Does forest extent affect salamander survival? Evidence from a long‐term demographic study of a tropical newt

Abstract Forest loss has been associated with reduced survival in many vertebrates, and previous research on amphibians has mostly focused on effects at early life stages. Paramesotriton hongkongensis is a tropical newt that breeds in streams but spends up to 10 months per year in terrestrial habitats. Populations are threatened by habitat degradation and collection for the pet trade, but the cryptic terrestrial lifestyle of this newt has limited our understanding of its population ecology, which inhibits development of a species‐specific conservation plan. We conducted an eight‐year (2007–2014) mark–recapture study on four P. hongkongensis populations in Hong Kong and used these data to evaluate relationships between forest cover, body size, and rainfall on survival and to estimate population sizes. Hong Kong has been subjected to repeated historic territory‐wide deforestation, and thus, we wanted to determine whether there was a link between forest extent as a proxy of habitat quality and newt demography. Annual survival was positively associated with forest cover within core habitat of all populations and negatively related to body size. Mean annual survival (~60%) was similar to that of other stream‐dwelling amphibians, but varied among years and declined substantially in 2012–2013, perhaps due to illegal collection. Despite the link between forest extent and survival, population sizes declined at the most forested site by 40% and increased by 104% and 134% at two others. Forest protection and consequential secondary succession during recent decades in Hong Kong may have been responsible for persistence of P. hongkongensis populations.


| INTRODUCTION
Forest loss has been associated with losses of a number of vertebrate taxa in tropical Asia (Castelletta, Sodhi, & Subaraj, 2000;Gibson et al., 2013;Laurance et al., 2012). Much of this previous work has focused on changes in species composition, and examinations of impacts to population demography are limited for the region. Many species with biphasic life cycles spend their early life stages in aquatic habitats and transition to terrestrial habitats as subadults. Core terrestrial habitat (sensu Semlitsch & Bodie, 2003) encompasses the upland habitat around breeding sites that many biphasic amphibians use for feeding, overwintering, and dispersal, and the importance of these habitats to maintaining viable populations has historically been underappreciated and overlooked (Semlitsch & Bodie, 2003). Proportions of intact core habitat around amphibian breeding sites are positively correlated with abundance and probability of occupancy, although the factors underlying this relationship, namely the linkage between habitat quality and survival, have yet to be clearly established (Gibbs, 1998;Homan et al. 2004;Peterman et al. 2011).
Terrestrial salamanders' response to deforestation/timber harvest in the United States is well documented: Relative counts of terrestrial plethodontids decreased by onefold to fivefold following clearcuts (Petranka et al. 1993;Sattler & Reichenbach, 1998;Knapp, Haas, Harpole, & Kirkpatrick, 2003); pond-breeding ambystomatids responded mostly negatively to clear-cuts and partial clear-cuts in terms of adult survival, juvenile survival, and water loss (Semlitsch et al., 2009); monthly survival of Plethodon shermani from harvested forest plots was 6% lower than those from unharvested plots (Connette & Semlitsch, 2015). Terrestrial habitat quality is likely to be especially significant for salamandrids, which, like many terrestrial plethodontids, spend substantial portions of their life cycles on land (Fu et al., 2013;Gibbs, 1998) and whose populations are highly susceptible to forest degradation (Schmidt et al. 2005;Cushman, 2006;Denoël, 2012 (Lau & Chan, 2004) and included on Appendix II of the Convention on International Trade in Endangered Species (CITES, 2017) because of concerns over population declines associated with habitat degradation and overexploitation for the pet trade. Forests throughout Hong Kong have been cleared repeatedly since the 16th century, with the most recent episode occurring in the 1940s during the Japanese occupation (Dudgeon & Corlett, 2011). However, postoccupation protection of the countryside, and the establishment of an extensive country park system in the 1970s, has allowed the regrowth of extensive tropical secondary forests, which are characterized by a mixture of native and non-native trees (e.g., Lophostemon confertus, Machilus spp., Schefflera heptaphylla; Dudgeon & Corlett, 2011).
Virtually nothing is known about effects of historic range-wide deforestation and contemporary forest recovery on P. hongkongensis, but its fidelity to breeding sites and predictable seasonal migration patterns (Fu et al., 2013) indicate that populations of this newt would be amenable to monitoring using mark-recapture methods. Studies of the demography of this newt in Hong Kong thus offer a unique opportunity to access the recovery potential of populations of a tropical forest-dependent amphibian.
Here, we present the results of a spatially replicated demographic study on a tropical salamandrid based on eight years of mark-recapture data. Our objectives were to identify factors that influence annual survival and determine population sizes of P. hongkongensis at these sites. Because P. hongkongensis spends up to 10 months per year (Fu et al., 2013) in core terrestrial habitat (i.e., riparian forest up to 113 m from breeding streams; Lau et al., 2017), and abundances of their insect prey on land tend to be positively associated with rainfall (Chan, Yu, Zhang, & Dudgeon, 2008;Yuen & Dudgeon, 2015), we hypothesized that annual survival of P. hongkongensis would be positively related to the extent of forest cover within its core habitat, as well as rainfall during their period spent on land each year.
Female newts typically oviposit on trailing, submerged bank-side vegetation, or among leaf litter and plant roots within slow-flowing stream pools. Larvae develop in water and migrate into surrounding terrestrial habitat following metamorphosis, which takes several months (Kong & Tong, 1986). Juvenile newts spend an estimated 1-3 years on land before returning to the stream as reproductive adults, which exhibit site fidelity and visit the same breeding pool repeatedly (Fu et al., 2013;Kong & Tong, 1986).

