Resource Availability, Propagule Supply, and Effect of Nonnative Ungulate Herbivores on madagascariensis Invasion1

Erin J. Questad,2,4 Amanda Uowolo,3 Sam Brooks,3 Robert Fitch,2 and Susan Cordell 3

Abstract: Nonnative invasive herbivores can create complex biotic interactions by differentially feeding on native and nonnative invasive . The her­ bivores may act as enemies of nonnative and prevent them from becoming invasive, or they may facilitate invasion by having a greater negative impact on native plants compared with nonnative plants. It is also possible that within the same ecosystem nonnative herbivores could either facilitate or inhibit invasion under different abiotic or biotic conditions. In this study we experimentally in­ vestigated how abiotic (soil nutrients) and biotic (propagule density) conditions infuence the effect of invasive generalist herbivores on Senecio madagascariensis, an invasive plant species in Hawaiian dry forest plant communities. We used fenced exclosures to manipulate presence or absence of invasive ungulates (feral goats and sheep), and we used seed addition to manipulate propagule supply of S. madagascariensis. The experiment was replicated in a recently burned and an unburned site to examine how a resource pulse following fre may alter plant­ herbivore interactions. There were very few seeds of S. madagascariensis in the seed rain of both sites, and recruitment was four times higher when seeds were experimentally added, suggesting that S. madagascariensis is dispersal limited in this area. Recruitment of S. madagascariensis was fve times higher in the burned site compared to the unburned site, suggesting that increased resources promote recruitment. Recruitment was three times higher when herbivores were present compared to when they were excluded, but plants were much smaller when her­ bivores were present. We conclude that herbivores can alter S. madagascariensis recruitment, even during dry conditions, and that propagule availability infu­ ences where S. madagascariensis can become established.

Keywords: propagule limitation, fre, feral ungulates, Hawai‘i, Senecio madagascariensis

1 This work was funded by a grant from the Depart­ Nonnative invasive herbivores can create ment of Defense (DoD) Strategic Environmental Re­ complex biotic interactions by differentially search and Development Program to S.C., J. M. Thaxton, feeding on native and nonnative plant species. and G. P. Asner; and funding from the DoD Environ­ In the case of nonnative plants, the herbivores mental Security Technology Certifcation Program to may act as enemies, controlling plant popu­ E.J.Q., S.C., and J. R. Kellner. Manuscript accepted 2 June 2017. lations and preventing them from becoming 2 Biological Sciences Department, California State invasive. Alternatively, herbivory may have Polytechnic University, Pomona, 3801 West Temple a greater negative impact on native plants, Avenue, Pomona, California 91768. compared to nonnative plants, facilitating 3 Institute of Pacifc Islands Forestry, USDA Forest Service, 60 Nowelo Street, Hilo, Hawai‘i 96720. the invasion of nonnative plant species (Sim­ 4 Corresponding author (e­mail: [email protected]). berloff and Von Holle 1999, Richardson and Pyšek 2006, Simberloff 2006). It is also pos­ sible that within the same ecosystem non­ native herbivores could either facilitate or in­ Pacifc Science (2018), vol. 72, no. 1:69–79 doi:10.2984/72.1.5 (Includes online supplements) hibit invasion under different biotic or abiotic © 2018 by University of Hawai‘i Press conditions. Here, we investigate how abiotic All rights reserved conditions infuence the effect of nonnative,

