Feral Burros and Other Influences on Desert Tortoise Presence in the Western Sonoran Desert

Globally, ownership and management of land, and decisions about uses, are subjects of discussion and conflict. In the United States, substantial land in the West (994,317 km²) is public land under the jurisdiction of the U.S. Department of the Interior's Bureau of Land Management (USBLM; USBLM 2018). Public land is managed for uses such as livestock grazing, mining, recreation, transportation and energy corridors and development, cultural resources, and wildlife (U.S. Congress 1976). Multiple-use management presents challenges to land managers when the requirements of one resource, such as a threatened species, conflict with other uses such as off-highway vehicle recreation (e.g., Bury and Luckenbach 2002).

One such case is the federally threatened and declining Agassiz's Desert Tortoise (Gopherus agassizii Cooper, hereafter desert tortoise), a species of the Mojave and western Sonoran deserts with ∼74% of habitat on federal lands (USFWS 1990, 1994). Over the past several decades, tortoise populations declined and habitats were lost or degraded from anthropogenic uses (Berry and Murphy 2019). From 2004 through 2014, adult tortoises declined ∼32% across the geographic range and by 2014, 76% of populations were below viability, defined as <3.9 adults/km² (USFWS 1994, 2015). Several predictive models for tortoise presence or survival, developed for parts of the Mojave Desert and associated with public lands, identified negative correlations with roads, recreational vehicle off-road use, livestock grazing, trash, denuded areas, predation, and proximity to human habitations (e.g., Bury and Luckenbach 2002; von Seckendorff Hoff and Marlow 2002; Kristan and Boarman 2003; Berry et al. 2014, 2015).

In the first recovery plan for the desert tortoise in 1994, the U.S. Fish and Wildlife Service (USFWS) identified the Chemehuevi population as the most robust and the most promising for recovery (USFWS 1994). This population is in the western Sonoran Desert and is relatively remote from cities, towns, and other human influences. The USFWS identified major threats to the Chemehuevi population as livestock grazing, including feral burros (Equus asinus), and mortality on roads. In 2009, we established a study area in the Chemehuevi Valley adjacent to a critical habitat unit for desert tortoises (USFWS 1994). Our ultimate objective was to identify positive and negative relationships between tortoise presence and the biotic, physical, and anthropogenic variables to understand and improve recovery efforts. We addressed four questions associated with tortoise presence, the environment, predators, and anthropogenic impacts: (1) how were tortoises distributed and what was the abundance? (2) did tortoises show evidence of diseases? (3) were tortoises associated with a vegetation type or major topographic feature, i.e., the Colorado River or Chemehuevi Wash? and (4) was tortoise presence positively or negatively correlated with any of several anthropogenic variables? For anthropogenic variables, we focused on those known to affect tortoises or likely to occur in the study area (e.g., feral burros, vehicle use).

The study area was on ∼93 km² of public lands administered by the USBLM within the eastern Chemehuevi Valley, San Bernardino County, California, in the western Sonoran Desert, 200–550 m above sea level (Fig. 1). The northern boundary was ∼1.4 km south of Havasu Lake Road, the eastern boundary was the Chemehuevi Reservation of the Chemehuevi Indian Tribe, the southern boundary was the Whipple Mountains Wilderness, and the western boundary abutted the Chemehuevi critical habitat unit for the tortoise (USFWS 2011). Most (93.6%) of the area was within the Chemehuevi Burro Herd Management Area (USBLM and CDFG 2002).

Perennial vegetation was composed of alliances of Creosote Bush (Larrea tridentata) with several species of shrubs: White Bursage (Ambrosia dumosa), Pima Rhatany (Krameria erecta), White Rhatany (Krameria bicolor), Brittlebush (Encelia farinosa), and several species of cactus and trees (CDFW 2018). Ephemeral stream channels supported Cheesebush (Ambrosia salsola), Desert Lavender (Condea emoryi), Sweetbush (Bebbia juncea), Smoke Tree (Psorothamnus spinosus), and Blue and Little-Leaved Palo Verde (Parkinsonia florida, Parkinsonia microphylla). Throughout, distribution of shrubs and trees was sparse, especially on desert pavements and between ephemeral stream channels. Cover and diversity of shrubs, trees, and grasses were highest in Chemehuevi Wash; trees and large shrubs were limited in secondary and tertiary ephemeral stream channels. Vegetation and other topographic features were comparable with the adjacent Chemehuevi critical habitat (Nussear et al. 2009). Plant nomenclature followed Jepson Flora Project (2018).