| Study sites and sampling methods
We conducted a mark-recapture study in four third-or fourth-order streams that are breeding sites for P. hongkongensis in Hong Kong, marked (by photographing its unique ventral pattern) and measured as described in Fu et al. (2013) (Figure 2). All newts captured were breeding individuals, and those captured only during one breeding season were considered to be transients (Kendall, Nichols, & Hines, 1997;Schmidt et al., 2007).
WildID (v1.0.1) was used for computer-assisted photograph identification to generate capture histories of individual newts (Bolger, Morrison, Vance, Lee, & Farid, 2012). Before photographs were analyzed by WildID, we extracted the ventral patterns (i.e., orange spots) of each newt from digital photographs using the "color range" function in Adobe Photoshop CS6 ( Figure 2). WildID, by default, presents pictures with the 20-highest similarity scores from the database to the user (Bolger et al., 2012). Based on our experience, the correct match occasionally does not appear among the top 20 candidates. To ensure correct identification and matching of individuals, the coding in WildID was modified by a computer programmer such that it presented the top 100 most similar individuals as potential matches.

| Modeling survival probabilities and population size
We used the robust design (hereafter RD; Kendall et al., 1997) implemented in Program MARK ver. 8.1 (White & Burnham, 1999) to model annual apparent survival (S), temporary emigration (G' and G"), capture probability (p, set equal to recapture probability (c) to assume no handling effects), and derived population size (N-hat). The basic structure of RD models includes multiple "closed" capture occasions (i.e., secondary occasions), between "open" survival intervals (i.e., primary occasions) (Kendall et al., 1997), which in this case correspond to sampling year (primary) and month (secondary). We did not include search effort as a parameter influencing capture rate because the effort at each site was standardized and remained constant over time (i.e., six surveys per year).
The primary parameters of interest were apparent survival, the covariates expected to affect survival (extent of forest cover; rainfall during the terrestrial phase; body length) and population size. We used a hierarchical approach to evaluate competing models to avoid unnecessary bias and imprecision in survival estimates (Lebreton, Burnham, Clobert, & Anderson, 1992;Lee et al., 2012). We started by building a fully parameterized global model with time and sex dependence in S,  G', G", p, and N-hat (Table S1). We first determined whether p was constant and/or varied over time (both by year and by month) and among sexes (e.g., p(.), p(sex), p(year)…p(sex + year + month). We then determined whether temporary emigration (i.e., the probability of being off the study area and unavailable for capture during a primary occasion) was "random" (G" = G') or "Markovian" (G'≠G") (Kendall et al., 1997) and whether it varied over time (secondary occasions within years) and among sexes. Lastly, we determined whether S varied by time and/or among sexes.
At all stages (i.e., before and after incorporating covariates), we identified the most parsimonious models using Akaike's information criterion, adjusted for small sample sizes (AICc/QAICs), as those with the lowest AICc/QAICc scores. Models within two AICc/QAICs units are considered to be indistinguishable from each other (Burnham & Anderson, 2002).
After determining the best model (i.e., model with the lowest ΔQAICc) for S, G", G', and p, we denoted it as a starting model and built additional models to examine the effects of forest cover extent within the core habitat, body size, and rainfall on apparent survival.
We incorporated different combinations of these factors as site, individual, or time covariates to the survival term in the starting model (e.g., S(forest cover + rainfall + year) vs. S(body size + year)) . Data from all four sites were pooled to estimate survival. Derived population size was estimated separately for each of the four sites. To minimize the number of parameters in our models, only simple linear relationships between apparent survival, core habitat forest cover, body length, and rainfall were considered, as more complex interactions between these factors were not expected.

