OUP user menu

Long-term study of population dynamics and habitat selection of rodents in the Negev Desert

Georgy Shenbrot, Boris Krasnov, Sergei Burdelov
DOI: http://dx.doi.org/10.1644/09-MAMM-S-162.1 776-786 First published online: 16 August 2010

Abstract

Population fluctuations of 13 rodent species in 5 habitats were monitored for 16 years in the central Negev Desert, Israel. Analysis of factors affecting population dynamics of 9 common and abundant species demonstrated that densities of most rodent species in the late summer, after the end of reproduction, were determined mainly by total precipitation during the previous rainy season. Rodent densities in the winter, before the reproductive season, were determined mainly by their densities in the previous (late summer) season. Rodent populations in dry river beds (wadi) demonstrated either no or negative correlations with total rainfall, suggesting episodes of population crash due to flash winter flooding. For all species occurring in more than 1 habitat, densities, at least in some habitats, were correlated with their contemporary densities in other habitats. For these species, processes of density-dependent habitat selection were indicated using isodars analysis. Generally, population dynamics of desert rodents were determined by the complex interactions of extrinsic (rainfall) and intrinsic mechanisms but were modified by density-dependent habitat selection.

Key words
  • desert rodents
  • habitat selection
  • isodar analysis
  • long-term study
  • population dynamics
  • rainfall

Water is the primary limiting resource in desert ecosystems. Because water comes to deserts in short-term rain events, precipitation is considered as the main factor limiting productivity of plants in these systems (Noy-Meir 1973). Relationships between productivity and rainfall differ among plant life forms and depend on habitat type and pattern of rain events (Southgate et al. 1996). Nonetheless, it has been shown that seed production of annuals and growth intensity of perennials both correlate directly with total rainfall (Gunster 1994; Kadmon 1993; Mauchamp and Janeau 1994). The amount of seed production of annuals is a direct determinant of habitat carrying capacity for granivorous rodent populations, and growth intensity of perennials is a direct determinant for folivorous rodents. Reproduction in desert rodents is thought to be stimulated by the appearance of newly vegetated annual plants after rain events (Beatley 1969, 1974; Fichet-Calvet et al. 1999; Reichman and Van De Graff 1975; Reynolds and Turkowski 1972; Van De Graff and Balda 1973). Fresh green vegetation provides rodents with water necessary for lactation and green plant factor, 6-methoxyben-zoxazolinone, which has been shown to be important for successful reproduction (Berger et al. 1987; Linn 1991; Negus and Berger 1977; Negus and Pinter 1966). These different aspects of food supply are together thought to be important, but we do not know their relative contributions to rodent population dynamics. Nevertheless, temporal fluctuations in desert rodent density are positively correlated with annual plant population densities of annuals (Brown and Heske 1990). Numerous studies have reported changes in densities of desert rodents following rainfall fluctuations (Beatley 1969; Dickman et al. 1999; Ernest et al. 2000; Masters 1993; Meserve et al. 1995; Newsome and Corbet 1975; Predavec 1994; Southgate and Masters 1996; Whitford 1976), although in some cases direct relations between rainfall patterns and rodent population fluctuations in long-term series were not found (Brown and Ernest 2002; Ernest et al. 2000).

Density fluctuations of different species at the same site are not usually synchronous (Brown and Heske 1990; Heske et al. 1997). The common explanation of this phenomenon is that individual species respond differently to changes in quality and quantity of food resources with consequent differences in seasonal patterns of reproduction. As a result, the time lag in response to rain events is species-specific (Dickman et al. 1999; Southgate and Masters 1996). Other extrinsic factors such as interspecific competition (Brown and Munger 1985; Heske et al. 1994) and predation (Meserve et al. 1996) also can be important in contrasting among-species differences in density dynamics. Moreover, rainfall in deserts usually has seasonal patterns, and rodent density dynamics within time periods without rainfall events can be determined exclusively by intrinsic factors such as population autocorrelative density-dependence (Lima and Jaksic 1999; Murua et al. 2003).

Population dynamics of desert rodents are closely correlated among sites in close proximity assuming they are driven by dispersal processes (Ernest et al. 2000). Cases of nonsynchronous density fluctuations in different habitats at the same site were described for nondesert rodents (Okansen et al. 1999). This phenomenon is usually explained by predator-prey interactions. However, the theory of habitat selection also can be involved in an explanation of this phenomenon because the theory assumes that average fitness of individuals in a population is density-dependent and that this fitness-density dependence is habitat-specific (Rosenzweig 1992). Moreover, if fitness (at a given density) in 1 habitat type is constant but fluctuates and sometimes drops close to 0 in another habitat type, it will lead to a situation described by the “source-sink” concept (Pulliam 1988; Pulliam and Danielson 1991).