The western Sonoran or Colorado Desert is hotter and drier compared with the Mojave Desert to the north (Rowlands 1995). Desert tortoises depend on free water and food from ephemeral annual plants stimulated to germinate and grow by fall–winter and summer rains (Henen et al. 1998; Jennings and Berry 2015). The nearest weather station was at Lake Havasu City, Arizona, 9.6 km east, where the 30-yr means for annual rainfall for the hydrological year (1 October–30 September) and fall–winter rainfall (1 October—31 March) were 97.5 mm and 73.7 mm, respectively (National Climate Data Center, National Oceanic and Atmospheric Administration 1981–2010). Droughts occur when precipitation is below the mean for 1 October to 31 March. During the fall, winter, and early spring of 1 October 2007 to 31 March 2008, precipitation was close to the long-term mean (72.6 mm). In the months preceding the start of the survey (1 October 2008 to 31 March 2009), precipitation was 90.7 mm, 23.1% above the long-term annual mean for the period. Thus, fall–winter rainfall was at or above the mean in the years before and during the survey year.

Following techniques used in Berry et al. (2014, 2020), we created a boundary around the area of interest in the Chemehuevi Valley (excluding 5.2 km² of state-owned and privately owned lands) and then developed a randomly sampled distribution of 200 points minimally spaced 500 m apart using geographic information systems (GIS, Environmental Systems Research Institute, Inc. [ESRI], Redlands, CA). We used Hawth's Tools extension in ArcGIS v9.0 to establish 200 center-point locations for plots (Beyer 2004) and then expanded each center point to a 1-ha plot.

We conducted the survey in spring, 3 April–30 June 2009, the most propitious season for locating tortoises. In addition, because rainfall for the fall–winter hydrologic year (1 October 2008 through 31 March 2009) exceeded the long-term mean there was a profusion of annual wildflowers (tortoise forage; Zimmerman et al. 1994; Henen et al. 1998). Team members were experienced in locating live tortoises and tortoise sign; identifying annual plants, shrubs, and trees; and recording evidence of human activities. Canopy cover of shrubs and trees was sparse, enhancing visibility of sign. Using methods described in Berry et al. (2014, 2020), the field team surveyed the 200 1-ha plots twice by walking along transects spaced at ∼10-m intervals, usually during the same day or within a few days. They recorded all signs of tortoise presence (cover sites, scats, drinking sites, live and dead tortoises) within each plot.

Vegetation.—For each plot, the field team rated abundance of perennial plants occurring on each plot numerically, in descending order, as abundant (5), common (4), sparse (3), rare (2), represented by one to two individuals (1), or absent (0; see Jepson Flora Project 2018 for definitions). They also noted abundance of the nonnative invading annual, Sahara or African Mustard, Brassica tournefortii (Minnich and Sanders 2000).

Live and dead tortoises and tortoise sign.—Because we anticipated low densities of tortoises, we processed every live and dead tortoise observed using standard protocols (Berry and Christopher 2001). We took measurements of size, identified sex, and examined tortoises for clinical signs of infectious diseases (e.g., upper respiratory tract disease caused by Mycoplasma agassizii, Mycoplasma testudineum, and herpesvirus (Jacobson et al. 2012, 2014)), as well as shell lesions (cutaneous dyskeratosis, shell necrosis), other diseases (e.g., urolithiasis, gout), and trauma (Jacobson et al. 1994; Homer et al. 1998). We processed dead tortoises and shell–skeletal remains following Berry and Woodman (1984). We recorded other signs of tortoises as well (burrows, rock shelters, pallets and caves, scat, tracks, drinking sites, courtship rings, and combat encounter sites; e.g., Keith et al. 2008; Berry et al. 2014).

