Abstract
Populations of small mammals often fluctuate dramatically in arid environments, persisting at very low density during drought and erupting briefly but dramatically after rain. We selected 3 species that exhibit such boom and bust dynamics in spinifex grassland in the Simpson Desert, central Australia, and asked how they survive prolonged dry periods when they are scarce or absent from the trapping record. We postulated that animals persist by retreating to small patches of open woodland that are embedded within the grassland matrix and predicted that these patches would provide more food and shelter for small mammals than the surrounding grassland during dry periods but not at other times. We also predicted that capture rates and activity of the study species would be higher, and risk of predation lower, in the woodland than in the grassland during drought, and that after heavy rain the rate of return of small mammals to spinifex grassland would be correlated with proximity to patches of woodland. Sampling provided partial support for these predictions. Some food resources (seeds) were more abundant in woodland patches irrespective of environmental conditions, but others (invertebrates) showed no clear pattern; shelter resources were mostly invariant between habitats and times. Of the 3 study species, only the sandy inland mouse (Pseudomys hermannsburgensis) behaved largely as we had anticipated, but even this species was not confined to woodland during droughts. Both the spinifex hopping-mouse (Notomys alexis) and brush-tailed mulgara (Dasycercus blythi) continued to use grassland more than woodland during drought and nondrought periods, although N. alexis showed a tendency to increase its activity in woodland during dry conditions. Trappability remained relatively constant among species, habitats, and times, indicating that the temporal patterns of habitat use we uncovered were real and not artifacts of changes in animal behavior or susceptibility to capture. We conclude that woodland patches contribute importantly to the persistence of P. hermannsburgensis during droughts, whereas N. alexis and D. blythi either use other unidentified refugia at such times or occur at such low densities in grassland habitat that their chances of being captured are very small.
Resumen
Es común que las poblaciones de mamíferos pequeños en ambientes áridos fluctúen dramáticamente, persistiendo en muy bajas densidades durante la sequía, y creciendo breve pero dramáticamente después de las lluvias. Seleccionamos 3 especies que exhiben dicha dinámica “auge y quiebra” en praderas de spinifex en “Simpson Desert,” Australia central, e investigamos como sobreviven periodos prolongados de sequía cuando se muestran escasos o ausentes del record de captura. Postulamos que estas especies persisten retirándose a pequeños parches de bosque que se encuentran embebidos dentro de la matriz de pradera, y predijimos que estos parches proveen de mas comida y resguardo, para estas especies, que la pradera circundante, durante los periodos de sequía pero no durante los periodos sin sequía. También predijimos que en el bosque, durante la sequía, la taza de captura y actividad de las especies estudiadas será mayor, y el riesgo de depredación menor, y que después de fuertes lluvias, la tasa de retorno de mamíferos pequeños hacia las praderas de spinifex estará correlacionada con la proximidad a los parches de bosque. El muestreo apoyó parcialment estas predicciones. Algunas fuentes de alimento (semillas), fueron más abundantes en los parches de bosque independientemente de las condiciones ambientales, pero otras (invertebrados) no mostraron patrones claros; fuentes de resguardo fueron en su mayoría sin variantes entre hábitats y periodos. De las 3 especies estudiadas, solamente Pseudomys hermannsburgensis mostró el comportamiento que anticipábamos, pero incluso esta especie no estuvo confinada a los bosques durante las sequías. Ambos, Notomys alexis y Dasycercus blythi continuaron usando las praderas en mayor medida que los bosques durante los periodos de sequía y sin sequía, aunque N. alexis mostró un aumento relativo en actividad en el bosque durante las condiciones de sequía. La capturabilidad permaneció relativamente constante entre especies, híbitats, y periodos, indicando que los patrones temporales en uso de hábitat que encontramos fueron reales y no un artefacto de cambios en el comportamiento de los animales o susceptibilidad a ser capturado. Concluimos que los parches de bosque contribuyen de manera importante a la permanencia de P. hermannsburgensis durante la sequía, mientras que N. alexis y D. blythi pueden estar utilizando refugios no identificados durante estos periodos, o se encuentran en densidades tan bajas en el habitad de pradera que la probabilidad de ser capturados es extremadamente pequeña.
Resources in many arid environments are patchy in space and time, and mammal populations that exploit them often show correspondingly marked variations in size. Large desert-dwelling mammals like the giraffe (Giraffa camelopardalis), for example, make extensive movements along riparian corridors while following changes in food resources (Fennessy 2009; Le Pendu and Ciofolo 1999). Others, such as the blue wildebeest (Connochaetes taurinus), will move short distances (<25 km) in the direction of localized thunderstorms to access the fresh pastures that result but also might exhibit large-scale directional migrations (>100 km) to escape drought-stricken areas (Talbot and Talbot 1963). Site-based population estimates of such mobile species thus vary dramatically depending on conditions, especially the availability of food, in both the local and regional environment (Ogutu and Owen-Smith 2003).
Small desert-dwelling mammals, by contrast, are more sedentary than their larger counterparts and usually have to cope with local shortfalls in food and other resources in situ (Randall 1993). Some evidence exists for long-range movements in Australian desert rodents (Finlayson 1939; Letnic 2002) and marsupials (Haythornthwaite and Dickman 2006b), but even the longest forays seldom exceed 10 km (Dickman et al. 1995). Many species instead use physiological mechanisms such as torpor to conserve energy and water (Boyer and Barnes 1999; Geiser 2004) and buffer themselves from climatic extremes by using deep burrows and emerging when conditions are most benign (Degen 1997; Withers et al. 2004). Populations of some desert-dwelling small mammals appear to be regulated both by the climate-resource axis and by endogenous factors such as competition (Ernest et al. 2000; Lima et al. 2008; Ojeda et al. 2011) and exhibit long-term stability or regular multiannual cycles (Brown and Zeng 1989; Zeng and Brown 1987). However, others appear to be dominated primarily by the effects of climate on food or shelter resources, or both, and fluctuate between dramatic population booms and busts (Fox 2011; Letnic and Dickman 2010; Letnic et al. 2011). Such dynamics are exhibited by both prolific, short-lived rodents and more K-selected species (Meserve et al. 1995; Previtali et al. 2009), with bust periods of low numbers, or even no animals, lasting for months or years (Predavec and Dickman 1994). How populations persist for long periods at very low density is not known. However, with global climate change likely to alter rainfall and resource distribution patterns in many arid regions, we have an increasing imperative to understand how resources influence spatial and temporal variation in the population sizes of small desert organisms (Meserve et al. 2011; Morton et al. 2011; Previtali et al. 2010; Thibault et al. 2010).
As resources shrink small mammals respond in several ways. If the resource shortage is perceived to be short term, animals might conserve energy by temporarily reducing their activity. If the resource shortage is prolonged, animals might slow or halt reproductive activity and begin to exploit alternative resources. They also are likely to take more risks in obtaining those resources; for example, venturing further or for longer periods into places where the risk of predation is relatively high (Brown 1988). Movements likely will increase to facilitate resource acquisition, and populations might undergo changes in social organization (Dickman et al. 2010). If resources decline further, animals can become restricted to localized “hot spots” in the landscape, where predator refuges and resources remain available, and persist there until the surrounding conditions again become favorable (Harrison 1991; Milstead et al. 2007; Morton 1990). In field studies it can be difficult to distinguish declines in activity (i.e., when trappability is reduced) from declines in population size, because both result in reduced captures. Also, it is seldom clear whether declines are driven by scarcity of food or other resources such as shelter (Letnic et al. 2004, 2005), although food is usually assumed to be the key resource.
In this paper we investigate temporal shifts in the habitat use of 3 species of small mammals in the northeastern Simpson Desert, central Australia, and also describe changes in food and shelter resources that might influence these shifts. Since 1990 the 3 study species have disappeared from regularly sampled trapping grids in grassland habitat for periods up to 4 years, especially during dry conditions, but have returned consistently within 3–6 months of heavy summer rainfall (Dickman et al. 1999b, 2001, 2010). The grasslands have been the focus of study, because they cover >70% of this desert region (Greenville et al. 2009). Although small mammals can respond to ephemeral patches of resources that generate after local rainfall (Newsome and Corbett 1975), the reliability of the 3 species' reappearance after summer rains in different sites across the Simpson Desert (Haythornthwaite and Dickman 2006a) suggests that they are moving short distances to the trapping grids from local refugia. Inspection of vegetation maps and Landsat imagery (Greenville et al. 2009; Purdie 1984) suggests further that these refugia could be small patches of open woodland that lie close to most of the trapping grids. These woodlands provide habitat and food resources for many species of lizards, birds, and bats (Frank 2010; Shephard 1999; Williams and Dickman 2004) and also could be important at times for small terrestrial mammals. The most extensively distributed type of woodland in the northeastern Simpson Desert is dominated by gidgee (Acacia georginae). This species is considered to be a poisonous or problem species on cattle properties (Alexander 2006) and is sometimes removed by managers; hence it is important to clarify its potential importance as a drought refuge for native species.
