Anthropogenic water sources and the effects on Sonoran Desert small mammal communities

Study area

We conducted this study within the Sauceda Mountains on the Barry M. Goldwater Range (BMGR-East), a 424,919 ha military training area 39 kilometers south of Gila Bend, in Arizona, USA. Multiple AWS were constructed to support desert bighorn sheep (Ovis canadensis nelsoni) populations and draw federally endangered Sonoran pronghorn (Antilocapra americana sonoriensis) away from military testing ranges. The presence of AWS and relatively undisturbed Sonoran Desert make BMGR-East an ideal site to investigate effects of AWS on vegetative structure and rodent communities. Study site elevations ranged from 425 to 730 m. Topography was characterized by large hills, valleys, and ephemeral washes with vegetation characteristic of Arizona Upland and Lower Colorado subdivisions of the Sonoran Desert scrub community (Brown & Lowe, 1980). Study sites within these two subdivisions and ephemeral washes supported plant species including, creosote bush (Larrea tridentata), triangular bursage (Ambrosia deltoidea), yellow paloverde (Parkinsonia microphylla), saguaro cactus (Carnegiea gigantean), cholla (Cylindroptunia spp.), Acacia spp, and ocotillo (Fouquieria splendens). Additional species present in xeroriparian areas were thornbush (Lycium spp.), velvet mesquite (Prosopis velutina), desert ironwood (Olneya testoa), and desert honeysuckle (Anisacanthus thuberi). Special use permit #2012-01 was issued for access onto the BMGR-East by the 56th Range Management Office of Luke Air Force Base, Arizona.

Mammal trapping

To determine differences in mammal abundance, biomass, richness, and diversity at AWS and non-AWS control sites, we live-trapped rodents during winter 2011 (October–January) and during spring 2012 (February–May). Because maximum daily temperatures exceed 41 °C during the summer (30-year average, NOAA weather station USC00023393 at Gila Bend, AZ), we trapped during cooler periods to reduce heat-stress on animals. We trapped rodents during four sessions (two sessions per season), with three trap-nights per session. Rodents were trapped at five AWS including four human-constructed sites and one modified natural site (tinaja—depression formed in bedrock carved by rainfall or seepage). Rodents were trapped at five control sites from the surrounding desert. Trapping sessions 1 and 2 during winter averaged 54 days apart; trapping sessions 3 and 4 during spring averaged 48 days apart. Trapping sessions between seasons (sessions 2 and 3) averaged 94 days apart. We trapped along 135 m transects with 10 traps per line. We used Sherman live traps (five traps of 8 × 9 × 23 cm and five traps of 8 × 9 × 33 cm) and alternated trap type along the transect (sensu Burkett & Thompson, 1994; Pearson & Ruggerio, 2003; Hopkins & Kennedy, 2004). Each site (AWS and CS) had four trapping transects placed randomly using ESRI ArcMAP 10 software (Environmental Systems Research Institute, Redlands, CA, USA). For example at AWS, we generated four random points within 50 m of the tank or tinaja and transects radiated away on random bearings (1–360) that did not intersect with other transects around the AWS (Fig. A1). Control sites were selected by generating random points using ArcMAP that occurred between 500 and 700 m from each AWS (Fig. A1). The orientation of control transects were also selected by choosing a random bearing. Transects were not revisited and new transects were established during each trapping session. Therefore, the sample size to estimate mammal abundance, biomass, richness, and diversity was n = 80 transects during winter and n = 80 transects during spring (i.e., 5 sites × 4 transects × 2 treatments × 2 sessions = 80).

Traps were placed before sundown and baited with apple wafer pellets (Manna Pro; St. Louis, MO, USA). Traps were checked starting 1.5 h before sunrise each day unless stormy weather and rain warranted checking traps earlier. Polyester or cotton batting was placed in each trap during winter trap sessions to reduce exposure and minimize trap mortality. Captured animals were identified to species (Hoffmeister, 1986), aged (e.g., juvenile or adult), sexed, and weighed. Animals were individually marked with self-piercing metal tags applied to ear pinnae using an applicator (National Band and Tag Company, New Port, KY, USA) in the left ear. We also marked ears with permanent ink marks, with individually unique patterns of our own design, in case tags tore from pinnae. Animals were released at their site of capture. Animals were handled and processed following Arizona State University Institutional Animal Care and Use Committee (IACUC) protocol #09-1051R.

Vegetation sampling

To assess how vegetation differed between AWS and CS and to relate vegetation to rodent occurrence, we measured plant species richness, density, and cover. To quantify plant density and species richness (Epple & Epple, 1995), we collected data during the spring season in two 4 × 25 m (0.01 ha) macro-plots randomly located along each trapping transect (Carpenter, 1999; Fig. 1). We recorded shrub cover along the midline of macro-plots using line intercept sampling techniques (Canfield, 1941). We placed (Daubenmire, 1959) frames (20 × 50 mm) at 5 m intervals along the line intercept to measure herbaceous cover. Grasses and forbs were estimated visually and placed into 12 cover classes (<1, 1–5, 6–15, 16–25, 26–35, 36–45, 46–55, 56–65, 66–75, 76–85, 86–95, >95%).