Predators and anthropogenic factors.—For each plot, the field crew recorded numbers of scats, marking sites, and dens of mammalian predators, as well as species and numbers of avian predators. They counted trash, balloons, casings from firearms and shooting targets, and survey markers. They measured surface areas (m²) disturbed by dirt roads and tracks and trails created by four-wheel vehicles, motorcycles, and other vehicles; feral burro scats, trails, and wallows; utility corridors; and mining pits and excavations.

Vegetation.—We performed k-means clustering analysis using R v3.4.3 (R Core Team 2017) on perennial plant data for trees, shrubs, herbs, cacti, grasses, and vines using the six ordinal categories of abundance to categorize plots by vegetation association (also described in Berry et al. 2014, 2020). We specified k = 3 clusters for the analysis. Then we verified that the three clusters corresponded to recognized natural vegetation associations by evaluating composition, relative abundance, and diversity within each (see Supplemental Data S1 in the Supplemental Materials available online). We compared each cluster to the list of natural communities and assigned a name to the vegetation association (CDFW 2018). Next, to characterize them, we conducted an analysis of variance of the vegetation associations using elevation as the dependent variable. Within each cluster, we considered species occurring on ≥70% of plots as common species and calculated the Shannon diversity index for each cluster (Oksanen et al. 2018). We also included African Mustard (presence/absence) in models.

Live and dead tortoises.—Following Berry and Christopher (2001), we assigned size–age classes to live and dead tortoises by carapace length in millimeters at the midline (MCL): juvenile, <99 mm; immature, 100–179 mm, and adult, ≥180 mm. Because numbers of live tortoises on plots were low, we used the bootstrap method to generate confidence intervals for tortoise density estimates. We calculated densities per square kilometer of all live tortoises and of adults only by using 20,000 iterations to compute the mean and 95% confidence interval (CI). For the CI computation, we applied the bias-corrected accelerated method (Davison and Hinkley 1997; Canty and Ripley 2017; R Core Team 2017).

For shell–skeletal remains, we determined MCL and sex, estimated time since death, and, where possible, assigned cause of or contributors to death (Berry and Woodman 1984). Using keys, we categorized time since death into two classes: ≤4 yr and >4 yr. We calculated the annualized death rate for the previous 4 yr for on-plot adult tortoises as 1 – (1 – D/N)^0.25, where D was the number of adults dead ≤ 4 yr and N is the sum of D and the number of live adult tortoises. To assign a cause of or contributors to death, we evaluated location, forensic evidence, and general appearance left by Common Ravens, Corvus corax (Boarman 1993), mammalian predators (Berry et al. 2013), vehicles (Homer et al. 1998), or gunshot (Berry 1986).

Models of tortoise presence.—We used all live and dead tortoises and sign counted on a plot as a surrogate for tortoise presence because a strong positive relationship exists between counts of live tortoises and their sign, such as cover sites and scats (Krzysik 2002). Using generalized linear models (GLMs), we examined the relationship between tortoise presence and the native perennial vegetation associations (generated by k-means clustering), African Mustard, mammalian predators, anthropogenic impacts, and topographic features. We treated tortoise presence as the binary response indicator variable where absence was designated categorically by 0 for each plot and presence by 1, which we modeled as a binomial distribution with a logit link function. We treated presence and absence of African Mustard and mammalian predators (hereafter Mammals) as indicator variables.

We acquired several GIS layers from the U.S. Census Bureau (human population census blocks, roads; available at http://www.census.gov/geo/www/tiger/), USBLM (land ownership, rights of way, vehicle routes; available at http://www.blm.gov/ca/gis/), and U.S. Geological Survey (USGS 2013). We used the Near tool in ArcGIS v10.2 (ESRI, Inc., Redlands, CA) to calculate distances from the nearest corner of each plot to: (1) the nearest location along the bank of the Colorado River; (2) the center-line features representing the axial valley wash, Chemehuevi Wash; (3) paved and dirt roads; and (4) the center of human-populated census block polygons (hereafter river, wash, road, and houses, respectively).