The theory of habitat selection (Rosenzweig 1981, 1991, 1995) explains density-dependent habitat selectivity based on the ideal-free distribution (Fretwell and Lucas 1970), which assumes that individuals select habitats to maximize fitness. At low density maximum per capita fitness is highest in the best (most productive) habitat, and all individuals should concentrate there. As density increases per capita fitness decreases, eventually to the point where some individuals can achieve the same or higher fitness reward by moving to the next best habitat. At equilibrium, individuals will be distributed among habitats such that average per capita fitness is equal across habitats. This process of attaining the density-dependent habitat distribution is known as density-dependent habitat selection.

The nature of habitat differences and corresponding types of density-dependent habitat selection (population regulation) can be inferred from data on habitat distribution and abundance of individuals using isodar analysis (Morris 1988, 1990). A habitat isodar is a curve (usually a straight line) plotted in density space (N1 plotted on N2) and connected points of density in the 1st habitat plotted versus density in the 2nd habitat at different density levels. At every point of this line the fitness of individuals in 1 habitat is equal to that of individuals in another. In practice, isodars are found by regressing population density in 1 habitat against density in another. The slope of an isodar indicates qualitative habitat differences, that is, habitats differing in conditions of resource acquisition, and the intercept of the isodar indicates quantitative differences in resource availability of 2 habitats. Nonsignificant regressions imply density-independent or no habitat selection.

Habitats can differ in food availability (quantitative difference) and structure, kinds of resources, and the interacting species they contain (qualitative difference). Depending on the combination of qualitative and quantitative habitat differences, 5 models of population regulation have been distinguished (Morris 1988). In the parallel regulation model habitats differ only quantitatively and produce an isodar with a positive intercept and a slope equal to 1. Two models of divergent regulation refer to situations where habitats differ qualitatively (isodar slope > 1). The intercept equals 0 in quantitatively equal habitats and is positive if the qualitatively better habitat also is quantitatively better. The models of convergent and crossover regulations describe the cases where the 1st habitat is quantitatively better but qualitatively less suitable than the 2nd habitat, resulting in isodars with positive intercepts and slopes < 1; these 2 models differ in relative values of carrying capacities of the 2 habitats and can be distinguished by the position of the point of equal densities on an isodar line, within (crossover model) or out of (convergent model) the range of observed densities. However, an isodar reflects habitat differences in an accurate manner only in the case of cost-free habitat selection (Morris 1992). Some technical limitations exist for isodar methodology, such as the arbitrary nature of the initial assignment of habitats to abscissa and ordinate axes, restriction to 2 habitat types, and its requirement for areas of habitat to be constant for the duration of the study.

One of the principal assumptions of the isodar approach is constant carrying capacity of habitats, and, thus, isodar analysis should be highly sensitive to fluctuations in resource abundance. Theoretically, temporal variation in an environment would produce isodars that differ in slope or intercept or both (Morris 1990). For desert rodents isodar differences related to seasonal and multiannual fluctuations in resource abundances were demonstrated in our studies (Shenbrot 2004; Shenbrot et al. 2006).

We present data collected over 16 years on population dynamics of 9 desert rodent species in different habitats in the Negev Highlands, Israel. We are testing the general hypothesis that population dynamics of desert rodents is a result of complex interactions between intrinsic and extrinsic regulation and processes of habitat selection. In desert environments water is a limiting resource that affects producers and consumers through linear trophic responses. Therefore, positive correlations should exist, with a time lag, linking precipitation and rodents (Brown and Ernest 2002; Ernest et al. 2000). More specifically, in the Negev Desert rainfall, which determines the level of plant production during the spring vegetation season, occurs in late autumn-early spring. The amount of green vegetation and level of seed production affect the intensity of rodent breeding that takes place in spring-early summer. Individuals of species occurring in several habitats redistribute among habitats via the processes of density-dependent habitat selection. Thus, we test 2 specific hypotheses. First, for each species at least 1 (optimal) habitat should exist where summer (postbreeding) density is determined mainly by rainfall with a 0.5-year lag, and winter (prebreeding) density is determined mainly by previous (summer) density due to the absence of rainfall events and reproduction between summer and winter rodent censuses. Second, for species occurring in several habitats their densities in other than the optimal habitats are determined mainly by the processes of dispersal and density-dependent habitat selection resulting in higher correlation of density in a given habitat with current densities in other habitats than density in a given habitat with rainfall or with previous density in the same habitat.

