Tropical insular fish assemblages are resilient to flood disturbance

Periods of stable environmental conditions, favoring development of ecological communities regulated by density-dependent processes, are interrupted by random periods of disturbance that may restructure communities. Disturbance may affect populations via habitat alteration, mortality, or displacement. We quantified fish habitat conditions, density, and movement before and after a major flood disturbance in a Caribbean island tropical river using habitat surveys, fish sampling and population estimates, radio telemetry, and passively monitored PIT tags. Native stream fish populations showed evidence of acute mortality and downstream displacement of surviving fish. All fish species were reduced in number at most life stages after the disturbance, but populations responded with recruitment and migration into vacated upstream habitats. Changes in density were uneven among size classes for most species, indicating altered size structures. Rapid recovery processes at the population level appeared to dampen effects at the assemblage level, as fish assemblage parameters (species richness and diversity) were unchanged by the flooding. The native fish assemblage appeared resilient to flood disturbance, rapidly compensating for mortality and displacement with increased recruitment and recolonization of upstream


INTRODUCTION
Many biotic interactions, such as competition and predation, are density dependent and develop as populations grow in number, while stochastic abiotic events, such as flooding and drought, periodically reset community structure and interrupt the outcome of biological interactions (Connell 1978, Jackson et al. 2001).Severe, temporary abiotic conditions (i.e., disturbances) result in the mortality or displacement of individuals (Sousa 1984), redistribution of habitat and substrate (Pickett and White 1985), and altered nutrient cycling (Pringle 1997).Abiotic factors set the physical template upon which ecological communities develop; species traits such as dispersal rate and environmental tolerances set the pool of available colonists from which communities are derived, and the disturbance regime (frequency, intensity, and predictability) modifies community composition over time (Poff and Ward 1989, Waide and Willig 2012, Mims and Olden 2013, Smith and Kwak 2014a).Knowledge of physical and biotic regulatory processes and their interaction is not only of interest for ecological conceptual development, but can enhance understanding of the distribution and abundance of organisms with direct conservation applications (Ludwig et al. 2001, Poff andZimmerman 2010).
Aquatic community structuring was framed as a dichotomy between deterministic biotic and stochastic abiotic processes as the field of stream disturbance ecology developed in the 1980s (Grossman et al. 1982, 1985, Matthews 1982, Yant et al. 1984).Investigators of this topic reported variable stream fish assemblage structure at small scales in patchy stream habitats (Grossman et al. 1982) and high assemblage stability across very large spatial scales (Matthews 1982).After years of empirical study, research findings conformed to a unifying paradigm that stream systems followed a predictable continuum from highly-disturbed, stochastically structured communities to rarely-disturbed, deterministically structured communities (Resh et al. 1988, Poff and Ward 1989, Strange et al. 1992).
Findings from lotic communities have strongly influenced theory on community disturbance ecology.Early research in stream disturbance ecology contributed to a general shift in ecological theory toward considering disturbance regimes in community structuring, and recent stream disturbance research provided tests for contemporary biodiversity theory (Stanley et al. 2010, Death 2010).Stream research has generally generated data inconsistent with the Intermediate Disturbance Hypothesis (Connell 1978), possibly due to the prevalence of highly dispersive larval stages in aquatic communities.While terrestrial community dynamics are driven by reproduction, aquatic community dynamics may be influenced by movement (Death 2008).Highly mobile larvae may generate patterns contrary to predictions of Hubbell's Neutral Theory of Biodiversity (Hubbell 2001); that is, larval dispersal tends to reduce the effect of community structuring by spatial separation (Thompson and Townsend 2006).Stream communities may also offer insight into community resilience (Gunderson 2000).For example, prolonged larval duration has been hypothesized as an adaptation conferring community resilience to frequently disturbed stream ecosystems (McDowall 2010).Stream disturbance research has demonstrated that the physical template and disturbance regime feed back on populations and communities over time, fostering the evolution of traits that modulate future responses to disturbances (Lytle andPoff 2004, Death 2010).
Understanding the specific role of disturbance in structuring communities leads to enhanced development and prediction of resource management success under various strategies.For example, many reservoir, lake, and pond ecosystems are successfully managed using a community-based strategy originally developed by Swingle (1956).This approach is often successful in lentic systems that are infrequently disturbed with fish populations that tend to reach high biomasses and become regulated by densitydependent, biotic interactions (Poff and Ward 1989).Conversely, density-dependent fishery assessment approaches have produced less reliable results in marine ecosystems (e.g., stockrecruit relationships; Hilborn and Walters 1992), where disturbance is frequent and intense.Environmental effects often take precedence in structuring these communities (Dayton 1971, Connell 1978, Harmelin-Vivien and Laboute 1986).Fish life histories have evolved to cope with different biotic and abiotic challenges under conditions of variable disturbance, and those life histories respond differently to exploitation and management (Winemiller 2005, Mims andOlden 2013).