69 70 PACIFIC SCIENCE · January 2018 generalist herbivores (feral goats and sheep) documented in Hawai‘i due to reduced defen­ on the invasive plant species Senecio madagas- sive traits in the native fora (Denslow 2003). cariensis Poir. Due to the presence of PAs in S. madagas- Senecio madagascariensis is native to south­ cariensis, it is likely that it is more resistant to ern Africa and has invaded areas of , herbivory than native species in Hawai‘i. , , Japan, and Hawai‘i (Sindel The abiotic conditions at the time of her­ 1989, 2009). It is a species of economic con­ bivore removal also could affect the response cern due to its production of pyrrolizidine al­ of invasive plant populations. In general, con­ kaloids (PAs) that are toxic to grazing animals. ditions of high resource availability can pro­ Senecio madagascariensis was accidentally in­ mote the spread of invasive species (Davis troduced to Hawai‘i Island in the early 1980s, et al. 2000). In addition, the removal of possibly through ground cover seed imported enemies may provide the greatest beneft to from Australia (Motooka et al. 2004). It has species that have high levels of resource use since become a target for eradication due to (Blumenthal 2006). Herbivore removal in a its ability to spread quickly to large areas of dryland ecosystem on Hawai‘i island led to in­ pastureland, its potential to reduce forage creased S. madagascariensis population growth yields, and its production of PAs that are poi­ and spread when the removal occurred during sonous to livestock (Sindel 1989, Motooka a higher rainfall year or time of high resource et al. 1999, Le Roux et al. 2006). Species in availability (Kellner et al. 2011). Here, we ex­ the genus Senecio are well defended by their plore whether enhanced resource availability PAs against many generalist herbivores, and following fre may also contribute to increased plants are capable of allocating more re­ invasion of S. madagascariensis, and whether sources toward PA production in their inva­ there are interactions between resource avail­ sive range, compared to their native range, ability and the removal of herbivores. We due to founder effects and evolution after in­ hypothesized that the removal of ungulate vasion ( Joshi and Vrieling 2005, Cano et al. herbivores would increase the population 2009). However, some herbivores, such as do­ growth of S. madagascariensis and that the mestic sheep and goats, can limit the growth plant invasion response to herbivore removal and spread of Senecio populations (Sindel would be greater in areas with higher resource 2009). For example, in a Hawaiian dry forest availability. We tested these hypotheses by dominated by Myoporum sandwicense and So- replicating ungulate fencing and S. madagas- phora chrysophylla (Myoporum-Sophora dry for­ cariensis seed addition treatments in a recently est) the removal of nonnative, feral goats had burned and unburned dry forest area on a positive effect on S. madagascariensis popula­ Hawai‘i Island. tion growth and no effect on native plant spe­ Species with high propagule densities are cies, suggesting that herbivore removal may likely to respond more rapidly to herbivore contribute to S. madagascariensis invasion and removal compared to species with limited that nonnative herbivores inhibit invasion propagule supply (Denslow 2003, Richardson (Kellner et al. 2011). and Pyšek 2006, Gurevitch et al. 2011). Na­ The effect of herbivores on plant invasions tive species in Hawaiian dryland ecosystems has received extensive study in the Hawaiian are primarily woody species with slow, con­ Islands, where native plant species evolved in servative growth that have a relatively lower the absence of ungulate herbivores and many propagule supply compared to an abundant, have a limited ability to defend themselves invasive, short­lived species such as S. mada- against native herbivores and pathogens or in­ gascariensis. Propagule supply was found to troduced herbivores like feral pigs, goats, and be important for determining the invasion sheep (Goergen and Daehler 2001, DeWalt rate of Senecio vernalis, a related species (Erf­ et al. 2004). Generalist herbivores may have a meier et al. 2013). We examined S. madagas- greater negative effect on native, compared to cariensis population dynamics in a site where nonnative, species if the nonnative species its abundance was low (less than 2% cover). It have superior defensive traits, a pattern well is thought that its abundance is kept low at Herbivores and Senecio madagascariensis Invasion · Questad et al. 71 this site by ungulate herbivory (Kellner et al. Myoporum sandwicense (Naio) and has an 2011); however, the possibility of propagule understory of shrubs, grasses, and herbaceous limitation in this location has not been tested. species (Shaw and Castillo 1997). Soils of We used an experimental seed addition to both sites are Keekee ashy loamy sand, on 0 to test whether S. madagascariensis was propagule 6% slopes (U.S. Department of Agriculture limited, and how propagule availability inter­ 2016). The substrate for both sites is of Pleis­ acted with herbivore removal. We did not tocene or Holocene origin from Mauna Kea expect S. madagascariensis to be propagule (Wolfe and Morris 1996). This community limited, but we expected invasion rates in re­ has a high value for conservation of critical sponse to herbivore removal to be faster when habitat for the single extant population of the propagule densities were increased through endangered bird Loxioides bailleui (Palila) seed addition. (Pratt et al. 1998). The area of our study is one of the last The Burned site was in a 1,300 ha area remaining tracts of Myoporum-Sophora dry where a high­intensity wildfre occurred in forest in the world and is designated as “criti­ August 2010. The fre was set by arson and cal habitat” for the endangered Palila honey­ appeared to destroy almost all of the vegeta­ creeper, Loxioides bailleui (Pratt et al. 1998). tion, including mature native trees. The ex­ Due to fres and other disturbance the eco­ perimental site contained 20 individual study system is in need of restoration, and invasive plots within a 100 × 150 m area centered on ungulates are being removed from targeted the location 19.734792° N, −155.504697° W. areas. Thus, understanding how abiotic and The Unburned site was of similar size and biotic conditions infuence the impact of non­ was located in an adjacent area that was not native herbivores on plant invasions can guide affected by the fre, centered on the location future ungulate management and restoration 19.731533° N, −155.507600° W. This site efforts in this critically endangered ecosystem was 300 m from the edge of where the fre oc­ and others where S. madagascariensis is a man­ curred and was 470 m from our Burned site. agement concern. We refer to this site as “unburned,” but due to the historical presence of infrequent fres in this region there is a possibility that this site materials and methods burned some time in the past (Kinney et al. Study Sites 2015). The study sites were located within the Experimental Design Phakuloa Training Area (PTA), a U.S. De­ partment of Defense training installation, in a We replicated the same experimental design subalpine region between the three highest in the Burned and Unburned site, using 20 volcanoes on the island of Hawai‘i (1,300 – 1 m × 1 m study plots in each site. Within 2,600 m elevation). Mean annual precipita­ each