Materials and Methods

Study area.—We studied the community of rodents in the Ramon erosion cirque and vicinity, Negev Highlands, Israel, from 1992 to 2008. The Ramon erosion cirque (30°35′N, 34°45′E, about 200 km2 total area) forms the southern boundary of the Negev Highlands. The northern and southern rims of the cirque are at elevations of 800 m and 510 m, respectively, and the lowest point of the cirque is 420 m elevation. Summer is hot and winter is relatively cold (mean monthly maximum and minimum air temperatures are 33.4°C and 19.8°C in August and 14.8°C and 6.6°C in January). Rain usually occurs from October to March. An abrupt decrease in mean annual rainfall occurs from 80 mm on the north rim of the cirque to 49 mm at the bottom (our data). This precipitation gradient also is expressed from west to east of the Ramon cirque. Data on rainfall at the upper part of the precipitation gradient were recorded automatically every half-hour using a Weather Monitor II weather station (Davis Instruments Corp., Hayward, California), stored in the memory of the station, and downloaded into the computer once a week. A rain gauge was installed at the lower part of the precipitation gradient on the most remote grid, and readings were taken at the end of each month during the rainy season. Fluctuations of annual rainfall during our observations at the upper part of the precipitation gradient (Mizpe Ramon) are presented (Fig. 1).

Fig. 1

Precipitation for rainy season (October–May) for the period of observations (1992–2008, Mizpe Ramon weather station) in the Negev Highlands, Israel.

The landscape of the central and eastern parts of the Ramon cirque and areas adjacent to its north rim ranges from sand dunes to limestone and sandstone rocks in the rims. Five main habitat types were distinguished within the study area based on rodent response to existing vegetation and substrate gradients (habitats from a “rodent point of view”—Krasnov et al. 1996b): sand dunes (hereafter referred to as sand) under the eastern wall of the cirque with cover of Echiochilon fruticosum (perennial vegetation cover 7.4%); flat gravel plains (hereafter referred to as hammada) of the eastern part of the cirque with sparse vegetation of Hammada salicornica, Anabasis articulata, and Gymnocarpos decandrum (cover 5.7%); limestone cliffs (hereafter referred to as rock) of the southern and central parts of the cirque with sparse cover (4.0%) of Zygophyllum dumosum, Helianthemum kahiricum, and Reaumuria hirtella; deep valleys filled by loess (hereafter referred to as loess) with densely vegetated wadi (cover 18.5%) among rocky hills partially covered with loess over the northern rim of the cirque with Anabasis articulata, Atriplex halimus, and Artemisia herba-alba; and wide wadi among hammada of the eastern and central parts of the cirque (hereafter referred to as wadi) with dense cover (26.0%) of Retama raetam, Moricandia nitens, Tamarix nilotica, and Artemisia monosperma. Seed abundance in soils was estimated in 2000–2005 by seed flotation in 50-g soil samples (6 samples per habitat per year) taken from the upper (0–3 cm) soil level. Average seed abundance per 1 kg of soil was 7.8 g in sand, 4.1 g in hammada, 38.8 g in loess, and 6.8 g in wadi; seed abundance was not estimated in rocks.

Rodent species.—Thirteen species formed the community of desert rodents in the Ramon cirque (Table 1). Of these species, 6 (bushy-tailed jird [Sekeetamys calurus], common spiny mouse [Acomys dimidiatus], golden spiny mouse [A. russatus], house mouse [Mus musculus], black rat [Rattus rattus], and Asian garden dormouse [Eliomys melanurus]) were omnivorous, 4 (Wagner's gerbil [Dipodillus dasyurus], lesser Egyptian gerbil [Gerbillus gerbillus], Henley's gerbil [G. henleyi], and Baluchistan gerbil [G. nanus]) were granivorous, 1 (fat sand rat [Psammomys obesus]) was strongly folivorous, and 2 (lesser Egyptian jerboa [Jaculus jaculus] and Sundevall's jird [Meriones crassus]) consumed a mixed diet of seeds and green parts of vegetation. Interspecific competition was not a widespread phenomenon in this rodent community and regularly occurred in only 3 pairs of species, D. dasyurus–G. henleyi, G gerbillus–G. henleyi, and D. dasyurus–M. musculus. In all of these cases competition was asymmetrical, with a significant effect of the 1st species listed on the distribution and density of the 2nd species but without an opposite effect (Shenbrot and Krasnov 2002).

View this table:
Table 1

Rodent species in the community, showing mean body weights and mean species densities in the 5 main habitat types in the Ramon Cirque, Negev Highlands, Israel, averaged for period 1992–2008 across all trapping sessions.

SpeciesBody weight (g)Density (individuals/ha)
SandHammadaWadiLoessRock
Dipodidae
Jaculus jaculus67.00.300.490.140.030.00
Muridae
Deomyinae
Acomys dimidiatus42.80.000.000.030.271.83
A. russatus51.40.000.020.710.000.79
Gerbillinae
Dipodillus dasyurus21.10.080.164.5011.681.62
Gerbillus gerbillus20.14.160.010.010.000.00
G. henleyi9.80.811.380.960.190.00
G. nanus22.30.030.000.010.000.00
Meriones crassus81.11.580.361.281.550.02
Psammomys obesus1660.010.000.218.250.00
Sekeetamys calurus52.80.000.000.000.000.81
Murinae
Mus musculus14.00.000.000.024.040.00
Rattus rattus76.50.000.000.000.010.00
Gliridae
Eliomys melanurus49.60.000.000.000.140.00