Based on these observations and the putative drought refuge function of gidgee patches, our main objective was to describe the habitats used by the 3 study species of small mammals and to test 6 hypotheses derived from the drought refuge model. We expected that i) food or shelter resources, or both, would be greater in gidgee woodland than in spinifex grassland habitats during drought but not during nondrought times; ii) trappability of small mammals would be constant between habitats and times; iii) capture rates of small mammals would be higher in gidgee woodland than in spinifex grassland habitats during drought but not during nondrought times; iv) activity of small mammals would be higher in gidgee woodland than in spinifex grassland habitats during drought but not during nondrought times; v) risk of predation on small mammals would be lower in gidgee woodland than in spinifex grassland habitats during drought but not during nondrought times; and vi) after drought has broken, rates of return of small mammals to spinifex grassland habitat would be correlated with their proximity to patches of gidgee woodland.
In testing these hypotheses we drew from a 20-year field study that covered extended dry periods punctuated by 3 very heavy summer rainfall events (Dickman et al. 2010). We used data from the year following each deluge to characterize nondrought periods and data from before or > 1 year after to characterize droughts. The term drought is used here in an operational sense only to connote times outside the non-drought periods when rainfall is >50% less than the long-term average. We introduce below the study species and the resources that they are likely to require, having used a combination of resource sampling methods, livetrapping, tracking, and giving-up-density experiments to test our hypotheses.
Materials and Methods
Study area.—Research was carried out mostly on Ethabuka station (now Ethabuka Reserve) in the northeastern Simpson Desert, western Queensland, Australia (23°46′S, 138°28′E), with limited additional sampling on Carlo station and Cravens Peak Reserve immediately to the north. All areas are dominated by long, parallel sand dunes that lie 0.6–1.0 km apart and rise to about 8–10 m high (Purdie 1984). Grassland, dominated by needle-leaved hard spinifex (Triodia base-dowii), covers the floors of the dune valleys to just below the dune crests. Patches of gidgee woodland covering <0.5 ha to >10 ha occur on low-lying clay soils in the dune valleys and cover some 7–16% of the study region (Frank 2010; Greenville et al. 2009). Chenopod shrubs such as Enchylaena tomentosa, Salsola kali, Atriplex spp., and Sclerolaena spp. grow under the open gidgee canopy, but spinifex is usually absent. Other shrubs such as Acacia ligulata, A. dictyophleba, Eucalyptus pachyphylla, E. gamophylla, Grevillea stenobo-trya, and G. juncifolia occur sporadically throughout the study area, and annual grasses, forbs, and herbs grow abundantly after rain.
The Simpson is classified as a hot desert (Williams and Calaby 1985). Daily temperature maxima in summer are 46–49°C, and minima in winter are −6°C (Purdie 1984). The regional climate is dominated by the El Niño Southern Oscillation and is highly irregular (Letnic and Dickman 2006). The study area lies along a north-south rainfall gradient, with the southern parts on Ethabuka receiving about 199 mm a year (n = 94 years, recorded at Marion Downs, 120 km distant) and the northern parts on Cravens Peak eceiving 262 mm (n = 123 years, recorded at Boulia, 135 km distant). Most rain falls over the austral summer between November and February. During this study 3 periods of above-average rain were experienced in the Simpson Desert in the summers of 1990–1991 (439 mm), 2000–2001 375 mm), and 2006–2007 (260 mm), with each resulting in a subsequent eruption of the 3 study species (Dickman et al. 2010). Summer rainfall in other years from 1990 to 2009 averaged 101 mm (range = 37–216 mm, data from Marion Downs until 1994, and from an on-site automatic weather station [Environdata, Warwick, Queensland, Australia] on Ethabuka from 1995). Wildfires can occur 1–2 years after heavy summer rains when vegetation has dried off and have a mean minimum return interval of 26 years in the study region Greenville et al. 2009). Sites used in the present study were ast burned some time prior to 1972.
Study species.—Of the 15 species of small mammals that have been recorded in the study area, 3 species were selected or study: the sandy inland mouse (Pseudomys hermannsbur-ensis, —12 g) and spinifex hopping-mouse (Notomys alexis, ∼30 g), both rodents, and the brush-tailed mulgara (Dasy-ercus blythi, ∼90 g), a dasyurid marsupial. The remaining pecies were captured either on all or almost all sampling ocasions and thus appeared either not to use drought refugia e.g., lesser hairy-footed dunnart [Sminthopsis youngsoni]) or to use such refugia too sporadically to be able to uncover patterns in their habitat use (Dickman et al. 1999b, 2001; Haythornthwaite and Dickman 2006a).
The 2 species of rodents are omnivorous, displaying some preference for invertebrates in captivity but also eating seeds and green plant material as part of their natural diet (Murray and Dickman 1994a, 1994b). Breeding can occur at any time of year, with populations erupting within 6 months of heavy rain and then collapsing ≤6–9 months later (Breed 1992; Dickman et al. 2010). Torpor has not been clearly demonstrated in any species of Australian arid zone rodent, although Predavec (1997b) described a condition resembling torpor in captive P. hermannsburgensis. The mulgara hunts invertebrates and small vertebrates, apparently taking prey items largely in relation to their availability (Chen et al. 1998; Fisher and Dickman 1993b). This species is more sedentary than the rodents and breeds only once a year, in spring (Masters 2003). Although D. blythi, like other dasyurid marsupials, is not considered to be an eruptive species, populations peak 7–9 months after heavy summer rains and decline less than a year later (Dickman et al. 2001). Animals can enter torpor and do so most readily in winter if invertebrates predominate in the diet (Pavey et al. 2009).
Livetrapping.—Pilot trials suggested that small mammals could be captured most readily in pitfall traps. These were constructed from polyvinyl chloride pipe (16 cm in diameter, 60 cm deep) and buried vertically flush with the soil surface. Trap success was increased by erecting a drift fence of aluminum wire screening (30 cm high, 5 m long) over the top of each trap to intercept and guide surface-active animals toward the trap opening (Friend et al. 1989). To prevent captured animals from digging their way out we covered the bottom ends of the pitfall traps with a floor of wire screening. Pitfall traps were capped with metal lids when not in use.
Pitfall traps were set in grid formation. Those in spinifex grassland covered 1 ha, with each grid comprising 6 lines of 6 traps spaced 20 m apart. To ensure that the topographic range of the dunefield system was sampled we set the top line of traps on the dune crest and the bottom line 100 m distant in the dune valley. Traps in gidgee woodland were set in 3 lines of 6 and were spaced 20 m apart, with the long axis of these grids running parallel to the sand dunes. This configuration allowed the traps to be contained entirely within the woodland patches and also provided a woodland buffer ≥ 10 m wide to reduce any edge effects. Five grids were established in grassland, with each spaced 0.5–2.0 km from its nearest neighbor, and 8 grids in woodland. The latter were set singly or in pairs flanking the grassland grids at distances of 50–500 m from them; we considered this to be close enough to allow movements to take place between the habitat types. These 13 grids were used to test our first 5 hypotheses. As part of other studies we had established more than 60 additional grids in spinifex grassland over some 5,600 km2 of the Simpson Desert (Haythornthwaite and Dickman 2000, 2006a); we used capture data from some of these to test our hypothesis vi. The grassland and adjacent woodland grids were opened simultaneously for 3 nights on 6 occasions between April 2008 and December 2009 as conditions were drying out, with traps checked in the mornings and sometimes also the afternoons. Captured animals were identified, weighed, inspected for reproductive condition, marked uniquely by ear-notching, and then released under vegetative cover near their point of capture. Captured lizards and frogs also were processed and released (Dickman et al. 1999a; Greenville and Dickman 2009; Letnic and Dickman 2005; Predavec and Dickman 1993). Livetrapping allowed us to quantify habitat use and movements of the 3 study species and to obtain individuals for tracking and assessment of activity. The grassland grids over the broader study region were established at various times since 1990 and trapped 2–6 times a year using the same protocols as above.
