Factors driving Arabian gazelles (Gazella arabica) in Israel to extinction: time series analysis of population size and juvenile survival in an unexploited population

The Yotvata NR is an alluvial fan area located at the edge of the Yotvata salt marsh in the southern Arava Valley at an altitude of 75 m above sea level (Evenari et al. 1971; Fig. 1). Within the Yotvata NR (32 km²), a fenced enclosure (3.5 km²) was established in July 2006, to protect the G. arabica population from wolf predation (Cohen et al. 2013). To date this fenced area of natural habitat represents the last stronghold of G. arabica outside the Arabian Peninsula. Depending on soil salinity, the Yotvata NR comprises six vegetation belts (Baharav 1980), of which only the Vachellia (formerly Acacia)-Anabasis community occurs within the G. arabica enclosure. The community is dominated by Vachellia raddiana and V. tortilis, accompanied by the semi-parasite Plicosepalus acaciae, and a number of shrubs such as Hammada salicornica, Anabasis articulata, Ochradenus baccatus, and Lycium shawii (Baharav 1980). Annual herbs and grasses are scarce or even absent in this community (Baharav 1980). The climate is hyper-arid with an annual mean temperature of 23.5 °C (max.: 38.7 °C, min.: 8.8 °C; En.climate-data.org 2017), and unpredictable rainfall (averaging 35 mm per year) that occurs only between October and May (IMS 2018). Heavy, short rains cause flash floods that cross the G. arabica enclosure from west to east, draining into the basin of the Yotvata salt marsh. In October and November 2012, as well as in February 2013, these flash floods destroyed the fence allowing wolves (Canis lupus)—the main natural predator of G. arabica—to enter the enclosure. In June 2013, the fence was repaired and access for wolves prevented again (Shalmon pers. obs.). The G. arabica population shares the habitat and the enclosure with 40 to 170 dorcas gazelles (G. dorcas), a species that directly competes with G. arabica for two major food plant species, i.e., Vachellia raddiana and V. tortilis (Breslau et al. 2019).

Until recently, G. arabica was merged with the mountain or Palestine gazelle (G. gazella; Groves 1985, 1997; Mendelssohn et al. 1995), but split into at least seven subspecies (Groves and Grubb 2011), one of them being Gazella ‘gazella’ acaciae described from the southern Arava Valley in Israel (Mendelssohn et al. 1997; Fig. 2a). Recent phylogenetic and morphological investigations, however, favoured the division of this species complex into two species: G. arabica and G. gazella (Wronski et al. 2010; Bärmann et al. 2013a, b; Lerp et al. 2013; Hadas et al. 2015).

Until 1956 the G. arabica population in Israel numbered probably several hundred individuals in the Vachellia groves of the northern (Hazeva) and southern Arava Valley (Yotvata-Eilat). The northern population was eradicated between 1956 and 1963 due to illegal hunting by soldiers of the Israeli Defence Forces (Mendelssohn 1974). After 1964, only the southern population survived, being annually assessed by the Israel Nature and Parks Authority (INPA; Yom-Tov and Ilani 1987). This small population experienced pronounced variation in population size, with maximum numbers never exceeding 68 individuals and a minimum population size of only 10 animals in 1996 and 11 in 2005 (this study). From 1964 to 1983 the populations’ range was reduced to the Arava Valley between Eilat in the south and Yotvata in the north. Since 1983 the population is restricted to its preferred habitat, i.e., the Vachellia groves of the Yotvata NR. The current population size (in 2019) is 30 individuals, with 10 males, 13 females and 7 juveniles (Shalmon pers. obs.).

From 1995 to 2005, annual counts of Arabian and dorcas gazelles (Fig. 3a, b) were carried out by the INPA as described in Yom-Tov and Ilani (1987). After fencing in 2006, total drive counts (sensu Bothma 2002) were conducted within the G. arabica enclosure during autumn (October to November). Three to five vehicles, each with a driver and one observer, were slowly (5 km/h) driven from south to north, driving gazelles towards the northern fence. Those gazelles trying to escape across the drive line between the vehicles were counted by one of the observers. Eye and radio contact between observers prevented double counting. Since 2010, two such counts were conducted; one in summer (June–July), another one in winter (December–February), and results were averaged across each year (Fig. 3). In 2017 the number of counts was increased to four, i.e., two in each season. Moreover, due to long-term observations by INPA staff most adult females were individually known (see below), contributing to the correct estimation of population size. Based on known female-male and female-juvenile ratios (Yom-Tov and Ilani 1987; Shalmon 1991), the missing count result of 1986 was estimated as 35. After 2006, dorcas gazelles outside the enclosure were not considered.