Data collection.—Rodents were trapped on 17 permanent 1-ha grids that were chosen to represent main substrate and vegetation gradients. Of these grids, 3 were placed in sand, 3 on hammada, 3 on rock, 4 in loess, and 4 in wadi. Distances between grids of the same habitat type varied from 400 m to 5 km. Cases of movement of marked individuals between grids were extremely rare, and, thus, these grids were considered as independent samples. We began establishing the system of sampling grids in the summer of 1992 and completed it in the summer of 1993. We sampled grids twice a year from July 1993 to August 2008, once in winter (January– February) and once in summer (July–August).

Each grid was subdivided into 25 plots of 20 × 20 m, centers of which were marked with numbered wire flags. Each grid (5 × 5 stations with 20-m intervals between stations) was sampled for 3 days using folding live traps (H. B. Sherman Traps, Tallahassee, Florida) baited with millet seeds and placed near the center of each plot. We chose the 3-day survey period as a trade-off between time investment and completeness of survey because preliminary trapping for 6–8 days demonstrated that for most common species (D. dasyurus, G. gerbillus, G. henleyi, M. crassus, A. dimidiatus, A. russatus, and M. musculus) 35–50% of all individuals were recorded in the 1st night, 60–70% in the first 2 nights, and 80–85% in the first 3 nights, and additional trapping nights added only 2–7% of individuals. The number of traps per plot varied from 1 to 3 (25–75 traps per grid), depending on rodent density, so that the number of traps per grid was at least twice the number of rodents to prevent trap competition. The initial number of traps placed on a grid was determined from results of the previous trapping session and adjusted accordingly with the results of the 1st trapping night of the current session.

Jaculus jaculus, which was not caught in live traps in summer, was caught with a net at night using a searchlight. We spent 0.5 h per grid between 2200 h and 2400 h during each summer trapping session searching for J. jaculus. For P. obesus, which also was not caught in Sherman live traps, we mapped all active burrows and then counted animals near these burrows using binoculars at the beginning of their morning surface activity (about 2 h after sunrise), when they tend to stand near their burrows. Later, they were trapped with Havahart 2-door cage live traps (model 1025; Havahart, Lititz, Pennsylvania), with 2 traps per burrow system and using fresh leaves of Atriplex halimus or succulent stems of Anabasis articulata as bait.

Each animal caught was sexed, weighed, marked, and released. From 1992 to 2000 animals were marked by toe clipping. From 2001 to 2008 they were implanted with a passive microsponder (Trovan ID-100; Trovan Ltd., Hessle, East Yorkshire, United Kingdom). We subsequently identified individuals by reading identification codes from the transponder using an Allflex portable RFID reader (P/N 930002-002E; Allflex USA, Inc., Dallas, Texas). Rodent densities were estimated as the minimum number of individuals known to be alive of each species per 1-ha grid (100 × 100 m). Note that for all species except P. obesus these estimates provide adequate indices of relative rodent densities rather than true density estimates.

In our field procedures we followed guidelines approved by the American Society of Mammalogists (Gannon et al. 2007). All fieldwork was done under permits from Israel Nature and Natural Parks Protection Authority, obtained each year.

Data analysis.—To estimate factors influencing rodent densities we analyzed data from summer and winter surveys separately for each species-habitat combination. Data were included in the analysis if a given species–habitat–season combination contained >50% nonzero values. We used stepwise multiple regression analysis with current density as the dependent variable and the previous density in the same habitat (with time lag 0.5 year) and total rainfall during the previous rainy season (time lag 0.5 year for summer and 1 year for winter) as independent variables. An independent variable was included into a regression equation if its F-value was > 1.00 and P < 0.05. Partial regression coefficients were standardized so that their relative importance in determining the current density could be assessed. Preliminary time-series analysis demonstrated that statistically significant partial autocorrelations, if recorded, were only for the time lag equal to 0.5 year. Such statistically significant partial autocorrelations were found for D. dasyurus in loess; G. gerbillus in sand; G. henleyi in hammada and wadi; M. crassus in sand dunes, hammada, and loess; S. calurus in rock; and P. obesus in loess. Data points in our analysis were taken from time series with the lag equal to 1 year (separate analyses for summer and winter densities) and, thus, can be considered as statistically independent. For species regularly captured in >1 habitat, current densities in each other habitat also were included in the analysis as independent variables. For cases for which we have direct or indirect evidences of interspecific competition (M. musculus-D. dasyurus, G. henleyi-G. gerbillus, and G. henleyi-D. dasyurus), current densities of potential competitors in the same habitat also were included in the analysis as independent variables. The time-series and multiple regression analyses were done using Statistica 7.0 software (StatSoft, Inc., Tulsa, Oklahoma).