site, the plots were grouped into 5 m × tion is less than 400 mm and occurs primarily 5 m blocks as two plots with at least a 1 m in the winter months (November to April). walkway around each plot. In December 2010, In general, soils are poorly developed due to fve blocks in each site were contained in recent deposition of substrates from volcanic exclosures 1.8 m high made from heavy­wire, sources (Rhodes and Lockwood 1995); how­ ungulate­resistant fencing to prevent grazing ever, our sites occur on alluvium substrates or browsing by feral goats and sheep. The from the Pleistocene and Holocene that are other fve blocks were in the open with no relatively old for this region (Sherrod et al. exclosure present. Each block was separated 2007). We delineated two study sites, Burned by 5 m, and fence treatments were randomly and Unburned, within a community that is assigned to blocks. characterized as Sophora­Myoporum shrubland We employed a S. madagascariensis seed with grass understory; it is dominated by addition treatment by adding 60 S. madagas- the trees Sophora chrysophylla (Mämane) and cariensis seeds to one plot in each block; the 72 PACIFIC SCIENCE · January 2018 other plot did not receive S. madagascariensis mesh)] sewn into 5 cm × 15 cm bags that were seeds. Seeds were added to the feld plots attached to the bottom of a funnel, to reduce on 12 January 2011. All plots, including plots loss from predation or wind (Cottrell 2004, without seeds added, were covered with coco­ Thaxton et al. 2010). Traps were placed on nut fber mats 1 cm thick (North American a rebar stand in the feld at a standard height Green, Poseyville, Indiana) to prevent seed of 50 cm above the ground surface to protect loss from high winds. Watering was necessary seeds from vertebrate seed predators (e.g., due to the drought conditions present during Mus musculus, Rattus sp., Pternistis erckelii ). the experiment. All plots were watered at a We collected seeds from each trap on 11 May rate of 9.5 liters/m2 after the initial seed addi­ 2011 and 8 August 2011. We identifed seeds tion and according to the following schedule: to species and counted the number of seeds of daily from 13 to 30 January 2011, three times all species collected. We analyzed soil inor­ − + per week from 31 January to 20 February ganic nitrate (NO 3 ) and ammonium (NH 4 ) 2011, twice per week from 21 February to 8 at each site by collecting one soil sample from May 2011, and once per week from 9 May to the center of each block on 31 August 2011. − + 13 June 2011. Watering was suspended dur­ For analysis of NO3 and NH4 , we extracted ing times of rainfall. samples with a 2.0 M KCl solution and sent them for analysis at the University of Hawai‘i at Hilo (UHH) Analytical Laboratory (Hilo, Sampling Hawai‘i) with a Pulse Autoanalyzer III with We sampled S. madagascariensis recruitment Autosampler IV (Saskatoon, Saskatchewan, and percentage cover in each plot on 5–6 May Canada). 2011 and 28 July–3 August 2011. We mea­ sured recruitment as the number of individ­ Statistical Analysis uals. We estimated percentage cover using a 1 m2 quadrat with an internal grid containing All data analyses were conducted in the R 25 points where we used a pin fag to deter­ statistical environment (R Core Team 2015). mine if S. madagascariensis was present at each The R package lme4 was used to analyze gen­ of the 25 points in the quadrat. With this eralized linear models (Bates et al. 2015). All method, only plants that touched the sam­ model validation procedures and diagnostics pling points in the quadrat were recorded. were conducted as described in Zuur and Ieno We ended the study and removed S. madagas- (2016). In Supplemental Appendix S1, we cariensis from the study sites in August 2011, present the diagnostic graphs for each model when S. madagascariensis began to repro­ used in the analyses. duce. At that time, we harvested the above­ Authors’ Note: Supplemental materials ground biomass of all plants in each plot, available online at BioOne (http://www dried the samples at 70°C for at least 72 hr .bioone.org/toc/pasc/current) and Project until at a constant mass, and then weighed the MUSE (http://muse.jhu.edu/journal/166). biomass. We also sampled the natural seed rain to: Comparisons among Burned and Unburned Sites (1) further investigate whether S. madagas- cariensis is dispersal limited in this area and We measured the difference between soil (2) ensure that our experimental exclosures nitrogen and S. madagascariensis recruitment, did not affect seed dispersal into the plots. cover, and biomass between the Burned and Senecio madagascariensis is an annual species Unburned sites. Low recruitment in the that sets seed continuously throughout the Unburned site and in plots where S. madagas- year as individual plants grow and mature. cariensis seeds were not added led to a heavily We collected seeds from seed traps over a zero­infated data structure as well as issues 6­month period by installing one seed trap in with homogeneity of variance between treat­ each block on 7 February 2011. Seed funnel ments. When S. madagascariensis did occur, traps were constructed of monoflament poly­ the total count of recruits was consistently ester fabric [33.9 threads per cm (no. 86 small (range was 0–8 seedlings per plot) and Herbivores and Senecio madagascariensis Invasion · Questad et al. 73 the data set contained numerous zeros; there­ Recruitment fore, we transformed the data into presence absence data. Cover data had similar prob­ Low recruitment in the Unburned site led to lems and were also transformed into presence a heavily zero­infated data structure; there­ absence data. Both data sets were analyzed fore, we analyzed the effects of our experi­ using a generalized linear model with a bino­ mental treatments only in the Burned site mial distribution using the standard logit link for recruitment. Thus, S. madagascariensis function. The fnal sampling date ( July 2011) recruitment was analyzed using a general was analyzed for recruitment data. Biomass linear mixed effects model via restricted had similar zero­infation problems that were maximum likelihood (REML). The model more diffcult to handle. Biomass models used for recruitment analysis was as follows: Recruitment ~ β intercept + β2Fence + β3Seed consistently resulted in poor model ft and + β + β × severe underdispersion; therefore, biomass Addition 4Date 5Fence Seed Addition + β6Fence × Date + β 7Seed Addition × Date + data could not be statistically analyzed. Soil β nitrate, ammonium, and total nitrogen were 8Block. Block was included as a random analyzed using a generalized linear model term. with a gamma distribution using the standard log link function. The gamma distribution Cover and Biomass was chosen for these data because the gamma distribution is appropriate for positive con­ We were unable to statistically analyze the tinuous data and provided the best model ft cover and biomass of S. madagascariensis compared to other distributions (Zuur and across all experimental treatments due to Ieno 2016). All of the between­site analyses zero­infation, violation of assumptions, poor were analyzed using the same model struc­ model ft, and models that were too complex for the data structure. The cover data set con­ ture: Response Variable ~ β intercept + β 2Site, tained 65% zeroes, and the biomass data set where β is the estimate of the effect of the contained 57% zeroes. factor (Site) on the response variable.

