Population Viability of the Narrow Endemic Helianthemum Juliae (CISTACEAE) in Relation to Climate Variability

BIOLOGICAL CONSERVATION 136 (2007) 552– 562

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/biocon

Population viability of the narrow endemic Helianthemum juliae (CISTACEAE) in relation to climate variability

Manuel V. Marrero-Go´meza, J. Gerard B. Oostermeijerb, Eduardo Carque´-A´ lamoa,A´ ngel Ban˜ ares-Baudeta,* aParque Nacional del Teide, Calle Emilio Calzadilla, no. 5 – 4Izq 38002, Santa Cruz de Tenerife, Canary Islands, Spain bUniversity of Amsterdam, Institute for Biodiversity and Ecosystem Dynamics, Kruislaan 318, 1098 SM Amsterdam, The Netherlands

ARTICLE INFO ABSTRACT

Article history: Narrow endemic plants are highly vulnerable to extinction as a result of human distur- Received 27 June 2006 bance and climate change. We investigated the factors affecting the population viability Received in revised form of Helianthemum juliae, a perennial plant endemic to the Teide National Park on Tenerife, 2 January 2007 Canary Islands. One population was demographically monitored from 1992 to 2001 and Accepted 5 January 2007 analysed using matrix projection models to determine ﬁnite rates of increase and critical stages in the life cycle. Lambda values varied between 0.697 and 1.740, and were highly positively correlated with annual precipitation, but not with temperature. Survival of adults Keywords: had the highest elasticity, and summed elasticities of the growth and fecundity transitions Endangered plants correlated positively with lambda and precipitation. Most of the mortality in the population Extinction seemed drought-related, and no other threats were identiﬁed. Deterministic simulations PVA showed population increase, but introducing environmental stochasticity by modelling Population dynamics variation in precipitation from existing data of the past 85 years revealed high extinction Conservation probabilities (0.74–0.83 in the next 100 years). This plant is likely to be at risk under scenar- Global warming ios of global warming. Our simulations suggest that augmenting the population would only Canary islands delay extinction. A more viable option for long-term conservation seems to be the introduction of populations at more humid locations within the Teide caldera. 2007 Elsevier Ltd. All rights reserved.

1. Introduction always received special attention from conservation organi- zations. This is certainly true for geographically highly One of the main tasks of the conservation botanist is to pro- restricted endemic plants, which are exceptional representa- tect endangered plant species from extinction. In a way, all tives of certain regions. In many cases, the presence of such plants face the risk of extinction due to various causes, such species has been the motivation for the founding of national as habitat destruction, habitat fragmentation, displacement parks, such as the Sierra Nevada National Park in Spain (Bo- by or hybridization with invasive exotic species, climate letı´n Oﬁcial del Estado, 1999). It is therefore not surprising change, overharvesting for economic purposes, etc. (Ooster- that the conservation of endemic plant species often has meijer, 2003). Rare plants, however, are particularly sensitive very high conservation priority in such parks (Marrero to these threats, because of their restricted distribution and et al., 2003a). low numbers of populations and individuals (Goodman, Yet, it is not always clear whether endemic species are in 1987; Menges, 1991). Because of this vulnerability, they have need of intensive conservation management (Ban˜ ares et al.,

* Corresponding author: Tel.: +34 922290129; fax: +34 922244788. E-mail address: [email protected] (A´ .Ban˜ ares-Baudet). 0006-3207/$ - see front matter 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.biocon.2007.01.010 BIOLOGICAL CONSERVATION 136 (2007) 552– 562 553

2001). After all, they have survived in their restricted distribu- del Obispo populations are located on small rocky ledges over tion areas for very long time periods without any special care, an area of 300 m2; the Risco Verde population is found on deb- so why should they need any now? It is clear that when they ris at the base of cliffs covering a surface of 200 m2. are subjected to any threats imposed by human actions (hab- At a distance of 800 m from the Can˜ ada de las Pilas pop- itat destruction and fragmentation, invasive exotic species, ulation, an experimental introduction (sensu Falk et al., overharvesting, etc.), conservationists should focus on 1996) was established in Can˜ ada de Diego Hernańdez in removing these threats as soon as possible. However, when 1989–1990. The introduction comprised a total of 350 human threats are not so obvious, as is the case with the ef- three-year old plants, planted over an area of 500 m2, which fects of climate change for example, removing them is also were grown ex situ from seeds collected in the Can˜ ada de las less straightforward (Oostermeijer, 2003). Pilas population. In 1992, 122 plants had survived and abun- Assessing the impact of different threats on the survival of dant recruitment by seedlings was observed. In this year, populations is the main aim of population viability analysis, the population was designated for demographic monitoring or PVA. In this paper, we report on such a PVA for the plant in order to evaluate the success of the reintroduction and species Helianthemum juliae Wildpret (CISTACEAE), which is determine the factors that affect population viability in this unique to a very small area on the crater wall of the Teide Na- species. tional Park on Tenerife, Canary Islands, Spain. Specifically, we The local microhabitat differences are minimal among the ask the following questions: native and the reintroduced populations, especially in relation with the factors that might affect soil water availability. (1) What is the risk of extinction of the study populations? The differences in altitude among the sites are less than (2) Which are the most critical life-stages in the life cycle? 100 m (2141–2236 m.a.s.l.), they all have a West-exposition (3) Which environmental variables are the most important (Azimuth 268–276), the slope is similar (36–42) and the sub- determinants of extinction risk? strate has identical characteristics: weakly developed soils (4) To what extent might climate variability have affected (Orthents) over salic geological materials. population dynamics in the past and how will it possi- bly affect future population viability? 2.2. Data collection and analysis (5) Which strategies are indicated for the conservation of this species? 2.2.1. Life stages Demographic censuses were made during 10 years. We in- A very useful tool to resolve these questions are stage- or stalled a 10 · 10 m2 monitoring plot protected with a 1.75 m size-structured matrix projection models, that can be con- high aluminium fence with the aim to exclude rabbit grazing. structed on the basis of medium- to long-term demographic All individuals were labelled with aluminium identification data (Menges, 1986; Morris and Doak, 1998; Caswell, 2001). If tags and were classified at each census in one of the following such data are available over a long period, it is possible to test four categories: Juveniles (small, vegetative individuals with a the relationships between demographic transitions and cli- height <15 cm and no signs of previous flower stalks); Vegeta- matic variables such as temperature and precipitation. For tive adults (individuals >15 cm with evidence of previous flow- H. juliae, we have performed demographic censuses for 10 er stalks but vegetative at the moment of censusing), young years between 1992 and 2001, and obtained 85 years of cli- reproductive adults (plants <20 cm tall with flower stalks) and mate data from a nearby meteorological station. mature reproductive adults (>20 cm tall with flower stalks). In the beginning, the plot enclosed 22 mature reproductive 2. Materials and methods adults, seven young reproductive adults, 10 vegetative and seven juvenile individuals. 2.1. Species and study area We devised this stage classification to characterize the population using a biological approach (Lefkovitch, 1965; Wer- The genus Helianthemum is represented on the Canary Islands ner, 1975), which relies on field observations of developmen- by 12 species, of which 7 are endemic to this archipelago tal states. First, individual plants were classified in the basic (Hansen and Sunding, 1993). They colonize most of the eco- stages, i.e. juveniles, vegetative and reproductive individuals. systems existing on these islands; from the xerophytic coast- Later, the reproductive individuals were divided into two cat- al landscapes to the high mountain shrublands. H. juliae is a egories based on the intensity of flowering. The first class small iteroparous shrub that is endemic to the high mountain (young reproductive adults) comprises the adults in their first ecosystem of Tenerife (Canary Islands). Adult plants are up to or second year of flowering. They have a small size (<20 cm) 30 cm tall and produce between 5 and 40 flower stalks with and produce less than 50 flowers. The second class (mature yellow flowers from their third or fourth year onwards. The reproductive adults) includes the larger plants (P20 cm tall) average fruit contains 45 seeds. The average life span of indi- with more than 50 flowers. This subdivision was based on vidual plants is relatively short (14 years). field observations of growth and survival rates of reproductive H. juliae is considered critically endangered (CR) (Marrero adults. We did not consider a seed stage because seeds germi- et al., 2003b). Only three small populations are known from nated directly after dispersal, so we assumed a short-lived the wall of the Can˜ adas del Teide caldera (Fig. 1): Can˜ ada de seed bank. Hence, introducing seeds into the life cycle model las Pilas (43 adults), Risco Verde (25 adults) and Mesa del Obis- would introduce a time-lag (Caswell, 2001). We recorded the po (49 adults). Globally, the species has only 868 individuals numbers of seedlings in the autumn and spring cohorts dur- (adults and immatures). The Can˜ ada de las Pilas and Mesa ing additional censuses in December and March. We mapped 554 BIOLOGICAL CONSERVATION 136 (2007) 552– 562

