland ecosystems across Hawai ʻi, and that are critical for restoration due to their degree of grass invasion and strategic location for fre suppression. The following three community types were selected: Lowland dry forest high productivity (HP) dominated by Diospyros sandwicensis (lama) and Metrosideros polymorpha ( ʻohī ʻa lehua), Myoporum-Sophora dry forest/woodland moderate productivity (MP) dominated by Myoporum sandwicense (naio) and Sophora chrysophylla (mamane), ¯and Dodonaea viscosa (a ʻali ʻi)-dominated shrubland low productivity (LP). Two of these communities were within the Pohakuloa ¯Training Area (PTA) and one was found at Pu ʻu Wa ʻawa ʻa (PWW; Figure S1).
Pu ʻu Wa ʻawa ʻa is a state forest reserve located on the western or leeward side of the Island of Hawai ʻi on the northern fank of Hualalai ¯volcano ( Figure S1). Lavas of Hualalai ¯are primarily Holocene in age, and the study area occurred on a 1500-3000-yearold ʻA ʻā fow [39]. The study site was at approximately 600 m elevation with a mean annual temperature around 20 • C and 480 mm annual precipitation on average (Table S1). Botanical records for Pu ʻ ʻ ʻ ʻ u Wa awa a date back almost 100 years. Native plant communities in this zone are among the most diverse in Hawai i, containing many rare and endangered species [40,41]. These woodlands have been greatly damaged by fre and feral animals during the past 150 years [41]. The lowland dry forest (LDF) site occurred within PWW with D. sandwicensis and M. polymorpha as the dominant tree species. Other less common trees include Psydrax odorata (alahe e), Erythrina sandwicensis (wiliwili), Reynoldsia sandwicensis ʻ ( ohe makai), and Colubrina oppositifolia (kauila). Pennisetum setaceum (fountain grass) was ʻ the dominant invasive species at the site. Due to the higher rainfall of this region, this site was considered to be high productivity. The Pohakuloa ¯Training Area, is the U.S. Army's largest training area in the Hawaiian archipelago located on the Island of Hawai ʻi, encompassing 44,045 ha in the saddle between Mauna Loa and Mauna Kea volcanoes ( Figure S1). Because of PTA's position largely above the inversion layer, its rainfall is considerably lower than the LDF zone at lower elevations. The average annual precipitation across the installation is 370 mm with the highest rainfall accruing in the winter months. The annual mean temperature is about 16 • C. Vegetative cover varies from barren lava to dense shrub and forest ecosystems but is collectively classifed as Subalpine Dryland [42]. The vegetation found in a given area is largely a function of the age of the lava fow on which it grows, as nutrient availability varies predictively with substrate age and type [43,44]. This area is biologically rich encompassing 24 vegetation communities. Twenty-two rare plant species have been documented with eleven of those listed as federally endangered and nine as species of concern [45]. Numerous rare and endangered fauna are also found at PTA. Critical habitat and areas of special concern in terms of their botanical composition and or habitat value for rare species have been designated at PTA [46]. The MP dry forest/woodland site occurred within PTA at Kipuka Alala at 1600 m elevation on a 3000-5000-year-old pahoehoe ¯substrate from Mauna Loa. The LP Dodonaea viscosa shrubland site occurred within PTA at Kipuka Kalawamauna ā t 1622 m on a 14-65 thousand-year-old a'a substrate from Mauna Kea [47][48][49]. Average rainfall in the four years prior to this study at the MP and LP was 404 mm and 297 mm, respectively (Table S1). Common invasive species at both sites include P. setaceum and Senecio madagascariensis (freweed). Delairea odorata (cape ivy) also occurs at the MP site.
The three sites represent a gradient of productivity based on rainfall and substrate age, as nutrient availability, especially phosphorous, declines as substrates age [43,44]. The HP site has the highest rainfall and youngest substrate; the MP site has intermediate rainfall and an intermediate substrate age; the LP site has lower rainfall and the oldest substrate (Table S1). During the present study, rainfall was exceptionally low due to historic drought conditions (Table S1). Aboveground carbon stocks, an indicator of primary production, were estimated to be 1090 Mg C ha −1 for the HP site and 1 Mg C ha −1 for the MP site [50]. Carbon stocks for the LP site were not estimated but are expected to be lower than the MP site due to a lack of tree canopy cover.