results Seed Rain Site Differences Seed rain of S. madagascariensis was too sparse to statistically analyze. To infer whether or The Burned site soils had higher total N and not the fencing treatment obstructed natural NH4 (Table 1, Supplemental Figure S1). seed rain into the plots, we analyzed seed rain Senecio madagascariensis recruitment, percent­ results for the two most abundant species, age cover, and biomass were higher in the Chenopodium oahuense and Erodium cicutarium. Burned site compared to the Unburned site, Seed counts were added from both trapping although the differences were not statistically periods for a total seed count for each plot. signifcant at α = .05 [P < .08 (Table 1)]. The One block from the Unburned site was dispersion statistic was 1.05 for the recruit­ omitted because the funnel disappeared dur­ ment and percentage cover models, indicat­ ing the second trapping period, leaving 10 ing good model ft (Zuur and Ieno 2016). samples from the Burned site and nine sam­ All attempted models for biomass ft poorly ples from the Unburned site. The model due to the zero­infated data structure. There­ chosen for these analyses was as follows: Seed fore, we report results for biomass as patterns Rain ~ β intercept + β 2Site + β 3Fence + β 4Site × only. Fence. Each species was analyzed separately. A generalized linear model with a negative Seed Rain binomial distribution using the standard log link function was selected because the nega­ Only eight S. madagascariensis seeds were col­ tive binomial distribution is appropriate for lected from the Burned site, and two seeds overdispersed count data that have large vari­ were collected from the Unburned site; there­ ation (Zuur and Ieno 2016). fore, we were unable to statistically analyze 74 PACIFIC SCIENCE · January 2018

