Survival of a long-lived single island endemic, the Raso lark Alauda razae, in relation to age, fluctuating population and rainfall

Estimating and understanding variation in survival rates is crucial for the management of threatened species, especially those with limited population sizes and/or restricted ranges. Using a capture-resighting dataset covering 2004–2017, we estimate adult survival in the Raso lark Alauda razae, a Critically Endangered single-island Cape Verdean endemic, whose population varied 25-fold during the study. Average annual adult survival was similar for males (0.813 ± 0.011) and females (0.826 ± 0.011) over the period. These values are high for a temperate passerine but not unusual for an insular tropical species like the lark. The oldest bird was recorded 13 years after first ringing. There was strong evidence that survival varied among years (between 0.57 and 0.95), being generally higher in wetter years. Survival, especially of males, was lower when the population was large, but only in drier years. Survival declined with age but there was no evidence that this decline was other than linear. High survival, even in the face of dry conditions, at least when the population is depressed, has probably contributed to the persistence of the species on its 7 km2 island home over several centuries.


Results
The average resighting probability was high, for both females (0.87 ± 0.01) and males (0.89 ± 0.01). Average annual survival probabilities were similar between the sexes (female: φ = 0.826 ± 0.011; male: φ = 0.813 ± 0.011), but there was substantial variation between years (Fig. 1). Annual survival probabilities were more similar in the latter half of the study, when population size was higher, and they were influenced by the number of birds in the population and by rainfall.
Rainfall alone accounted for ~23% of this variation (Table 2), with survival being higher following years of greater precipitation (β = 0.22 ± 0.04), and survival of males being greater in wetter years (β = 0.32 ± 0.06) than females (β = 0.13 ± 0.05). Population size, by itself, explained less of the variation in survival probabilities (~13%), although again the effect was strongly sex-specific with males (β = −0.00062 ± 0.00015) having lower survival in years with high population size, while females were seemingly not affected (β = 0.00017 ± 0.00016). As a result, in   Table 1), annual population change was more closely related to male (r = 0.86, p = 0.003) than female survival (r = 0.54, p = 0.13). The impact of population size on survival probabilities was most marked in dry years (Fig. 2); in combination, rainfall and population size accounted for about two-thirds of the annual variation in survival probability in the population (Table 2). Individual survival probabilities declined with time since marking (Fig. 3), and presumed older birds (those with damaged claws on ringing) had a lower survival probability ( Table 3). The rate of decline in survival with time since marking was higher for females (β = −0.15 ± 0.05) than for males (β = −0.10 ± 0.04), and female individuals with damaged claws consequently exhibited a greater lowering of survival (β = −0.39 ± 0.20) compared to those with undamaged claws than did their male counterparts (β = −0.26 ± 0.19), though the estimated difference was small. The decline in survival with time since marking among the claw-damaged birds was slightly higher than among (presumed) younger individuals (β = −0.05 ± 0.07), suggesting the possibility of an accelerated decline with increasing age, but including a quadratic TSM term in the model did not improve the fit (Table 3).
An estimate of the survival of individuals of known age indicated that survival of first-year birds (φ = 0.63 ± 0.05) was lower than that of adults, and supported a decline in survival with age (Table 4), although the estimate of the rate of decline was small (β = −0.01 ± 0.17).  www.nature.com/scientificreports www.nature.com/scientificreports/ Despite the marked sexual dimorphism, there was no discernible effect of bill or wing length on annual survival of males and females, and that remained true regardless of rainfall (Table S1).