Experimental Design and Data Collection
We used a factorial feld experiment with three factors: habitat quality (high or low), weed control (yes or no), and species addition (none, seeding, or outplanting). Within each community type, we established six 15 x 25 m blocks with similar elevation and substrates in fenced ungulate exclosures. Three of these blocks represented high-quality habitat patches and three represented low-quality/degraded habitat. We used remote sensing data and imagery analysis to locate high-quality and degraded habitat areas. For the HP and MP sites, we used LiDAR data to fnd high-quality areas with dense tree canopy cover (upper 25th percentile of canopy cover for each ecosystem), and we located low-quality blocks in nearby open areas. Areas under trees were considered higher quality due to shade that reduces solar stress and inhibits invasive grass growth as well as lower wind speeds which can be desiccating to young seedlings. Due to the low stature of vegetation in the LP, we used analysis of microtopography to locate high-and low-quality areas. Higher-quality areas were protected from desiccating winds and were in topographic depressions where soil accumulates more quickly and moisture is retained for longer periods [36].
Each block was divided into six 4.5 x 4.5 m plots with a buffer of at least 5 m in between. Experimental weed control and species addition treatments were applied in a fully factorial combination to the six plots in each block. In 2009, the plots receiving weed control had all invasive species removed by spraying 2% glyphosate herbicide, followed by mechanical weed trimming and hand clearing two weeks later. Plots were monitored regularly for the duration of the study and weeded as necessary.
Our frst outplanting (cohort 1) of native species occurred in November and December 2009 (Table S2). Outplants were grown from seed collected from each site in a growing medium of one part cinder and three parts Sunshine Media Mix #3 (Sun Gro Horticulture Distribution Inc., Agawam, MA, USA) at the Institute of Pacifc Islands Forestry nursery (Hilo, HI, USA). Outplants were grown for three to six months, depending on the species, and were hardened in the feld at each site for at least two weeks before outplanting. The size of plants varied but was approximately 20 cm tall on average. The outplants were planted in a random order in a seven-by-seven grid, equally spaced within each plot. The species selected for each site represented common species occurring in nearby intact plant communities. As is typical with large-scale restoration projects, the number of individuals of each species varied depending on propagation success of the species (Table S2). While this led to an unequal number of each species planted it refects a standard practice in restoration used to maximize the total number of seedlings added to a site. Seeds of all species were hand broadcast to seeded plots in January 2010. The number of seeds added per species varied depending on seed viability, determined with tetrazolium testing and germination trials; therefore, more seeds were added for species with lower viability to equalize germination potential (Table S2). Due to low survival of cohort 1 outplants a second outplanting (cohort 2) of 20 D. viscosa seedlings per plot occurred in March 2011. This species was selected due to its common abundance at all three sites. Plants were grown by Future Forests Nursery (Kailua-Kona, HI, USA) from seeds collected at each site. Plants were planted evenly spaced in each plot, and evenly spaced with any existing outplants from cohort 1.
The effectiveness of restoration treatments was measured as survival of outplants; increases in native species cover, richness, and diversity, and decreases in invasive species cover. Outplants in cohort 1 were surveyed for survival in Spring 2010 during April, Fall 2010 during August, and Spring 2011 in February. Outplants in cohort 1 and 2 were surveyed in Fall 2011 in October, and Spring 2012 in February and March. Species cover estimates were collected from three 4.5 m parallel transects in each plot, point sampled every 50 cm with a 1.9 cm diameter pole. The percent cover of each species was calculated as the number of points it occurred at in the plot divided by the total number of sampling points. Only understory species were sampled and large canopy trees were excluded. We calculated total native and invasive cover by summing the percent cover of all species in each category, native species richness (S) as the number of native species recorded, and native species diversity as the inverse Simpson's index (1/D).