TABLE 1 Analyses between the Burned and Unburned Sites

Response Variable Burned Site Mean Unburned Site Mean Estimate (SE) P

S. madagascariensis seedling recruitment 1.65 (0.59) 0.3 (0.21) −1.57 (0.88) .073 (no. of seedlings per plot) S. madagascariensis abundance (% cover) 4.8 (1.89) 0.4 (0.4) −2.09 (1.13) .065 S. madagascariensis biomass (g) 25.60 (10.50) 0.66 (0.37) *** *** Total available soil N 64.60 (24.10) 13.42 (2.63) −1.73 (0.36) <.005 Soil NH4 51.70 (23.90) 4.21 (0.82) −2.70 (0.42) <.005 Soil NO2 + NO3 12.92 (4.16) 9.22 (1.94) −0.34 (0.38) .384

Note: Estimates (SE) and P values are reported from the generalized linear models. A negative estimate indicates that the response variable was lower in the Unburned, compared to the Burned, site. Means ± SE are calculated for all experimental plots in each site ( N = 20 per site for Senecio madagascariensis recruitment, abundance, and biomass data; N = 10 per site for soil nitrogen data). *** Data could not be statistically interpreted due to heavily zero­infated data structure.

seed rain results for S. madagascariensis due to TABLE 2 the heavily zero­infated data structure (79% Experimental Treatment Effects on Senecio zeros). Chenopodium oahuense, a native shrub, madagascariensis Recruitment in the Burned Site and Erodium cicutarium, a nonnative invasive forb, were the most abundant species sampled Factor Estimate (SE) P in our seed traps. There was no difference in seed rain of either species inside and outside Intercept −0.15 (0.81) .85 Date: July 2011 0.300 (0.86) .73 the fenced exclosures [no signifcant Fence ef­ Fenced: Open 0.300 (1.09) .78 fect (P > .70) (Supplemental Figures S2, S3)]. Seed: Added 1.900 (0.86) .02 Seeds of native species other than C. oahuense Fenced: Open × Seed: 4.000 (0.99) <.0001 rarely occurred. Added Date: July 2011 × Fenced: −0.600 (0.99) .55 Open Experimental Treatment Effects Date: July 2011 × Seed: −0.600 (0.99) .55 Added We analyzed the effect of our experimental Note: Betas (SE) and P values are reported from the general­ treatments for data from the Burned site, but ized linear model. There was a signifcant increase in recruitment not the Unburned site, due to the low recruit­ when seeds were added (Seed Addition). When seeds were added, ment of in the Unburned plots that were not fenced had higher recruitment, but this effect S. madagascariensis of fencing was not apparent when seeds were not added (Fenced site (Table 1, Supplemental Figure S4). Sene- Open × Seed Addition). cio madagascariensis recruitment was signif­ cantly seed limited [signifcant Seed Addi­ tion Effect (Table 2, Figure 1a)]. Senecio browsing animals increased madagascariensis percentage cover and bio­ S. madagascarien- mass seemed to be affected by seed addition, sis recruitment. In contrast, percentage cover but we could not evaluate the statistical sig­ and biomass showed a trend of lower values nifcance of the pattern due to poor model ft when herbivores were present (Figure 1b). (Figure 1b). In plots where was S. madagascariensis discussion experimentally added, its recruitment was higher in unfenced, compared to fenced, Dispersal Limitation treatments [signifcant Seed Addition × Fence interaction (Table 2, Figure 1a)]. These re­ Overall we found that seed availability had sults suggest that once dispersal barriers were the greatest impact on the recruitment of S. removed by adding seeds, the presence of madagascariensis (Table 2, Figure 1a). It is sur­ Herbivores and Senecio madagascariensis Invasion · Questad et al. 75

cariensis to occur more abundantly in the seed rain. The areas used for this experiment occur in a large valley between Mauna Loa and Mauna Kea volcanoes, where winds are not impeded by landforms and wind speeds are high. Interpolated, 30­m wind speeds for our sites range from 22 to 25 kmph (U.S. Depart­ ment of Energy 2014). These high winds may limit the ability of S. madagascariensis seeds to become established, and instead seeds may be blown to other areas where they are de­ posited. However, seeds of other species, Chenopodium oahuense and Erodium cicutarium, were captured by our seed traps. These spe­ cies produce many small seeds and occur in high numbers at our sites, suggesting that it may take more time for S. madagascariensis to naturally invade this area from its surround­ ing populations. It is also possible that invasive vertebrate seed predators could affect invasion rates (e.g., Mus musculus, Rattus sp., Pternistis erck- elii ). Our seed traps were designed to protect collected seeds from vertebrate seed preda­ tors, which can negatively infuence seedling recruitment and survival and also increase seed dispersal (Cuddihy and Stone 1990, Shiels and Drake 2011). These predators are typically considered a barrier to the regenera­ tion of native plant species in Hawai‘i; how­ ever, their potential to control or disperse S. madagascariensis deserves consideration and study.

Figure 1. Recruitment and biomass as a function of seed Resources and Plant-Animal Interactions addition and herbivore removal. Bars show mean number of individuals for seed addition (+/−) treatments crossed Resource availability can alter invasive plant­ with herbivore removal treatments (fence/no fence) for (a) S. madagascariensis recruitment (number of seedlings) animal interactions. The fuctuating resources and (b) S. madagascariensis biomass (g). Error bars show hypothesis proposes that periods of high 1 SE. Recruitment of S. madagascariensis was greatest resource availability facilitate colonization when seeds were added and herbivores were present. Bio­ (Davis et al. 2000). The resource­enemy­ mass of S. madagascariensis was greatest when seeds were release hypothesis (R­ERH) expands on this added and herbivores were removed. idea and suggests that species adapted to high resource availability (high resource species) will be favored more by enemy release than prising that S. madagascariensis was strongly species with slower growth that are adapted dispersal­limited in light of its abundance in to lower resource availability (low resource the region and wind dispersal typical of other species) (Blumenthal 2006). The result of the species in the . It occurs in high R­ERH is that enemy release and resource abundance at sites downwind within 15 km of availability may act together to facilitate the the study area, so we expected S. madagas- invasion of high resource species. 76 PACIFIC SCIENCE · January 2018