Lotic ecosystems may be so frequently disturbed that equilibrium is rarely attained, and the resulting system is in perpetual recovery from the previous disturbance event (Resh et al. 1988).Stochastic disturbance regimes favor rapidlymaturing, short-lived opportunistic species, while low levels of predictable disturbance favor late-maturing, long-lived equilibrium species (Winemiller 1995).Cycles of intense disturbance and favorable environmental conditions foster the development of high fecundity and low percapita investment in offspring (i.e., periodic life history).Opportunistic, periodic, and equilibrium species are each expected to respond differ-ently to common population management strategies (Winemiller 2005).Similarly, management of the stream environment with particular attention to the timing of pulsed flood events, affects fish populations variably, based their life history strategy (Zeug and Winemiller 2011), because life histories and adaptations evolve to optimize reproductive success under a given disturbance regime and other environmental conditions (Winemiller 1995).Applying environmental management approaches, such as augmented minimum flows or restoration of natural flow regimes, can regulate stream flow to favor the life histories of stream community components and can account for disturbance characteristics (Annear et al. 2004).An understanding of the interaction of native fishes with natural flow regimes may allow managers to restore or conserve native fish assemblages by simply restoring or maintaining ecological flows (Marchetti andMoyle 2001, Kiernan et al. 2012).
Ecological theory predicting the effects of disturbance on stream communities has been developed primarily from examples in temperate regions (e.g., Grossman et al. 1982, Matthews 1982, Poff and Ward 1989, Death 1996, Townsend et al. 1997).Disturbance regimes in tropical systems are contrastingly frequent, intense, and stochastic and could exert major ecological influences on stream communities (Fitzsimons and Nishimoto 1995, Covich et al. 1996, Ramı ´rez et al. 2009); however, few investigators have sought to describe the specific effects of disturbance on individual behavior and its role in structuring tropical insular stream communities.To supplement this information gap in disturbance ecology, we conducted unique research to quantify the effects of a major flood disturbance, Hurricane Irene (August 2011), on the native riverine fish assemblage of a tropical island river, Rı ´o Mameyes, Puerto Rico, USA.Through a combination of population estimates, two approaches to tagging and monitoring individual fish, and instream habitat surveys before and after dramatic flooding, we characterize the role of a major flood event in shaping individual behavior and fish assemblage dynamics of an insular tropical river system.Our goal was to provide insight into the individual-and population-level mechanisms and processes that result in assemblage change or resilience.

Study system
Rı ´o Mameyes is a fourth-order, free-flowing river draining the Luquillo Mountains of northeastern Puerto Rico, USA, the eastern-most island of the Greater Antilles in the Caribbean Sea (Fig. 1).The Luquillo Mountains are of volcanic origin and reach heights greater than 1,000 m within 20 km of the island coast (Pike et al. 2010).The steep topography interacts with northeasterly trade winds and frequent tropical storms to produce extremely high rainfall (averaging nearly 5 m annually; Lugo et al. 2012) and flood-dominated rivers, typical in hydrology to other Antillean rivers.Spates 50 times greater than base flow are common in the region, and river hydrographs are flashy, often peaking and returning to near base flows within 24 hours (USGS 2011).
Rı ´o Mameyes is one of only a few undammed rivers in Puerto Rico (Cooney and Kwak 2013) with continuous aquatic connectivity from the headwaters to the ocean.It has a steep, short drainage, with an origin at 850 m above sea level, flowing north just 12 km to the Atlantic Ocean.The upper watershed consists of protected secondary forest within the El Yunque National Forest, and the lower watershed contains primarily secondary forest, agricultural land, and small urbanized areas, including the villages of La Vega, Palmer, and Fortuna (Martinuzzi et al. 2007).Rı ´o Mameyes' unrestricted connection to the ocean, short total length, and generally undeveloped watershed render it an ideal system to study a tropical aquatic community with minimal anthropogenic influence.