Statistical Analysis
Due to low survival of cohort 1 at all sites (Table 1) survival data from only the Spring 2010 census were analyzed. For the HP site, survival of each outplant from cohort 1 was analyzed with a generalized linear mixed effects model with a binomial distribution. The model had fxed factors for species identity, habitat quality, weed control, and species addition, and random factors for plot and block. All two-way interaction terms were included in the model. Due to very low survival, the species M. polymorpha was excluded from the analysis. Due to even lower survival of cohort 1 at the LP and MP sites, survival of each species (presence/absence within a plot) was analyzed with a generalized linear mixed effect model with a binomial distribution using the same fxed and random effects and interaction terms as for the HP site.
Survival of cohort 2 was calculated as the proportion of D. viscosa outplants surviving in each plot and was analyzed separately for each site. Survival was analyzed with a linear mixed effects model with fxed effects for sampling date, habitat quality, weed control, and species addition, and random factors for plot and block. All two-way interaction terms were included in the model.
Total invasive cover, total native cover, S, 1/D, and E were analyzed separately for each site with a linear mixed effect model. The model included fxed factors for date, habitat quality, weed control, and species addition, and random factors for plot and block, as well as all two-way interaction terms. All analyses were conducted using R and the lme4 package [51,52]. Least square means were calculated for pairwise comparisons between levels of fxed factors and interaction terms using the package emmeans [53].

Outplant Survival
Survival of the frst cohort of seedlings was low, especially in the MP and LP sites (Table 1). At the HP site, cohort 1 survival of Chenopodium oahuense was greatest among the species (Table S3, Figure 2), and survival of all species declined over time ( Figure S2a). There was a trend of greater survival in high-quality habitats, and survival of C. oahuense, S. fallax, S. chrysophylla, and O. anthyllidifolia was 1.35, 1.55, 2.96, and 1.94-times higher in high-quality habitats, respectively; however, the difference was signifcant only for S. chrysophylla (Table S3, Figure 2). When cohort 1 survival data were analyzed for the MP and LP sites, no signifcant treatment effects were found at either site (p > 0.05). By the fnal survey in 2012, survival of cohort 2 ranged from 15% to 100% among plots, with all plots having three or more of the twenty planted individuals surviving. Survival of cohort 2 declined over time at all three sites and was lowest overall at the HP site ( Table 2, signifcant date factor; Figure S2b). There were no other signifcant treatment effects on survival of cohort 2 at any site.

Percent Cover
Non-native cover was highest in 2011 in the HP and MP sites and was highest in 2012 in the LP site (Table 3 signifcant year factor). Weed control reduced non-native species cover at all sites (Table 3 signifcant weed control factor; Figure 3C). The MP site had the greatest increase in non-native cover over time ( Figure 3A). Interestingly, the interaction of habitat quality with the weed control treatment was different between the MP and LP sites (Table 3, Figure 4). In plots without weed control in the MP site, non-native species cover was 1.5-times greater in high-quality, compared to low-quality, treatments; however, in the LP site, it was 1.8-times greater in low-quality, compared to high-quality, treatments ( Figure 4, Table 3 signifcant habitat quality x weed control interaction). There were also differences in which invasive species were abundant in each site; S. madagascariensis was the most abundant invasive species in the MP site, whereas P. setaceum was the most abundant species in the LP and HP sites (Table S4). In the plots without weed control in the LP site, non-native species cover was 53% in the no-species-addition treatment and 58% in the seed-addition treatment, and it was reduced to 41% in the outplanting treatment (Table 3 signifcant weed control x species addition interaction; Figure S3).   Native cover was lowest in the HP site and highest in the MP site, and it increased over time in the HP and LP sites and stayed the same in the MP site (Table 3 signifcant year factor, Figure 3B). In the MP site, the seed-addition treatment had slightly higher native cover than the other treatments; however, this was likely due to the native cover conditions at the start of the experiment (Table 3, Figure 3D). The infuence of the experimental treatments on native cover was most evident in the HP site, where native cover was approximately three-times greater in outplanted treatments compared to no-addition treatments, in 2011 and 2012 (Table 3 signifcant species addition x year interaction, Figure 3D). Plots receiving weed control in the HP site had low native cover in 2010, but in 2011, the native cover was greater than unweeded plots; however, by 2012, native cover was equal among weed control treatments (Table 3 signifcant weed control x year interaction, Figure S4).