We replicated our study in a Burned and for invasive, but not native, species (Vilà and Unburned area to examine how resources and Weiner 2004). Senecio madagascariensis cover disturbance may change the relationship be­ and biomass were higher in fenced treatments tween invasive ungulate enemies and S. mada- (Figure 1b), but S. madagascariensis seedling gascariensis invasion. Recruitment and plant recruitment was lower in fenced treatments biomass for S. madagascariensis were very low (Figure 1a). These differing responses to at the Unburned site and increased at the fencing may occur due to intraspecifc com­ Burned site (Table 1). Native species and petition (i.e., self­thinning) that is greater other nonnative species also had greater when herbivores are not present. As individ­ recruitment and biomass in the Burned site uals grow larger in the absence of herbivores, (E. Questad, A. Uowolo, S. Brooks, and S. they compete with each other and reduce the Cordell, unpubl. data). This increase in inva­ overall number of individuals in the popula­ sion may be the result of the observed pulse of tion. This competition is absent when the increased nitrogen (N) that was made avail­ seedlings are kept small by herbivory. There­ able after the fre; however, our study also fore, the overall effect of herbivores on S. coincided with a period of extreme drought madagascariensis is to reduce the size of plants, in the region. The relatively low recruitment but increase the number of individuals. This at both sites suggests that drought likely had fnding contrasts with other studies in which a strong, limiting effect on plant population the presence of herbivores increased intra­ dynamics. Even though all plots were watered specifc competitive interactions for invasive during the experiment, volumetric soil water plant species (Steets et al. 2006, Russell and content remained low and did not exceed Spencer 2010). It also suggests that herbi­ 0.16 m3/m3 during the time of the study, vores may alter the population growth of including on days with rainfall or when S. madagascariensis by reducing intraspecifc water was added (E. Questad, A. Uowolo, S. competition, but limiting the overall size and Brooks, and S. Cordell, unpubl. data). Plants reproductive output of individuals. were likely water­limited and could not fully take advantage of any soil nutrient or other Wind resource pulses after the fre. Therefore, it is still diffcult to isolate the role of resources Although we did not explicitly test the effects in S. madagascariensis invasion from this of wind on invasion, results of this and other experiment. studies suggest that wind may play an im­ In our study system, S. madagascariensis is portant role in invasion dynamics in this eco­ a fast­growing high resource species, and the system. Wind has substantial effects on the native species are slower­growing low re­ behavior and spread of fres across a landscape source species. An earlier study during a (Beer 1991, Fendell and Wolff 2001). The year with high rainfall found that when inva­ high winds in our study region make wildfres sive ungulates were removed from a large diffcult to suppress and lead to fres extend­ management area, S. madagascariensis cover ing over large areas. Wind accelerates erosion increased (Kellner et al. 2011), similar to our and can lead to nutrient loss, especially after a fnding of a trend toward increased biomass fre when high levels of erosion occur. The when ungulates were removed. These pat­ combination of high winds with the volcanic terns support the role of the enemy release ash substrate in this ecosystem accelerates hypothesis in explaining S. madagascariensis erosion and may lead to a depletion of topsoil, invasion in both high and low resource organic matter, and benefcial microorgan­ conditions. isms from the ecosystem. Feral goats in this Our results further suggest that Senecio ecosystem spend a greater amount of time may experience greater intraspecifc competi­ in areas protected from the prevailing winds tion when ungulates are removed. In pairwise (Cordell et al. 2016). These areas also have studies of competition between native and in­ the most suitable microclimates for plant vasive plant species, intraspecifc competition growth due to reduced water stress (Questad is often higher than interspecifc competition et al. 2014). Wind patterns can also strongly Herbivores and Senecio madagascariensis Invasion · Questad et al. 77 affect seed dispersal and thus plant establish­ layers of stems, for biological control of S. ment and community composition. Specif­ madagascariensis on Hawai‘i Island is a step cally convective updrafts, turbulence, and toward achieving control and could make it landforms that impede wind movement are possible to remove ungulates from areas with important determinants of geographic disper­ high levels of S. madagascariensis invasion sal patterns for wind­dispersed seeds (Tac­ without increasing the spread of S. madagas- kenberg 2003, Damschen et al. 2014). Here we cariensis (HDOA 2008, 2013). found a surprisingly low number of S. mada- When the removal of feral ungulates is gascariensis seeds in the seed rain, suggesting necessary for conservation or restoration, it that wind patterns in this region may not may be more effective for native conservation create dispersal opportunities to these sites. if it occurs in an area with low S. madagas- Thus, wind, and its interaction with land­ cariensis seed rain. In such areas, explosive forms, fre, nutrients, ungulate behavior, and population growth of S. madagascariensis fol­ dispersal dynamics may be an important, but lowing ungulate removal is less likely to occur; little­studied, force controlling plant commu­ therefore, there would be a lower investment nity interactions and invasion in this ecosys­ needed in S. madagascariensis control. In fact, tem (Swanson et al. 1988). our study area and the saddle region sur­ rounding it are good candidates for ungulate removal because seed rain Management Applications S. madagascariensis is low and the conservation value of this re­ The observed effects of dispersal and abiotic gion is high. This area is one of the last large conditions on S. madagascariensis abundance remnants of Myoporum-Sophora dry forest, have implications for managing this invasion. the critical habitat for the conservation of First, it will be important to focus S. madagas- the endangered palila honeycreeper (Loxioides cariensis control efforts in areas where its seed bailleui ). Removal of feral ungulates is less rain is high and where site conditions are con­ likely to promote S. madagascariensis invasion ducive to its growth. Focusing control where in this region compared to other areas with existing populations are large and resources greater seed rain. are readily available should be a priority. Fur­ ther study of seed rain patterns surround­ acknowledgments ing existing populations could help determine additional areas for control. Domesticated We thank R. Moseley, H. K. Hinds, J. Diep, goats and sheep have been used to control S. and C. Perry for assistance in the feld. We madagascariensis in other places where it is also thank the Center for Environmental invasive (Sindel 2009). This practice is com­ Management of Military Lands at Colorado patible with highly managed systems such as State University and the Natural Resources grazed pastures or rangeland but is less com­ Offce of PTA for logistical support. We patible in ecosystems where native species will gratefully acknowledge Commanders W. also be negatively infuenced by browsing. Richardson, R. C. Niles, and E. P. Shwedo for However, intensive control efforts either with support at PTA. managed ungulates or with herbicide applica­ tion could be focused in areas where invasion Literature Cited is most likely to occur or where S. madagas- cariensis seed rain is already high. Bates, D., M. Machler, B. M. Bolker, and If areas with high S. madagascariensis seed S. C. Walker. 2015. Fitting linear mixed­ rain need to be fenced for conservation, fenc­ effects models using lme4. J. Stat. Softw. ing during a dry period may reduce the level 67:1–48. of S. madagascariensis invasion. In addition, Beer, T. 1991. The interaction of wind and a follow­up plan to manage S. madagascariensis fre. Boundary­Layer Meteorol. 54:287– will be necessary for those areas. The release 308. of the phytophagous moth Secusio extensa, Blumenthal, D. M. 2006. Interactions whose larvae feed on leaves and strip the outer between resource availability and enemy 78 PACIFIC SCIENCE · January 2018