Native amphidromous fauna dominate Puerto Rico freshwater streams with uninterrupted connectivity to the estuary, whereas exotic fauna dominates streams above larger obstructions to aquatic migration, such as waterfalls and large dams (Holmquist et al. 1998, Kwak et al. 2007, Hein et al. 2011, Cooney and Kwak 2013).Thus, native amphidromous fish and shrimp fauna are the dominant aquatic community members in the free-flowing Rı ´o Mameyes.Amphidromous larvae temporarily occupy estuaries and river mouths before migrating back into freshwater as metamorphosing juveniles (McDowall 1988, Smith andKwak 2014b).Amphidromous shrimp, Xiphocaris elongata, Atya spp., and Macrobrachium spp., and amphidromous fishes, mountain mullet (Agonostomus monticola), bigmouth sleeper (Gobiomorus dormitor), sirajo goby (Sicydium spp.), river goby (Awaous banana), and smallscaled spinycheek sleeper (Eleotris perniger) are the most common native macrofauna.A sixth diadromous fish, the American eel (Anguilla rostrata) follows a catadromous life history and is also common among Puerto Rico streams (Kwak et al. 2007).Globally, diadromous fish of the Gobiidae, Eleotridae, and Anguillidae families are dominant fish assemblage members on tropical volcanic islands and are in need of comprehensive conservation strategies (McDowall 1988(McDowall , 1999)).Thus, Rı ´o Mameyes can be considered representative of the undeveloped streams with native diadromous fish assemblages and intense flood regimes, found on tropical islands throughout the world, and our findings may be applied toward broad-scale conservation.

Riverine habitat
Before and after Hurricane Irene, standardized instream habitat surveys of a 200-m coastal plain river reach were conducted to describe the magnitude of physical effects from the storm (Kwak et al. 2007).The surveyed reach was selected as representative of the lower Rı ´o Mameyes watershed, with all three major macrohabitats, rifle, run, and pool.The purpose of the habitat survey was to make inferences regarding the potential for redistribution of habitat during the disturbance rather than quantifying habitat throughout the entire basin.Water depth, meancolumn velocity, and dominant substrate were measured at 10 evenly spaced points along crosssectional transects that were evenly separated by a distance of two mean river widths.At points of less than 0.5-m depth, velocity was measured at 60% of total depth, and at points greater than 0.5m depth, velocity was measured at 20% and 80% of total depth then averaged.A profile of discharge since 1970 was developed from the U.S. Geological Survey's streamflow gauge on Rı ´o Mameyes (USGS 2011).Disturbance frequency was estimated by averaging the time between hydrograph peaks equal to or greater than the peak occurring during Hurricane Irene.The dominant substrate (areal coverage) within 1 m Fig. 1.Land cover map of northeastern Puerto Rico, showing Rı ´o Mameyes and passive integrated transponder array and locations.The coastal plain array was located 2.2 km upstream of the ocean (river km 2.2; 2 m above sea level), and the foothills array was located at river km 6.4 (42 m above sea level).
v www.esajournals.org of each sampling point was classified according to a modified Wentworth particle size scale (Bovee and Milhous 1978).Habitat parameter data were tested for normality with a Shapiro-Wilk test, and normal data were tested for a change in the mean using two-sample t-tests, interpreted at a ¼ 0.05 (R Development Core Team 2012).

Fish density
Fish density of all species was estimated using a standardized three-pass removal procedure (Hayes et al. 2007, Kwak et al. 2007) in randomized 100-m river reaches from just upstream of the estuary (river km 2) to the highest accessible site (river km 10.7).Fish were sampled using two Smith-Root LR-24 backpack electrofishers (400 V, 30% pulse width, 0.2-0.3A) and three sequential passes through the 100-m reach.Either block nets or natural barriers (e.g., steep riffles and cascades) were used to close each reach to fish movement during sampling.All fish were collected without replacement during each 30-minute pass and then measured and weighed before release subsequent to the final pass.
During the summers of 2009 and 2010 and the winter of 2010, 22 pre-disturbance removal fish density estimates were completed, and during the two weeks following Hurricane Irene (August 26-September 9, 2011), six post-disturbance estimates were completed (Fig. 2).Sites were randomized among 51 possible 100-m reaches without replacement during each seasonal sampling period.Candidate models included all possible combinations of models of constant, seasonal, or variable capture probabilities and all possible combinations of seasonal densities (i.e., different seasonal densities set equal).All candidate models were compared using Akaike's information criteria (AIC; Akaike 1974).Each species was stratified into three size categories, juveniles, sub-adults, and adults, based on estimates of size at maturation (W.E. Smith, unpublished data).Mountain mullet males reach much smaller asymptotic sizes compared to females, and no external characters distinguish the sexes.Therefore, we considered the sub-adult size to be representative of sub-adult female and adult male mountain mullet.The presence of American eels in collecting tanks increased mortality among other fishes, which interfered with fish tagging, so American eels were only collected during the three-pass removal samples of summer 2009.In samples after the storm, American eels were enumerated only in the first pass, and a single-pass estimate of density was generated.