For each plot, we created a variable for total surface disturbance by summing areas (m²) of all disturbances measured during the field collection of data: feral burro trails, wallows, and scat; dirt roads, old and recent vehicle trails, tracks, and routes; mining operations; utility corridors; and evidence of shooting. Similarly, we created a variable (total burro disturbance) by summing areas of burro trails, wallows, and scat, and created a variable (total vehicle disturbance) by summing areas of old and recent vehicle routes, trails, and tracks. We inspected the histograms of all continuous predictor variables for the presence of outliers. When outliers occurred, we reviewed data records for accuracy and log₁₀-transformed the data (see below). We calculated Spearman's rank correlation among all disturbance variables (burro trails; burro scat; vehicle tracks—recent; vehicle tracks—old), distances to features (river, wash, road) and counts of trash and balloons, and ordinal classification of vegetation clusters (vegetation association), ordered from lowest to highest mean elevation. We excluded separate variables for ravens, mammals, firearms (e.g., casings, shells, and targets), mines, the utility corridor, and survey markers from statistical analysis because they were each present on <20 plots (<10% of total plots). We conducted all statistical analyses using R statistical software v3.4.3 (R Core Team 2017).

For predictors of tortoise presence in GLMs, we log₁₀transformed the areas covered by burro trails, burro scat, total burro disturbance, vehicle tracks—recent and vehicle tracks—old, total vehicle disturbance, and total surface disturbance to normalize their distributions and mitigate excessive leverage of unusually large values. We also log₁₀transformed counts of balloons and trash. Before transforming each variable, we replaced zero values with 1/10th the minimum nonzero value of that variable. Additional predictors included a categorical classification of vegetation association, presence–absence indicator for African Mustard, and distances to river, wash, roads, and houses.

To address issues of concern for wildlife and land-use managers, we modeled the relationship between tortoise presence and predictor variables using multiple models based on almost all combinations of predictors, except we never combined variable totals with any of its constituent parts. For example, total burro disturbance and burro trails were never in the same model. This restriction led predictors to occur in unequal numbers of models. We used Akaike's information criterion corrected for small sample size (Burnham and Anderson 2002) to perform a preliminary comparison of models to determine whether variable totals (total burro disturbance, total vehicle tracks, or total surface disturbance) or their constituent parts (e.g., burro trails, burro scat, vehicle tracks—recent, vehicle tracks—old) made better predictors of tortoise presence. We subsequently removed models with variable totals from further analysis, leaving 12 predictors evenly represented across 4096 unrestricted model combinations. We evaluated variable importance for each predictor by summing Akaike proportional weights across models containing that predictor and considered that predictor important when its Akaike weight exceeded the proportion of models containing it, i.e., 0.5.

We predicted spatially explicit probabilities of tortoise presence by applying every model to the predictor variables measured in every plot and then taking the weighted average of predictions across all models according to their Akaike weights (Mazerolle 2017). To develop a spatial map of the probability of tortoise presence, we conducted a kriging analysis in Geostatistical Analyst to interpolate estimates of the probability of tortoise presence derived from model averages for each plot (ArcMap v10.5.1, ESRI, Inc.). We selected ordinary kriging with a probability surface output and threshold value of zero and allowed ArcMap to generate the layers using the program defaults for semivariogram, nugget, and lag size/number. Finally, we calculated Spearman's rank correlations between the model-averaged probabilities of tortoise presence and predictor variables, then constructed a path diagram showing relationships between correlated variables.

We identified 29 species of shrubs, 8 species of cacti, 4 species of trees, 5 species of grasses, and many species of herbs. We assigned each plot to one of three vegetation associations on the basis of perennial shrubs, trees, and grasses: (1) Ambrosia dumosa (hereafter White Bursage) with two abundant species; (2) K. erecta, K. bicolor (hereafter Rhatany) with six abundant species; and (3) E. farinosa (hereafter Brittlebush) with nine abundant species (Table 1, Fig. 2; Supplemental Data S1). Creosote Bush (L. tridentata) and A. dumosa were the top two species occurring on plots within each vegetation association. Vegetation associations differed in elevations (F_2,197 = 43.68, P < 0.001), with Brittlebush occurring at higher elevations than the other two vegetation associations. Diversity was highest in Brittlebush (2.87), followed by White Bursage (2.60) and Rhatany (2.48). African Mustard occurred on 50% (100/200) of plots and higher counts occurred on plots in the northern part of the study area. Annual wildflowers from the fall–winter rains were abundant.