Discussion
Using a capture-resighting dataset of colour-ringed individuals observed from 2004 until 2017, we show that annual survival of adult Raso larks is generally high, but fluctuates in response to a combination of population density and rainfall. Additionally, survival was age-related, with older birds having lower survival rates, especially amongst females.
Previous studies (usually of insectivorous passerines) have shown that survival may be increased in wetter 13,14 or drier 15 times. Survival in the Raso lark, primarily a granivore during drier periods, was lower in drier years, in common with other granivores living in arid environments 16,17 , suggesting food resources may be a critical mediating factor 18 . Since rainfall also promotes breeding activity and larger clutches 8 , the implication is that there is little or no trade-off between reproduction and survival 19,20 in this species. Instead, the Raso lark seems to elevate its reproductive effort to take advantage of wetter years with more resources, avoiding this trade-off. Moreover, Estimates are for birds without claw damage; those exhibiting damaged claws had a lower intercept (β = −0.35 ± 0.14). Shading indicates 95% confidence limits about the regression. Reading from the top, the four grey-shade boundaries represent male upper confidence limit, female upper confidence limit, male lower confidence limit, and female lower confidence limit.  www.nature.com/scientificreports www.nature.com/scientificreports/ the impact of rainfall per se on survival was more pronounced than that of population size, which nonetheless interacted as a factor with rainfall: in wetter years, survival was little affected by population size despite an order of magnitude variation in population size. The absence of a signal of density-dependence in these wetter years was unexpected. However, in drier years, when food was presumably scarcer, survival was depressed when the population was higher.
This study bears on two discussion strands concerning avian survival. The first focuses on the characteristics of islands and the distinct impact that they could have on the survival rates of island species. Researchers frequently describe a pattern whereby island species and island populations of widespread species have higher survival rates than their continental counterparts 21,22 . Compared to continental birds, island species may shift more resources towards self-maintenance as a means of increasing survival in order to maximize lifetime reproductive success; they may be able to do this because of the generally more stable, milder climates on islands and the lower prevalence of parasites and predators 22 . These arguments are applicable to the island of Raso, which has a fairly stable temperature, little annual variation in daylength, very few predators for the lark and, in all likelihood given its aridity, few lark parasites 23 .
The Raso lark's annual adult survival rate is certainly high compared to the majority of continental passerine bird species. Its closest relative, the skylark Alauda arvensis, has a trans-Palaearctic breeding distribution and a much lower annual survival rate, estimated by different researchers between 0.39 and 0.78, with most studies placing it around 0.50-0.60 24 . The Raso lark's survival rate is also higher than that of the continental Dupont's lark Chersophilus duponti 25 , and much higher than that of most other continental passerines. Blake and Loiselle 26 estimate average survival rates of forest species in Eastern Ecuador at 0.58. Peach et al. 27 estimate average survival rates of granivorous southern African passerines at 0.54, and of insectivorous and nectarivorous passerines at 0.72. In Nigeria, McGregor et al. 28 estimate the average survival rate of birds at 0.60. Another survey reported a mean survival rate of 0.53 in North American passerines 29 . However, island passerines seem generally to have higher survival rates, comparable to that of the Raso lark 21,30-32 .
The second (highly debated) conjecture is that tropical passerines have higher survival rates than their temperate counterparts 26,29,33 . This conjecture is based on the fact that birds in the tropics generally lay smaller clutches than in the temperate zones, and on the inverse relationship between fecundity and survival 29 . It is doubtful whether this argument can be applied to the Raso lark whose survival (this study) and fecundity 8 are both higher when rainfall is higher.
Although average survival of males and females was similar, the most supported model ( Table 2) includes an interaction term between year and sex, suggesting that the influence of sex on adult survival varies over time. Indeed, while males seem to have had higher survival at the beginning of the study, this trend reversed between 2011 and 2015 (Fig. 1). Coupled with the population expansion derived from the recruitment of equal numbers of young males and females to the population, the reversal abolished the sex ratio bias that so prominently favoured males in the early years of the study 11 (Table 1). One potential factor explaining variation in the relative survival of males and females, body size, had no discernible effect (Table S1).
Interestingly, survival of females was less influenced by both of the environmental factors we explored (rainfall and population size) than that of males. A similar result has been shown recently in a Neotropical wren, indicating that another limiting factor can over-ride local environmental effects, perhaps through the increased cost of reproduction incurred by females 34 . Consistent with this, the survival of females declined more with age than did that of males (Fig. 3). This implies that if there is a sustained period of low reproduction, then a male bias (as was observed in the early years of the study; Table 1) will arise demographically if recruitment is insufficient to counterbalance the higher female mortality. Thus, these results suggest that the male bias commonly reported in threatened populations 12 may result from sustained low reproductive success rather than, necessarily, being induced directly by some environmental factors affecting female survival, at least in longer-lived species.
These results also suggest a possible explanation for why male survival was lower than that of females during 2011-2015. This could be a direct consequence of the reduction of the survival of males, but not of females, caused by the large population then on the island. Why the high population size should differentially impact males is not clear: one possibility is that their larger body size and metabolic needs place them at a disadvantage compared to smaller females when the population is high and competition for resources presumably most intense (although there was no evidence of differential survival between smaller and larger individuals). The phenomenon is unlikely to be caused by the cost of intrasexual competition for mates or territorial defence in males, as when the population is large and the environment is dry, birds generally forgo breeding and territory defence, and instead forage in flocks.
Older birds experienced lower survival than younger birds, a finding replicated in a number of other passerine studies [35][36][37] . Despite the fair size of our dataset, it was not possible to determine whether the decline in survival was more or less linear or whether there was an age at which the decline in survival with age abruptly steepened 38,39 . Anecdotally, the fact that five birds have been observed 12 years after first capture and only one subsequently at 13 years suggests that this could be an age at which survival probabilities deteriorate sharply.

conclusion
The Raso lark is an island-dwelling tropical species, typically with high survival. Only when a very dry year (or years) coincides with a high population is survival substantially depressed. These features enable the species to persist through several years of drought, investing in survival and only making the additional investment in reproduction when conditions once more become relatively favourable 40 . Such a strategy is viable because, at least for the first ten years of life, senescence approaches slowly. The features also clarify how the species has survived for several centuries on a single small island with a population at times falling below 100. Additionally, this suggests that, in this and similar species, conservation management actions may best target reproductive potential or, www.nature.com/scientificreports www.nature.com/scientificreports/ indeed, increase the range of the species, if this has contracted historically. With this in mind, and the Raso lark facing an uncertain future, we are currently attempting to (re-) establish a second population on Santa Luzia 41 .

