Dead Sea level decline pushed a sensitive ecosystem out of equilibrium, causing the relocation of a colonial bird breeding site

ABSTRACT Capsule Dead Sea Sparrows Passer moabiticus transferred their historical breeding colony location, following geo-ecological and hydrological changes, to a new, less arid zone within a nature reserve. Aims To find biotic and abiotic factors affecting the choice of nest sites by Dead Sea Sparrows, in order to locate potential breeding areas and plan in advance for their protection. Methods Mapping old and new nest sites of Dead Sea Sparrows within Einot Tzukim nature reserve, Dead Sea Valley, and using anthropogenic and geo-ecological GIS layers, we created a map of the potential preferred breeding colony area. Results We found the biotic and abiotic factors affecting nest site preference of Dead Sea Sparrows, which included areas far from human activity, areas close to historic springs, and areas affected by a major fire. New areas in the reserve exposed by the retreating Dead Sea were inhabited by the birds, which gradually transferred their breeding centre to a less arid zone within the reserve. Conclusion When managing colonial breeding sites, it is crucial to understand the role of both natural and anthropogenic influences in order to prioritize sites for conservation and management. Producing a preference map, based on such data, can help managers locate potential breeding areas and plan in advance for their protection.

Breeding is a critical life stage for birds, which are typically constrained to their nests while incubating eggs and caring for young (Ibáñez-Álamo et al. 2015, Natusch et al. 2017. Availability of food and the intensity of predation play critical roles in habitat and nest site selection (Eberl & Picman 1993). Nest placement may influence both the ability of predators to detect nests and the degree to which nests are sheltered from extreme environmental conditions (Nelson & Martin 1999). Habitat degradation can force animals to relocate to new areas, where they would need to adjust to an unfamiliar resource landscape and find new breeding sites (Bernat-Ponce et al. 2020). Relocation may be costly and could compromise reproduction (Lima 2009, Kavelaars et al. 2020. Birds that exhibit a high degree of natal and breeding philopatry, and normally breed in stable environments, may suffer costs of philopatry if the quality of their habitat deteriorates (Ganter & Cooke 1998, Haran et al. 2019. Indirect impact of faraway anthropogenic activities can cause a chain reaction, leading to habitat loss or degradation (Rockwell et al. 2003).
The breeding colony of Dead Sea Sparrows Passer moabiticus in Einot Tzukim has been located near the Dead Sea for decades (Inbar 1975). The aquatic habitat in Einot Tzukim relies on springs and is not directly related to the Dead Sea itself. However, a continuous indirect impact, which has resulted from the drop in the level of the Dead Sea through hydrogeological processes, is evident within the endemic ecosystem of Einot Tzukim; among the impacts seen have been shifts in the nest locations of Dead Sea Sparrows (Burg et al. 2016, Haran et al. 2021. Increasing volumes of water withdrawn from the Jordan River for consumption in Israel and Jordan, together with water pumped for the potash industry, are the main causes for the drop in the level of the Dead Sea (Bowman et al. 2011). The drop began gradually in the 1940s and was followed by a reduction in the ground water level in the northern part of Einot Tzukim nature reserve, pushing the hydrogeological system that feeds the springs out of equilibrium. Previous studies (Yechieli & Magal 2003, Mallast et al. 2013) raised the concern that the sea level decline may have an effect on the location of the springs and, consequently, on the extremely sensitive ecosystem that relies on them (Burg et al. 2006, Burg et al. 2016. Indeed, some springs in the northern areas of the reserve began to dry up and followed the withdrawing sea eastward and southward. The rapid decline in the level of the lake, by approximately 1 m/ year, subsequently caused morphological processes in Einot Tzukim nature reserve: increased distance from the Dead Sea shoreline, exposure of clay mud plains without vegetation, and development of deep erosional flow creeks that cut through the mud plains toward the receding lake (Bowman et al. 2011, Vachtman & Laronne 2014) (online Figure S1). By the mid-nineties, the springs in the north-western part of the reserve were permanently dry in their original locations and had shifted eastward following the withdrawing sea. The aquatic vegetation was replaced by plants with deep roots, such as Nile Tamarisk Tamarix nilotica, the sea-blite Suaeda monoica, and Mediterranean Saltbush Atriplex halimus, and the whole area became a saltmarsh (Burg et al. 2006).