Fig. 1 – Distribution of Helianthemum juliae on the island of Tenerife. The asterisk shows the locality of the planted population at Diego Herna´ndez (1) and the circles indicate the localities of the natural populations (2, Risco Verde; 3, Can˜ ada de las Pilas; 4, Mesa del Obispo). The shaded area represents the Teide National Park.

the position of all recent seedlings, and tagged them with ual. Secondly, we estimated the proportional contribution of small plastic sticks. each reproductive life stage to the total number of seeds pro- Each year, at the end of the summer, just before the first duced in the plot. Thirdly, the total number of juveniles autumnal rains, new individuals emerging in the plot were counted in the following year was allocated to the two repro- tagged and the life stage and size (number of stalks, flowers ductive life stages according to their proportional contribu- and fruits) for new and old the plants were recorded. The tion to the total reproductive effort. Finally, the number of transitions between the four different life stages observed in recruits per reproductive life stage was divided by the total the study population are shown in the life-cycle diagram number of plants in each stage to obtain fecundity. (Fig. 2). We first used Lefkovitch matrices based on the classification The collection of demographic data is sometimes consid- of individuals into life stages (Caswell, 2001). The basic matrix ered a threat to small populations that grow in restricted model is given by: n(t+1) = An(t); where nt and n(t+1) are vectors areas. Trampling of seedlings and small individuals may oc- whose elements, aij are the number of individuals that belong cur, and researchers may cause soil compaction. We avoided to the ith category at time t and t + 1, respectively, and A is a these negative effects on the natural populations by working non negative square matrix, whose elements, aij represent in the experimentally introduced population, which we as- the transitions or contributions from individuals in the jth cat- sumed experienced highly similar environmental conditions. egory to the ith category after one time step (Caswell, 2001). We used the Pop-Tools 2.4 spreadsheet for Excel (Hood, 2.2.2. Analysis of demographic data 2002) to calculate the main demographic parameters from Transition probabilities were obtained by calculating the pro- the projection matrices: the finite rate of increase (k) of the portion of individuals in each category experiencing each spe- population, the stage-specific reproductive values and the cific fate from one year to the next. Fecundity was calculated stable-stage distribution, and the elasticity matrices. The lat- as follows. First, we annually counted the total number of ter were used to study the relative importance of different capsules produced by each plant, and removed two capsules phases of the life cycle for the population growth rate (de from each plant to count the seeds, obtaining the average Kroon et al., 1986). The bootstrap method (Caswell, 2001; Kal- number of seeds per capsule and, by multiplying this value isz and McPeek, 1992) was used to establish 95% confidence with the total number of capsules, the total seeds per individ- intervals for k. We used an R-script that randomly resampled BIOLOGICAL CONSERVATION 136 (2007) 552– 562 555

F12 F13

G24 S42

G G 21 Young 32 Mature Juveniles Vegetative adults reproductive adults reproductive adults (J) (V) (Y) (M)

S11 S22 S33 S44

Stage at time t JYMV

JSFF0 + 1 11 12 13 t

YGS0G21 22 24

M0 GS32 33 0 Stage at time Stage

V0S042 S 44

Fig. 2 – Life cycle graph for Helianthemum juliae and the corresponding transition matrix. Circles indicate life stages, arrows represent the possible transitions among them, and letters show the connection between each transition and its matrix entry (F, fecundity; G, growth; S, survival).