Native Richness and Diversity
Native species richness ranged from zero to six species per plot. Richness and diversity were lowest at the HP site and highest at the MP site ( Figure S5). Outplanting signifcantly increased richness and diversity at the HP and LP sites, but not the MP site (Tables S5 and S6 signifcant species addition x year interaction; Figure 5). While there were some signifcant treatment effects on richness and diversity at the MP site, they were diffcult to interpret in the context of this study.  A,B) and HP (C,D) richness and diversity by species addition. In the LP site, the outplanting treatment on the fnal sampling date had the highest species richness. In the HP site, the outplanting treatment on the last two sampling dates had the highest richness. HP native diversity was greatest in the outplanting treatment over all sampling dates. Letters indicate signifcant differences among factor levels (p < 0.05).

Discussion
Removing or controlling invasive species is often the frst step in the restoration of invaded ecosystems and its effectiveness has been demonstrated previously in LDF restoration at other sites [29,30,33,34,[54][55][56]. Our weed control treatment was successful in reducing non-native species at all sites but had the greatest impact at the LDF/HP site, where the difference in invasive cover between weeded and non-weeded treatments was greatest. Therefore, if the goal of restoration is to reduce the abundance of invasive species and fne fuels, weed control would be successful across dryland communities; however, we found that if the goal is to increase native biodiversity, more intensive restoration is needed.
The restoration prescriptions that were the most effective at increasing native biodiversity varied across sites ( Table 4). As predicted (Figure 1), outplanting, the most expensive and labor-intensive treatment, had positive impacts on restoration, at both the low-and high-productivity sites and had more modest effects at the MP site. At the LP and HP sites, outplanting increased native species richness and diversity, although the increase in richness due to outplanting was modest, ranging from approximately 0.67 to 2.1 species more than sites receiving no outplanting or seed addition ( Figure 5 fnal sampling date). At the LP site, outplanting also reduced non-native cover, and at the MP and HP sites, it increased native cover but only by a small amount. Table 4. Site-specifc pathways conceptualizing relevance to future management based on restoration metric, treatments applied, and initial site conditions. Restoration success is considered as outplant survival, a decrease in non-native cover, or an increase in native cover, species richness, or species diversity. A (+) symbol indicates a signifcantly positive response between the treatment and the desired outcome; a (++) symbol indicates a very large, signifcant positive response; a (-) symbol indicates a signifcant negative response. Responses for the seed-addition treatment were not signifcant and are not included in the table. The larger increase in native richness with outplanting at the HP site is likely due to the greater survival of cohort 1, which included several species, compared to cohort 2, which included only D. viscosa. There was no beneft of outplanting on richness or diversity in the MP site, likely because this site had higher native richness and abundance than the other sites before restoration. Further research testing the establishment of a variety of species with a variety of functional traits would improve our understanding of which species are the most useful for restoration activities in different dryland communities.

Initial
Seeding has been used with moderate success in prior LDF restoration experiments but had no effect on restoration at any of our study sites [29,30]. This result could be explained by the extreme drought conditions that occurred during the study. In the year that seeding occurred, rainfall was 44, 68, and 105 mm at the LP, MP, and HP sites, respectively (Table S1). These values are 15-22% of the typical average rainfall and are among the lowest levels ever recorded at these sites. Seeding has been shown to be effective at increasing native cover in less stressful conditions [30]; therefore, seeding could be an effective restoration treatment in wetter years, but more study is needed to understand its impact across dryland ecosystems. However, the positive impact of outplanting but lack of effect of seeding suggests that there are biotic and abiotic flters during community assembly of these communities. First, the biotic flter of propagule limitation of native species was expanded by both treatments. The lack of impact of seeding suggests that addressing this flter alone was not suffcient and that other flters, such as seedling desiccation stress or granivory, also limit the establishment of native plants in these communities.