In cases where the above analyses indicated that density in 1 habitat was correlated significantly with density in another habitat, we applied isodar analysis (Morris 1988, 1990) to estimate the type of density-dependent habitat selection. Isodar regressions were calculated using densities in each habitat pair. Because isod,ar analysis is highly sensitive to fluctuations in resource abundance (Shenbrot et al. 2006), data across years were pooled separately into years of high and low productivity (that is, years with rainfall above or below average). Isodar equations were estimated using Model II multiple regressions (Legendre 2000) because density estimates in 2 different (although adjacent) habitats cannot be considered as independent and dependent variables (Morris 1987). To determine whether the slopes were >1, <1, or =1, and whether intercepts were >0 or =0, 95% confidence intervals (95% CIs) around the slopes and intercepts were calculated. These analyses were done using Model II regression software (Université de Montréal, Montreal, Quebec, Canada).

Results

Patterns of species composition and habitat use.—In total, 5,606 rodent individuals were captured 9,886 times. Most rodent species were present in all seasons. However, R. rattus was recorded only once in 32 trapping sessions, G. nanus during 3 sessions, and E. melanurus during 10 sessions. Analysis of densities of rodent species in different habitat types demonstrated that 4 rodent species were habitat specialists. G.gerbillus occurred in sand, S. calurus in rock, and E. melanurus and M. musculus in loess. Three species, D. dasyurus, G. henleyi, and M. crassus, were habitat generalists; D. dasyurus avoided only sand, whereas the other 2 species avoided rock. Other species occupied 2 or 3 habitat types, demonstrating clear habitat preferences. J. jaculus preferred hammada, A. dimidiatus favored rock, A. russatus preferred rock and wadi, and P. obesus had a preference for loess. G. nanus and R. rattus were trapped infrequently, and, therefore, we could not determine their habitat preferences (Fig. 2; Table 1).

Fig. 2

Density (individuals per hectare) for 1) winter and 2) summer of 9 rodent species in the central Negev Desert, Israel, in various habitats (rock, wadi, loess, sand, and hammada). Densities were estimated each winter and summer from winter 1993 to summer 2008. Note differences in scales of y-axes.

Stepwise multiple regression analyses demonstrated that in 12 of 17 cases analyzed for species-habitat combinations, summer density (Table 2) was determined mainly by total rainfall during the previous winter season. In most cases these relations were positive, but they were negative for D. dasyurus and M. crassus in wadi. Species density of the previous season was the most important factor in 3 cases (A. dimidiatus in rock, A. russatus in wadi, and G. gerbillus in sand) and the 2nd most important in 3 other cases (G. henleyi and M. crassus in wadi and P. obesus in loess). However, current density in other habitats was the most important factor determining summer density in 5 cases (D. dasyurus in rock and wadi and G. henleyi in 3 habitats) and the 2nd most important in 3 other cases (A. russatus in 2 habitats and M. erassus in sand).

View this table:
Table 2

Stepwise multiple regression analysis with summer density as the dependent variable for rodents in the Negev Highlands, Israel. Asterisks indicate significance level: * P < 0.05, ** P < 0.01, *** P < 0.001.

Standardized regression coefficients
SpeciesHabitatPrevious density 0.5 year beforeRainfall 0.5 year beforeDensity in other habitatsOther habitatFd.f.Adjusted R2
Acomys dimidiatusRock0.64785.7361,140.283*
A. russatusWadi0.57270.4111Rock17.2282,130.697***
Rock0.82340.5703Wadi36.9692,130.809***
Dipodillus dasyurusWadi−0.67461.1571Rock10.4232,130.574**
Rock0.44160.3708Wadi
0.3735Loess29.0883,120.857***
Loess0.69958.5561,140.386*
Gerbillus gerbillusSand0.64597.1791,140.452**
G henleyiSand0.43270.6029Hammada36.5902,130.826***
Hammada0.8504Sand36.5631,140.703***
Wadi0.44270.7958Hammada13.8932,130.632***
Meriones crassusSand0.65790.5448Wadi19.9322,130.716***
Hammada0.555211.6371.140.385**
Wadi0.4191−0.58268.2612,130.659**
Loess0.55895.1091,140.354*
Sekeetamys calurusRock0.623841.4661,140.861***
Psammomys obesusLoess0.52550.742735.4472,130.821***
Mus musculusLoess0.586328.6551,140.633***

Winter density (Table 3) in 12 cases was determined mainly by the density of the previous season. Total rainfall during the previous winter season (1 year before) was the most important factor in 2 cases (G. henleyi in wadi and M. crassus in sand) and the 2nd most important in 2 other cases (A. dimidiatus and G. gerbillus). Density of the current season in other habitats was the most important factor only for G. henleyi in sand and hammada and the 2nd most important for A. russatus in rock. Densities of potential competitors were not significant factors in any season.