Here we studied the nest site preference of the Dead Sea Sparrow in relation to changes in the sea level of the Dead Sea. Dead Sea Sparrows prefer to breed in loose colonies near aquatic habitats, as is common in other colonial bird species (Rolland et al. 1998). The ideal breeding conditions for Dead Sea Sparrows are a combination of high temperatures, proximity to water, and suitable trees on which to construct their exceptionally large nests (Yom-Tov & Ar 1980, Jamadi & Darvishi 2008. Some of the nests are repaired/reconstructed and reused year after year (Haran et al. 2021).
We sought to understand the impact of the sea level drop on the Dead Sea Sparrow breeding colony, as seemingly opposing processes were observed. Along with degradation of their original breeding habitat, there was also enlargement of potential new breeding areas within the reserve. Our hypothesis was, that despite their fidelity to the original breeding area, the degradation of their habitat will eventually force them to relocate. We expected to find a new centre of their breeding colony. The potential benefits and costs of colonial breeding are known, but the evolutionary role of coloniality in birds is still unresolved (Danchin & Wagner 1997, Rolland et al. 1998. When managing colonial breeding sites, it is crucial to understand the role of both natural and anthropogenic influences and prioritize sites for conservation and management (Tsai et al. 2016). Producing a colonial species breeding area preference map, based on such data, can help managers locate, plan, and protect the species.

Study area
Einot Tzukim is a large oasis nature reserve in the Dead Sea Valley (6 km long, 100-2700 m wide), located along the coast of the Dead Sea (31.72°N 35.45°E) and at the foot of the Judean Mountains ( Figure 1). The Dead Sea marks the eastern side of the reserve, and its retreat causes the territory of the reserve to grow accordingly. To the west, a major road (Road 90) runs along the reserve's north-south axis; parallel to it, a major vehicle maintenance trail runs inside the reserve. To the north of the reserve are agricultural fields and greenhouses, and to the south a Date Palm Phoenix dactylifera orchard ( Figure 1).
Along the western side of the reserve, hundreds of permanent springs flow with a total discharge estimated to be about 70 MCM/year, with salinities between 1000 and 3000 mg Cl/L (Burg et al. 2016). Tamarisks Tamarix sp., Date Palm and Christ's Thorn Jujubue Ziziphus spina-christi are the main trees in the reserve, while the main bushes include shrubby saltbushes Atriplex sp., Nitre Bushes Nitraria retusa, Ravennagrass Tripidium ravennae, Arabian Rush Juncus rigidus and the sea-blite Suaeda monoica. In addition, Common Reed Phragmites australis covers large areas of the reserve.
Throughout the reserve, Dead Sea Sparrow nests are spread in clusters with varying densities. Some of these loose colonies seem to be abandoned and some are active.
In the centre of the reserve, 20 ha are dedicated to public recreation with an array of wading pools and artificial shading. On average, 150,000 people visit the reserve every year (National Parks Authority data) ( Figure 1). In June 2008 a major fire occurred in Einot Tzukim and more than 200 ha were burned (National Parks Authority data) (online Figure S2). Many burnt Tamarix trees stayed standing and attracted Dead Sea Sparrows to construct their nests on them.