Resource sampling.—We sampled seeds and surface-active invertebrates to gain an indication of food resources for the small mammals and vegetation and litter cover to index the available shelter. Invertebrates were sampled using plastic vials (4 cm in diameter, 10 cm deep) filled to the three-quarter level with 3% formalin solution. Twelve vials were positioned on 4 grids in each of the 2 habitats on 2 occasions during drought (March and June 1990) and on 2 occasions after heavy summer rainfall (April and July 1991). All vials were set for 3 nights per sampling period and then removed. Seeds were collected in samples of sand using a circular metal quadrat (105 mm diameter) to a depth of 2 cm, the depth to which the study rodents commonly dig for food (Predavec 1994). Six samples were taken from 4 grids in each of the 2 habitats on 2 occasions during drought (March and June 1993) and on the same 2 occasions after summer rainfall, as for the invertebrates. All samples were stored and returned to the laboratory for sorting.
Shelter resources were assessed visually by 1 observer by scoring the percentage cover of dead spinifex, live spinifex, dicotyledonous plants < 1 m high and ≥1m high, leaf litter, and woody debris within circular quadrats of radius 2.5 m. Assessments were made at 6 locations on 4 grids in each of the 2 habitats during 2 periods of drought (June 1990 and June 1999) and 2 corresponding periods after heavy rain (July 1991 and August 2001).
Trappability.—To evaluate trappability we compared the numbers of animals captured in traps with 2 independent indexes of their presence: tracks and burrows. In the 1st instance, we selected pitfall traps randomly in spinifex grassland and gidgee woodland and established transects (5 m long, 1 m wide) of cleared, smoothed sand within a 5-m radius of each trap. For 3 successive mornings we checked the traps and inspected the associated transects for animal tracks. The tracks of P. hermannsburgensis could be distinguished by their small size (pes < 20 mm) and imprint of 5 hind toes and those of N. alexis by their length (>30 mm), imprint of the heel, ridge across the hind foot, and the impression of usually just 3 toes (Triggs 1996; K. Moseby, University of Adelaide, pers. comm.). Although the desert mouse (Pseudomys desertor) and introduced house mouse (Mus musculus) co-occur in the study area and can make similar tracks, we consider bias from these species to be unlikely because of their rarity throughout the course of the study. The tracks of D. blythi are larger and deeper than those of co-occurring species because of their large size. Footprint transects were run during 3 sampling sessions after heavy summer rain in 1991 and twice during drought in 1993.
We also compared captures of animals in traps with counts of their burrows. During summer P. hermannsburgensis excavates shallow (30 cm) sloping burrows near the base of shrubs or hummocks of spinifex, whereas N. alexis excavates tunnels and nest chambers to 1 m deep, with shafts that ascend vertically to open areas on the surface (Breed and Ford 2007; Finlayson 1941). The burrow systems of D. blythi have multiple entrances that are usually clustered within 1–2.5 m around the raised sandy base of spinifex hummocks (Woolley 1990); at 4- to 5-cm diameter at the entrance they are twice as wide as the burrows of sympatric small rodents. Burrows were counted along 6 transects (100 m long, 4 m wide) on each of 6 grassland trapping grids on 2 sampling sessions in 1991 (nondrought) and on 2 sessions in 1994–1995 (drought); no woodland grids were sampled on these occasions. Only active burrows were counted; activity, and the identities of burrow occupants, were confirmed by checking footprints on smoothed sand at burrow entrances.
Tracking.—We used 3 methods to assess animal activity. Fluorescent pigment tracking and spool and line tracking (Dickman 1988; Miles et al. 1981) were used to identify short-range directions of animal movement following their capture on grassland grids, and radiotracking was used to determine the relative amount of activity in the 2 habitats over periods of several days. We selected individuals at random for fluorescent pigment tracking, applied a different color pigment (Radglo Radiant Colors; Radiant Color N.V., Houthalen, Belgium) to different individuals tracked on the same grid, and released them in the evening near the point of capture. Animals were dusted with pigment on the dorsal, ventral, and flank fur but not on the head. The next night we followed the animals' paths using a handheld black light (Eveready Pty., Perth, Australia). Tracking spools of 2-ply thread (Coats Australia Pty., Sydney, Australia) were attached to the hair of animals' napes using cyanoacrylate glue, with the end of the spool unwinding from the rear. Spools weighed 1.0–5.2 g, ≤7% of the mass of any animal to which they were attached. Animals were checked to ensure that spools were fitted correctly and that the thread unwound from inside the bobbin with minimal impedance, and then they were released after dusk near their capture point. The free end of the thread was tied to a stake, and animal paths were followed the next day. We mapped both the fluorescent pigment and spool-thread paths using measuring tape and compass, discounting the first 10 m of paths as a flight distance (Cox et al. 2000).
For radiotracking we used simple tags with single-stage circuits (i.e., with no amplifier) with 10-cm trailing whip antennas that were attached to animals using plastic cable tie collars. Tags varied in mass from 0.4 to 2.5 g, the smallest being attached to P. hermannsburgensis and the largest to D. blythi, so that no individuals experienced a load gain > 5% of their body mass. We fitted animals with the collars after dusk, held them for 1–2 h in a soft cotton bag to ensure correct attachment, and then released them at the point of capture 3–4 h after dark. Tracking began the following day and continued for 5–11 days and nights. Animals were trapped again before batteries failed, and collars were removed. All tags operated at 150–151 MHz and were supplied by Biotrack (Wareham, United Kingdom), Holohil Systems Ltd. (Carp, Ontario, Canada), or Titley Electronics (Ballina, New South Wales, Australia) at different times during the study.
We sought to determine where animals were moving above ground and also where burrows were located, so tracking was carried out throughout 24 h. Radiosignals were detected using a 3-element handheld yagi antenna (Titley Electronics) and either a Regal 1000 (Titley Electronics) or TR-2 (Telonics Inc., Mesa, Arizona) receiver. Animals were tracked on foot at night every 1–4 h. If they were in open sandy areas, we could often see them by using a dim white or red flashlight and thus determine their location precisely. If animals could not be approached, we estimated locations by triangulation of 2 or 3 bearings taken in the direction of the peak signal. Bearings were taken using a prismatic compass either by a single observer 1–2 min apart or simultaneously by 2 or 3 observers. Bearings were taken from known triangulation points with angles constrained to be >20° and <120° from each other (Kenward 2001). Pilot trials using tags placed at known locations showed that bearings were usually accurate to within ±5° up to distances of — 110 m (Dickman et al. 2010). By day, animals in burrows were located using similar methods, with precise burrow locations pinpointed (to within 1–2 m) by removing the yagi antenna and following the signal using just the coaxial cable and receiver. Burrows and animal locations were flagged using marker tape and subsequently placed on fine-scale maps, using methods outlined by Dickman et al. (2010). All tracking was carried out during the same study periods as the resource sampling.
Giving-up-density experiments.—This approach creates small patches where food is locally enriched for the target species and measures the trade-off between foraging returns and the costs (energetic, predation, and missed opportunity costs of foraging—Brown 1988, 1992) of staying at the food patch in terms of how much food remains after foraging. This residual food is the giving-up density (GUD). We expected that, if the risk of predation on small mammals is lower in woodland than in grassland habitat during drought, animals would allocate relatively more time at feeding stations in woodland and hence return the lowest GUDs there. Barn owls (Tyto alba) occur in small numbers in the Simpson Desert and could be expected to pose some risk to all 3 species, and European red foxes (Vulpes vulpes) and feral cats (Felis catus) also are present (Dickman et al. 1991; Mahon et al. 1998).
Food patches were plastic bowls (15 cm in diameter, 4 cm deep) that were provisioned with either 20 peanut quarters or 10 live mealworms (larvae of Tenebrio molitor) mixed into 200 ml of sifted sand. The peanuts and mealworms were expected to provide high-quality food resources for the rodents (Kotler et al. 1998) and for D. blythi (Haythornthwaite and Dickman 2000), respectively. The bowls were half buried in the sand and smeared on the undersurface with a mixture of Vaseline and Coopex insecticide powder (Bayer Ltd., Pymble, New South Wales, Australia) to deter ants from accessing the food sources. We established 10–24 food patches in open areas > 2 m away from the nearest cover in woodland and grassland habitat, with each patch separated from its neighbor by >25 m. In general, experiments targeting the rodents and D. blythi were carried out either on different field trips or at the same time but in different parts of the study area to reduce the risk of D. blythi learning to use the peanut-enriched patches as ambush sites to hunt the rodents.