| Extent of forest cover
We hypothesized that survival of P. hongkongensis is influenced by the quality of terrestrial habitat around their breeding sites, and survival will be higher in sites with higher extent of forest cover. We assumed high forest cover to be a reasonable proxy for habitat quality because adult P. hongkongensis remained active throughout the nonbreeding season only at sites that had extensive forest cover . We quantified forest cover by calculating the mean satellite-derived normalized difference vegetation index (NDVI) of the terrestrial habitat within a 113 m radius of each breeding pool.
This value represents core terrestrial habitat occupied by P. hongkongensis during its nonbreeding season . NDVI is a remote sensing index used to differentiate vegetated and nonvegetated land cover types (Glenn, Huete, Nagler, & Nelson, 2008;Pettorelli et al., 2005) and is commonly used as a measure of forest cover (Carlson & Ripley, 1997;Carreiras, Pereira, & Pereira, 2006 (Pettorelli et al., 2005). As we did not have access to multispectral satellite images from 2007 to 2014, we used images taken in January 2015 to calculate NDVI. However, because three of four sites are located within protected areas and most of the forests are >40 years postrecovery, we assumed forest cover in these sites remained relatively stable over the study duration. We verified this by comparing Google Earth images taken from 2007 to 2014 (Fig. S1). In KP, because the two stream pools are less than 10 m apart, we used the same NDVI value for both pools.

| Rainfall during the terrestrial phase
To quantify rainfall at study sites, we obtained monthly rainfall records from the Hong Kong Observatory (http://www.hko.gov.hk). We calculated mean cumulative rainfall for the six-month period (April-September, when ~90% of the annual rainfall in Hong Kong was accumulated) preceding a breeding season, using data from the three weather stations nearest our four study sites. For example, for the primary occasion t 2 , the 6-month cumulative rainfall recorded between t 1 and t 2 was used.

| Goodness of fit
To verify survival and recapture rate estimates obtained from RD models and to obtain breeding pool-specific survival and recapture probability estimates, we collapsed the encounter histories to only  (Lebreton et al., 1992).  (Table 1). In total, 957 newts were recaptured (return rate = 25.6%) on two or more primary occasions, while 2,716 newts were transients. Recaptured individuals provided the data from which annual survival was modeled.

| Factors influencing survival
The extent of forest cover around breeding sites and body length (SVL) was the most important factors that influenced survival in P.
hongkongensis, as models that included NDVI and SVL consistently ranked higher than those without ( Table 2). The most parsimonious RD model was [S(year + NDVI + SVL) G"(sex + month) = G' p(sex + year + month) = c f 0 (sex + year)], which indicates that apparent survival (S) of adult P. hongkongensis varied between years and was positively associated with the extent of forest cover within core habitat (β NDVI = 15.46, SE = 1.46) and was negatively related to body size (SVL, β SVL = −0.03, SE = 0.01). Newt S (survival) at the site with the highest forest cover was almost 30% higher than at the site with the least cover (Figure 3a), and S of the smallest newt was almost 30% higher than that of the largest individual (Figure 3b). S was stable across most years (average = 0.58 ± 0.04, range = 0.28-0.77) but was considerably reduced (to >0.30) during 2012-2013 (Figure 3c).
Median c-hat procedures conducted on breeding pool-specific CJS models showed a reasonable fit (mean c-hat across all 7 pools = 1.38, range = 0.80-2.52). Survival and capture probability estimates from RD models were similar to those estimated from the CJS model but were more precise with smaller margins of error. Survival estimates (Phi) from CJS models that did not include forest cover as a covariate yielded similar survival estimates to S (Figure 3a). Capture probability (p) was typically <0.4 and was both time-(vary with sampling month and year) and sex-dependent. Capture probability (p) of females was lower than that of males early in the breeding season (October-November), but from December-March, p was similar between sexes ( Figure 4a). Temporary emigration was random (i.e., G" = G'), meaning that probability of a newt being in a breeding pool during one survey was not dependent on whether it was in the breeding pool during the previous survey, but varied with month and between sexes similarly to p (Figure 4b).
Of the 957 individuals that were captured in multiple seasons, 306 (32%) were not captured in study pools during one breeding season, 79 (8%) for two seasons, and 11(1%) for at least three seasons. One T A B L E 1 Number and average body length of Paramesotriton hongkongensis marked and recaptured from four sites in Hong Kong (2007Kong ( -2014  T A B L E 2 Top ten ranked robust design models (based on ΔAICc), using data from 3,674 Paramesotriton hongkongensis marked and recaptured from four sites in Hong Kong (2007Kong ( -2014 testing the effects of core habitat forest cover (NDVI), body length (SVL), and nonaquatic period rainfall (RF) on apparent survival (S)