Nest survey
In winter 2014/2015 we conducted a survey of all active and inactive nests of Dead Sea Sparrows throughout the reserve. These nests are up to 70 cm in height and so solidly constructed that they can stay intact for years; additionally, they are typically very visible and can be seen from far away. The nests are not occupied during winter. We marked all nest locations and registered them using a mobile global positioning system (GPS) device with an average accuracy of 8 m. In addition, we recorded trees with nests in them as being green, dry or burnt, in order to understand the choice of nest site location by Dead Sea Sparrows. A second nest survey was performed in spring 2015 to locate all new and previously unmarked active nests (online Figure S3).

Data analysis
We conducted the data analyses in R 3.6.1 (R Core Team 2019). The dataset included all the nests we located during our surveys within the nature reserve (N = 470) and an equal number of randomly distributed points (N = 470) in locations without nests within the nature reserve, generated using the random point generator tool in ArcMap software (ESRI. ArcGIS desktop: release 10.8.1. Environmental Systems Research).
We used a generalized linear model (GLM, using the GLM function included in the base R software stats package) with a binomial response variable (0 = no nest, 1 = nest), in relation to different variables. The anthropological variables included Euclidian distance from the nearest road, trail, agricultural field, or public area (recreation area where visitors are allowed to camp and enter water pools) and percentage of fire area in a 100 m radius. We predicted that nests would be found further away from these elements. The geoecological environmental variables included Euclidean distance from the nearest gully, Palm tree, historical spring, pond, and 1940, 1980, and 2015 shorelines of the Dead Sea. Additional variables were the normalized difference vegetation index, slope in degrees calculated from the digital elevation model using the slope function in the Spatial Analyst included in the ArcMap software (ESRI. ArcGIS desktop: release 10.8.1. Environmental Systems Research Institute), and percentage of reed cover within a 100 m radius. All these elements were affected by the hydrological and other changes caused by the withdrawing sea and in turn affect the choice of breeding site by Dead Sea Sparrows (Table 1).
We constructed two GLMs: one based on the entire dataset (N = 940) and the second based on only new nests discovered in the second survey (spring 2015) and a similar number of randomly distributed points (N = 164). We obtained digital maps of the environmental and anthropological features of the reserve from the Israel Nature and Parks Authority in order to characterize the attributes of nest locations. For each nest and each random point, we calculated the variables listed in Table 1 (i.e. distance from a feature, or proportional cover of a feature) using ArcGIS 10.6 software (www.esri.com).
To avoid multicollinearity, we chose the most meaningful variable from pairs of variables with a Spearman rank correlation |ρ| > 0.6 (Zuur et al. 2009(Zuur et al. , 2010. This ensured that all the predictors in the GLMs had a variance inflation factor (VIF) < 4 (online Tables S1 and S4). We then tested all combinations of remaining variables (Table 1) in the global model and ranked the selected models according to the Akaike information criterion (AIC; Burnham & Anderson 2002) using an automated stepwise model selection procedure (the 'dredge' function from R package MuMIn) (Barton 2009) in which models are fitted through repeated evaluation of modified calls extracted from the model containing all the meaningful variables (Sugiura 1978). Furthermore, we averaged all models with ΔAIC < 7 (Burnham et al. Tables S2 and S5) and used Akaike weights (w i ; Anderson et al. 2000, Anderson et al. 2001 to assess the relative importance of the different variables (online Tables S3 and S6). We selected the model with the lowest AIC as the best model. Using the trainControl function from the R package CARET (Kuhn 2008), we used 10-fold cross-validation with 10 repetitions, where the model was trained on 70% of the data and then applied to the remaining 30% of the data. These data subsets were chosen randomly for each repetition (Hastie et al. 2009, Meijer & Goeman 2013. From the repeated cross-validation, we reported the ability of the best model to distinguish between the presence and absence of nests using: (a) the area under the curve (AUC) of the receiver-operating characteristic curve (with standard deviation); (b) the logistic regression accuracy (defined as the ratio between the sum of correct predicted cases of presence and absence of nests and the sum of correct  (Fawcett 2006). In order to visualize the areas with different chances to locate nests of Dead Sea Sparrows, we created a coloured map where darker colours represent higher probabilities of finding nests. This was based on the predictive logistic regression model (best model), applied to a cell size of 15 m × 15 m. Locations of the observed (true) nests and border lines of the Dead Sea in 1880, 1940, and 1980 were included for reference.