The food patches usually were prebaited for 1 night to allow animals to become accustomed to them and then run for 3 or 4 consecutive nights. In the late afternoons, prior to these nights, the 20 peanut quarters or 10 mealworms were added to all food patches and the surrounding sand smoothed so that foot tracks could be read. The food patches were checked at dawn after each night. We identified tracks in or near the bowls using the characteristics described above and counted the numbers of peanut quarters or mealworms remaining per bowl to obtain the GUD. If both P. hermannsburgensis and N. alexis had foraged in a bowl, we used the GUD for that patch only if we could distinguish which species had been the last forager; if in doubt, the data for that patch were discarded. Similarly, for D. blythi we often found bowls where smaller species of insectivorous dasyurids (Ningaui ridei and Sminthopsis spp.) also had foraged. Data were included for these bowls only if we could see the tracks of the larger species overprinting those of its smaller counterparts.
All procedures from 1990 to the present were carried out under license from the Queensland National Parks and Wildlife Service (now Queensland Environmental Protection Agency), with approval from the University of Sydney Animal Ethics Committee. In addition, procedures were consistent with guidelines approved by the American Society of Mammalogists (Sikes et al. 2011).
Data processing and comparisons.—To test our 1st hypothesis concerning food resources (see the “Introduction”) we 1st processed the field samples to ensure that we quantified only edible invertebrates and seeds. Invertebrates were decanted from sampling vials, placed into 70% ethanol, and then sorted in petri dishes under a low-power dissection microscope. We removed ants, collembolans, mites, and other invertebrates < 2 mm long from consideration, because these are not eaten by the study species (Chen et al. 1998; Fisher and Dickman 1993a, 1993b; Murray and Dickman 1994a), but counted all remaining invertebrates. Seeds were extracted from sand samples using a 500-µm sieve. We removed seeds with very hard seed cases such as some Eremophila spp. and Stylobasium spathulatum, because these do not appear to be eaten by the 2 species of study rodents (Murray and Dickman 1994a, 1997; Murray et al. 1999), but included all others. After sorting, pilot samples of seeds and invertebrates were dried in an oven (80°C) to constant mass.
Inspection of the raw resource data revealed that many invertebrate vials and individual seed samples were empty, so the numbers of items in each food category were 1st summed within grids and then pooled within the drought or nondrought sampling periods. Two-factor analyses of variance (ANOVAs) were computed separately on the summed seed and invertebrate data, with period (drought and nondrought) and habitat (woodland and grassland) as factors. Levene's test indicated that variances were equal for both data sets and the Kolmogorov-Smirnov test confirmed normality, so no transformations were carried out. Results for the numbers and dry masses of both resource categories were very similar; hence we show the count results only. Data on shelter resources were treated in a similar manner to those for food but were subjected to arcsine transformation prior to analyses.
Our hypothesis ii, on trappability, was tested using correlation analyses. To investigate trappability at the trap level we compared the numbers of captures of each study species in each habitat per trip with the numbers of associated transects that had footprints of the same species. Because the data were not distributed normally and results from drought periods had many zero values, we used Spearman's rank correlations (rs—Zar 1974). To investigate trappability at the grid level we compared counts of the total numbers of individuals of each species per grid with the associated numbers of their burrows on those grids. Data from within both drought and nondrought periods were combined to increase sample sizes and compared using Spearman's rank correlations. Tests for each species were subjected to sequential Bonferroni correction to reduce the chance of committing type I errors (Quinn and Keough 2002).
To test our hypothesis iii we 1st tallied the numbers of captures per grid in both habitats. Because trapping effort on the woodland grids (18 traps) was only one-half that on the grassland grids (36 traps), we doubled the numbers of captures obtained on the woodland grids to allow visual comparison. We then constructed a generalized linear model to investigate if capture rates differed between habitats (habitat) and between drought and nondrought periods (trip). Captures were pooled from the grassland and the woodland grids; a negative binomial model with a log-link function then was chosen to best fit the overdispersed capture data with an offset (traps) used to take into account the unequal trapping effort between habitats. The factors habitat, trip, and the interaction between trip and habitat were included in the model. Analysis of deviance was used to test the overall model terms, with the deviance approximately following the chi-squared distribution (Zuur et al. 2009). Autocorrelation plots of residuals from the models provided no indication of correlation between trips, so we did not build a temporal autocorrelation structure into the models. Diagnostic plots (predicted values against the residuals, normal Q-Q plots of residuals, and leverage plots) were inspected to confirm that the assumptions of the generalized linear models were not violated.
Tests of our hypothesis iv took 2 forms. Because animals equipped with spool and line devices or dusted with fluorescent pigments were captured in grassland located too far away for their tracks to be followed into woodland, we arranged the data to indicate whether animals moved toward the nearest patches of woodland, or in other directions, between drought and nondrought periods. The prevailing direction taken by each tracked individual was determined by linear best-fit methods and the compass direction recorded. From the point where each animal track began, we then drew 2 further lines originating from the same point but at a 45° bearing on either side of actual direction of travel. This allowed a 90° cone to be described in the general direction of travel. We then scored whether tracked animals would have encountered a patch of woodland if they had continued moving in the direction prescribed by the cone. In most cases the paths of tracked animals terminated within 100 m of woodland. Although individuals might have used resources en route to the woodland, 30–100 m is within the range of nightly movement of each of the study species; hence it seems reasonable to infer that each tracked individual could have reached the woodland. Frequencies of animals moving toward woodland and in other directions were pooled between drought and nondrought periods and compared using chi-square contingency tests for each species.
Animals equipped with radiotags provided longer-term tracking results, allowing us to determine the relative numbers of fixes that occurred in woodland compared with grassland. Animals that were observed, and locations of burrows, were scored as being in either grassland or woodland. For triangulated animals positions were calculated from the compass bearings and positions of the triangulation points using an iterative maximum-likelihood estimator (Location of a Signal software; Ecological Software Solutions, Sacramento, California), which minimizes angular error between bearings and location of the signal (Lenth 1981). Positions were accepted as being located in grassland or woodland only if the distance to the nearest boundary with the other habitat was greater than the bearing error; positions within the bearing error were discarded. For each animal tracked we calculated the percentage of all fixes that were scored in woodland and transformed these by arcsine. Data from the different drought and nondrought periods were pooled, and differences in use of woodland between these periods were identified using unpaired t-tests, separately for burrow and active locations, for each species. If Levene's test showed that variances were not equal, we used t-tests without the assumption of equal variances.
The GUD data, collected to test hypothesis v, were averaged for each food patch over the 3 or 4 nights that the patches were set. GUDs were included in analyses if food patches had been visited by the study species, as judged by foot tracks in the sand matrix of the bowls, even if no food had been removed. Food patches that had been disturbed by other foragers, such as Australian ravens (Corvus comnoides), or had not been visited were omitted from analyses. GUD data from different drought and nondrought periods were pooled, and differences between period and habitat were detected using 2-factor ANOVA. These analyses were performed on untransformed data because we could not obtain equal variances even after transformation. For these tests we therefore accept significance at α = 0.01 but use α = 0.05 in all other tests (Underwood 1997).
To test our hypothesis vi we collated data from 33 of the >60 one-hectare trapping grids that we had established in spinifex grassland over the broader study region and measured the distance from each grid to the nearest patch of gidgee woodland. Measurements were made using aerial photographs, Google Earth (Google, Mountain View, California) imagery, and ground-truthing (Greenville et al. 2009). Monthly changes in the capture rate were calculated for each species for each grid from the period of lowest numbers immediately before heavy summer rainfall to the peak in capture rate when it occurred later in the same year. We plotted rate of increase, and also the maximum capture rate attained, against distance to the nearest gidgee patch. Capture rate was standardized as mean captures per 100 trap nights (1 trap night = 1 trap open for 1 night) per field trip, with grids used as replicates. Capture data obtained from the same grid over 2 or 3 population eruptions of the study species (Dickman et al. 2010) were averaged to obtain a single value and log transformed to improve normality. Maximum capture rates were rounded to the nearest integer. Generalized linear models with normal distributions were used to evaluate the rate of increase versus distance data and generalized linear models with negative binomial regressions the maximum capture rate versus distance data. These were constructed with log-link functions, with diagnostic plots confirming that model assumptions were not violated.