Percent fawn survival was calculated as the number of new-borns that survived for more than 5 months (i.e., the time when young males leave the maternal herd; Mendelssohn et al. 1995), as a proportion of all fawns born in the respective year. Month of birth and death were obtained during regular (daily) patrols in and around the enclosure carried out by INPA staff from 1995 to 2017. Fawns were identified through their mothers, who were individually known based on their characteristic face markings and horn shapes (Fig. 2a).

Rainfall recordings were obtained from IMS (2018), taken at Yotvata settlement, 1.5 km north of Yotvata NR, while annual mean maximum temperature data were obtained from IMS (2018) for Eilat, the nearest weather station to the study area that recorded temperature for the relevant years. The annual mean maximum temperature was calculated as the average of monthly (January–December) maximum temperatures for each year.

From 2000 to 2009 the food availability of two major food plants was measured as Vachellia tortilis and V. raddiana foliage greenness. Based on reference photos previously taken of both species in five different foliage stages, the greenness index was established by eyesight, using a green scale reaching from zero (no foliage) to five (maximum foliage). To determine greenness, photographs of both species were taken once a year in autumn at the beginning of the rainy season, using the same camera, location, distance, angle and camera settings. Average greenness was established for 30 trees of each species and then summed to obtain an overall greenness index for both species in each year. Missing values, from 1995 to 1999, were interpolated from adjacent values, i.e., the average from 2000 to 2008, that period in which data were obtained by visual estimation. After 2009, greenness was measured using near infra-red photographs to calculate the Normalized Difference Vegetation Index (NDVI).

Both, G. arabica and G. dorcas (Fig. 2a, b) feed mainly on the leaves, flowers and fruits of Vachellia trees (Shalmon 1991), suggesting a pronounced interspecific competition for food. Competition with G. dorcas was simply expressed as the number of dorcas gazelles present in the study area. From 1995 to 2017 both species were counted as described above (Yom-Tov and Ilani 1987).

In the Yotvata NR, wolves represent the only predator of both gazelle species (Shalmon 1991). Leopards (Panthera pardus nimr) are locally extinct, and caracals (Caracal caracal), golden jackals (Canis aureus) and striped hyenas (Hyaena hyaena) occur only sporadically in the study area. Red foxes (Vulpes vulpes) may scavenge on gazelle carcasses and afterbirths, but are too small to actively prey on gazelles (Mendelssohn 1974; Shalmon pers. obs.). From 1995 to 2005 a wolf encounter rate was established as a proxy of predation pressure in the respective year. Hereby, the number of wolves encountered (3–65 encounters/year) during INPA patrols was divided by the number of observer days in the study area (41–271 days/year). After fencing the G. arabica habitat in 2006, wolves were still able to enter the enclosure—although at a much lower frequency (Shalmon pers. obs.)—and to prey on the fawns and adults of both gazelle species (Fig. 2c). From 2006 to 2017 the annual wolf encounter rate was calculated as the number of wolves photographed by two camera traps permanently (365 days/year) located at two artificial water sources within the enclosure divided by the number of trapping days. To avoid overestimation of wolf encounter rates obtained from camera trapping, pictures taken on the same day were counted as only one encounter, unless two or more individuals were trapped on the same image (Fig. 2d). Individuals could not be distinguished by either survey method and wolf encounter rates may thus be biased by repeated sightings.

In all cases, measurements were averaged across samplings to obtain one value for each year (1995–2017). To achieve normal distribution, all absolute data, i.e., G. arabica population size, G. dorcas population size, greenness index, annual mean maximum temperature, rainfall and the wolf encounter rate were ln(x + 1) transformed. Relative data such as percentage fawn survival were arcsine (square root) transformed.

Initially, two generalized least squares (GLS) models (nlme package in R; Pinheiro et al. 2012) without any autocorrelation process (AR), i.e., GLS-AR(0) models were conducted to explore the relationships between two dependent variables (G. arabica population size and percentage fawn survival) and five environmental variables (i.e., G. dorcas population size, greenness index, annual mean maximum temperature, rainfall and wolf encounter rate). As our dataset contains longitudinal time series observations, we examined whether autocorrelated residuals occurred within initial GLS regressions by using Auto Correlation Function (ACF) and Partial Auto Correlation Function (PACF) of dependent time series.