each annual matrix with replacement 5000 times. The log- Performance Systems, 1998). The mathematical basis of this likelihood ratio, G (Sokal and Rohlf, 1981; Zar, 1984) was used model was identical to the matrix model described above. to determine significant differences in transition probabilities Stella software allows a greater deal of flexibility in relating between data sets. transitions to other variables than software designed purely In order to test for possible relationships between k and for the analysis of matrix models. With the aim to avoid col- climatic variables, we conducted two regression analysis linearity problems, we chose to introduce stochasticity by using k as dependent variable and annual total precipitation randomly drawing an entire matrix (from the nine available and annual average temperature as independent variables. matrices) for each simulated time-step. All simulations pro- At the same time we used the climatic data of the last 85 jected population dynamics over 100 time-steps (years) and years to detect historical trends that could be used for predic- each simulation was iterated 100 times. tive models. For an analysis of precipitation at a short term To calculate the stochastic growth rate, the numeric values we used the data of the Diego Hernańdez meteorological sta- for the population size (nt) generated for each year of a simulation, situated very close to the monitoring plot. Analysis of tion were used. It was obtained from the average growth rate long-term trends in temperature and precipitation was based over a long simulation (Dennis et al., 1991; Caswell, 2001; Garcıá on climatic data of Izan˜ a meteorological station (at 6 km dis- and Iriondo, 2002) as follows: logks = (lognt logn0)/t; where nt tance), with records since 1916; and Diego Hernańdez meteo- and n0 are the population size at the end and at the beginning of rological station (at 500 m distance) with data from 1990. the simulation, respectively, and ‘t’ is the total duration of the Using the data available from these two stations, we estab- simulation (100 years). Stochastic growth rate is presented as lished a correlation between the two stations for the last 10 logks (Caswell, 2001) and ks (Tuljapurkar et al., 2003). Addition- years, and obtained an equation that could be used to calcu- ally, the probabilities of quasi-extinction were calculated using late for the last century the amount of precipitation in the an approximation of the matrix model to a diffusion model area where the species grows. The obtained equation was: (Tuljapurkar and Orzack, 1980), where we considered two dif- 2 PDH = 0.664PI + 48.8 (r = 0.821), where PDH is the annual pre- ferent quasi-extinction scenarios: nt =5 (nq = 5) and nt =10 cipitation at the Diego Hernańdez station, and PI is the annual (nq = 10). So we can define a quasi-extinction threshold (h)as precipitation observed at Izan˜ a station. nq/n0 and the probability of eventual quasi-extinction (Pq)as

the probability that nt/n0 ever falls below (h); this is 1 if logks 6 0, 2 2.2.3. Stochastic simulations of population dynamics and exp((2logks logh)/r ) if logks P 0(Caswell, 2001). In this

To simulate environmental stochasticity we constructed a model the quasi-extinction time (Tq) is deﬁned as the ﬁrst time demographic model using the software package Stella (High that nt reaches nq and its distribution responds to a inverse 556 BIOLOGICAL CONSERVATION 136 (2007) 552– 562

Gaussian distribution (Folks and Chhikara, 1978) with mean (Table 1). Higher values (>1) were obtained from 1993 to 1998.

[E(Tq)] and variance [V(Tq)] as follows: E(Tq)=logh/jlogksj; In the most recent years, k’s were <1. We observed an increase 2 3 V(Tq)=r logh/jlogksj . in k during the first four years of monitoring, reaching a max- A process of restricted random selection to simulate the imum of 1.740 in 1995–1996. From that year on, k decreased to climatic variables was introduced in the Stella model. From 0.697 in 2001. the analysis of the climatic and demographic data we estab- The stable stage structure obtained from the average ma- lished ‘good years’ in which the annual precipitation was trix differed significantly from the observed initial distribu- higher than 350 mm, because we found that in these years k tion (G = 51.936; P < 0.010) for all stages, and shows lower was higher than 1. Consequently, ‘bad years’ were imposed values for juveniles (0.152) and young reproductive adults when annual precipitation was lower than 350 mm and (0.152), and higher values for mature reproductive adults k < 1. From the climatic data of the last 85 years, we calculated (0.478) and vegetatives (0.217). As expected, the highest repro- the percentage probability of good years (pgy) that was used to ductive value corresponded to the reproductive adult stages introduce a Montecarlo function in the model. If the Monte- and the lowest to the juvenile and vegetative adult stages carlo function selected a good year, a second random function (Table 1). picked a matrix from the set of the five matrices that had The highest elasticities (Table 1) were associated with the k > 1. For cases in which the Montecarlo function selected a survival of mature reproductive adults and the transitions bad year, a random selection occurred from the set of the four from juvenile to small reproductive adult. observed matrices with k <1. A diagram of the relative contribution of Growth (G), Finally, we used an additional model of climate-related Survival (L) and Fecundity (F) to the lambda of each matrix stochasticity, in which k was first calculated directly from (Silvertown et al., 1993) shows that all matrices are the randomly selected precipitation value using the equation concentrated towards the bottom right-hand of the triangle that relates k to precipitation in Izan˜ a and Diego Hernańdez (Fig. 5). This pattern corresponds to high elasticities for (see Section 2.2.2), and then subsequently used to calculate survival (L), and low for fecundity (F) elements. Higher k’s the new population size. corresponded to years with higher elasticities for fecundity and growth, and smaller k’s were associated with years 2.2.4. Extrapolation to the natural populations with higher elasticities for survival. Likewise, years In order to translate the observations from our experimental with a precipitation >350 mm corresponded with high study population to the natural situation, we used a set of fecundity and growth elasticities, and years with precipita- parameters that was easily measurable without disturbing tions <350 mm corresponded with high elasticities for the populations. In 2000, we determined population structure, survival. seedling mortality and fruit production of the natural and planted populations. To determine fruit production, the num- 3.2. Stochastic simulation model ber of capsules per individual was counted in each population. To assess seedling mortality, 100 recent seedlings were When environmental stochasticity was introduced by ran- labelled in December. Survivors were counted in November domly drawing (Fig. 4) from the nine transition matrices, of the next year. We tested differences between the planted the stochastic growth rate of the population (logks) was 0.12 and the natural population of Can˜ ada de las Pilas with a G- (ks = 1.13), indicating a very low extinction risk. The initial test. We selected only this population for three reasons: (1) population size used in the simulations was that observed to minimize the disturbance of the other natural sites intro- in 2001 (N = 313 individuals). The quasi-extinction probabili- duced by our visits; (2) our field observations indicated that ties are practically zero: Pq(5) = 1.33 · 1021; Pq(10) = 1.02 · only this population had a sufficient number of seedlings; 1017. However, these results did not correspond with the and (3) the population of Mesa del Obispo had not been dis- field observations, where the populations apparently did covered yet. not increase. This is probably due to the fact that even though we studied the population for 10 years, we encountered a rel- 2.2.5. Retrospective analysis atively high proportion (55%) of good years (i.e. years with Using the correlation between the precipitation record of the precipitation >350 mm and k > 1) in comparison to the fre- Izan˜ a and Diego Hernańdez stations, we calculated the theo- quency of years with high precipitation observed during the retical annual precipitation in the last century for the area past 85 years in the extrapolated data for Diego Hernańdez where the species grows. Then, these data were used to calcu- (30%). late a theoretical value of k for each year. With these k values Population growth rate increased with annual precipita- and the average population size (268 individuals), we calcu- tion during the study period [(k = 0.0032 * P 0.0018), lated the hypothetical population size at the beginning of r2 = 0.934, P = 0.0001]. However, temperature was not signifi- the 20th century. cantly correlated with k (r2 = 0.002, P = 0.856). We observed that when the annual precipitation fell below 350 mm, k val- 3. Results ues were <1, and vice versa. Accordingly we used the 350 mm threshold as the limit between good and bad years. The fre- 3.1. Deterministic transition matrix analysis quency of years with >350 mm precipitation during the last 85 years was 30%. The probability of a sequence of three con- The k value of the average matrix was 1.247. However, the k secutive good years in these data was 10%, which is consider- values were quite variable among the nine transition matrices ably lower than the probability of randomly drawing three BIOLOGICAL CONSERVATION 136 (2007) 552– 562 557