Most analyses were calculated using SPSS 15.0 (SPSS Inc. 2006). Generalized linear models were implemented using the MASS package (Venables and Ripley 2002) in R, version 2.10.1 (R Development Core Team 2009).
Results
Resources.—The numbers of edible invertebrates showed no clear pattern between habitats and drought compared to nondrought periods (Fig. 1A), with neither habitat (F1,28 = 0.71, P = 0.405), period (F1,28 = 0.40, P = 0.531), nor their interaction (F1,28 = 0.01, P = 0.916) approaching significance. In contrast, more seeds were recorded in woodland than in grassland (F1,92 = 7.46, P = 0.008) and after rain than during drought (F1,92 = 4.97, P = 0.028), but there was no habitat × period interaction (F1,92 = 0.19, P = 0.661; Fig. 1B).
Mean (± SE) numbers of A) pitfall-trapped invertebrates and B) seeds that are known to be eaten by small mammals collected in spinifex grassland and gidgee woodland habitats during non-drought and drought periods in the northeastern Simpson Desert, central Australia. Sample sizes appear above the bars.
Mean (± SE) numbers of A) pitfall-trapped invertebrates and B) seeds that are known to be eaten by small mammals collected in spinifex grassland and gidgee woodland habitats during non-drought and drought periods in the northeastern Simpson Desert, central Australia. Sample sizes appear above the bars.
Most shelter resources were invariant between habitats and times, with the coverage of live and dead spinifex averaging 29% and 12%, respectively, in the grassland compared to ∼1% for each component in the woodland and woody debris averaging 14% in the woodland compared with <1% in grassland. Dicotyledonous plants ≥ 1 m high and leaf litter averaged 11% and 46% cover, respectively, in woodland, compared with coverage values of 3% and 9% in the grassland. All main factor effects for habitat were significant for these habitat components (F1,92 values ranged from 6.75 to 78.49, P values all <0.01), but no period or habitat × period effects were significant. The only shelter component to vary by period and by habitat was dicotyledonous plants < 1 m high (Fig. 2), which achieved patchy but generally much higher coverage in both habitats during nondrought compared with drought periods (habitat: F1,92 = 5.85, P = 0.018; period: F1,92 = 47.10, P < 0.001; interaction: F1,92 = 0.21, P = 0.649).
Mean(± SE) percent cover of dicotyledonous plants < 1 m high assessed on circular quadrats of radius 2.5 m in spinifex grassland and gidgee woodland habitats during nondrought and drought periods in the northeastern Simpson Desert, central Australia. Sample sizes appear above the bars.
Mean(± SE) percent cover of dicotyledonous plants < 1 m high assessed on circular quadrats of radius 2.5 m in spinifex grassland and gidgee woodland habitats during nondrought and drought periods in the northeastern Simpson Desert, central Australia. Sample sizes appear above the bars.
Trappability.—Captures of animals in traps generally were correlated positively with the numbers of their tracks on the sand transects. Except for 1 sampling session for D. blythi in grassland, when rs = 0.48, P = 0.20, and 2 further sessions when no activity of this species was recorded in woodland, rs for all species, habitat, and period comparisons ranged from 0.39 to 0.82, with associated P-values < 0.05 and all but 1 P-value meeting the more stringent Bonferroni-corrected a < 0.005. Comparisons of burrow counts with captures for each species in grassland between drought and nondrought periods were similarly positive. The rs values ranged from 0.60 to 0.77, with associated P-values < 0.05, with only the smallest rs value failing to meet the corrected α < 0.025. These results suggest that trappability was consistent between habitats at different times for all species.
Capture rates.—For P. hermannsburgensis the capture rate fell as conditions dried in both habitats until December 2009 (Fig. 3A). The rate of decline tended to be faster in the grassland than in the woodland (Fig. 3B), with captures reaching a low in April 2009 and producing a significant trip × habitat interaction (Table 1). For N. alexis the capture rate declined rapidly in both habitats, but animals were not captured in woodland for the last 2 sampling sessions (Fig. 4A). The generalized linear model showed that capture rates were consistently higher in grassland than in woodland, but rates of decline were similar (Fig. 4B) and no trip × habitat interaction was evident (Table 1). No clear pattern of captures could be discerned for D. blythi in woodland, but capture rates appeared to increase over time in the grassland (Fig. 5A); no trip × habitat interaction was found (Table 1; Fig. 5B). The deviance explained by the models was generally high: 98% for P. hermannsburgensis, 91% for N. alexis, and 85% for D. blythi.
Captures of Pseudomys hermannsburgensis during increasing drought conditions from April 2008 to December 2009 in spinifex grassland and gidgee woodland habitats in the northeastern Simpson Desert, central Australia, showing A) mean (± SE) captures per grid, and B) predicted captures (± SE) derived using a negative binomial model with a log-link function. Numbers of captures in woodland have been doubled in A to standardize trapping effort between grids in the 2 habitats but modeled in B with an offset to account for unequal trapping effort between habitats. Data are drawn from 5 trapping grids in grassland and 8 grids in woodland.
Captures of Pseudomys hermannsburgensis during increasing drought conditions from April 2008 to December 2009 in spinifex grassland and gidgee woodland habitats in the northeastern Simpson Desert, central Australia, showing A) mean (± SE) captures per grid, and B) predicted captures (± SE) derived using a negative binomial model with a log-link function. Numbers of captures in woodland have been doubled in A to standardize trapping effort between grids in the 2 habitats but modeled in B with an offset to account for unequal trapping effort between habitats. Data are drawn from 5 trapping grids in grassland and 8 grids in woodland.
Captures of Notomys alexis during increasing drought conditions from April 2008 to December 2009 in spinifex grassland and gidgee woodland habitats in the northeastern Simpson Desert, central Australia, showing A) mean (± SE) captures per grid, and B) predicted captures (± SE) derived using a negative binomial model with a log-link function. Numbers of captures in woodland have been doubled in A to standardize trapping effort between grids in the 2 habitats but modeled in B with an offset to account for unequal trapping effort between habitats. Data are drawn from 5 trapping grids in grassland and 8 grids in woodland.
Captures of Notomys alexis during increasing drought conditions from April 2008 to December 2009 in spinifex grassland and gidgee woodland habitats in the northeastern Simpson Desert, central Australia, showing A) mean (± SE) captures per grid, and B) predicted captures (± SE) derived using a negative binomial model with a log-link function. Numbers of captures in woodland have been doubled in A to standardize trapping effort between grids in the 2 habitats but modeled in B with an offset to account for unequal trapping effort between habitats. Data are drawn from 5 trapping grids in grassland and 8 grids in woodland.
Captures of Dasycercus blythi during increasing drought conditions from April 2008 to December 2009 in spinifex grassland and gidgee woodland habitats in the northeastern Simpson Desert, central Australia, showing A) mean (± SE) captures per grid, and B) predicted captures (± SE) derived using a negative binomial model with a log-link function. Numbers of captures in woodland have been doubled in A to standardize trapping effort between grids in the 2 habitats but modeled in B with an offset to account for unequal trapping effort between habitats. Data are drawn from 5 trapping grids in grassland and 8 grids in woodland.
Captures of Dasycercus blythi during increasing drought conditions from April 2008 to December 2009 in spinifex grassland and gidgee woodland habitats in the northeastern Simpson Desert, central Australia, showing A) mean (± SE) captures per grid, and B) predicted captures (± SE) derived using a negative binomial model with a log-link function. Numbers of captures in woodland have been doubled in A to standardize trapping effort between grids in the 2 habitats but modeled in B with an offset to account for unequal trapping effort between habitats. Data are drawn from 5 trapping grids in grassland and 8 grids in woodland.