Methods fieldwork on raso. After trial visits in 2002 and 2003, the present study commenced in 2004 and has contin-
ued annually to 2017 with single visits each year lasting 12-20 days. These visits occurred in November or early December. This is towards the end of the period, August-November, when birds are most likely to be breeding following rainfall 8 . However breeding is rain-dependent and may occur at other times of year, rain permitting. This fieldwork schedule meant we encountered numerous breeding attempts but have no data on the number of attempts individual birds may make in a year.
All birds were captured and ringed under permits issued by the Direcção Nacional do Ambiente (Environment Ministry), Cape Verdes. This catching was done by M. de L. B., a fully-licensed bird ringer (British Trust for Ornithology permit A 1871 MP). Thanks to the species' approachability, Raso larks can be captured individually by two people carrying a mist net, fully extended and horizontal, on two poles. When a target bird is sighted, it is approached downwind and the net dropped over the bird, which is then extracted immediately. Each bird received an individually-numbered metal ring and a unique combination of three Darvic colour rings. During the 2-3 week visit to Raso in November or early December, the 2-person team caught and ringed new flying birds, recorded the colour-ring combinations of surviving birds ringed in previous years, and also ringed nestlings and juveniles (<3 months old). The latter were readily recognized by their browner plumage with broader pale feather edgings.
Towards the end of each year's visit, the sustained reading of colour-rings attached in previous years consistently generated resighting rates approaching 90% (see Results). We therefore knew with fair accuracy the number of colour-ringed birds on the island, the sum of those ringed in the current year plus those ringed in previous years. The number of colour-ringed birds was corrected the following year to account for the small number (usually 15-20) of colour-ringed individuals that were spotted then but had been missed in the previous year, thus accounting for the incomplete detection of individuals. This correction is applied to the population values presented by Brooke 7 that are used in our analyses. Transects conducted across the island then allowed an estimate of the proportion of birds that were colour-ringed, from which we calculated an overall population estimate. The population has varied greatly in size during the study: in the early years (until 2009), the population did not  (Table 1).
Survival modelling. Survival of birds over the period 2004-2017 was estimated by fitting Cormack-Jolly-Seber (CJS) models to datasets with marked individuals 42 using RMark 43,44 . Initial goodness-of-fit testing (using Program RELEASE) suggested that the data fitted the CJS model acceptably (χ 2 41 = 55.8, p = 0.062), but closer inspection revealed evidence of significant 'trap-dependence' (Test 2.CT: χ 2 10 = 30.1, p < 0.001) with males, in particular, being more likely to be re-encountered if they had been seen the previous year. Consequently, we model resighting probabilities as sex-specific and dependent on whether the individual had been encountered in the previous year or not (i.e. four parameters). Including a time component in the resighting model did not change the survival estimates substantially. During the field period, an attempt was made to find all colour-ringed birds present on the island over a similar length of time, so we assume equal resighting rates across years. Our notation follows that of Lebreton et al. 42 , with a '*' indicating an interaction between covariates and a '+' indicating that covariates are additive (i.e. that they vary in parallel); also we use capitalised names to denote linear covariates and lower case to denote factorial covariates (thus, in our study, 'Year' explains 1 d.f. and 'year' 13 d.f.).

Annual variation in survival.
We were interested in the determinants of survival between years and considered three variables: sex, population density and rainfall. All post-juvenile birds (known juvenile birds are excluded from this analysis) could be sexed on size; indeed the Raso lark is one of the most sexually dimorphic lark species, with, for example, male bills 20-25 percent longer than those of females 11,45 . Rainfall has not been consistently directly measured on Raso, so we extracted an annual measure from the remote-sensed NCAR TRMM Multi-satellite Precipitation Analysis dataset (TMPA v7 46 ), which is available daily at 0.25° resolution. We downloaded the monthly accumulated combined microwave-infrared data for the 5° square centred on Raso and bounded by 14°N 22°W, 19°N 27°W for the period of the study from https://pmm.gsfc.nasa.gov/data-access/ downloads/trmm; this smooths out some of the stochasticity inherent in estimation at finer resolutions. On Raso, most precipitation occurs in the latter half of the calendar year so we summed daily totals for the period August -November and related these to the survival of larks for the following year, assuming survival to be related to food resources that are determined by rainfall in this period.
We were less interested in identifying the most parsimonious model (i.e. our 'best' estimate of annual survival), than the relative importance of the different covariates in determining the observed variation in annual survival. Consequently, following Grosbois et al. 47 , we calculate the amount of deviance explained relative to our null model, survival constant over time but differing by sex, and the full model, survival estimated separately for each sex in each year; this measure is then analogous to the R 2 metric familiar from linear regression.
Age-specific variation in survival. We looked for age-specific variation in two ways. First, with the dataset of adult captures, we looked for a relationship with time since marking (tsm), as this will correlate with age. In common with most lark species, Raso larks undergo a complete post-juvenile moult, probably when they are about three months old, after which birds of different ages cannot be distinguished, so tsm is not perfectly correlated with age as most individuals were ringed at unknown ages. It is worth noting, though, that, in years of major population expansion (e.g. a three-fold growth from 2010-2011), at least two-thirds of the birds in the population