Removal data were modeled in a Bayesian framework using OpenBUGS software (Lunn et al. 2009) and the multiple-pass Bayesian removal model described by Wyatt (2002;Appendix and Supplement).Removal models estimated two parameters, abundance and capture probability, representing a hierarchy of state and observation processes (Williams et al. 2002).Abundance estimates were converted to density (number per unit area), which in turn were considered draws from a median density for each size class and species.Proportion change in fish density Dd associated with the flood was calculated as where d was the median density of each species and size class.Pre-and post-disturbance estimates of fish density were each summed across size classes to calculate community parameters, species richness, assemblage heterogeneity, and the rank order abundance of each species (Kwak and Peterson 2007).Species richness was estimated as the total number of species observed in each sampled reach, and t-tests identified changes in mean species richness after the disturbance.Change in heterogeneity was quantified as the difference of post-and pre-hurricane Shannon-Weaver index values (DH'; Shannon andWeaver 1949, Kwak andPeterson 2007).Derived parameters of the fish density model included proportional change in density and change in community parameters (pre-versus post-disturbance).Credible intervals for these population and community parameters that did not contain zero indicated significant departures from expected densities and assemblage heterogeneity.

Fish movement
Fish movement was monitored at two spatial scales, a broad, basin scale using large numbers of PIT-tagged fish, and a small, reach scale using a smaller number of fish tagged with radio transmitters.Fish were collected from two sequential passes of nine randomized 150-m reaches in both June and July 2011 as described above.Only mountain mullet and bigmouth sleeper were large enough for radio transmitter implantation, and only mountain mullet larger than 120 mm total length and bigmouth sleeper larger than 130 mm total length were PIT-tagged.No fish smaller than 200 mm total length received a radio transmitter.
Fish receiving only a PIT tag were lightly anesthetized in aerated river water containing 80 mg/L tricaine methane sulphonate (MS 222) for 2 minutes before transfer to a tank containing a maintenance level of anesthetic (30 mg/L MS 222).Total length, weight, sex, and capture location were recorded for each fish.Half-duplex PIT tags (Texas Instruments, Dallas, Texas, USA [23 mm, 0.6 g]) were implanted intra-abdominally via a 4-mm incision posterior to the pelvic fin.
Fish longer than 250 mm total length were implanted with larger 32-mm tags (0.8 g).Before each surgery, all surgical equipment and PIT tags were disinfected in a solution of Benzall surgical disinfectant.Each PIT tag implantation lasted approximately one minute and total sedation time was 7.5 minutes on average, after which fish recovered in a 30-L live well, with flowing river water, for 1-2 hours before release.Only fish in apparently good condition were released.PIT tag retention was assumed to be 100% (Smith and Kwak 2014a).
Fish tagged with radio transmitters were captured and allowed to recover following the same procedures for PIT tagging.Fish receiving a radio transmitter were anesthetized for 4 minutes in 80 mg/L MS 222.A small radio transmitter (Advanced Telemetry Systems, Isanti, Minnesota, USA [Model F1545, 0.9 g]) and a 23-mm PIT tag were implanted intra-abdominally through a 12- v www.esajournals.orgmm incision that was closed with two sterile sutures.Trailing antennas were coiled and implanted intra-abdominally (not inserted through body wall).Each surgery lasted approximately 10 minutes.Greater measures were required to ensure that each fish survived radio tag implantation in good condition, including closure of the incision wound and artificial circulation of water containing 30 mg/L MS 222 over the gills with a hand siphon.
To monitor the movement of PIT-tagged fish, pairs of PIT antennas (i.e., arrays) were installed in Rı ´o Mameyes across the entire river channel, one in the coastal plain and one in the foothills (Fig. 1).Each antenna was driven by a multiplexed reader (Oregon RFID, Portland, Oregon, USA) and constructed from a single loop of 8gauge stranded copper cable in a horizontal, pass-over design.The use of paired antennas enabled a determination of the direction of fish movement; however, we elected to avoid assumptions regarding detectability of movement direction by summing all movement detections by day to quantify general fish activity levels (Aymes andRives 2009, Smith andKwak 2014a).Radio-tagged fish were manually tracked weekly, and a hand-held Global Positioning System unit was used to record each fish location.Fish locations were later converted to river km using Google Earth (Google 2011).
The location of radio-tagged fish in relation to the storm was quantified two ways, the change in location proximately after the river returned to base flow and the change in mean location during the weeks before and after the flood.Locations during the four weeks prior to the flood were averaged as an estimate of mean preflood location, and locations during the two weeks after the flood were averaged to generate an estimate of mean post-flood location.Change in location immediately following reduced flow and the mean change in location in the weeks following the storm were pooled by species and tested for before-after differences using a Mann-Whitney test (R Development Core Team 2012).

RESULTS
Hurricane Irene struck Puerto Rico on 22 August 2011, and rain persisted for four days.The heaviest rainfall occurred during the first day of the storm and produced a peak daily discharge in Rı ´o Mameyes of 39 m 3 /s (Fig. 2), representing the highest flow among the previous 7 years; the 30-year median August flow was 1.6 m 3 /s.Discharges equal to or greater than the Hurricane Irene peak flow have occurred in Rı ´o Mameyes on average every four years over the last 30 years that the streamflow gauge has operated.During Hurricane Irene, the hydrograph peak was brief, but elevated water levels persisted for six days.