Summary statistics for generalized linear model analyses on capture rates of three species of small mammals in spinifex grassland and gidgee woodland habitats (habitat) during drought and nondrought periods (trip) in the northeastern Simpson Desert, central Australia. Generalized linear models used negative binomials with log-link functions, with model terms tested by analysis of deviance, and deviance following the chi-square distribution.
| d.f. | deviance | d.f. residual | Residual Deviance | P | ||
|---|---|---|---|---|---|---|
| Pseudomys | ||||||
| hermannsburgensis | ||||||
| Null | 61 | 3,332.3 | ||||
| Trip | 5 | 827.86 | 56 | 2,504.4 | <0.001 | |
| Habitat | 1 | 2,452.93 | 55 | 51.5 | <0.001 | |
| Trip × habitat | 5 | 12.21 | 50 | 39.3 | 0.008 | |
| Notomys alexis | ||||||
| Null | 61 | 674.1 | ||||
| Trip | 5 | 27.39 | 56 | 646.7 | <0.001 | |
| Habitat | 1 | 582.93 | 55 | 63.8 | <0.001 | |
| Trip × habitat | 5 | 4.69 | 50 | 59.1 | 0.49 | |
| Dasycercus blythi | ||||||
| Null | 61 | 161.6 | ||||
| Trip | 5 | 15.22 | 56 | 146.4 | <0.001 | |
| Habitat | 1 | 119.49 | 55 | 26.9 | <0.001 | |
| Trip × habitat | 5 | 3.37 | 50 | 23.6 | 0.28 |
| d.f. | deviance | d.f. residual | Residual Deviance | P | ||
|---|---|---|---|---|---|---|
| Pseudomys | ||||||
| hermannsburgensis | ||||||
| Null | 61 | 3,332.3 | ||||
| Trip | 5 | 827.86 | 56 | 2,504.4 | <0.001 | |
| Habitat | 1 | 2,452.93 | 55 | 51.5 | <0.001 | |
| Trip × habitat | 5 | 12.21 | 50 | 39.3 | 0.008 | |
| Notomys alexis | ||||||
| Null | 61 | 674.1 | ||||
| Trip | 5 | 27.39 | 56 | 646.7 | <0.001 | |
| Habitat | 1 | 582.93 | 55 | 63.8 | <0.001 | |
| Trip × habitat | 5 | 4.69 | 50 | 59.1 | 0.49 | |
| Dasycercus blythi | ||||||
| Null | 61 | 161.6 | ||||
| Trip | 5 | 15.22 | 56 | 146.4 | <0.001 | |
| Habitat | 1 | 119.49 | 55 | 26.9 | <0.001 | |
| Trip × habitat | 5 | 3.37 | 50 | 23.6 | 0.28 |
Summary statistics for generalized linear model analyses on capture rates of three species of small mammals in spinifex grassland and gidgee woodland habitats (habitat) during drought and nondrought periods (trip) in the northeastern Simpson Desert, central Australia. Generalized linear models used negative binomials with log-link functions, with model terms tested by analysis of deviance, and deviance following the chi-square distribution.
| d.f. | deviance | d.f. residual | Residual Deviance | P | ||
|---|---|---|---|---|---|---|
| Pseudomys | ||||||
| hermannsburgensis | ||||||
| Null | 61 | 3,332.3 | ||||
| Trip | 5 | 827.86 | 56 | 2,504.4 | <0.001 | |
| Habitat | 1 | 2,452.93 | 55 | 51.5 | <0.001 | |
| Trip × habitat | 5 | 12.21 | 50 | 39.3 | 0.008 | |
| Notomys alexis | ||||||
| Null | 61 | 674.1 | ||||
| Trip | 5 | 27.39 | 56 | 646.7 | <0.001 | |
| Habitat | 1 | 582.93 | 55 | 63.8 | <0.001 | |
| Trip × habitat | 5 | 4.69 | 50 | 59.1 | 0.49 | |
| Dasycercus blythi | ||||||
| Null | 61 | 161.6 | ||||
| Trip | 5 | 15.22 | 56 | 146.4 | <0.001 | |
| Habitat | 1 | 119.49 | 55 | 26.9 | <0.001 | |
| Trip × habitat | 5 | 3.37 | 50 | 23.6 | 0.28 |
| d.f. | deviance | d.f. residual | Residual Deviance | P | ||
|---|---|---|---|---|---|---|
| Pseudomys | ||||||
| hermannsburgensis | ||||||
| Null | 61 | 3,332.3 | ||||
| Trip | 5 | 827.86 | 56 | 2,504.4 | <0.001 | |
| Habitat | 1 | 2,452.93 | 55 | 51.5 | <0.001 | |
| Trip × habitat | 5 | 12.21 | 50 | 39.3 | 0.008 | |
| Notomys alexis | ||||||
| Null | 61 | 674.1 | ||||
| Trip | 5 | 27.39 | 56 | 646.7 | <0.001 | |
| Habitat | 1 | 582.93 | 55 | 63.8 | <0.001 | |
| Trip × habitat | 5 | 4.69 | 50 | 59.1 | 0.49 | |
| Dasycercus blythi | ||||||
| Null | 61 | 161.6 | ||||
| Trip | 5 | 15.22 | 56 | 146.4 | <0.001 | |
| Habitat | 1 | 119.49 | 55 | 26.9 | <0.001 | |
| Trip × habitat | 5 | 3.37 | 50 | 23.6 | 0.28 |
Activity.—Of 12 P. hermannsburgensis tracked during drought using spool and line devices and fluorescent pigments, 10 moved in the direction of woodland patches and the other 2 in other directions. In nondrought periods only 25 of 61 tracked animals moved toward woodland, indicating a directional travel bias toward woodland during dry times (χ12, P = 0.008). No patterns were evident for 60 tracked N. alexis (χ12, P = 0.09) or for 28 tracked D. blythi (χ12 = 2.67, P = 0.10).
Radiotracked animals showed different patterns of habitat use between species. In general, none of the study species used burrows very often in woodland. However, P. hermannsburgensis roughly doubled its use of burrows in woodland during drought compared with nondrought periods (t43 = 2.14, P = 0.041), whereas no clear pattern was observed for either N. alexis (t34 = 0.80, P = 0.436) or D. blythi (t12 = 0.23, P = 0.822; Figs. 6A–C). Active fixes increased dramatically during drought compared with nondrought periods for P. hermannsburgensis (t43 = 3.37, P = 0.002), with a similar trend noted for N. alexis (t34 = 2.06, P = 0.059), but no clear pattern was evident for D. blythi (t12 = 0.23, P = 0.819; Figs. 7A–C).
Mean (± SE) numbers of A) Pseudomys hermannsburgensis, B) Notomys alexis, and C) Dasycercus blythi radiotracked to burrows during nondrought and drought periods in the northeastern Simpson Desert, central Australia, expressed as the percentages of burrows that were located in gidgee woodland habitat. Sample sizes appear above the bars.
Mean (± SE) numbers of A) Pseudomys hermannsburgensis, B) Notomys alexis, and C) Dasycercus blythi radiotracked to burrows during nondrought and drought periods in the northeastern Simpson Desert, central Australia, expressed as the percentages of burrows that were located in gidgee woodland habitat. Sample sizes appear above the bars.
Mean (± SE) percent of radiofixes of radiotracked A) Pseudomys hermannsburgensis, B) Notomys alexis, and C) Dasycercus blythi while active above ground that were in gidgee woodland habitat during nondrought and drought periods in the northeastern Simpson Desert, central Australia. Sample sizes appear above the bars.
Mean (± SE) percent of radiofixes of radiotracked A) Pseudomys hermannsburgensis, B) Notomys alexis, and C) Dasycercus blythi while active above ground that were in gidgee woodland habitat during nondrought and drought periods in the northeastern Simpson Desert, central Australia. Sample sizes appear above the bars.
Giving-up densities.—The GUDs for P. hermannsburgensis were lower during drought than nondrought periods (F1,102 = 168.04, P < 0.001) and lower in woodland than in grassland (F1,102 = 37.39, P < 0.001), with very low GUDs in woodland during drought producing a habitat × period interaction (F1,102 = 25.06, P < 0.001; Fig. 8A). GUDs for N. alexis also were lowest during drought (F1,124 = 68.86, P < 0.001), but no habitat effect was observed with a set at 0.01 (F1,124 = 5.98, P = 0.016), and no habitat × period interaction was revealed (F1,24 = 2.59, P = 0.11; Fig. 8B). The pattern for D. blythi was very different, with consistently low GUDs in grassland producing a strong habitat effect (F1,90 = 55.45, P < 0.001) but no period effect (F1,90 = 0.02, P = 0.889) or interaction (Fl,90 = 0.12, P = 0.725; Fig. 8C).