Table 1 – Projection matrices and elasticity matrices for the annual transitions from 1992 to 2001 and the average matrix for the planted population of Helianthemum juliae and their dominant eigenvalues (k), conﬁdence intervals (CI), reproductive value and stable stage distribution k CI Demographic matrix Stable-stage Reproductive Elasticity matrix distribution value

0.931 0:818–1.021 1992–1993 J Y M V J Y M V J 0.286 0.000 0.045 0.000 0.060 0.174 0.010 0.000 0.022 0.000 Y 0.429 0.429 0.000 0.500 0.066 0.262 0.022 0.024 0.000 0.006 M 0.000 0.286 0.909 0.000 0.858 0.357 0.000 0.022 0.886 0.000 V 0.000 0.143 0.000 0.300 0.015 0.207 0.000 0.006 0.000 0.003 1.358 1.000–1.471 1993–1994 J 0.667 0.000 1.364 0.000 0.518 0.119 0.129 0.000 0.134 0.000 Y 0.333 0.455 0.000 1.000 0.206 0.246 0.134 0.073 0.000 0.011 M 0.000 0.455 1.000 0.000 0.262 0.453 0.000 0.134 0.375 0.000 V 0.000 0.091 0.000 0.000 0.014 0.181 0.000 0.011 0.000 0.000 1.146 0.925–1.242 1994–1995 J 0.656 0.000 1.333 0.000 0.577 0.076 0.124 0.000 0.092 0.000 Y 0.188 0.500 0.000 1.000 0.194 0.198 0.092 0.083 0.000 0.014 M 0.000 0.200 0.963 0.000 0.212 0.553 0.000 0.092 0.487 0.000 V 0.000 0.100 0.000 0.000 0.017 0.173 0.000 0.014 0.000 0.000 1.740 1.605–1.819 1995–1996 J 0.596 2.000 2.393 0.000 0.647 0.115 0.131 0.139 0.112 0.000 Y 0.404 0.417 0.000 1.000 0.205 0.326 0.252 0.082 0.000 0.009 M 0.000 0.500 1.000 0.000 0.138 0.372 0.000 0.112 0.152 0.000 V 0.000 0.083 0.000 0.000 0.010 0.187 0.000 0.009 0.000 0.000 1.545 1.315–1.616 1996–1997 J 0.320 2.241 4.882 0.000 0.663 0.068 0.067 0.173 0.082 0.000 Y 0.336 0.483 0.000 1.000 0.246 0.248 0.255 0.136 0.000 0.044 M 0.000 0.138 0.912 0.000 0.053 0.524 0.000 0.082 0.118 0.000 V 0.000 0.241 0.000 0.000 0.038 0.160 0.000 0.044 0.000 0.000 1.375 1.317–1.422 1997–1998 J 0.539 0.070 2.000 0.000 0.552 0.099 0.096 0.005 0.145 0.000 Y 0.362 0.263 0.000 0.857 0.199 0.229 0.150 0.039 0.000 0.016 M 0.000 0.421 1.000 0.000 0.224 0.529 0.000 0.145 0.387 0.000 V 0.000 0.175 0.000 0.000 0.025 0.143 0.000 0.016 0.000 0.000 0.898 0.813–0.966 1998–1999 J 0.577 0.000 0.000 0.000 0.000 0.234 0.000 0.000 0.000 0.000 Y 0.291 0.479 0.000 0.600 0.000 0.258 0.000 0.000 0.000 0.000 M 0.000 0.261 0.898 0.000 1.000 0.287 0.000 0.000 1.000 0.000 V 0.000 0.151 0.000 0.200 0.000 0.222 0.000 0.000 0.000 0.000 0.916 0.859–0.974 1999–2000 J 0.480 0.087 0.333 0.000 0.324 0.163 0.098 0.013 0.076 0.000 Y 0.331 0.339 0.000 0.500 0.246 0.215 0.089 0.069 0.000 0.029 M 0.000 0.173 0.798 0.000 0.360 0.460 0.000 0.076 0.510 0.000 V 0.000 0.189 0.000 0.250 0.070 0.162 0.000 0.029 0.000 0.011 0.697 0.595–0.786 2000–2001 J 0.520 0.000 0.000 0.000 0.000 0.183 0.000 0.000 0.000 0.000 Y 0.110 0.463 0.000 0.414 0.000 0.293 0.000 0.000 0.000 0.000 M 0.000 0.326 0.697 0.000 1.000 0.210 0.000 0.000 1.000 0.000 V 0.000 0.000 0.000 0.310 0.000 0.314 0.000 0.000 0.000 0.000 1.247 Average matrix J 0.516 0.489 1.372 0.000 0.540 0.111 0.114 0.046 0.116 0.000 Y 0.309 0.425 0.000 0.763 0.228 0.262 0.162 0.094 0.000 0.019 M 0.000 0.307 0.908 0.000 0.206 0.450 0.000 0.116 0.312 0.000 V 0.000 0.130 0.000 0.118 0.026 0.177 0.000 0.019 0.000 0.002 consecutive good matrices during the simulations, which of the iterations resulted in a population size below one indi- turned out to be c. 35%. vidual within 100 years. logks = 0.050 (ks = 0.950) and the