release in plant invasion. Ecol. Lett. 9:887– Goergen, E., and C. C. Daehler. 2001. Info­ 895. rescence damage by insects and fungi in Cano, L., J. Escarre, K. Vrieling, and F. X. native pili grass ( Heteropogon contortus) ver­ Sans. 2009. Palatability to a generalist her­ sus alien fountain grass (Pennisetum seta- bivore, defence and growth of invasive and ceum) in Hawai‘i. Pac. Sci. 55:129–136. native Senecio species: Testing the evolu­ Gurevitch, J., G. A. Fox, G. M. Wardle, tion of increased competitive ability hy­ Inderjit, and D. Taub. 2011. Emergent pothesis. Oecologia (Berl.) 159:95–106. insights from the synthesis of conceptual Cordell, S., E. J. Questad, G. P. Asner, K. frameworks for biological invasions. Ecol. Kinney, J. Thaxton, A. Uowolo, S. Brooks, Lett. 14:407–418. and M. W. Chynoweth. 2016. Remote HDOA (Hawai‘i Department of Agricul­ sensing for restoration planning: How the ture). 2008. Draft environmental assess­ big picture can inform stakeholders. Resto­ ment: Field release of Secusio extensa ration Ecology Early View Online. (Butler) (Lepidoptera: Arctiidae), for bio­ Cottrell, T. R. 2004. Seed rain traps for forest logical control of freweed, Senecio mada- lands: Consideration for trap construction gascariensis Poiret (: Asteraceae), and study design. BC J. Ecosyst. Manage. in Hawai‘i. Honolulu. 5:1–6. ———. 2013. Biocontrol moths released to Cuddihy, L. W., and C. P. Stone. 1990. fght invasive freweed. Honolulu. Alteration of native Hawaiian vegetation: Joshi, J., and K. Vrieling. 2005. The enemy Effects of humans, their activities and in­ release and EICA hypothesis revisited: troductions. University of Hawai‘i Press, Incorporating the fundamental difference Honolulu. between specialist and generalist herbi­ Damschen, E. I., D. V. Baker, G. Bohrer, R. vores. Ecol. Lett. 8:704–714. Nathan, J. L. Orrock, J. R. Turner, L. A. Kellner, J. R., G. P. Asner, K. M. Kinney, Brudvig, N. M. Haddad, D. J. Levey, and S. R. Loarie, D. E. Knapp, T. Kennedy­ J. J. Tewksbury. 2014. How fragmenta­ Bowdoin, E. J. Questad, S. Cordell, and tion and corridors affect wind dynamics J. M. Thaxton. 2011. Remote analysis and seed dispersal in open habitats. Proc. of biological invasion and the impact of Natl. Acad. Sci. U.S.A. 111:3484–3489. enemy release. Ecol. Appl. 21:2094–2104. Davis, M. A., J. P. Grime, and K. Thompson. Kinney, K. M., G. P. Asner, S. Cordell, O. A. 2000. Fluctuating resources in plant com­ Chadwick, K. Heckman, S. Hotchkiss, M. munities: A general theory of invasibility. Jeraj, T. Kennedy­Bowdoin, D. E. Knapp, J. Ecol. 88:528–534. E. J. Questad, J. M. Thaxton, F. Trusdell, Denslow, J. S. 2003. Weeds in paradise: and J. R. Kellner. 2015. Primary succession Thoughts on the invasibility of tropical on a Hawaiian dryland chronosequence. islands. Ann. Mo. Bot. Gard. 90:119–127. PloS ONE 10:e0123995. DeWalt, S. J., J. S. Denslow, and K. Ickes. Le Roux, J. J., A. M. Wieczorek, M. M. 2004. Natural­enemy release facilitates Ramadan, and C. T. Tran. 2006. Resolving habitat expansion of the invasive tropical the native provenance of invasive freweed shrub Clidemia hirta. Ecology 85:471–483. (Senecio madagascariensis Poir.) in the Ha­ Erfmeier, A., L. Hantsch, and H. Bruelheide. waiian Islands as inferred from phyloge­ 2013. The role of propagule pressure, netic analysis. Divers. Distrib. 12:694–702. genetic diversity and microsite availability Motooka, P., G. Nagai, K. Onuma, M. Du­ for Senecio vernalis invasion. PloS ONE Ponte, A. Kawabata, and G. Fukumoto. 8:e57029. 1999. Control of freweed (Senecio mada- Fendell, F. E., and M. F. Wolff. 2001. Wind­ gascariensis). University of Hawai‘i Press, aided fre spread. Pages 171–223 in E. A. Honolulu. Johnson and K. Miyanishi, eds. Forest Motooka, P., G. Nagai, K. Onuma, M. Du­ fres: Behavior and ecological effects. Aca­ Ponte, A. Kawabata, G. Fukumoto, and demic Press, San Diego. J. Powley. 2004. Control of Herbivores and Senecio madagascariensis Invasion · Questad et al. 79