| DISCUSSION
Annual survival of P. hongkongensis increased with greater extent of forest cover around breeding pools, suggesting that this factor was  (Connette & Semlitsch, 2015;Otto, Roloff, & Thames, 2014;Rittenhouse, Harper, Rehard, & Semlitsch, 2008;Rothermel & Semlitsch, 2006;Todd & Rothermel, 2006) and consequential increases in risk of population extinction where forest cover has been reduced substantially , as well as life stage-specific responses to different timber management practices (Semlitsch et al., 2009). As population dynamics in longerlived amphibians, such as P. hongkongensis, are mediated largely by the survival of adults (Berven, 1990;Gill, 1978;Wilbur, 1980), protection of undisturbed forests around breeding sites is essential for population viability. This applies particularly to species such as P. hongkongensis that spend, on average, over 10 months of the year away from water (Fu et al., 2013; the present study).
In contrary to our initial prediction of a positive effect of body length on survival, smaller newts had higher survival than larger individuals. Survival rates of salamander have been shown to be positively correlated with age and size . Unfortunately, we were not able to reliably age individual newts in the present study, as size has been proven to be a poor indicator of age for other species of newts, and skeletochronology may not be a reliable method for aging tropical species (Halliday & Verrell, 1988;Kusrini & Alford, 2006). Consistent with what was reported in an earlier study (Fu et al., 2013), newts from well-forested sites (MTL and HC) were longer than those from less-forested sites (KP and PNS) ( Table 1). We suspect the negative effect of size on survival and lower survival observed in a well-forested site could be related to the difference in age structure among our sites. This could also explain the population decline observed in HC.
Alternatively, or in addition, the decline of the HC population may be linked to its proximity to channelized streams. Hong Kong has an extensive network of artificial channels constructed to divert rain water from streams into reservoirs to control flooding and secure the water supply of the city's seven million occupants (Dudgeon, 1996).
These channels are a source of mortality for wildlife that become entrapped as a result of their steep-sided, smooth concrete surfaces, and are flushed downstream by high flows following heavy rain. During the breeding season, adults Hong Kong newts are often found within these artificial channels. However, these channels could be ecological traps to the newts as they contain no natural substrates and are generally shallower than natural streams where newts typically breed in.
Newts trapped in these artificial channels could be more susceptible to being washed out by floods and to desiccation if the channels dry out.
The estimated survival rate of adult P. hongkongensis (~60%) was similar to that of T. cristatus in Europe (Griffiths et al., 2010) and a sympatric stream-dwelling frog from Hong Kong (Quasipaa spinosa, Dicroglossidae;Chan, Shoemaker, & Karraker, 2014), but considerably lower than that of other temperate newts (Taricha torosa: >90% in most years, Petranka, 1998;Salamandra salamandra: 86%, Schmidt, Itin, & Schaub, 2014). Our apparent survival estimates may be biased low due to the high proportions of transients (>70%) in our populations (see Schmidt et al., 2007), which is not uncommon among capturemark-recapture studies of salamanders (52% over 3 years, Schmidt  49% over 5 years, Lee et al., 2012;93% over 3 years, Unglaub, Steinfartz, Drechsler, & Schmidt, 2015). Given our relatively low capture probability (~40%), the presence of large numbers of transients may be due to some newts entering and leaving pools without being captured. Alternatively, individuals may have bred in other pools (Bucciarelli, Green, Shaffer, & Kats, 2016) or skipped breeding when environmental conditions were unfavorable (Fontenot, 1999). hongkongensis are cryptic and rarely observed in the wild (only 24% of 117 newts found during terrestrial transect surveys were subadults; Lau et al., 2017), and little is currently known about their habitat requirements or growth rates. Juvenile survival rate of salamanders is notoriously hard to estimate directly because of difficulty in relocating juveniles on an annual basis, but it is an important parameter in staged-based demographic models built to predict population trends Trenham & Shaffer, 2005). These models have shown that amphibian population growth can be sensitive to variation in terrestrial juvenile and adult survival rates (Biek et al. 2002;Vonesh & De la Cruz, 2002), and relatively small changes in adult survival can have a large impact on population growth rate (Homyack & Hass, 2009).
Although interyear fluctuation in survival was apparent P. hongkongensis (Kolby et al., 2014). Although Hong Kong newt populations within country parks are supposedly protected by law from poaching, these parks are not patrolled at night and are easily accessible by poachers. Because of the tendency of P. hongkongensis to aggregate in the same breeding pools year after year, poaching could present a serious threat to the long-term persistence of wild populations (Lau & Chan, 2004). Indeed, such vulnerability is a serious concern for other species of tropical salamandrids exhibiting similar breeding habits (e.g., Laotriton laoensis; Phimmachak et al., 2012). It is unclear whether the sudden decline in survival observed in 2012-2013 was related to collection during summer 2012 when newt shipments were imported to the United States, but the exported animals were unlikely to be captive bred-something that is not known to occur on any scale in Hong Kong or southern China.
Although P. hongkongensis is, in theory, protected from poaching