The field team observed five live tortoises—four adults and one immature—each on a different plot; four additional adult tortoises were observed off plots. The relative ages of all adults (based on growth rings, wear, and condition of bones and scutes) included young, middle-aged, and old individuals. On-plot density estimates (± SE) were 2.5 tortoises ± 1.1/km² (95% CI = 0.5–5.0 tortoises/km²) for all sizes of tortoises and 2.0 tortoises ± 1.0/ km² (95% CI = 0.5–4.0 tortoises/km²) for adults only.

We conducted health evaluations for six of the nine tortoises, obtained partial observations for a seventh tortoise, and were unable to extract two tortoises from burrows because air and substrate temperatures exceeded the maximum allowed for handling tortoises under our USFWS permit no. TE 006556–14. All six tortoises showed signs of recent foraging; beaks and faces were matted with plant parts or sap. In all tortoises, crusts from plant sap or ocular discharge were evident in the periocular area. The patency of nares ranged from open (one tortoise) to partially or severely occluded (five tortoises) with colored plant materials, coupled with dust or dirt (or both). The condition of the face, eyes, and nares—typical of a spring with abundant wildflowers (and forage)—inhibited interpretation of potential clinical signs of upper respiratory tract disease. No tortoise was lethargic or debilitated; three tortoises had evidence of injuries from attack(s) by a mammalian predator, and a fourth had a scarred right foreleg. Three tortoises showed mild clinical signs of cutaneous dyskeratosis or another disease of the integument, e.g., shell necrosis.

Nine shell–skeletal remains occurred on plots and seven remains were observed off plot. Within plots, all remains were of adults: three adults had died ≤ 3 yr, whereas the other six died > 4 yr before our study. Mammalian predators probably killed or scavenged the three recently dead tortoises, evidenced by chew and gnaw marks and broken scutes and bones. Mammalian predators left scat on or adjacent to four shell–skeletal remains: kit fox scat was on the upturned plastron of one dead tortoise and mixed with two bobcat scats on another set of remains. A coyote scat was intermingled with a third set of remains. Burro scats were intermingled with or occurred within 5 m of four remains. The annualized death rate was 1 – (1 – 0.429)^0.25 or 13.1%/yr for the 4 yr before our study. Of the seven remains observed off plots, six were of adults, dead > 4 yr before our study, and one was a juvenile that died within a year of the survey, with signs of death from a Common Raven.

We observed 156 tortoise signs. Overall, 22.0% of plots (44/200) had ≥1 sign. More sign of all types occurred in the southern part of the study area, and the probability of observing sign was ≥0.939 on 33.74 km² of the 93-km² study area, or 36.3% of the study area (Fig. 3). More plots with tortoise sign (30.0%, 18/60) occurred on plots in Brittlebush than on Rhatany (25.0%, 18/72) and White Bursage vegetation associations (11.8%, 8/68). Scats (55.9%) and cover sites (32.4%) were more frequent than other forms of sign. Of 95 scats observed, 93.7% were deposited <1 yr before the study. Of 55 cover sites observed, 98.2% were in good, usable condition, and 11 (20.0%) showed evidence of use during the survey. We also recorded five drinking sites (one tortoise was observed drinking) and one set of tracks.

We observed three species of avian predators on plots: Loggerhead Shrikes (Lanius ludovicianus, n = 9), Red-Tailed Hawks (Buteo jamaicensis, n = 3), and Great Horned Owls (Bubo virginianus, n = 3). Additional avian species observed near or adjacent to plots were Common Ravens, Burrowing Owls (Athene cunicularia), Northern Harriers (Circus cyaneus), and Greater Roadrunners (Geococcyx californianus). Most Common Ravens were within 1 km of the power-line road (Fig. 1). We detected dens and marking sites of mammalian predators on 10 plots from Kit Foxes (Vulpes macrotis), Coyotes (Canis latrans), canids (species undetermined), and either canids or American Badgers (Taxidea taxus).

Most plots (91.0%, 182/200) had one or more human-related impacts, with feral burro scat the most prevalent (Table 2). Other impacts in descending order of prevalence were vehicle tracks—recent, trash, burro trails, and vehicle tracks—old. Other types of use occurred on ≤16.5% of plots. Total surface disturbance/plot ranged from 0 to 5450 m² or 54.5% of the surface area of a plot. More important, many vehicle tracks were off designated routes of travel and were unauthorized. Vehicle tracks also formed the majority of total surface disturbance.