Introducing in the model a random selection of matrices quasi-extinction probabilities for Nq = 5 and Nq = 10 were both

(Fig. 4) with a Montecarlo function restricted to drawing good 1, with mean times of extinction Tq(5) = 81, and Tq(10) = 68. years maximally 30% of the time (reflecting the data of the In the alternative model of climate-related stochasticity Diego Hernańdez station inferred from the Izan˜ a station), where k was first calculated directly from randomly selected the results of the simulations are dramatically different. precipitation values, the results were similar, but less dra- The population fluctuates severely, with maximum popula- matic. Population size below one within 100 years occurred tion sizes of 1.500 individuals. Under these conditions, 83% in 74% of the iterations; logks = 0.032 (ks = 0.967) and the 558 BIOLOGICAL CONSERVATION 136 (2007) 552– 562

Annual Precipitation (mm) Number of individuals 1800

1600

1400

1200

1000

800

600

400

200

0 1916 1922 1928 1934 1940 1946 1952 1958 1964 1970 1976 1982 1988 1994 Year

Fig. 3 – Reconstruction of the possible historical development in population size using the simulation model. The basis of the simulation is an average population of 268 individuals. The solid line represents the theoretical precipitation in the Diego Herna´ndez zone as inferred from the Izan˜ a Weather Station. The dashed line shows the number of individuals projected from k values obtained as a function of the precipitation data.

1400 quasi-extinction probabilities for Nq = 5 and Nq = 10 were both

1, with mean times of extinction above 100 years: Tq(5) = 115, 1200 and Tq(10) = 105.

1000 3 3.3. Comparisons between natural and planted populations 800 No differences between the demographic structure of planted and natural populations were observed (G = 4.996, P = 0.172). 600 In addition, there were no signiﬁcant differences in fruit pro- Population size duction (G = 0.238, P = 0.625) and seedling mortality (G = 2.028, 400 1 P = 0.154). These results suggest that the dynamics of the natural populations was quite similar to that of the newly 200 2 4 founded population.

0 3.4. The species in the past 0 50 100 150 200 250 Years The retrospective simulations indicated that the species prob- Fig. 4 – Population dynamics of the planted study population ably experienced better, but also worse, moments in the past. of Helianthemum juliae as predicted by the Stella simulation In the ﬁrst decades of the 20th century, H. juliae most likely model under different scenario’s. Line 1 presents the mean had stable populations with a relatively high number of indi- population trajectory if the model was based only on viduals (near 1500). However, between 1935 and 1945, several random selection of the nine matrices, without considering consecutive years with low precipitation probably nearly the climatic conditions in the last 85 years. Line 2 shows the drove the species to extinction. After 1955, a slow recovery population trajectory considering the climate conditions to the current situation must have occurred when good and observed during the past 85 years (the probability of good bad years alternated (Fig. 3). years is 30%). Line 3 presents the mean trajectory considering the climate conditions observed during the past 4. Discussion 85 years, but with a reinforcement of the population with 1000 individuals. Line 4 shows the development of a Population Viability Analysis examines the risk of extinction of population planted newly in a more humid environment populations in response to a variety of factors. Simulating the (i.e. probability of good years 40%). population dynamics under a set of relevant environmental BIOLOGICAL CONSERVATION 136 (2007) 552– 562 559

0 1.0

0.2 0.8

0.4 L G 0.6

P > 350 mm 0.6 96-97 0.4 5 4 97-98 95-96 6

2 93-94 0.8 0.2 94-95 3 8 99-00

P < 350 mm 92-93 1 1.0 0 00-01 9 7 98-99 1.0 0.8 0.6 0.4 0.2 0

Fig. 5 – Triangular ordination diagram representing the position of the nine matrices for Helianthemum juliae between 1992 and 2002 with respect to their relative contribution (=summed elasticities) of fecundity (F), Growth (G) and survival (L) to the population growth rate, k. The matrices have been chronologically numbered from 1, 1992–1993 to 9, 2001–2002. Shaded areas enclose matrices that correspond to two precipitation classes, P < 350 mm, P > 350 mm. conditions and/or experimental treatments allows managers Sensitivity or elasticity analyses are prospective and quantify to evaluate the efficacy of specific management actions and the change in population growth rate given a specified change to project what will happen if a certain management strategy in one or more elements of the matrix. Retrospective analy- is implemented (Song, 1996). However PVA models must be ses, by life table response experiments, can be used to deter- used with caution, because they are just models, or ‘‘carica- mine life-history stages or demographic transitions that tures of reality’’ (Beissinger and Westphal, 1998). Accordingly, contributed to differences between populations observed dur- one must be aware of the limitations of the models, especially ing the observation period (Horvitz et al., 1997; Caswell, 2000). when the results are to be used for decisions on practical spe- In this paper we resorted principally to prospective analyses cies conservation and management. In the case of the model because they are the most suitable approach to predict how presented here, it is necessary to point out that all data comes expected levels of stochasticity will interact with conserva- from a single monitoring plot and this lack of replication does tion measures to influence the future population growth rate not allow us to discern local from environmental variability. of threatened or endangered species (Menges, 1990; Heppell However, we were forced to restrict our monitoring to a single et al., 1994). The retrospective analyses used in this paper location, in order to avoid endangering the species by our own do not clearly coincide with the classical approach. Here, we activities. In addition, comparisons between the native and mainly used it to get an idea of the most likely historical fluc- introduced populations were performed for only one year, tuations in population size. and relevant vital rates such as the survival of mature adults Elasticity analysis revealed that the most critical stage in were not compared. However, we believe that our comparisons the life cycle of H. juliae is the survival of mature reproductive of the demographic structure, fruit production and seedling plants, which showed the highest elasticity (31%) in the aver- mortality were sufficient to suggest that the dynamics of the age matrix. The survival of this stage in natural conditions natural and planted populations were quite similar. We also was 90.8%, and did not show much variation between good assume that herbivores are not a significant factor, because and bad years. For management reasons, we periodically vis- we did not observe any damage that could be attributed to ited the natural populations from 1985 onwards, and the introduced (rabbits, mouflons) or natural herbivores (lizards, planted population from 1989 onwards, when the individuals insects, etc.), and all observed deaths in the field were caused were three years old. Our field observations indicated that by summer drought or natural senescence. large reproductive plants generally died of senility after 14 In PVA’s, demographic responses to variable environments years. In contrast, juvenile recruitment was highly variable, can be analyzed by prospective or retrospective analyses. with very high values (4.88) in good, and low values (0) in 560 BIOLOGICAL CONSERVATION 136 (2007) 552– 562