Mean (± SE) giving-up densities (GUDs) of A) Pseudomys hermannsburgensis\ B) Notomys alexis, and C) Dasycercus blythi in the northeastern Simpson Desert, central Australia, shown for each species during nondrought and drought periods in spinifex grassland and gidgee woodland habitats. GUDs represent the numbers of peanut quarters left by rodents in food patches from 20 quarters that were provided in A and B, and the numbers of mealworms (Tenebrio molitor) left by D. blythi in food patches from 10 that were provided in C. Sample sizes appear above the bars.
Mean (± SE) giving-up densities (GUDs) of A) Pseudomys hermannsburgensis\ B) Notomys alexis, and C) Dasycercus blythi in the northeastern Simpson Desert, central Australia, shown for each species during nondrought and drought periods in spinifex grassland and gidgee woodland habitats. GUDs represent the numbers of peanut quarters left by rodents in food patches from 20 quarters that were provided in A and B, and the numbers of mealworms (Tenebrio molitor) left by D. blythi in food patches from 10 that were provided in C. Sample sizes appear above the bars.
Recovery after rainfall—Rates of recovery of P. hermannsburgensis on grassland grids after rain tended to be faster if the grids were close to patches of woodland rather than farther away, although the statistical outcome was only marginally nonsignificant (F1,16 = 3.36, P = 0.086); maximum rates of capture also were higher on grids close to woodland (Z = 2.08, P = 0.037), although the deviance explained by the latter model was only 15% (Figs. 9A and 9B). No clear patterns of rate of increase or maximum capture rate were observed on the grassland grids after rain for either N. alexis (F1,18 = 0.001, P = 0.97 and Z = 0.54, P = 0.56, respectively; Figs. 10 A and 10B) or D. blythi (FU9 = 0.003, P = 0.95 and Z = 0.04, P = 0.97, respectively; Figs. 11A and 11B).
Capture rates of Pseudomys hermannsburgensis, standardized as captures per 100 trap nights (1 trap night = 1 trap open for 1 night) per field trip, in the northeastern Simpson Desert, central Australia, showing in A) the monthly rate of change in capture rate from lowest to highest in the year after rainfall on trapping grids set in grassland habitat at different distances from the nearest gidgee woodland, and B) the maximum capture rates obtained in the same years and sites. Symbols represent means for different trapping grids.
Capture rates of Pseudomys hermannsburgensis, standardized as captures per 100 trap nights (1 trap night = 1 trap open for 1 night) per field trip, in the northeastern Simpson Desert, central Australia, showing in A) the monthly rate of change in capture rate from lowest to highest in the year after rainfall on trapping grids set in grassland habitat at different distances from the nearest gidgee woodland, and B) the maximum capture rates obtained in the same years and sites. Symbols represent means for different trapping grids.
Capture rates of Notomys alexis, standardized as captures per 100 trap nights (1 trap night = 1 trap open for 1 night) per field trip, in the northeastern Simpson Desert, central Australia, showing in A) the monthly rate of change in capture rate from lowest to highest in the year after rainfall on trapping grids set in grassland habitat at different distances from the nearest gidgee woodland, and B) the maximum capture rates obtained in the same years and sites. Symbols represent means for different trapping grids.
Capture rates of Notomys alexis, standardized as captures per 100 trap nights (1 trap night = 1 trap open for 1 night) per field trip, in the northeastern Simpson Desert, central Australia, showing in A) the monthly rate of change in capture rate from lowest to highest in the year after rainfall on trapping grids set in grassland habitat at different distances from the nearest gidgee woodland, and B) the maximum capture rates obtained in the same years and sites. Symbols represent means for different trapping grids.
Capture rates of Dasycercus blythi, standardized as captures per 100 trap nights (1 trap night = 1 trap open for 1 night) per field trip, in the northeastern Simpson Desert, central Australia, showing in A) the monthly rate of change in capture rate from lowest to highest in the year after rainfall on trapping grids set in grassland habitat at different distances from the nearest gidgee woodland, and B) the maximum capture rates obtained in the same years and sites. Symbols represent means for different trapping grids.
Capture rates of Dasycercus blythi, standardized as captures per 100 trap nights (1 trap night = 1 trap open for 1 night) per field trip, in the northeastern Simpson Desert, central Australia, showing in A) the monthly rate of change in capture rate from lowest to highest in the year after rainfall on trapping grids set in grassland habitat at different distances from the nearest gidgee woodland, and B) the maximum capture rates obtained in the same years and sites. Symbols represent means for different trapping grids.
Discussion
Resources provided by gidgee woodland.—Our hypotheses about how small mammals would use patches of gidgee woodland were predicated on the expectation that the woodland would provide more food or shelter resources for the study species during drought than at other times. This expectation was partly borne out. Although invertebrate numbers were similar throughout, the density of edible seeds was greatest in woodland, especially during nondrought periods. Shelter resources differed between habitats but remained largely unchanged between periods, although rainfall-triggered germination and growth of low-growing herbs and forbs ensured that these plants provided more cover in wet periods than during drought. Despite these differences, we found no habitat × period interaction for any of the resources sampled. Thus, although woodland furnishes more shelter and more food, at least for the granivores, it does so at all times. Why then did the study species not make more extensive use of woodland habitat?
In the 1st instance small mammals likely respond not just to the amount of particular resources in the environment but also to resource quality. Although we sampled food resources that are known to be eaten by the study species (Chen et al. 1998; Murray and Dickman 1994a), it is not clear how the distributions of preferred foods change over time between habitats. Spinifex seed appears to be selected when it becomes available after heavy summer rain, and it also might assist in triggering reproduction in the 2 study rodents (Ricci 2003). It is more readily available in the grassland than the woodland habitats at these times. Likewise, D. blythi most likely prefers large invertebrates that are accessed more readily in open sandy areas near dune crests (Fisher and Dickman 1993a). Shelter resources also are likely to vary in quality. For example, hummocks of needle-leafed spinifex are available only in the grassland and probably provide more effective protection from predators than do piles of leaf litter and woody debris that provide the main sources of cover in woodland patches.
Second, although we sampled resources that are potentially available to active small mammals, we did not sample the shelters available to animals at rest. Burrows are critically important to the 3 study species for the maintenance of social structure and protection from extreme conditions on the soil surface (Dickman et al. 2010; Woolley 1990) and probably are energetically less costly to excavate in firm sandy soil in grassland than in the compacted clay soils that underlie the woodland. The shallowest digger, P. hermannsburgensis, used burrows in woodland more than the other 2 species (Fig. 6), supporting the notion that clay generally precludes very deep excavations. Although shallow burrows might be less protected from the thermal environment than deep ones, shade from the trees and shrubs ≥ 1 m high in woodland probably help to reduce subsurface temperatures.
Third, until 2004 cattle were present in the study area. These large herbivores aggregate under gidgee trees, using them for food and shade (Frank 2010), and probably degrade the integrity of woodland patches by scattering and breaking woody debris, preventing the regeneration of shrubs, and trampling and compacting the soil. These activities probably reduce the availability of resources and the suitability of the woodland patches generally for small mammals but especially for N. alexis and D. blythi. Although the deep burrows of these species can shield them from the immediate impacts of trampling, the energetic costs of continually repairing damaged burrows or excavating new ones might exceed any benefits gained by remaining in the gidgee. The lower costs of repairing or excavating shallow burrows again could be a factor allowing greater burrowing activity of P. hermannsburgensis in the woodland patches. Cattle were removed in 2004; although this would have removed the direct effects of trampling on burrows, the recovery time for the compacted soils and depleted resources is likely to be in the order of decades (Croft et al. 2007; Frank 2010).
These considerations suggest that gidgee woodlands provide important resources for small mammals, but not enough for them to serve as complete refugia at any time in the drought-wet cycle. Despite this, the 3 study species used the woodland patches very differently over time, suggesting that they perceive and respond to the habitat matrix in quite different ways.
Spatial dynamics in drought and nondrought times.—Of the 3 species studied, P. hermannsburgensis conformed more closely to our initial expectations and provided some support for the predictions of hypotheses iii-vi (see the “Introduction”). Thus, as the study region became progressively drier from 2008 through 2009, captures of this species declined, but the rate of decline was slower in the woodland than the grassland habitat. Movements were directed toward woodland during drought, and animals were roughly twice as active in woodland during drought as after rain. After drought populations of P. hermannsburgensis tended to recover more quickly and to a greater extent in grassland if they were in close proximity to gidgee. These findings indicate that P. hermannsburgensis made relatively greater use of woodland as conditions dried and food resources diminished in the surrounding grassland. In contrast, N. alexis and D. blythi used grassland habitat most extensively at all times, although the former species tended to increase its aboveground activity in woodland during drought. These findings provide little support for hypotheses iii–vi.