The path diagram of Spearman's rank correlations between tortoise presence and anthropogenic, topographic, and vegetation variables illustrates the correlations that were ∣r∣ > 0.4 and P < 0.05; pairs with small correlations were excluded (∣r∣ <0.4; Fig. 4). Tortoise presence correlated positively with vegetation association (ordered from lowest to highest elevation), particularly Brittlebush, which had the highest mean elevation. The positive correlation between tortoise presence and the tetrad of distances to the Colorado River, Chemehuevi Wash, roads, and houses followed the pattern of elevation and vegetation association. In contrast, tortoise presence was negatively correlated with burro scat and trash.

Four variables exceeded a weight of 0.5 in variable importance, suggesting that models with these variables were more likely than those without (Table 3). The top two variables predicting presence of tortoise sign in the models were burros (burro scat) and vegetation association, with weights of 0.84 and 0.77, respectively (Table 3; Supplemental Data S1; Supplemental Data S2 in the Supplemental Material available online). Brittlebush, the vegetation association with the most tortoise sign and the highest mean elevation, was more distant from the Colorado River than other vegetation associations (Fig. 2). Trash and distance to Colorado River were also somewhat important, with importance weights of 0.58 and 0.51, respectively.

The findings of our analyses are correlative. Under ideal conditions, research on the effects of anthropogenic uses to a species would be designed as before–after–impact–control or involve experimental studies of cause and effect. Unfortunately, opportunities for such studies of wild tortoises are limited or don't exist because (1) historical baselines for tortoise demography, characteristics of vegetation, and anthropogenic uses are limited or unavailable; (2) multiple anthropogenic uses occur in many areas and have different histories, thereby creating difficulties in discerning which uses have the more negative or positive effects; (3) tortoise population densities are very low and declining, limiting robust sample sizes; and (4) some types of experiments are unlikely to be approved by permitting agencies because the tortoise is a threatened species. As a result, we were limited in the types of research designs and analyses of multiple variables.

Few tortoises and sign occurred on plots, despite excellent conditions for observations. Our predictive models revealed a higher probability of observing tortoises and sign primarily on the western and southern edges of the study area, adjacent to critical habitat and wilderness (Figs. 1, 3). Throughout the study area, adult tortoise densities were very low but comparable with figures reported by the USFWS using distance sampling (USFWS 2015). In the decade between 2004 and 2014, adult populations declined 64.7% in the adjacent Chemehuevi critical habitat unit, resulting in a density of 2.8 tortoises/ km² (SE = 1.1) in 2014. The rate of decline in Chemehuevi was two times higher than the average recorded throughout the geographic range of the species (USFWS 2015). The 2009 density of adults in our study (2.0 tortoises ± 1.0/ km², 95% CI = 0.5–4.0) was so low and the mortality rate so high that the population was probably nonviable (USFWS 1994). Survival of prereproductive tortoises was limited: only one live and one dead tortoise were observed. We found no evidence of moderate to severe clinical signs of disease that might have contributed to recent high mortality rates. However, trauma, most likely from predators or scavengers, was evident on most live tortoises and remains. Because tortoise numbers were so low, the loss of even a few individuals—especially adults—to predation likely adds to downward trends (see Kristan and Boarman 2003; Esque et al. 2010). Predation by carnivores is often coupled with drought (Esque et al. 2010). Drought alone can contribute to an increase in deaths (e.g., Turner et al. 1984; Berry et al. 2002). The 2009 figures and 10-yr decline summarized by the USFWS (2015) for critical habitat are in sharp contrast to the 1988 densities of 86 tortoises (all sizes)/km² on a long-term plot 12.8 km to the west, also in critical habitat (USFWS 1994:F1).