In case autocorrelated errors were detected, coefficients were estimated for the first-order Autoregressive model AR(1) in combination with Restricted Maximum Likelihood (REML) of variance components. Subsequently, we expanded the autocorrelation process to another 5 years’ time lag effect (i.e., every 1 year) and calculated the GLS-AR(2-6) models, before finally comparing nested models using Akaike Information Criteria (AIC). All environmental variables, as well as their two-way interaction terms were included in the initial model but removed using a stepwise backwards elimination procedure, to identify the best fit model with the smallest AIC and lowest residual standard error (SE). Unstandardized beta (B), t-test statistics (t) and variance inflation factor (VIF) were established and applied as described in Pinheiro et al. (2012) and Pekár and Brabec (2016). Inspection of model residuals did not indicate violations of model assumptions, i.e., normal error distribution and homoscedasticity. All data analyses were carried out in RStudio (version 3.5.1).

The initial GLS-AR(0) model revealed mean annual maximum temperature to have a significant positive effect on G. arabica population size (B = 10.57, t = 2.65, p = 0.02; Table 1). The GLS-AR(0) model residuals varied distinctly through time suggesting a substantial autocorrelation (Supplementary Material Fig. S1a). The ACF autocorrelogram of GLS-AR(0) model residuals showed a sinusoidal pattern with a negative autocorrelation (r = − 0.42) at time lag 6 (Supplementary Material Fig. S1b). The PACF autocorrelogram of model residuals showed no significant autocorrelations (|r| < 0.4; Supplementary Material Fig. S1c). Considering all potential autoregression processes from AR(1) to AR(6), GLS-AR(3) and GLS-AR(6) were identified as the best autocorrelation models (Supplementary Material Table S1). The GLS-AR(3) model revealed that the G. dorcas population size had a significant positive effect on the G. arabica population size (B = 0.40, t = 2.80, p = 0.01, Supplementary Material Table S1). The GLS-AR(6) model further confirmed that the G. dorcas population size had a significant positive effect on the G. arabica population size (B = 0.48, t = 6.86, p < 0.001), while the wolf encounter rate showed a significant negative effect (B = − 0.33, t = − 0.36, p = 0.003; Supplementary Material Table S1). After applying a stepwise backwards elimination procedure, only wolf encounter rate and G. dorcas population size were retained as predictors for the G. arabica population size in the final models (Table 2). Both final models revealed that the wolf encounter rate had a significant negative relationship with the G. arabica population size (GLS-AR(3): B = − 0.37, t = 2.73, p = 0.01; GLS-AR(6): B = − 0.29, t = − 2.75, p = 0.01; Table 2, Figs. 3a, 4), while G. dorcas population size had a significant positive effect (GLS-AR(3): B = 0.41, t = 2.80, p = 0.01; GLS-AR(6): r = 0.44, t = 6.39, p < 0.001; Table 2, Fig. 3b).

The initial GLS-AR(0) model revealed wolf encounter rate to have a significant negative effect on percentage fawn survival (B = − 0.66, t = − 2.31, p = 0.03, Table 1). The GLS-AR(0) model residuals showed a stochastic distribution, suggesting no autocorrelation (Supplementary Material Fig. S2a). Both, the ACF and PACF autocorrelogram of GLS-AR(0) model residuals did not reveal any significant autocorrelation (Supplementary Material Fig. S2b, c). After applying a stepwise backwards elimination procedure, the best fit GLS-AR(0) model retained only wolf encounter rate and the mean annual maximal temperature as predictors (Table 3). Wolf encounter rate was found to have a significant negative correlation with the percentage fawn survival (B = − 0.7, t = − 3.80, p < 0.01; Table 3, Fig. 5a), while mean annual maximal temperature was significantly positive related to percentage fawn survival (B = 4.62, t = 2.11, p = 0.05; Table 3, Fig. 5b).