Habitat surveys revealed that water in the coastal plain reach was significantly deeper and faster after the flood compared to before the flood, despite greater river discharge during the initial survey than during that after the flood (USGS 2011).All habitat parameters were normally distributed (P , 0.01), and comparisons were made using t-tests.A deep channel was scoured along the downstream portion of the reach, where water velocity (n ¼ 293; t ¼À5.8;P , 0.01) and depth (n ¼ 294; t ¼À3.3;P , 0.01) were greater after the flood.Further, a large 30 3 75 m gravel bar was deposited along the shore, resulting in a narrower channel (n ¼ 4; t ¼ 1.9; P ¼ 0.06) in the downstream portion of the reach.Substrate varied widely from large boulders to silt.Dominant substrate of large and small cobble, however, was not significantly different after the disturbance (n ¼ 301; t ¼À1.1;P ¼ 0.25).
Averaged among species and size classes, a 46% decline in fish density was observed after the flood disturbance (Fig. 3).AIC of three-pass removal models indicated that the best model included densities that only changed after the disturbance and capture probabilities that varied by season (AIC of best model ¼ 8,964; AIC of second best model ¼ 8,977).AIC support of this model indicated that densities were not significantly different during summer 2009, winter 2010, and summer 2010, but densities were different after the flood disturbance than any of these time periods.Most examined fish groups showed a large reduction in density after the flood.Only sub-adult bigmouth sleeper, subadult mountain mullet, and juvenile sirajo gobies were found at expected or higher than expected densities after the flood (Fig. 3).Adult sirajo gobies were generally absent from post-hurricane samples.Although size distributions of some species changed, species richness did not appear v www.esajournals.orgto change during the disturbance, as all common species were present before and after the storm.The disturbance did not extirpate any common fish species from Rı ´o Mameyes; furthermore, there was no change in fish assemblage heterogeneity (DH' 95% credible interval: À0.030, 0.035).The rank order of species density was similar before and after the storm, but American eel declined from the most abundant species to the second most abundant species (Table 1).
Detections of PIT-tagged fish indicated peaks in activity following the flood disturbance, but no major downstream displacement of fish.Totals of 280 bigmouth sleeper and 179 mountain mullet were PIT-tagged and released during June and July 2011 and available for detection at automated PIT arrays.Including both the coastal plain and foothills PIT arrays, 8 PIT-tagged bigmouth sleeper and 14 mountain mullet were detected during the month prior to the flood, and 10 bigmouth sleeper and 28 mountain mullet were detected during the two weeks after the storm.PIT arrays were disabled during the flood and reinstalled 5-11 days later.After the flood, peaks in the activity of PIT-tagged fish occurred in both species (Fig. 4).Post-flood activity levels were especially high for bigmouth sleeper in the coastal plain and for mountain mullet in the foothills.Most bigmouth sleeper detected (70% before, 87% after) were tagged locally, within 2 km of the array where they were detected.A smaller fraction of PIT-tagged bigmouth sleeper detected appeared to be transient, having been tagged greater than 2 km away from the array.However, the proportion of these transient fish was greater before the storm than after it (30% before, 13% after).A small fraction of all fish appeared to be transient, and some individuals of both species were detected after moving long distances downstream to the coastal plain (.3 km; four bigmouth sleeper and six mountain mullet detected at the coastal plain array), demonstrating a low level of downstream displacement during the disturbance.The most notable dispersal observed in tagged fishes was a large number of mountain mullet initially captured and tagged 3-5 km downstream in the coastal plain that were observed in the foothills (22 PIT-tagged fish, 82% of all mountain mullet detections in the foothills), moving upstream following the flood.Upstream movement was assumed because these fish were not detected during the period between being tagged downstream and being observed at the upstream v www.esajournals.orgPIT array.This pulse of migrating fish represented more than 10% of all PIT-tagged mountain mullet.
Radio-telemetered fish showed moderate mortality and movement associated with the flood disturbance.Nine bigmouth sleeper and 12 mountain mullet were tagged with radio transmitters in early June and survived until Hurricane Irene.During the storm, two mountain mullet (17% of total tagged fish) and one bigmouth sleeper (11% of total tagged fish) died.Mortality was confirmed by retrieving radio transmitters from the streambed.Among all radio-telemetered fish, neither immediate location nor average location was significantly different after the disturbance (P .0.05).Movement direction was generally stochastic.The most notable change in location associated with the storm was one mountain mullet found in a large pool 1 km upstream from its pre-flood location on the first day that fish were relocated after the disturbance.This fish returned downstream to its area of core use within one week.A final exceptional observation of fish movement occurred outside of the study period during a brief spate, three weeks subsequent to Hurricane  Irene.A mountain mullet previously located in the mountains near river km 9 (82 m above sea level) was detected 1.25 km downstream (36 m above sea level).Two days later, the same fish was relocated upstream in its core use area (river km 9).