Two apparent drivers of tortoise presence in 2009, based on variable importance, were burros and vegetation (Table 3). The rank correlations between variables revealed that tortoise presence increased with increasing distances from the Colorado River, houses, and Chemehuevi Wash, and decreased with increasing area of burro scat, the metric for burro use and activity. The relationships between burros and the Colorado River and Chemehuevi Wash relate to the dependence of burros on water, shade during the heat, and forage. Burros change use of vegetation type and location in the Chemehuevi Valley seasonally. From July to October they spend most of their time in major ephemeral stream channels, e.g., Chemehuevi Wash and tributaries (secondary washes), and from January through May they are on the bajadas (Woodward and Ohmart 1976). They use major washes to travel to the Colorado River for water at approximately 3-d intervals. Hanley and Brady (1977) measured browse utilization at distances up to 5.0 km from Lake Havasu in the Colorado River and reported greater utilization closer to the Colorado River than 5.0 km away. They also noted that overgrazing affected secondary washes more than the primary washes (Chemehuevi Wash) and open desert. The results of our study, conducted >30 yr later, indicated that burros traveled ≥12 km from the Colorado River, probably in search of preferred foods. Tortoise sign was also more common at greater distances from the Colorado River in Brittlebush, the vegetation association with higher mean elevation and thus more common on the western and southern parts of the study area. Tortoise sign was rare in White Bursage, the vegetation association prevalent near the river where burros come to water (Woodward and Ohmart 1976).

The Chemehuevi Valley has a long-standing designated herd management area that encompasses and extends beyond study area boundaries (USBLM 1980). As early as 1957, the Colorado River was identified as a focal point for several concentrations of burros (McKnight 1958; Abella 2008). In 1980 the USBLM estimated a population of 1200 burros (USBLM 1980) and set 108 burros as the appropriate management level (AML) for the herd. The AML is the level “at which the wild horse and burro populations are consistent with. . . the land's capacity to support them and other mandated uses. . ., including protecting ecological processes and habitat for wildlife and livestock” (USBLM 2020a).

Feral burros have high fecundity and recruitment (Woodward and Ohmart 1976; Wolfe et al. 1989; Garrott 2018). Populations increase rapidly, frequently exceeding the designated AML, requiring periodic removal of multiple burros. In September of 2009, the year of our survey, the USBLM gathered and removed 67 burros from the herd, leaving an estimated 119 burros within the herd (R. Pawelek, personal communication). Between summer of 2011 and 2017, the USBLM conducted counts and removed some burros but in insufficient numbers to keep the herd at 108 animals. By 2017, 970 burros were in the herd, almost nine times above the AML (USBLM 2020a).

Management of rapidly increasing wild burro and horse populations on public lands is a major challenge for the USBLM, in part because a segment of the public is not supportive of reducing herds (U.S. Congress 1971; Santini 1981; USBLM 2020a). By 2017, wild herds exceeded AMLs by threefold in the West, and by 2018, wild horses and burros kept in off-range holding facilities reached 46,000 individuals. Growth of wild populations has continued unchecked. According to a report from the USBLM to Congress, numbers have increased “because between 2007 and 2014, removals had to be kept low in order to avoid off-range costs from spiraling out of control” (USBLM 2020b:3). Further, “program funding. . .limits. . .how many animals can be removed and held” (USBLM 2020b:4). Approximately 62% of the agency's annual budget for the program goes to care of horses and burros kept in holding facilities (Garrott and Oli 2013; USBLM 2020a). Throughout the West, overpopulation by burros and horses resulted in deterioration of range and water sources (Beschta et al. 2013; USBLM 2020a). The combinations of overpopulation, drought, and lack of food and water have led to dehydration, starvation, and deaths of horses and burros in some areas, e.g., Garrott (2018).

Feral burros, like other livestock, can have negative effects on tortoises through overlap of forage species, trampling of tortoises and burrows, long-term changes in composition and structure of vegetation, and disturbance to the substrate (e.g., Avery and Neibergs 1997; Keith et al. 2008; Berry et al. 2014; Tuma et al. 2016). Burro tracks and trails degrade tortoise habitat (e.g., Ostermann-Kelm et al. 2009). Burros browse on shrubs that are important sources of protective cover for the tortoises from extremes of temperatures and predators: White Bursage (A. dumosa), Blue Palo Verde (P. florida), Cheesebush (A. salsola), and Rhatany (K. erecta; Burge 1978; Berry and Turner 1986). Feral burros grazed heavily on White Bursage (Woodward and Ohmart 1976; Hanley and Brady 1977; Bonham and Brown 2002). In a study of burros near the Colorado River, Hanley and Brady (1977) reported that the most pronounced impacts to vegetation in secondary washes reduced density and canopy cover of almost all shrub and tree species. Abella (2008) summarized the diet, which consisted of forbs, grasses, and shrubs. Several of the more abundant species in the diet were also among food sources for the desert tortoise: Indian Wheat (Plantago ovata), Globemallow (Sphaeralcea ambigua), and Bush Muhly (Muhlenbergia porteri; e.g., Oftedal 2002; Abella 2008).