bad years. As expected (Pfister, 1998; de Kroon et al., 2000) this Some authors suggest that the estimates of ks are likely to temporally variable transition had comparatively low be imprecise even with sets of 10 matrices. Following the elasticities (between 0% and 14%). In terms of conservation N = (5–10)T rule (Fieberg and Ellner, 2000), we would only be management, it is therefore most effective to protect the able to produce reliable predictions of extinction probability adult reproductive plants from damage by any human-related for a 2 years forecast interval with our set of nine matrices. factors (Oostermeijer, 2003). In the case of H. juliae these fac- However, for conservation managers and policy makers, pre- tors are restricted to trampling by tourists and occasional her- dictions over such short intervals (2–3 years) are not very use- bivory by introduced animals (rabbits and mouflons). ful, and developing censusing programs that last for 50 years Juvenile recruitment is generally strongly correlated with or more generally extend well beyond the time-frame of prac- climatic variability, and more specifically with precipitation, tical conservation. because water is the principal limiting resource in arid regions Fieberg and Ellner (2001) suggest that modelling with envi- (Harrington, 1991). Drought effects on fruit production and ronmental covariates may help overcome the requirements germination, as shown for another Mediterranean Helianthe- with respect to the sampling intensity using the environmen- mum species (Helianthemum squamatum, Escudero et al., tal variables that have major effects on vital rates. Through 1999), were not observed. Seedlings are generally highly sus- the analysis of these variables we can model the relationship ceptible to drought (Chabot and Mooney, 1985). In bad years between demographic rates and environment, and knowing (precipitation <350 mm) seedling mortality of H. juliae was the distribution over time of the environmental variables we high (nearly 100%), but in good years (precipitation >350 mm) can obtain better estimates of population viability. For Helian- mortality decreased to around 90%. themum juliae, we have followed this alternative approach. We Probably owing to its extreme fluctuations, related to the show that the population growth rate is strongly correlated variation in precipitation, juvenile recruitment has low elas- with the annual precipitation and censused during a suffi- ticity (Pfister, 1998; de Kroon et al., 2000). This strong depen- cient number of years to collect several good as well as bad dency of recruitment on the amount of precipitation is years. In addition to the standard approach with random probably the normal situation for H. juliae. Regular recruit- selection of matrices, we conducted an alternative random ment is necessary to maintain viable populations on the long sampling strategy based on the frequencies of good and bad term. However, conservation management can do very little years observed during the last century. to stimulate recruitment. So, under a scenario of climate Introducing stochastic variability by a random selection of change, problems with this vital rate might arise. matrices offered results similar to the deterministic analysis, The method to simulate stochasticity used in this paper is and predicted a very low extinction risk. When we simulated the simplest: randomly sample from a set of transition matri- the population dynamics in relation to climatic variability, the ces and project the population for each year (Beissinger and results were quite different. Under the climate scenario of the Westphal, 1998). We preferred this method because it avoids past 85 years (30% chance of good years) the simulations indi- problems of correlation among vital rates (Greenlee and Kaye, cated a high extinction risk over the next 100 years. Simulat- 1997; Kaye and Pike, 2003). One of the most important aspects ing the future tendency for the three natural populations in these models is the duration of demographic monitoring using this model, we obtain a regressive dynamic too (Fig. 6) that is needed to adequately estimate environmental variabil- Of course, this result needs to be considered with caution be- ity (Bierzychudek, 1999). The consequences of censusing pop- cause our model did not include the total complexity of the ulations during an insufficient number of years have not been climatic variables and its future changes. fully explored (Fieberg and Ellner, 2001) but it quite likely Anthropogenic climate change is a rapid process (N.R.C., leads to unreliable predictions. 2002; Overpeck et al., 1991; Root et al., 2002), and the survival

500 450 Risco Verde 400 Cañada delas Pilas 350 Mesa del Obispo 300 250 200 150

Number of individuals 100 50 0 0 10 20 30 40 50 60 70 80 90 100 years

Fig. 6 – Population dynamics of the three natural populations of Helianthemum juliae as predicted by the Stella simulation model under the scenario that consider the climate conditions observed during the past 85 years (the probability of good years is 30%). BIOLOGICAL CONSERVATION 136 (2007) 552– 562 561

of many plant species may depend on the rates at which they the founding of new populations in suitable areas could be a are able to migrate to more favourable areas (Graham and good strategy to spread risks of local extinction (Kutner and Grimm, 1990; Kutner and Morse, 1993). Morse, 1993). For this strategy to work, the new populations Our exploration of the climatic data revealed a significant should preferably not be subject to the same climatic fluctua- increase in temperature (r = 0.480, P < 0.05) over the past 85 tions. For H. juliae, this means that selecting new sites with years that agrees with the global trend of climate warming adequate climatic conditions is more difficult, because avail- (Graham and Grimm, 1990; Schneider, 1989). Precipitation, able options are restricted to the Teide caldera, to which the however, did not change significantly over time (r = 0.038, species is endemic. However, sites with a probability of good P > 0.5). Our analysis showed that k was only significantly cor- years >40%, or sites with lower evapotranspiration can proba- related with precipitation and not with temperature. bly be found in the Teide caldera. Such sites might be found at Several models predict changes in precipitation and tem- the northern slopes or in sectors in the Southwest and South- perature in the Canary Islands climate zone for the next 100 east of the National Park, where a lower altitude and more years. The UK Met Office Handley Centre for Climate Change favourable orientation permit a greater incidence of clouds (http://www.met-office.gov.uk/research/hadleycentre) pre- and fog than inside the Teide Caldera. Such locations form dicts a 3–5 C increase in the average annual temperature. Fu- good candidates for founding new viable populations, which ture changes in precipitation are less clear (0.2 to 0.2 mm/ would significantly increase metapopulation viability (Fig. 4). day), but are predicted to be low. Hence, we could argue that global change will not affect precipitation in the Teide Acknowledgements National Park. However, even if precipitation remains the same, a temperature rise will most likely increase evapotrans- This work was supported by the Organismo Autońomo de Par- piration, and thus lower water availability in the soil (Laven- ques Nacionales through the recovery project for the threa- der et al., 1998; Rosenzweig and Hillel, 2000). Using the more tened flora of the Teide National Park. We thank Manuel common equations (Thornthwaite, Turc, Coutagne, Penman, Durbań Villalonga, Director of the Teide National Park for etc.) to calculate the theoretical evapotranspiration (Sańchez, his encouragement and to the Park staff for their invaluable 1992), a 3–5 C temperature increase could result in a 20–35% field assistance. We also thank the staff of the Department increase in evapotranspiration over the current values. Ecology of Complutense University of Madrid for their advice Our retrospective simulations suggest that H. juliae is cur- and help with the simulation models. We gratefully acknowl- rently recovering after a strong decline induced by several edge Dr. Diethart Matthies for his help with the bootstrap consecutive bad years at the start of the 20th century. Further analyses and for his valuable comments on the manuscript. climate warming could stop this recovery process. Under these conditions, we predict a problematic future for H. juliae, and most likely this is also true for other endemic species in REFERENCES this vulnerable ecosystem.