On 28 August, six days after the hurricane's eye passed and on the first day of reduced discharge in Rı ´o Mameyes, a large recruitment migration of postlarval sirajo gobies was observed at the location of the coastal plain PIT array (river km 2).For approximately 8 hours, a 0.5-m wide column of postlarval sirajo gobies (22-mm mean total length) was observed moving upstream 1 m from the shore.This recruitment event was similar to those described by previous researchers studying sicydiine gobies and goby fry fisheries (Erdman 1961, Bell 1994).

DISCUSSION
The flood produced by Hurricane Irene on Rı ´o Mameyes was a phenomenal event.Minor floods following frequent thunderstorms are routine in Puerto Rico rivers and rarely surpass stream banks; however, the Hurricane Irene flood greatly exceeded the river banks, flowing onto rarely inundated coastal floodplain near the village of Fortuna.Observed from the most downstream bridge, the normally clear and gently flowing 20-m wide river was filled with muddy, surging water extending over 200 m wide.Downstream PIT array equipment was completely scoured and washed downstream.Observed from the most upstream bridge, the mountainous portion of the stream was intimidating, with rushing water rising to the underside of the shaking bridge, and whole trees swept past rapidly.
Hurricane Irene represented an intense disturbance event in a tropical insular river, and substantial biotic effects resulted at the population level.However, rapid recovery processesrecruitment and recolonization of upstream habitats-were evident within populations and reduced the effects of the disturbance at the assemblage level.Our individual fish tagging results reveal fish mortality, displacement, and increased activity that occurred due to the disturbance, providing insights into higher population-and assemblage-level processes.Tagging studies of amphidromous fishes are extremely rare in the primary literature (e.g., Koster et al. 2013), and we are not aware of any other that elucidated effects of disturbance.We documented reduced fish densities among species and size classes, indicating that significant mortality or displacement occurred, and the mortality of radio-telemetered fishes during the disturbance confirms mortality as a consequence.Anecdotally, a large recruitment immigration of postlarval sirajo gobies was observed in the receding floodwaters, and high densities of postlarval sirajo gobies were present throughout the river during removal sampling.Episodic recruitment migrations into riverine habitats are a regular feature in large tropical insular river systems (Erdman 1961, Bell 1994), and our observation of recruitment migration after a major flood is consistent with juveniles induced to migrate upstream by flooding or receding floodwaters.The higher than expected densities of sub-adult mountain mullet in the coastal plain and foothills reaches surveyed after the flood were likely displaced from upper elevations, and the small number of PIT-tagged fish moving downstream after the disturbance confirmed the potential for a low level of downstream displacement.Displacement and crowding of diadromous fishes in downstream reaches following hurricanes have been associated with increased flows in other aquatic systems (Fitzsimons andNishimoto 1995, Houde et al. 2005).High fish densities in lower reaches in our study were subsequently associated with upstream migration into the headwaters, observed in PIT-tagged mountain mullet moving through the foothills.Population parameters changed significantly after the storm, but community parameters, species richness, heterogeneity, and species rank order changed only minimally.Thus, we conclude that native fish assemblages are resilient to the effects of flood disturbance in tropical insular rivers.
Hurricane Irene was a major flood disturbance that persisted over several days and significantly altered riverine habitat.Floods of equal or greater magnitude occur on average every four years in Rı ´o Mameyes, approximately equal to or slightly less than the maximum life span of fishes in the assemblage, based on growth and survival rates (Smith and Kwak 2014a).Thus, the native fish assemblage has adapted to this type and frequency of disturbance, and we should expect those adaptations to manifest in the native assemblage response to disturbance (Lytle and Poff 2004).Among fish populations, two general patterns were observed in fish density and movement-low densities in the foothills and coastal plain and little evidence of displacement (pattern 1) and high densities in lower reaches and greater evidence of displacement (pattern 2).All species were reduced in number, with the exception of sub-adult bigmouth sleeper and juvenile sirajo goby, which did not change in density, and sub-adult mountain mullet, which increased in density.Further, among tagged fishes, only mountain mullet were observed migrating after the disturbance.We hypothesize that all species suffered significant mortality due to catastrophic displacement.Mortality as a consequence of the flood disturbance was documented by the death of a small number of radiotelemetered bigmouth sleeper and mountain mullet.Most species followed pattern 1, but mountain mullet, consistent with pattern 2, suffered displacement as well but were able to survive the event and migrate into presumably vacated headwater habitats.The unchanged density of sub-adult bigmouth sleeper and observations of long distance downstream movement after the flood suggested that bigmouth sleeper may have also followed pattern 2, but to a lesser extent than mountain mullet and without subsequent movement into upper river reaches.