Data analyses

We calculated relative rodent abundance (hereafter, abundance) as the number of unique individuals captured per 100 trap nights. Analyses were done at the level of the transect. We determined transects to be independent because transects did not overlap in space or time and we did not recapture marked animals across transects. Species richness was the average number of species captured per transects per treatment. Mammal biomass was calculated using the mean mass for each individual (if an individual was encountered more than once during a three-night trap session, mass was averaged) and then summing the mass for each species on each transect. Simpson’s diversity index (Simpson, 1949) was calculated to examine rodent diversity between treatments. Occurrence was determined as presence/absence of each species per transect. Where data did not meet assumptions of normality, we utilized a non-parametric multivariate analysis, called non-parametric multi-response permutation procedures (MRPP; Biondini, Mielke & Berry, 1988), to investigate differences in rodent community attributes (i.e., abundance, biomass, and species richness) between the two treatments (AWS and CS). A Sidak correction was utilized to adjust for type I error across multiple MRPP tests (Abdi, 2007).

We summarized variation in vegetation between treatments using a principal component analysis (PCA), using IBM SPSS version 20 (IBM Corp, Armonk, NY, USA). PCA is a multivariate technique to reduce many correlated independent variables into a set of uncorrelated axes called principal components (Legendre & Legendre, 1998). To interpret each component of the PCA, we considered vegetation variables that loaded high (>0.500) in the component matrix. We used eigenvalues and scree plots, which are explained variances, to discriminate the relative importance of each component. Principal component scores and vegetation variables were compared between treatments using a Mann–Whitney rank sum test.

To explain species-habitat relationships, we used species’ occurrences as the response variables because species were not ubiquitous in the study area. We correlated occurrence with principal component scores (predictor variables) using logistic regression (Legendre & Legendre, 1998). Because vegetation was only measured during the spring (and not during the winter) and we wanted to relate mammals captured along transects to where vegetation was sampled, we only used mammal capture data from the spring in habitat models (n = 80).

Small mammals

During our study, we captured 370 individuals representing three genera and eight species of rodents across 4,800 total trap nights. The most common species encountered was the desert pocket mouse (Chaetodipus penicillatus, 198 captures). Other species encountered included the rock pocket mouse (C. intermedius, 67 captures), Bailey’s pocket mouse (C. baileyi, 42 captures), Merriam’s kangaroo rat (Dipodomys merriami, 28 captures), white-throated woodrat (Neotoma albigula, 22 captures), cactus mouse (Peromyscus eremicus, 11 captures), Arizona pocket mouse (Perognathus amplus, 1 capture), and Harris’s antelope squirrel (Ammospermophilus harrisii, 1 capture; Table A1). Cactus mouse, Arizona pocket mouse, and Harris’s antelope squirrel were excluded from species-habitat analyses due to their limited number of captures. The remaining five species represented more than 90 percent of captures and were included in analyses.

We did not detect seasonal (winter vs spring) differences in rodent community attributes (standard errors in parentheses, n = 160). Rodent abundance was 8.9 individuals per 100 trap nights (±1.0) in winter and 6.4 individuals per 100 trap nights (±0.8) in spring (MRPP, P = 0.054). Rodent biomass was 63.9 g (±7.8) in winter and 62.4 g (±11.1) in spring (MRPP, P = 0.306). Richness was 1.3 species (±0.1) in winter and 1.2 species (±0.1) in spring (MRPP, P = 0.567). However rodent community abundance and biomass differed between treatments (MRPP, P < 0.001 for both metrics); with abundance almost twice as high at AWS compared to CS (Table 1). Rodent diversity was similar between treatments with Simpson diversity indices of AWS and CS equal to 2.859 and 2.971, respectively. Similar species were encountered at both treatments; however, richness per trapping transect was greater at AWS compared to CS (MRPP, P < 0.001; Table 1). Nearly 40% of CS transects encountered either no animals or only a single species (Fig. 2).

Only two rodent species showed differences in abundance and biomass between AWS and CS. Desert pocket mouse abundance was greater at AWS (Table 2) and biomass at AWS was nearly twice that of CS (Fig. 3). Biomass of white-throated woodrat was over five times greater at AWS (Fig. 3).

Vegetation characteristics

We reduced 11 vegetation variables into five principal components which accounted for 91.1% of variation at the AWS and CS sites (Table A2). Variables associated with the presence of water (i.e., distance to AWS or distance to wash) did not explain a large percentage of variation and were not included in the final PCA. We interpreted principal component 1 as ground cover; principal component 2 as shrub cover; principal component 3 as tree overstory; principal component 4 as cactus density; and principal component 5 as shrub density (Table A2). Overall, AWS and CS were very similar in vegetation characteristics in terms of cover and vegetation density (Table 3).

Of the five species of rodents included in species-habitat analyses, three showed significant relationships with principal components (Table 4). Bailey’s pocket mouse occurrence was positively related to areas with higher cactus density (PC4). Merriam’s kangaroo rat occurrence was negatively related to areas with high tree and shrub density (PC5) and high tree cover (PC3). Rock pocket mouse occurrence was negatively influenced by greater herbaceous ground cover (PC1).

Management implications

The use of water as a management tool for endangered or game species is popular and has increased in recent history. Even with debate about its effectiveness as a management tool (Broyles, 1995) state, federal, and private agencies have allocated large sums of resources to install and maintain AWS. Over a decade ago, Arizona was spending $750,000 annually on AWS (Rosenstock, Ballard & Devos, 1999). The result of this study, even with a short sampling period and focus on organisms that experience fluctuating populations (Krebs & Myers, 1974), suggests that some species of wildlife may increase in abundance near AWS because of changes to vegetation, utilization of human-modified structures, or perhaps changes in food resources. Our trapping data may contribute to understanding patterns of small mammal use of AWS during a dry winter, that will likely become more common as the environment gets hotter and drier in the Southwest (Ye & Grimm, 2013). Combined with past research on AWS, our results will help managers make informed decisions about construction and maintenance of AWS as a management tool.