2011) (online
This visualization, together with historic Dead Sea border lines, Einot Tzukim reserve border lines, and a latitude line, enabled us to define two distinct areas which together hold most of the nests of Dead Sea Sparrows. We compared the different quantities and densities of nests (old and new separately) in these areas. We calculated the mean centre of collective true nests locations in each of these areas using the ArcGIS, and calculated the distance between these centres.

Results
Four hundred and seventy Dead Sea Sparrow nests were located throughout Einot Tzukim national reserve during the survey of winter 2014/2015 ( Figure S1), and in addition, 81 new nests were found in spring 2015. Although 90% of the Tamarisks in the reserve were green, (9% dry, 1% burnt), more nests were found on dry and burnt trees than on green trees in winter 2014/2015 (χ 2 = 2854, df = 3, P < 0.001; Figure  S2).⍰ Of the new nests in spring 2015, 86.4% were found on green Tamarisks.
Five variables were found to greatly influence the probability of finding a Dead Sea Sparrow nest in the reserve as indicated by their high coefficient values (best model, Table 2). The probability of finding a Dead Sea Sparrow nest in the reserve is greater where the percentage of the burnt area was higher and closer to the 1940 shoreline, additionally, it increases with greater distance from public places and agricultural areas, while the presence of large reedbeds reduces the probability of finding nests.
The results of the separate logistic regression model applied only on the new nests found during the spring 2015 survey are reported in Table 3. Only three variables were found to be significant in relation to finding a new nest on the reserve: being close to the 2008 fire zone, far from agricultural areas and close to the 1940 sea shoreline. These variables are the same as in the main logistic model. A probability map for only new nests is presented in online Figure S2.
The two distinct areas which held most of the Dead Sea Sparrow nests were the northern historic part of the reserve, west of the 1940 shoreline and the more recently exposed area, west of the 1980 shoreline (Figures 2 and 3). The highest densities of Dead Sea Sparrow nests (total and also new nest subset, not shown) per hectare were found in the area west of the 1980 shoreline, south of latitude X:625000 (Figure 3) both in 80-100%, and 60-80% probability areas (Table 4). The distance from the centre of this area to the centre of the historic centre in the north was 1300 m (Figure 3).

Discussion
The biotic and abiotic variables affecting the position of the breeding colonies of Dead Sea Sparrows The results of the logistic model (GLM of all data) present a number of variables which raise the  probability of finding Dead Sea Sparrow nests. A higher proportion of nests were found in areas influenced by the historic 2008 fire, particularly on dry or burnt Tamarix trees ( Figure S2). As nests cannot escape a fire, these nests were most likely built after the fire.
Dead Sea Sparrows prefer to construct or reconstruct their nests on burnt or dry trees. Although 90% of the Tamarisks in the reserve were green, more nests were found on dry and burnt trees than on green trees in winter 2014/2015. Furthermore, new nests on trees/ branches in leaf are much less stable, especially during the spring winds that occur early in the season (pers. observation). Many nests fall or are abandoned unfinished after a windy event, and we speculate that these nests are probably built by inexperienced birds early in the season.
Predators may be attracted to bird breeding colonies by the increased visual, auditory, and olfactory cues (Varela et al. 2007). Nest site choice is a trade-off between the need for concealment and the need to maintain an ability to maintain vigilance around the nesting area (Götmark et al. 1995). Dead Sea Sparrows might be attracted to nest in post fire zones on leafless trees in order to enable them to detect predators, such as snakes and raptors, near to their nests (Yom-Tov & Ar 1980). Snakes are more exposed on the leafless trees and can be easily detected, at which point the adult birds can either flee or mob the snakes. A similar phenomenon of nonrandom tree choice was also seen in Metallic Starling Aplonis metallica which prefers trees without low branches and with smooth bark. Their choice of nesting colony site was found to be linked to defence against snake predation (Natusch et al. 2017).