View this table:
Table 3

Stepwise multiple regression analysis with winter density as the dependent variable for rodents in the Negev Highlands, Israel. Asterisks indicate significance level: * P < 0.05, ** P < 0.01, *** P < 0.001.

Standardized regression coefficients
SpeciesHabitatPrevious density 0.5 year beforeRainfall 1 year beforeDensity in other habitatsOther habitatFd.f.Adjusted R2
Acomys dimidiatusRock0.57560.393466.3982,120.903***
A. russatusWadi0.479111.8231,130.404*
Rock0.48000.4796Wadi26.0262,120.758***
Dipodillus dasyurusWadi0.508514.9991,130.483**
Rock0.520616.1891,130.539**
Loess0.832629.3861,130.670***
Gerbillus gerbillusSand0.48250.44187.5262,120.482**
G. henleyiSand0.7805Hammada20.2611,130.579***
Hammada0.6029Sand
0.4038Wadi18.9192,120.705***
Wadi0.635116.9661,130.517**
Meriones crassusSand0.917068.6671,130.829***
Hammada0.500811.9041,130.421**
Wadi
Loess0.820426.7571,130.648***
Sekeetamys calurusRock0.78977.8291,130.477**
Psammomys obesusLoess0.537013.3101,130.638***
Mus musculusLoess0.561821.5731,130.733***

Significant isodars were found for different species–habitat combinations at various productivities (Table 4). In all cases isodar intercepts were significantly positive, indicating quantitative habitat differences. Isodar slope did not differ significantly from 1 for wadi–loess isodars for D. dasyurus or for sand–hammada at low productivity levels and hammada-wadi isodars for G. henleyi, indicating similar habitat quality. Isodar slopes were significantly > 1 for wadi–rock and loess– rock isodars for D. dasyurus and for sand-hammada isodar of M. crassus, indicating that quantitatively better habitat also was qualitatively better. Slopes were significantly <1 for wadi-rock isodar for A. russatus, sand-hammada at high productivity level and sand–wadi isodars for G. henleyi, and hammada–wadi isodar for M. crassus, indicating that quantitatively better habitat was qualitatively poorer.

View this table:
Table 4

Estimation of isodar parameters for pairs of habitats for years of low and high productivity for rodents in Negev Highlands, Israel. The slope of an isodar indicates qualitative habitat differences, and the intercept of the isodar indicates quantitative differences in resource availability of 2 habitats.

SpeciesHabitatsProductivitySlope ± 95% CIIntercept ± 95% CIR2
Acomys russatusWadi–rockLow0.403 ± 0.2170.454 ± 0.2450.44
Dipodillus dasyurusWadi–rockLow2.339 ± 1.2043.484 ± 1.6080.23
Loess–rockLow3.399 ± 2.9495.465 ± 3.9400.24
Loess–rockHigh3.903 ± 1.5296.144 ± 3.9520.68
Loess–wadiHigh1.130 ± 0.9496.251 ± 7.6440.29
Gerbillus henleyiSand–hammadaLow0.869 ± 0.2880.724 ± 0.2760.69
Sand–hammadaHigh0.537 ± 0.2861.065 ± 0.6270.57
Sand–wadiHigh0.347 ± 0.3550.916 ± 0.7760.23
Wadi–hammadaHigh0.903 ± 0.3800.586 ± 0.5690.66
Meriones crassusSand–hammadaHigh1.418 ± 0.6491.738 ± 0.5790.65
Wadi–hammadaHigh0.669 ± 0.4470.834 ± 0.3990.45

Discussion

Changes in desert rodent density related to rainfall fluctuations have been observed in numerous studies (Beatley 1969; Dickman et al. 1999; Masters 1993; Meserve et al. 1995; Newsome and Corbet 1975; Predavec 1994; Southgate and Masters 1996; Whitford 1976). A simple qualitative model for water resource regulation of desert rodent populations assumes that precipitation leads to germination, growth, and reproduction of plants, and the resulting increases in food supply in the form of seeds, fruits, and leaves lead to increases in rodent populations (Brown and Ernest 2002). The 1st hypothesis we tested was that for each species at least 1 (optimal) habitat should exist where its summer (postbreeding) density is determined mainly by rainfall with 0.5-year lag and where its winter (prebreeding) density is determined mainly by its own previous (summer) densities, due to the absence of rainfall events and reproduction between summer and winter rodent censuses. This hypothesis was supported for 6 of 9 tested species in summer and for 8 of 9 tested species in winter.

All 3 cases for which our hypothesis concerning summer density regulation was not supported were related to a violation of our assumption of 0.5-year lag in rodent density response to rain events. In these cases positive correlation of population density with precipitation was found but with a 1-year lag. In 2 of these 3 cases, however, precipitation was not the main but rather the 2nd-order factor in density regulation. The rodent density response to rain events is characterized by a time lag that varies from 2 to 15 (usually 4–6) months (Dickman et al. 1999; Southgate and Masters 1996). In the case of rodent density and annual vegetation abundance, response time varies from 0 to 12 months (Brown and Heske 1990).