Ragwort (aka Madagascar freweed, Senecio Simberloff, D., and B. Von Holle. 1999. Posi­ madagascariensis). College of Tropical Ag­ tive interactions of nonindigenous species: riculture and Human Resources, Univer­ Invasional meltdown? Biol. Invasions sity of Hawai‘i at Mänoa. Publ. WC­2. 1:21–32. Pratt, T. K., P. C. Banko, S. G. Fancy, G. D. Sindel, B. 1989. The ecology and control of Lindsey, and J. D. Jacobi. 1998. Status and freweed (Senecio madagascariensis Poir.). management of the Palila, an endangered University of Sydney, NSW, Australia. Hawaiian honeycreeper, 1987–1996. Pac. ———. 2009. Fireweed in Australia: Direc­ Conserv. Biol. 3:330–340. tions for future research. Bega Valley Fire­ Questad, E. J., J. R. Kellner, K. Kinney, weed Association, Bega, NSW, Australia. S. Cordell, G. P. Asner, J. Thaxton, J. Steets, J. A., R. S. Salla, and T. L. Ashman. Diep, A. Uowolo, S. Brooks, N. Inman­ 2006. Herbivory and competition interact Narahari, S. A. Evans, and B. Tucker. to affect reproductive traits and mating 2014. Mapping habitat suitability for at­ system expression in Impatiens capensis. Am. risk plant species and its implications for Nat. 167:591–600. restoration and reintroduction. Ecol. Appl. Swanson, F. J., T. K. Kratz, N. Caine, and 24:385–395. R. G. Woodmansee. 1988. Landform ef­ R Core Team. 2015. R: A language and envir­ fects on ecosystem patterns and processes. onment for statistical computing. R Foun­ BioScience 38:92–98. dation for Statistical Computing, Vienna, Tackenberg, O. 2003. Modeling long­ Austria. distance dispersal of plant diaspores by Rhodes, J. M., and J. P. Lockwood. 1995. wind. Ecol. Monogr. 73:173–189. Mauna Loa revealed: Structure, composi­ Thaxton, J. M., T. C. Cole, S. Cordell, R. J. tion, history, and hazards. American Geo­ Cabin, D. R. Sandquist, and C. M. Litton. physical Union, Washington, D.C. 2010. Native species regeneration follow­ Richardson, D. M., and P. Pyšek. 2006. Plant ing ungulate exclusion and nonnative grass invasions: Merging the concepts of species removal in a remnant Hawaiian dry forest. invasiveness and community invasibility. Pac. Sci. 64:533–544. Prog. Phys. Geogr. 30:409–431. U.S. Department of Agriculture. 2016. Natu­ Russell, F. L., and M. Spencer. 2010. Com­ ral Resources Conservation Service soil bined effects of folivory and neighbor survey for the island of Hawai‘i area, plants on Cirsium altissimum (tall thistle) Hawai‘i. rosette performance. Plant Ecol. 208:35– U.S. Department of Energy. 2014. Hawaii 46. 30­meter residential­scale wind resource Shaw, R. B., and J. M. Castillo. 1997. Plant map (https://windexchange.energy.gov/ communities of Pohakuloa Training Area, maps­data/161). Hawaii. Fort Collins, Colorado. Vilà, M., and J. Weiner. 2004. Are invasive Sherrod, D. R., J. M. Sinton, S. E. Watkins, plant species better competitors than na­ and K. M. Brunt. 2007. Geologic map of tive plant species? Evidence from pair­wise the state of Hawai‘i, Sheet 8. Island of experiments. Oikos 105:229–238. Hawai‘i. In U.S. Geological Survey Open­ Wolfe, E. W., and J. Morris. 1996. Geologic File Report 2007­1089. map of the island of Hawaii. Miscellaneous Shiels, A. B., and D. R. Drake. 2011. Are Investigations Series I­2524A. U.S. Geo­ introduced rats (Rattus rattus) both seed logical Survey, Reston, Virginia. predators and dispersers in Hawaii? Biol. Zuur, A. F., and E. N. Ieno. 2016. Beginner’s Invasions 13:883–894. guide to zero­infated models with R. Simberloff, D. 2006. Invasional meltdown 6 Highland Statistics Ltd., Newburgh, years later: Important phenomenon, un­ United Kingdom. fortunate metaphor, or both? Ecol. Lett. 9:912–919.