Approximately 10% of the green area of the reserve is dedicated to public recreation ( Figure 1) and Dead Sea Sparrows tend to stay away from this area. Humaninduced disturbance includes noise, vehicle traffic, and more, and can have a significant negative effect on breeding success, as a result of nest abandonment and increased predation (Remacha et al. 2016, Bernat-Ponce et al. 2021. Outside the breeding season, human recreation reduces the use of sites by birds (Hockin et al. 1992). Furthermore, the public areas of the reserve are also home to a community of House Sparrows Passer domesticus which tend to usurp Dead Sea Sparrows from their nests, and use them for breeding themselves (Yom-Tov & Ar 1980, R.H. pers info).
The area east of the 1940 shoreline was previously covered by the Dead Sea (Figure 4). The 1940 shoreline is highly correlated with historical spring locations and with proximity to Road 90 which serves as a main road south to Eilat, Israel (both variables removed from analysis to prevent multicollinearity). The historical springs provided water to the aquatic habitat of Einot Tzukim natural reserve. Most nests close to the 1940 shoreline were found west of it, even after the retreat of the sea, and that is where the historic springs were located. Many studies provide evidence of strongly reduced densities of many species adjacent to busy roads with traffic noise being the most critical factor (Reijnen et al. 1997, Polak et al. 2013. The potential negative influence of the vehicles passing on Road 90 on the nest preference of the Dead Sea Sparrows was not checked separately. Most nests were found more than 150 m away from the road, nearer to the historical spring locations. Dead Sea Sparrows preferred to nest away from the agricultural areas in the north and away from large reedbeds. There are no observations of Dead Sea Sparrows foraging in the agricultural areas north of Einot Tzukim (Pers. Info). As the whole system of springs shifted southward with the retreat of the Dead Sea, the Dead Sea Sparrows followed these springs southward, away from the agricultural fields, to the new emerging aquatic ecosystem. Large reedbeds in Einot Tzukim are homogeneous and rarely host trees ( Figure 5). Dead Sea Sparrows build their nests on Tamarisks which do not grow in large reedbeds (Yom-Tov & Ar 1980). Some of the areas with high probability of finding Dead Sea Sparrow nests, as predicted by the model, were actually vacant ( Figure 2). Colonially breeding species that show strong site fidelity are likely to occupy only a portion of the breeding habitat available to them (Matthiopoulos et al. 2005). In addition, some of these areas are dominated by low-growing Tamarisks which might suffer from lower ground water levels and therefore may not attract Dead Sea Sparrows.