With our study we envisaged to define the factors shaping the population viability of Arabian gazelles in southern Israel, probably the only, currently unexploited—not hunted—population worldwide (Mallon and Kingswood 2001; IUCN/SSC Antelope Specialist Group 2017) and thus an ideal study population to identify reasons for their decline without having to consider the impact of poaching. However, at this point it should be also highlighted that a fenced population containing only 30 individuals is not ideal to examine natural processes. Five environmental variables were tested for their effects on two dependent variables representative for population viability (i.e., population size, percentage fawn survival). Our time series analysis (GLS models, Table 2) revealed two environmental variables influencing the G. arabica population size: (i) a negative effect of wolf encounter rate suggesting that predation by wolves has a strong impact on the population size of G. arabica and (ii) a positive effect of G. dorcas population size indicating an analogous population growth of both species, rather than competition. This is surprising since it was predicted that Arabian gazelles compete with sympatric dorcas gazelles for limited food resources (Shalmon 1989, 1991). Instead, it rather appears that the population growth of both species is subject to the same environmental stressors, i.e., predation by wolves (Fig. 3). Resource partitioning and reduced competition were reported from other ungulate communities such as swamp deer (Cervus duvauceli) and hog deer (Axis porcinus), two sympatric deer species from Nepal that consumed almost equal proportions of the same woody browse species (Wegge et al. 2006), or from the renowned grazing succession of migrant ungulates in the Serengeti Ecosystem (Sinclair and Norton-Griffiths 1982). Here, niche segregation has developed among grazing ungulates, allowing a relaxed co-existence but facing a similar predation pressure (Sinclair 1985; Putman 1996; Arsenault and Owen-Smith 2002). Our result unrevealed not only a similar predation pressure on both gazelle species, but also two periodic population cycles, i.e., 3 and 6 years’, reflected by the two time lag effects of GLS-AR(3) and GLS-AR(6) models. Such fluctuating population size patterns are known from other coupled predator–prey cycles (e.g. snowshoe hare, Lepus americanus and the Canadian lynx, Lynx canadensis; Elton and Nicholson 1942; Keith 1963; Krebs et al. 1995), suggesting bottom-up control of wolf populations by prey availability when gazelle population size (both species together) is high, and top-down control by wolves when wolf abundance is high. Likewise, predators consume more prey items per capita when prey densities are high and less when prey densities are low (Adamcik et al. 1978). In the present case, this predator–prey cycle could be 6 years, whereby the 6 years’ time lag represents high gazelle densities and the 3 years’ time lag high wolf densities or vice versa. Unless further time series analysis include longer sampling periods to confirm such a cycle, this remains speculative and possibly biased by management actions such as culling of wolves and fencing the prime habitat of the gazelles.

Finally, food availability, measured as Vachellia greenness, and rainfall had no significant effect on the G. arabica population size (Supplementary Material Table S1). This finding refutes the prediction that the Arava population occurs in sub-optimal habitat at the limit of the species’ original distribution and is thus more vulnerable to climate change, i.e., increasing temperatures and decreased precipitation than other populations on the Arabian Peninsula (Yom-Tov and Ilani 1987; Dolev and Perevolotsky 2004; Polack, T., pers. com.).

Percentage fawn survival of Arabian gazelles in the Yotvata NR was determined by a strong negative effect of wolf encounter rate (Table 3, Fig. 5a) and a weaker positive effect of mean annual maximum temperature (Table 3, Fig. 5b), suggesting that higher temperatures and lower wolf encounter rates benefit the survival of G. arabica fawns. This finding corresponds to our results on population size, providing another strong argument that wolf predation shapes the G. arabica population size in the study area, at least since it was protected by the fence in 2006. Apart from habitat destruction and fragmentation, predation by carnivores—mainly wolves, but also jackals and straying dogs—is considered the major reason for the decline of gazelle populations in Israel (Gingold et al. 2009; Yom-Tov 2013). This applies likewise to the G. gazella populations in northern Israel and the Golan Heights (Reichmann and Saltz 2005; Gingold et al. 2009; Manor and Saltz 2004; Geffen et al. 1999; Shamoon et al. 2017), to the G. dorcas population in the Negev Desert and the Arava Valley (Shalmon 1986; Margolis et al. 2008; Barocas et al. 2018), but also—as our study has shown—to the G. arabica population. Canine predators in Israel have noticeably increased in recent decades (Gingold et al. 2009; Borkowski et al. 2011; Talmon 2015), mainly due to foraging on human discard, unharvested or dropped agricultural products and undisposed carcasses of domestic livestock (Shalmon 1986; Reichmann and Saltz 2005; Margolis et al. 2008; Gingold et al. 2009; Talmon 2015; Barocas et al. 2018), leading to an increased predation on their natural prey species. By contrast, Cohen et al. (2013) reported that there is no direct evidence of wolf predation on dorcas and Arabian gazelles in the Negev Desert (only 6.3% of scat samples contained gazelle components), and Barocas et al. (2018) demonstrated that gazelle occurrence in Israel is not a predictor for habitat selection and movement patterns of wolves.