Our results support the conclusion that upstream movement of mountain mullet following disturbance is associated with atypically higher densities in lower reaches.Two hypotheses could explain this phenomenon.Mountain mullet may have become crowded in lower reaches and responded with density dependent emigration from the coastal plain; these facultatively predatory fish may have followed migrating juvenile sirajo gobies that we observed at lower elevations or both species may be cued to migrate upstream by flooding.Studies of fish migration and movement associated with a flood event during a season when postlarvae do not migrate en masse (e.g., winter-spring) could separate these confounding influences during summer floods.
Movement analysis was enhanced by incorporating both broad-scale, basin-wide data from PIT-tagging and remote monitoring with fine-scale reach data from radio-telemetry.Both tagging approaches indicated that most fish movement was local (,2 river km), with the exception of mountain mullet migrating into upper elevations following the disturbance.Approximately 10% of all PIT-tagged mountain mullet were observed migrating upstream.Only 10 of 12 radio-telemetered mountain mullet survived the disturbance, and only one moved a short-distance upstream (1 km).Chance alone could explain the lack of telemetered mountain mullet migrating at the scale observed in PITtagged fishes (.3 km), and the incorporation of large numbers of PIT-tagged fish facilitated the quantification of a small fraction of migrating fish.
Mountain mullet may have experienced greater difficulty than other species in locating hydrodynamic refuge during the flood disturbance, resulting in their downstream displacement.Morphology and the potential for downstream displacement can limit the upper altitude distribution of native stream fishes in flashy tropical streams like Rı ´o Mameyes (Schoenfuss and Blob 2007).Among the six native freshwater fishes found in Rı ´o Mameyes, mountain mullet are the only water-column dwelling species; all others are demersal, making them less susceptible to involuntary displacement.The potential for extensive downstream displacement of mountain mullet was observed in six PIT-tagged mountain mullet after the disturbance and in one radio-telemetered individual during a brief but intense spate outside of the study period.Furthermore, only mountain mullet were observed at high densities in the coastal plain and foothills reaches we surveyed after the disturbance.These findings suggest a higher probability of downstream displacement in mountain mullet than other fish species.
The flood disturbance effects on an amphidromous fish assemblage documented here are consistent with previous studies showing similar changes in the spatial distribution of amphidromous shrimp fauna in another northeastern Puerto Rico watershed following Hurricane Hugo in 1989 (Covich et al. 1996).At least one shrimp genus, Atya, appeared to follow a similar pattern as that of mountain mullet, with a 20% reduction in abundance at high elevations followed by greater than a 100% increase in abundance at lower elevations.Further, at broad, annual time scales another shrimp genus, Macrobrachium, was relatively unaffected by flood disturbances (Covich et al. 2006).We suggest that short-term alterations to the distribution of Macrobrachium may have occurred but were not observed because populations were monitored at a coarse time scale.Thus the conclusions of these studies on amphidromous shrimp (Covich et al. 1996(Covich et al. , 2006) ) are consistent with ours on amphidromous fishes.Amphidromous populations experience frequent, acute effects due to flood disturbance, but amphidromous populations and assemblages are chronically resilient to flood disturbances in the long term.
The removal methods we employed to estimate fish densities are recognized to be biased in some populations by heterogeneity of capture probability associated with the number of removal passes, fish size or age, and environmental variation (Williams et al. 2002, Peterson et al. 2004, Dauwalter and Fisher 2007).Previous assessment determined that three removal passes were sufficient sampling effort to accurately assess fish densities in Puerto Rico streams (Kwak et al. 2007).We directly accounted for variability in capture probability related to size or age by estimating size-specific capture probability.Estimates of capture probability in the removal model were highly variable, and some estimates were imprecise, indicated by wide credible intervals.Such variability in capture probability likely reflects habitat heterogeneity.Most fish did not rise to the water surface after being stunned by the electrofisher; they instead sank to the bottom, where they were more difficult to locate under cobble and boulders.Larger substrates exacerbated this variability in capture probability.Imprecision in capture probability estimates resulted in imprecision in estimates of the proportional change in fish density.Despite imprecision in capture probabilities, most estimates of fish density change were significantly different from zero (i.e., credible intervals did not include zero), and the overall decline in proportional change in fish density was consistent among species and size classes.