We found that rodent densities in most cases were positively correlated with total precipitation during the previous rainy season. Such a positive correlation was demonstrated by all species in loess, rock, and sand. In loess, time lag was 0.5 year in all cases. In rock, 3 species demonstrated a 0.5-year time lag and 1 species a 1-year time lag, whereas in sand 1 species demonstrated a 0.5-year time lag and 2 species a 1-year time lag. However, a positive correlation of density with rainfall in hammada habitat was found in 1 of 2 cases (with time lag 0.5 year) and in only 1 of 3 cases in wadi (with time lag 1 year). Thus, a positive correlation of density with rainfall was found in all but 3 of 17 species–habitat combinations. The 0.5-year time lag was found in 10 species–habitat combinations and the 1-year lag only in 3 combinations; in 1 case (M. crassus in sand) significant positive correlation between population density and precipitation was found with both 0.5- and 1-year lags.

The reasons for delayed (1-year) lag in rodent response to precipitation might be species-specific or habitat-specific or both. Rainfall acts indirectly through plant production and rodent breeding in affecting rodent density. This chain of reasons and consequences can be broken or distorted at any link. The influence of rainfall on plant production can be habitat-dependent (Kadmon 1993). Furthermore, desert rodent reproduction in some cases is not induced by the appearance of fresh green vegetation (White and Bernard 1999; White et al. 1997). This is not the case for most of the species in our study, in which the beginning of reproduction coincides with the 1st heavy rains (Krasnov et al. 1996a; Shenbrot et al. 1994, 1997, 1999). However, A. dimidiatus is an exception. We observed that this species begins breeding in midsummer, which could explain its delayed response to precipitation. The 2 other cases of the delayed response to precipitation are strongly granivorous species (G. gerbillus and G. henleyi) in sand. Intensity of reproduction of these species could be determined by the abundance of food resources (seed soil bank), which is dependent on precipitation during the previous rainy season (1 year before). Summer (postbreeding) density of these species should be dependent on precipitation with a 1.5-year lag, but this dependence can be masked by action of other factors. If winter (prebreeding) density of these species is determined mainly by survival of individuals born during the previous breeding season, and if this survival is mediated by food resource abundance (seed soil bank) that is dependent on precipitation during the previous rainy season, it can explain the observed 1-year lag.

The relative importance of intrinsic and extrinsic factors in rodent population regulation has been the subject of continuing debate (Krebs 1996; Wolff 1997). Most cases of presumable intrinsic regulation concern temperate populations of microtines (Krebs 1996; Tamarin 1985). For desert rodents this type of regulation has never been considered as predominant but rather as acting only in combination with extrinsic factors. Theoretical consideration of behavioral prerequisites necessary for exclusively intrinsic type of population regulation occur only under a limited set of conditions, and, therefore, most mammal populations are probably controlled mainly by extrinsic factors (Wolff 1997). Theoretically, a strong correlation between external influence (for example, precipitation) and population density could occur only if the population is regulated strongly near an equilibrium level by density-dependent factors (Royama 1992). Density fluctuations in some South American semiarid rodents and in Chihuahuan desert Dipodomys are the result of complex interactions between population-level (direct density-dependent), community-level (delayed density-dependent), and exogenous (rainfall, associated primary production, competition, and predation) processes (Lima and Jaksic 1998, 1999; Lima et al. 2002, 2006, 2008).

According to our hypothesis we found autocorrelation of winter density with a time lag of 0.5 year in 8 of 9 tested species. For 6 species density in the previous summer was the only factor determining winter density, whereas for 2 other species it acted together with precipitation. Density autocorrelation with 0.5-year lag also was recorded in summer in 6 of 9 species. However, in only 2 species density in the previous winter was the only factor determining their summer density, whereas in other cases density autocorrelation acted not as an independent agent of regulation but together with rainfall or densities in other habitats.

The 2nd hypothesis we tested was that for species occurring in several habitats densities in other than optimal habitats are determined mainly by processes of dispersal and density-dependent habitat selection. If valid, this should have resulted in higher correlation of density in a given habitat with current densities in other habitats than with rainfall or with previous density in the same habitat. This hypothesis was supported only partially by our data. We found a significant correlation of density in a given habitat with current densities in other habitats in 9 of 12 observed species-habitat combinations, but this correlation was higher than the correlation with rainfall or with previous density in the same habitat in only 5 combinations.