The Dead Sea level drop chain reaction effect
The historic Dead Sea 1940Sea , 1980Sea , and 2005 shorelines were chosen to represent the impact of the withdrawal of the Dead Sea on Dead Sea Sparrow breeding area preferences. Using these shorelines, we created a simulation of how the Dead Sea withdrawal affected the Einot Tzukim nature reserve by exposing new terrain (Figure 4). For a century or more (until Table 4. Einot Tzukim nature reserve Dead Sea Sparrow nest densities in relation to the 1940 and 1980 shorelines. The data relates to the categories of 80-100% and 60-100% chances of finding a nest (Figure 3 1940), the level of the Dead Sea was relatively stable, fluctuating between −391 and −399.5 m below sea level (Klein 1986). During this period, Dead Sea Sparrows could only breed along the north-west side of the reserve, as the rest of the area was covered by the sea (Figure 4). Forty years later, in 1980, and 25 more years later, in 2005, we can see the effect of the retreating Dead Sea on Einot Tzukim (Figure 4). It created two different gradual geo-ecological chain effects: 1. in the north of the reserve, the drying of springs, followed by the drying of the aquatic vegetation, the rise of new vegetation of salt marsh, and eventually, the gradual abandonment of the historic (Inbar 1975) Dead Sea Sparrow breeding colony ( Figure 3, Table 4, and also in numerous bird ringing sessions over the years when progressively fewer Dead Sea Sparrows were captured in the north; data in Israel Bird Ringing Center Archive); 2. in the south, the gradual effect of shoreline recession caused the exposure of new, rather narrow strips of land and the shift of springs towards the shorelines. The rise of new vegetation created new Dead Sea Sparrow breeding sites. The ArcGIS software and the logistic model enabled us to define specific areas in the reserve and determine the respective densities of nests of Dead Sea Sparrows (Table 4). The areas we chose to focus on are presented in Figure 3. They were cut out of the preference map ( Figure 2) and overlaid with the historic Dead Sea 1940 and 1980 shorelines. The southern area, presented in Figure 3, is the area with the highest density of nests in both categories of 80-100% and 60-100% chance of occurrence (Table 4). More new nests are being built there than in the old colony in the north. The densities of total nests in the old northern colony are lower in both categories of density, and almost half when counting new nests alone (Table 4). It seems that the centre of the breeding colony of the Dead Sea Sparrows in the reserve has shifted southward, as of 2015, from the original area (Israeli Transverse Mercator N:625806 E:243314), to the centre of the reserve as of 2015 (N:624489 E:243004; Figure 3). The distance between these centres is 1350 m.
At the time of this study, 20 years have elapsed since the complete drying of the northern springs. Deciphering the map of Dead Sea Sparrow nests ( Figure 2) in 2015 is complicated as we do not know when each nest was built or reconstructed. This is an important limitation of this study, but using the Dead Sea historic shorelines on the map of nests, gives a more accurate understanding of when nests in a given area could not have been built, as the area was covered by the sea (Figures 2-4). Some small, deserted breeding sites of Dead Sea Sparrows that were found at the eastern edge of the vegetation zones are probably testimonies of the temporary locations of springs following the retreating Dead Sea. These breeding sites were likely formed by dispersing young birds and were abandoned when the springs shifted again.
The desertion of the northern part of the reserve by the Dead Sea Sparrows is gradual. Young birds might prefer to settle along a colony's periphery and their settlement pattern might lead to a long-term shift in a colony location as a whole (Ganter & Cooke 1998). Contradictory elements play a role in this phenomenon. The benefits of Dead Sea Sparrows fidelity to their nests and to their breeding colony (Haran et al. 2021) is being challenged by the degradation of their habitat, especially in terms of fresh water available on the surface. They would now need to fly about 1 km from their original colony to the closest source of fresh water. Lack of fresh surface water might also decrease abundance of arthropods, which make up part of the Dead Sea Sparrow diet during the breeding season (McCluney & Sabo 2012, Frampton et al. 2000.

Implications for management
The set of variables influencing the preference of a breeding colony location might be complex. This study showed that spatial analysis and logistic regression can produce a meaningful model for sometimes opposing or unpredicted variables, influencing the preference of a breeding colony location; in Einot Tzukim, for example, Dead Sea Sparrows are attracted to nest near the historic springs but probably away from a busy road which runs along the line of springs. Producing a breeding area preference map, based on such data can help managers locate and protect it. As geohydrological processes continue to shift habitats and ecosystems in the future, together with the impact of climate change, it will be advantageous to perform surveys and modelling, using old and new variables, to locate new potential breeding areas. In this case, further research is needed, as the latest surveys are already showing that major breeding activity of Dead Sea Sparrows is found along the new gullies leading to the Dead Sea. A collaboration is needed with hydrologists, in order to follow the advancement of seepages and springs towards the Dead Sea and their impact on the endemic ecosystem along these gullies and the Dead Sea Sparrow breeding colony in particular.