At this point it should be highlighted that the wolf encounter rate, as established in our study, was based on two different survey methods, i.e., on sightings obtained during daily patrols and on camera trapping images recorded at two artificial water sources. Although sightings and trapping events were standardised by taking the observation time into consideration, our results might be biased by this discrepancy. Nevertheless, the statistical support of our models seems to be robust enough to preclude the possibility that this methodological shortcoming has biased our overall result. During the 2012/13 flash floods and the subsequent collapse of the fence (Figs. 3, 5), wolves entered the enclosure and killed about 40 Arabian and 50 dorcas gazelles within a few weeks (Shalmon pers. obs.). Such surplus killing (sensu Kruuk 1972) is well described for canine predators (Andelt et al. 1980; Miller et al. 1985; DelGiudice 1998) and has—without much doubt—strongly contributed to the significant predation effect in our model. There are two main reasons why our findings are not applicable to populations on the Arabian Peninsula: the decline of Arabian gazelles on the Arabian Peninsula is mainly attributed to illegal hunting (Thouless et al. 1997; IUCN/SSC Antelope Specialist Group 2017), and secondly, wolves suffer exactly the same threat, i.e., severe persecution (Cunningham et al. 2009; Cunningham and Wronski 2010). Only, if human persecution of prey and predator can be controlled, the findings of our study would become applicable.

Interestingly, fawn survival of Arabian gazelles in the Yotvata NR benefited from increasing annual maximum temperatures, rather than suffering from adverse climatic effects instigated by global warming (Williams et al. 2012). On the contrary, G. arabica fawns kept in captivity are known to be prone to infections (e.g. brucellosis) when weather conditions are cold, wet and windy (Soares et al. 2019). The steady increase of both, fawn survival and temperature, over time might be an artefact owned to an autocorrelation among the variables included into our model. However, possible autocorrelations were not detected within our GLS (Supplementary Material Fig. S2), and are thus unlikely to be the cause for this positive relationship. Moreover, the overall study period of 20 years may have been too short to detect a long-term pattern in our data. Given that G. arabica seems to be well adapted to the hot desert climate of the Uruq Bani Ma’arid Protected Area in southern Saudi Arabia, where Arabian gazelles were reintroduced into a hyper-arid environment and population size increased rapidly (Islam et al. 2011; Wronski et al. 2013), it is unlikely that fawn survival in the equally hot climate of the Arava Valley is negatively affected by these climatic variables. Moreover, the largest stronghold of G. arabica is the Farasan Islands population that is thriving relatively well in the hot, muggy climate of Red Sea islands (Wronski 2013). Again, the prediction that the Arava population occurs at the northern most limit of G. arabica’s original distribution and may be thus more prone to climate change (i.e., increasing mean annual maximum temperature) than populations on the Arabian Peninsula (Yom-Tov and Ilani 1987; Dolev and Perevolotsky 2004; Polack, T., pers. com.), was therefore rejected.

Two major threats to Arabian gazelles were identified by the IUCN Red list (IUCN/SSC Antelope Specialist Group 2017), i.e., illegal hunting for meat and live capture for pets, and competition with domestic livestock. Since goat and sheep grazing as well as illegal hunting were prevented since 1949 and 1960 respectively (Mallon and Kingswood 2001), none of these threats applies to the current G. arabica population in the Arava Valley. Instead, the effect of wolf predation appears to be the dominant factor shaping the population of both gazelle species. It is thus imperative for the survival of the G. arabica population to sustain anti-predator protection at a maximum, i.e., keeping the fence in place and ensuring its regular maintenance. Furthermore, the disposal of human discard, such as garbage and agricultural products need to be reduced (or banned), and if necessary wolves need to be culled at low rates after careful discussion with all stakeholders involved. Once, the G. arabica population size has reached a self-sustaining, stable level these measures should be reassessed and adjusted to the new situation. It is further recommended to investigate the population genetics of G. arabica in the Yotvata NR with regard to population viability, inbreeding and a male biased natal sex ratio. With reference to previous and recent suggestions (Mallon and Kingswood 2001; Polack, T., pers. com.), we also recommend to move a small portion of the G. arabica population to a captive breeding unit that is entirely protected from predation by wolves and that could facilitate juvenile survival leading to a rapid increase of overall population size. Young adult individuals should be then reintroduced to supplement the remaining ‘wild’ population. One important ecological aspect that was not considered in our study is the health status of Arabian and dorcas gazelles in the study area. It is thus recommended to look into this aspect and carry out a parasite screening with focus on gastro-intestinal parasites, protozoan infections, zoonotic diseases and a possible cross-infection between dorcas gazelles and the G. arabica population. Despite the positive relationship between G. dorcas and G. arabica population size, it is recommended to remove all dorcas gazelles from the enclosure. This will ensure that the carrying capacity inside the enclosure is unlikely to be reached and that food and water only benefit the Arabian gazelles. These recommendations and research priorities are in line with the INPA strategic plan for the conservation of G. arabica within the next 5 years (Polack, T., pers. com).