4.1. Conservation strategies Ban˜ ares, A., Marrero, M.V., Carque´, E., 2001. Biologıá de la conservacio´ n de la flora amenazada en los Parques Nacionales To test possible conservation strategies, we used the Stella Canarios. In: Centro de Estudios Ramo´ n Areces (Ed.), model with climatic restrictions to simulate a worst-case sce- Conservacio´ n de especies vegetales amenazadas del a´rea nario in which the probability good years is 30%, reflecting the mediterrańea: una perspectiva desde el fin de siglo, Madrid, data for Diego Hernańdez extrapolated from the Izan˜ a mete- pp. 35–62. orological station in the last 85 years, and incorporate several Beissinger, S.R., Westphal, M.I., 1998. On the use of population reinforcement campaigns in this scenario. The demographic models of population viability analysis in endangered species management. Journal of Wildlife experimental augmentation of the population with subadult Management 62, 821–841. individuals did not result in a significant reduction in the Bierzychudek, P., 1999. Looking backwards: assessing the extinction risk, but only increased the time to extinction projections of a transition matrix model. Ecological (Fig. 4). This implies that population reinforcement is not an Applications 9 (4), 1278–1287. effective tool to increase population viability. However, Boletıń Oficial del Estado, 1999. Ley 3/1999 de 11 de enero, por la increasing the time to extinction can perhaps provide more que se crea el Parque Nacional de Sierra Nevada. Boletıń opportunities for incidental good years, which would yield a Oficial del Estado n 11 de 13-01-1999, pp. 1512–1525. Caswell, H., 2000. Prospective and retrospective perturbation more substantial reduction of extinction risk. analyses: their roles in conservation biology. Ecology 81, 619– Another management strategy could be to increase the 627. number of populations of H. juliae on Tenerife. Seeds of H. ju- Caswell, H., 2001. Matrix Population Models: Construction, liae do not disperse easily, judging from the observation that Analysis and Interpretation, second ed. Sinauer Associates, most seedlings are found in the direct neighbourhood of Sunderland, Massachusetts. (their) mother plants. Hence, we consider the natural ability Chabot, B.F., Mooney, M.A., 1985. Physiological Ecology of North American Plant Communities. Chapman & Hall, New York. to colonize other sites with suitable climatic conditions to de Kroon, H., Plaisier, A., van Groenendael, J., Caswell, H., 1986. be limited. Elasticity: the relative contribution of demographic In situ propagation of threatened species is becoming a parameters to population growth rate. Ecology 67, 1427–1431. more important conservation strategy (Maunder, 1992; Falk de Kroon, H., van Groenendael, J., Ehrlen, J., 2000. Elasticities: a et al., 1996; Oostermeijer, 2003). In a rapidly changing climate, review of methods and model limitations. Ecology 81, 607–618. 562 BIOLOGICAL CONSERVATION 136 (2007) 552– 562