Ecological and management implications
We observed a stochastic mortality and dispersal event that demographically and spatially restructured tropical stream fish populations.Consistent with the predictions of Poff and Ward (1989), populations experiencing intense, stochastic disturbance at a moderate frequency (approximately once every four years in this case) were structured by abiotic events.Restructuring occurred only at the population level, manifesting as mortality and altered dispersal patterns.The disturbance, however, appeared to trigger rapid recovery processes that were related to biotic effects, including recruitment that replaced diminished adult populations and crowding or predatory aggregation in lower reaches.Population recovery processes appeared to dampen the assemblage level effects of the disturbance, as assemblage parameters did not change significantly.To fully understand the forces structuring populations and communities in tropical streams, we rely on an interpretation of the traits that amphidromous species have evolved to cope with frequent and intense disturbance (Lytle andPoff 2004, Death 2010).
The amphidromous life history has been interpreted as an adaptation for colonizing distant island stream habitats then recolonizing those same habitats following catastrophic disturbance (McDowall 2010).The critical life stage in this colonization adaptation is the marine larval dispersal phase (Fitzsimons and Nishimoto 1995).Larval dispersal can seed newly-formed volcanic islands across vast areas of ocean, and at the same time, marine larvae can act as a local pool of colonists that are relatively protected from the floods that occur in freshwater habitats.The movement information that we generated from tagging individual fish yielded insight into mechanisms and processes acting at both individual-and population-levels that explain the observed assemblage resiliency to flood disturbance.These findings add to the theory of amphidromous dispersal that adults of some species are also capable of recolonizing headwater habitats after downstream displacement and crowding.
Among theoretical models that may explain the observed native fish assemblage, mass effects models (Shmida and Wilson 1985) and neutral models (Hubbell 2001) of metacommunity dynamics are particularly relevant.Mass effects or source-sink models imply that heterogeneous communities result from interspecies variation in dispersal and reproduction, while neutral models assume that heterogeneity in community structure is primarily the result of variation in habitat quality (Falke and Fausch 2010).Consistent with McDowall's (2010) review of the amphidromous life history and stream communities' previously noted departure from Neutral Theory (Thompson and Townsend 2006), the amphidromous stream fish assemblage we examined appeared to be structured by dispersal and recolonization rather than habitat variability.Thus, the dynamics we observed in Rı ´o Mameyes are best described by a mass effects model.
The finding of a highly resilient native stream fish assemblage in Rı ´o Mameyes contrasts with others rivers on the island that are dominated by exotic fish fauna (Kwak et al. 2007).Other rivers with prolific exotic fish populations contain dams somewhere in the watershed, including more permeable low-head dams that do not restrict native amphidromous fauna, or extensive floodplains that may provide lentic habitat during flooding (Johnson et al. 2008); Rı ´o Mameyes is one among only a few free-flowing, unregulated streams on the island and supports minimal occurrences of exotic species (Hein et al. 2011, Cooney andKwak 2013).We hypothesize that the flood disturbance regime of tropical streams like Rı ´o Mameyes may provide biotic resistance to establishment of non-native species, rendering flooding a critical element in the conservation of native amphidromous stream communities.Future research may further elucidate the role of disturbance in maintaining native tropical stream fish communities by comparing disturbance effects on community dynamics in free-flowing and regulated streams.
Results from our study may inform conservation and management of diadromous fish assemblages on tropical islands globally.Our findings indicate the importance of high flows in regulating fish assemblage structure and the resilience of native species.Managing water releases from dams on regulated rivers has long been applied as a conservation tool for diadromous species (Annear et al. 2004) and more recently as a management practice for limiting the occurrence and spread of nonnative fish species, with variable and unexpected results (Fausch et al. 2001, Valdez et al. 2001, Cross et al. 2011).Dam construction has facilitated the invasion of aquatic habitats by species that are not resilient to the natural disturbance regime (Poff et al. 1997, Johnson et al. 2008), but regulation of water released from dams can be a valuable tool in the maintenance and restoration of native aquatic communities and biodiversity.We conclude that an ecological flow regime, providing minimum flows for continuous connectivity between fresh and marine waters and periodic high flow events to facilitate native aquatic communities is among the most important management strategies for conserving native fish assemblages on tropical islands.

Fig. 2 .
Fig. 2. Flow over time in Rı ´o Mameyes in relation to fish sampling events and Hurricane Irene.The solid line represents daily discharge in Rı ´o Mameyes over 2.5 years.Hatched boxes indicate dates of 28 three-pass depletion samples.Triangles and bars indicate monthly median flows and standard deviations estimated from 30 years of streamflow data collected at the U.S. Geologic Survey gauge station on Rı ´o Mameyes (USGS 2011).

Fig. 4 .
Fig. 4. Activity indices, total number of movement detections, for two passive integrated transponder tagged species, bigmouth sleeper (Gobiomorus dormitor) and mountain mullet (Agonostomus monticola), before and after a major flood disturbance, Hurricane Irene.Hatched areas indicate periods when passive integrated transponder arrays were disabled.

Table 1 .
Rank order of fish density estimates in Rı ´o Mameyes before and after the disturbance of Hurricane Irene.