All cases with an absent or negative relationship between rainfall and rodent density were recorded in wadi, and in all of these cases density of the same species in other habitats was among the significant factors driving population fluctuations. Possible reasons for this phenomenon can be episodes of population crash due to flash winter flooding and immigration from other habitats. Extreme flooding events can cause catastrophic, species-specific mortality in desert rodents (Thibault and Brown 2008). However, influences of less harsh but more regular disturbance events have not been analyzed. Our results demonstrated that population dynamics after such episodes were determined by interactions with populations from adjacent habitats via processes of density-dependent habitat selection.

Postreproductive density of a species would be related to precipitation only if reproductive efficiency (e.g., fitness) is determined directly by per capita resource abundance. However, according to the basic statements of the modern theory of habitat selection, fitness within a species is density-dependent, and this is habitat-specific (Rosenzweig 1992). Using an isodar (comparing density estimates in a 2-habitat system across time periods) or paraisodar (comparing density estimates in a 2-time period system across a habitat gradient) approach (Morris 1988, 1990; Shenbrot and Krasnov 2000), 1 of the main assumptions determining the properties of population dynamics models is the nature of habitat differences. If habitats differ only quantitatively, density changes will be correlated across habitats and will be proportional to changes in per capita resource abundance. This, in turn, should be correlated with total rainfall. However, if habitats differ in quality, density changes will remain correlated for each pair of adjacent habitats but will be proportional to changes in per capita resource abundance only for the best-quality habitat (or the best-quality part of a habitat gradient). As a result, density changes in a habitat of relatively low quality will be correlated more strongly with density changes in a high-quality habitat than with changes in resource abundance.

Significant isodars indicating processes of density-dependent habitat selection were recorded for all species occurring in more than 1 habitat. However, only 9 of 11 isodars described habitat selection between adjacent habitats. Three isodars indicated processes of habitat selection between habitats that have no direct contact, such as wadi and loess or wadi and sand. These results support our earlier suggestion (Shenbrot et al. 2006) that these processes in desert landscapes occur by a stepwise flow of individuals across a network of adjacent habitat patches in the whole landscape rather than by direct migration between the 2 sites.

In all cases isodar intercepts were significantly positive, which indicates quantitative habitat differences. This is in agreement with our previous data (Shenbrot 2004; Shenbrot et al. 2006), which demonstrated existence of significant differences among habitats in abundance of food resources.

Only 3 of 11 isodars had slopes not significantly different from 1, which indicated similar habitat quality. At least in 1 of these 3 cases (wadi and loess for D. dasyurus) a similar habitat quality was demonstrated experimentally (Shenbrot et al. 2006). In all other cases isodar analysis indicated significant differences in habitat quality. Differences in habitat quality can be a result of conditions for burrowing due to habitat-dependent differences in mechanical structure of soils (Shenbrot et al. 2002), to habitat-specific distribution of competitors (Shenbrot and Krasnov 2002), or to other factors. Existence of such differences can produce a stronger dependence of population level on density changes in other habitats than on changes in rainfall.

Recurring disturbances in wadi lead to a situation described by the “source-sink” concept (Pulliam 1988; Pulliam and Danielson 1991). In this case, for the short time period after the disturbance, density in a source habitat is correlated with resource abundance, whereas density in a sink habitat is correlated with density in the source habitat but not with resource abundance. Moreover, resource-independent fluctuations of density due to disturbance events can create a condition for activation of the processes of density-dependent habitat selection. This statement is supported by the observation that all species involved in the processes of density-dependent habitat selection (A. russatus, D. dasyurus, G. henleyi, and M. crassus) occurred in wadi.

Interspecific relationships can modify the response of a species to environmental variation. Competition holds a central place in ecological and evolutionary theory, particularly in community ecology (Southwood 1987). A significant role of competition in the structuring of desert rodent communities was demonstrated in a few relatively well-studied cases (Brown and Munger 1985; Heske et al. 1994; Meserve et al. 1996; Ziv et al. 1993). However, in our rodent community interspecific competition is not a widespread phenomenon. Negative interactions were recorded in a few pairs of species and in some seasons only (Shenbrot and Krasnov 2002). Nevertheless, interspecific interactions were not found to be important factors of population dynamics, even for species that are known to be involved in such interactions based on observational data and field experiments. It seems that interspecific interactions regulate individual distribution at the microhabitat scale but are not able to drive population dynamics at the macrohabitat scale. In general, we conclude that population dynamics of desert rodents are determined by the complex interactions of extrinsic (rainfall) and intrinsic mechanisms and are modified by the processes of density-dependent habitat selection.

Acknowledgments

We thank A. Degen (Ben-Gurion University of the Negev) and 2 anonymous reviewers for helpful comments. Financial support was provided by Israel Ministry of Science, Local Council of Mizpe Ramon, by seed money grants from Ben-Gurion University of the Negev, and by Israel Science Foundation grants 414/02-1 and 297/07. This is publication 677 of the Mitrani Department of Desert Ecology, Ben-Gurion University of the Negev.

Footnotes

  • Special Feature Editor was Barbara H. Blake.

Literature Cited

View Abstract