Dennis, B., Munholland, P.L., Scott, J.M., 1991. Estimation of Lefkovitch, L.P., 1965. The study of population growth in growth and extinction parameters for endangered species. organisms grouped by stages. Biometrics 21, 1–18. Ecological Monographs 61, 115–143. Marrero, M.V., Ban˜ ares, A´ ., Carque´, E., 2003a. Plant resource Escudero, A., Somolinos, R.C., Olano, J.M., Rubio, A., 1999. Factors conservation planning in protected natural areas: an example controlling the establishment of Helianthemum squamatum,an from Canary Islands, Spain. Biological Conservation 113, 399–410. endemic gypsophyle of semi-arid Spain. Journal of Ecology 87, Marrero, M.V., Carque´, E., Ban˜ ares, A., 2003b. Helianthemum juliae. 290–302. In: Ban˜ ares, A´ ., Blanca, G., Gu¨ emes, J., Moreno, J.C., Ortı´z, S. Falk, D.A., Millar, C.I., Olwell, M., 1996. Introduction. In: Falk, D.A., (Eds.), Atlas y Libro Rojo de la Flora Vascular amenazada de Millar, C.I., Olwell, M. (Eds.), Restoring Diversity: Estrategies for Espan˜ a. Direccio´ n General de Conservacio´ n de la Naturaleza, Reintroduction of Endangered Plants. Island Press, Center for Madrid, pp. 284–285. Plant Conservation, Missouri Botanical Garden, pp. xiii–xxii. Maunder, M., 1992. Plant reintroduction: an overview. Biodiversity Fieberg, J., Ellner, S.P., 2000. When is it meaningful to estimate an and Conservation 1, 51–61. extinction probability? Ecology 81, 2040–2047. Menges, E.S., 1986. Predicting the future of rare plant populations: Fieberg, J., Ellner, S.P., 2001. Stochastic matrix models for demographic monitoring and modeling. Natural Areas Journal conservation and management: a comparative review of 6, 13–25. methods. Ecology Letters 4, 244–266. Menges, E.S., 1990. Population viability analysis for an Folks, J.L., Chhikara, R.S., 1978. The inverse Gaussian distribution endangered plant. Conservation Biology 4, 52–62. and its statistical application, a review. Journal of the Royal Menges, E.S., 1991. The application of minimum viable population Statistical Society, Series B 40, 263–289. theory to plants. In: Falk, D.A., Holsinger, K.E. (Eds.), Genetics Garcıá, M.B., Iriondo, J.M., 2002. Modelos matriciales de and Conservation of Rare Plants. Oxford University Press, pp. proyeccio´ n poblacional. In: Ban˜ ares, A´ . (Ed.), Biologıá de la 47–61. Conservacio´ n de Plantas Amenazadas. Organismo Auto´ nomo Morris, W.F., Doak, D.F., 1998. Life history of a long-lived de Parques Nacionales. Se´rie Tećnica, Madrid, pp. 45–54. gynodioecious plant, Silene acaulis (Caryophyllaceae), inferred Goodman, D., 1987. The demography of chance extinction. In: from size-based population projection matrices. American Soule´, M.E. (Ed.), Viable Populations for Conservation. Journal of Botany 85, 784–793. Cambridge University Press, pp. 11–34. N.R.C. (National Research Council), 2002. Abrupt Climate Change: Graham, R.W., Grimm, E.C., 1990. Effects of the global climate Inevitable Surprises. National Academy Press, Washington, change on the patterns of terrestrial biological communities. DC. p. 230. Trends in Ecology and Evolution 5 (9), 289–292. Oostermeijer, J.G.B., 2003. Threats to rare plant persistence. In: Greenlee, J., Kaye., T.N., 1997. Stochastic matrix projections: a Brigham, A.C., Schwartz, M.W. (Eds.), Population Viability in comparison of the effect of element and matrix selection Plants, Conservation, Management, and Modeling of Rare methods on quasi-extinction risk for Haplopappus radiatus Plants. Springer, Berlin, pp. 17–58. (Asteraceae). In: Kaye, T.N., Liston, A., Love, R.M., Luoma, D., Overpeck, J.T., Bartlein, P.J., Webb, T., 1991. Potential magnitude of Meinke, R.J., Wilson, M.V. (Eds.), Conservation and future vegetation change in Eastern North America: Management of Native Plants and Fungi. Native Plant Society comparison with the past. Science 254, 692–695. of Oregon, Corvallis, Oregon, pp. 66–71. Pfister, C.A., 1998. Patterns of variance in stage-structured Hansen, A., Sunding, P., 1993. Flora of Macaronesia. Checklist of populations: evolutionary predictions and ecological vascular plants, 4 rev. ed. Sommerfeltia 17, pp. 1–295. implications. Proceedings of the National Academy of Harrington, G.N., 1991. Effects of soil moisture on shrub seedling Sciences of the United States of America 95, 213–218. survival in semi-arid grassland. Ecology 72, 1138–1149. Rosenzweig, C., Hillel, D., 2000. Soils and global climate change: Heppell, S., Walters, J.R., Crowder, L.B., 1994. Evaluating challenges and opportunities. Soil Science 165, 47–56. management alternatives for Red-cockaded Woodpeckers: a Root, T.L., Price, J.T., Hall, K.R., Schneider, S.H., Rosenzweig, C., modeling approach. Journal of Wildlife Management 58, 479–487. Pounds, D., 2002. Fingerprints of global warming on wild High Performance Systems, Inc., 1998. Stella 5.1.1. for Windows, plants and animals. Nature 421, 57–60. Hannover. Sańchez, M.I., 1992. Me´todos para el estudio de la evaporacio´ ny Hood, G., 2002. Poptools v. 2.4. http://www.cse.csiro.au/CDG/ evapotranspiracio´ n. Cuadernos Tećnicos de la Sociedad poptools. Espan˜ ola de Geomorfologıá n 3, 1–36. Horvitz, C., Schemske, D.W., Caswell, H., 1997. The relative Schneider, S.H., 1989. The changing climate. Scientific American ‘‘importance’’ of life-history stages to population growth: 290 (9), 70–79. prospective and retrospective analyses. In: Tuljapurkar, S., Silvertown, J., Franco, M., Pisanty, I., Mendoza, A., 1993. Caswell, H. (Eds.), Structured-population Models in Marine, Comparative plant demography: relative importance of life Terrestrial, and Freshwater Systems. Chapman and Hall, New cycle components to the finite rate of increase in woody and York, pp. 247–271. herbaceous perennials. Journal of Ecology 81, 67–73. Kalisz, S., McPeek, M.A., 1992. Demography of an age-structured Sokal, R.R., Rohlf, F.J., 1981. Biometry, second ed. W.H. Freeman annual: resampled projection matrices, elasticity analyses, and Company, San Francisco. and seed bank effects. Ecology 73, 1082–1093. Song, Y.L., 1996. Population viability analysis for two isolated Kaye, T.N., Pyke, D.A., 2003. The effect of stochastic technique on populations of Haianan eld’s deer. Conservation Biology 10, estimates of population viability from transition matrix 1467–1472. models. Ecology 84, 1464–1476. Tuljapurkar, S.D., Orzack, S.H., 1980. Population dynamics in Kutner, S.L., Morse, L.E., 1993. Reintroduction in a changing variable environments I. Long-run growth rates and climate. In: Falk, D.A., Millar, C.I., Olwell, M. (Eds.), Restoring extinction. Theoretical Population Biology 18, 314–342. Diversity: Strategies for Reintroduction of Endangered Plants. Tuljapurkar, S.D., Horvitz, C., Pascarella, J.B., 2003. Island Press, Center For Plant Conservation, Missouri The many growth rates and elasticities of populations in Botanical Garden, pp. 23–48. random environments. American Naturalist 162 (4), 489–502. Lavender, B., Smith, J.V., Koshida, G., Mortsch, L.D., 1998. Werner, P.A., 1975. Predictions of fate from rosette size in teasel Binational Great Lakes-St. Lawrence Basin Climate Change (Dipsacus fullonum L.). Oecologia 20, 197–201. and Hydrologic Scenarios Report. Downsview. Environmental Zar, S.H., 1984. Biostatistical Analysis, second ed. Prentice-Hall,, Adaptation Research Groop, Environment, Canada. Englewood Cliffs, NJ.