It was interesting that cohort 1 had the highest survival at the HP site and extremely low survival at the other two sites, whereas cohort 2 had the lowest survival at the HP sites compared to the other two ( Figure 3). The low survival of cohort 1 at the drier sites can be explained by the extreme drought conditions during the study. However, the lower survival of cohort 2 at the HP site is harder to explain. One possibility is our use of volunteers to aid with planting at this site, but not the other sites due to the limited access of civilians in the Army training area. Volunteers were trained and supervised by research staff, but it is possible that survival was lower if volunteers were not as careful as paid employees during planting activities. The use of volunteers in restoration planting is ubiquitous and cost effective, but few studies have reported associated survival results, indicating an area that needs more study [57][58][59].
Outplanting in high-quality habitats led to greater survival rates for several species at the HP site (Figure 2), but we did not fnd strong associations between habitat quality and restoration outcomes at the other sites. It is possible that the drought conditions during the study reduced any fne-scale abiotic differences between high-and low-quality habitats as we defned them. These differences could be more apparent during years with more typical rainfall. Therefore, it is important to continue studies of various microsite conditions and how they affect plant establishment across years with varying conditions. However, we did fnd that without restoration, invasive species cover was higher in high-quality habitats at the LP site but was higher in low-quality habitats at the MP site. This result occurred due to how we defned habitat quality at each site and due to the identity of the dominant invasive species. At the LP site, P. setaceum was the most abundant invasive species, and its occurrence was greater in more extreme topographic microsites. This species is a C4 grass that is adapted to warmer temperatures and grows on steep hillsides; therefore, it may be able to take advantage of space in these more stressful areas, where native growth is reduced [36]. The dominant invader at the MP site was S. madagascariensis, a low-statured forb occurring in open areas. This species was less abundant in high-quality sites, under the canopy of existing trees, where it potentially was light-limited. Therefore, our designation of shadier microsites as "high quality" is somewhat species-specifc, especially when applied to invasive species that are adapted to high-light environments. However, this result supports the pattern also found in LDF restoration, where competition with invasive species is reduced in shadier microsites with greater soil moisture [33]. Therefore, restoration could be planned strategically in shadier sites if competition with invasive species is a concern.
In summary, several important abiotic and biotic flters were identifed by replicating this study among three dryland plant communities. Native dispersal limitation was apparent at the LP and HP sites but was not observed in the MP site, which had overall greater native diversity and abundance. Outplant survival was greater in high-quality habitats at the HP site, likely due to reduced abiotic stress. Invasive species were found in greater abundance in certain types of microsites at the LP and MP sites, suggesting that shade or topography can be used to plan restoration and weed control activities. Overall, active restoration methods improved outcomes at the high-and low-productivity sites, and less so at the moderately productive site. Weed removal and outplanting were effective restoration prescriptions at the LP and HP sites, and habitat quality could also be used to increase survival at the HP site. Active restoration could be a lower priority for moderately invaded, moderate-productivity communities, which have the capability to maintain a native ecosystem state.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/ 10.3390/su14095421/s1, Table S1. Annual rainfall totals for each site during the years of the study and the average of the four years prior to the study, Table S2. Number of cohort 1 outplants and seeds added to each plot at each site, Table S3. Generalized linear model results for cohort 1 at the HP site, Table S4. Mean abundance of common understory species in each site, Table S5. Linear model results for analyses of species richness, Table S6. Linear model results for analyses of native diversity, Figure S1. Map of research sites on the Island of Hawai ʻi, Figure S2. Outplant survival over time for cohort 1 and cohort 2, Figure S3. LP non-native cover by weed control and species-addition treatments, Figure S4. HP native cover by weed control and date, Figure S5. Richness (A) and diversity (B) by date among sites.