Our assessments indicated that trappability (hypothesis ii) remained relatively constant among species, habitats, and times and hence that the different temporal patterns of habitat use we observed for the 3 species were real and not artifacts of changes in animal behavior or susceptibility to capture. Several explanations can be advanced to account for these interspecific differences.
First, in other deserts interspecific competition occurs commonly and can have strong effects on the demography and habitat use of smaller species in the system (Heske et al. 1994; Kelt 2011; Meserve et al. 1996; Ziv et al. 1993). However, although P. hermannsburgensis is the smallest of the 3 study species, interspecific competition is an unlikely explanation for the shift of this rodent to woodland habitat. Shifts occurred when populations of N. alexis and other potential competitors were in decline and competitive interactions would have been weakest (Dickman et al. 1999b, 2010). Density-dependent intraspecific processes also are common in other deserts (Kelt 2011; Shenbrot et al. 2010) but seem unlikely to shape small mammal dynamics in the Simpson Desert. Again, this is because habitat shifts would be expected to occur at high rather than low population densities and not when, as in the present study, numbers are in decline. Second, woodland patches seldom cover >10 ha and occupy only 7–16% of the study region (Frank 2010; Greenville et al. 2009) and thus might provide areas that are too small to provide effective refuge for small mammals over extended periods. As the smallest of the 3 study species, P. hermannsburgensis probably has smaller absolute requirements for food than its larger counterparts and can persist for longer in woodland patches before moving on; thus its temporal use of woodland would be relatively greater than that of N. alexis and D. blythi as resources diminish during drought. In an analogous argument Morton (1990) proposed that small mammals survived European-induced habitat change in arid Australia better than medium-sized species such as bandicoots and wallabies, because they still could meet their modest resource requirements after the integrity of their refuge habitats had been diminished.
Third, temporal changes in the distributions of the study species among habitats can be influenced by other species such as ants that use common resources. Although not quantified here because they are not eaten by small mammals, ants were the numerically dominant invertebrates captured in the sampling vials and composed 90–95% of the catch in grassland and 99% in woodland (Gregory 2008; C. R. Dickman, pers. obs.). Many taxa in the study area are omnivorous and harvest seeds, other plant products, and insects (Reid 1995) and show high beta-diversity among habitats (Morton 1982). If ant activity declines during drought, especially among the smaller taxa that tend to predominate in woodland (Reid 1995), this might reduce demand on small seeds in woodland patches at this time and hence also reduce the likelihood of competition with rodents. Because P. hermannsburgensis is the smaller and more granivorous of the 2 rodent species and includes many small seeds in its diet (Murray and Dickman 1997), it is possible that this species benefits more than its counterpart by shifting to woodland patches during drought. Ants are strong competitors with rodents for seeds in some desert systems (Abramsky 1983; Brown and Davidson 1977) and highly efficient at removing seeds in the present study area (Predavec 1997a); hence, the potential for exploitative competition appears strong. Ants prey upon other invertebrates and occasionally small vertebrates, but it is unlikely that they deplete food resources to such an extent that they compete with D. blythi.
Finally, and perhaps most plausibly, the study species might perceive different risks of predation over time in the different habitats. The GUD results showed that the 2 species of rodents spent more time at the food patches during drought, when seeds are less abundant in the environment, and also that the lowest GUDs were exhibited by P. hermannsburgensis in woodland at this time. Neither N. alexis or D. blythi showed this latter pattern, instead quitting food patches at lower GUDs in grassland irrespective of environmental conditions. These results suggest that P. hermannsburgensis, but not N. alexis or D. blythi, perceived a reduced risk of predation in woodland during drought. The most commonly encountered predators in the study area are feral house cats (F. catus) and red foxes (V. vulpes), both of which use gidgee trees for shade by day and some hunting activity at night. Because both predators become scarce during prolonged droughts (Mahon 1999), their hunting activities would decline in woodland at such times. This in turn most likely would favor increased foraging by P. hermannsburgensis, which would otherwise be at high risk away from the protective cover of the surrounding spinifex grass. Both D. blythi and N. alexis, which is also saltatorial, are larger and faster moving than P. hermannsburgensis and might not perceive much difference in predation risk between habitats. If food and other resources that these species use are readily available in the grassland, little benefit would accrue if they moved into the woodland patches. Other potential predators such as the dingo (Canis lupus dingo) and barn owl (T. alba) also occur in the desert but are encountered relatively infrequently and are unlikely to contribute much to the observed patterns of habitat use in the study species.
Drought refugia in the desert.—Only 1 of the study species, P. hermannsburgensis, provided support for our initial hypotheses about the refuge role of woodland patches, and even then it was not confined wholly to woodland during dry periods. Some evidence existed for disproportionate use of woodland by this species as conditions dried; approximately 35% of active fixes of radiotracked animals were located in woodland during drought, even though this habitat covers no more than 16% of the study region. This finding provides support for the suggestion that at least some patches of woodland be fenced to keep them intact and free from the damaging effects of cattle (Frank 2010). Neither N. alexis or D. blythi showed a substantial or consistent increase in their use of woodland during drought, and their rates of recovery in grassland habitat were independent of proximity to gidgee. Where, then, do these species seek refuge during drought when they are effectively absent from our trapping grids in the grassland?
Radiotracking has shown that both species of rodents are more mobile at low density when it is dry than after rainfall, and also that variance in movement is high during drought (Dickman et al. 2010). Inspection of the raw data on animal locations during these periods suggests that a few individuals are consistently mobile, whereas others spend up to 4–5 days near tall shrubs or trees other than gidgee before moving rapidly to new and distant sites. The tall shrubs included mallees (E. gamophylla and E. pachyphylla) and large specimens of sandhill wattle (A ligulata), which occur as widely scattered isolates or in small groves throughout the spinifex grassland. Examination of limited data suggests that these shrubs provide deep leaf litter, stockpiles of seeds, and invertebrates that could be used by small mammals (Sato 2006). Although the combined cover of such shrubs is no more than 6% of the study region (Greenville et al. 2009), we speculate that they might help sustain animals during periods of drought. If this is correct, and animals are active locally around these small sites before moving rapidly to other such sites, our chances of capturing animals during drought would be low and confined largely to intercepting occasional dispersers. Other habitat types occur within the study region but are either not exploited by the study species (e.g., tussock grassland) or cover such small areas (e.g., samphire scrub [0.8%] and chenopod scrub [1.3%]—Greenville et al. 2009) that they are unlikely to play a refuge role. Surveys along riparian corridors and around the edges of ephemeral swamps in the study region (Free 2009; Haythornthwaite and Dickman 2006a) suggest similarly that these landscape features provide little or no spatial refuge for small mammals.
We are uncertain whether D. blythi moves between scattered microsites during drought, as might be the case for the rodents, but the generally sedentary nature of this species and its apparent reluctance to make long-range movements (Körtner et al. 2007; Masters 2003) make this unlikely. It is conceivable instead that the density of D. blythi becomes so low during drought that animals have only a very small probability of encountering or overlapping with our 1-ha trapping grids.
Acknowledgments
We thank D. and P. Smith and family for support and access to Ethabuka in the early years of this study and Bush Heritage Australia for support, logistical assistance, and access to the reserve since 2004. We also are indebted to hundreds of volunteers, students, and assistants for help in the field, especially P. Banks, C.-L. Beh, A. Frank, A. Glen, D. Gregory, A. Haythornthwaite, P. Higgs, F. Koch, M. Letnic, G. Madani, P. Mahon, G. McNaught, T. Parratt, L. Pastro, M. Predavec, F. Quails, S. Ricci, and M. Tischler, and to M. Price and N. Waser for helpful suggestions about the use of transects to assess trappability. B. Fox and an anonymous reviewer provided helpful comments on the manuscript, and M. Bedoya Perez kindly provided the Spanish translation for the “Resumen.” The Australian Research Council provided funding via Small, Large and Discovery Grants to CRD and GMW.
Literature Cited
Author notes
Special Feature Editor was Barbara H. Blake.