Rapidly growing and expanding populations of feral burros and horses are management challenges elsewhere in the United States and the world (e.g., Carrion et al. 2007; Taggart 2008; Nimmo 2018). Negative impacts to the related giant tortoise (Geochelone elephantopus vandenburghi; now Chelonoidis vicina Günther) from burros were documented on the Galápagos Islands (Fowler de Neira and Roe 1984; Fowler de Neira and Johnson 1985). Trampling or disturbance to 18.2% of giant tortoise nests reduced hatching success (Fowler de Neira and Roe 1984). The diets of giant tortoises and feral burros shared 72% of plants in common and one genus was a staple for both species (Fowler de Neira and Johnson 1985). The severity of negative impacts to giant tortoises and other rare and endangered species on the Galápagos Islands led to removal of feral burros from several islands (Carrion et al. 2007).

Although we focused attention on the two top importance variables, trash also deserves mention because of the negative rank correlation (r = –0.40) and the importance value (0.576). Negative relationships between trash and desert tortoises were described in other studies (e.g., Berry et al. 2006, 2020; Keith et al. 2008). Trash indicates human activity, is often associated with vehicle routes and unauthorized cross-country travel, and can attract predators (Berry et al. 2014). Tortoises eat trash and balloons, with fatal consequences (Donoghue 2006; Walde et al. 2007). The other variables were not unimportant (e.g., vehicle use, firearms, shooting; Berry 1986; Berry et al. 2014), but we did not find enough samples on plots to study the effects.

With this study, we provided new information on negative correlations between feral burro activity and presence of tortoises in the western Sonoran Desert. Feral burros (and horses) present major challenges to land-use managers. Highly desirable would be new information on grazing pressure on the native flora, e.g., whether shrubs are overutilized and, if so, how far from the Colorado River. Use of the native annual flora is also a concern. If feral burros are expanding westward into desert tortoise critical habitat, both a reduction in the AML for the Chemehuevi herd and more frequent reductions in herd size could be effective as part of adaptive management strategies. In the first recovery plan, the USFWS (1994) recommended that the Chemehuevi herd be managed at a zero-population level, but that recommendation did not result in a change in management. In addition, climate change and the forecast of extended periods of drought in the American Southwest support reevaluation of AMLs for herds in the desert regions to meet long-term objectives of protecting natural resources, including the tortoise (Beschta et al. 2013; Garfin et al. 2014; Garrott 2018).

We thank K. Anderson, T. Bailey, S. Boisvert, C. Furman, K. Kermoian, P. Kermoian, R. McGuire, and T. Ose for field assistance; N. Newman and H. Schneider for preliminary analyses; and W. Perry and M. Tuma for maps. We thank S. Abella, S. Jones, and two anonymous reviewers for helpful comments. The authors have no competing or conflicts of interests. The USBLM and California Department of Parks and Recreation provided funds through an interagency agreement to the U.S. Geological Survey for fieldwork and initial analyses and reporting. The U.S. Geological Survey supported analyses and manuscript preparation. The funding sources had no role in the study design; collection, analysis, and interpretation of data; writing the manuscript; and selecting the journal. The funding sources encouraged publication with no requirements other than high-quality science. Permits for handling tortoises were to KHB from the California Department of Fish and Game (Memorandum of Understanding, SC–003623), and USFWS (TE–06656–14). The U.S. Geological Survey's Animal Care and Use Committee approved the study plan. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Supplemental material associated with this article can be found online at https://doi.org/10.1655/10.1655/Herpetologica-D-20-00023.S1; https://doi.org/10.1655/10.1655/Herpetologica-D-20-00023.S2.