Timing of precipitation in an arid environment: Effects on population performance of a large herbivore

Abstract Climate models predict that shifts in temperature and precipitation patterns are likely to occur across the globe. Changing climate will likely have strong effects on arid environments as a result of increased temperatures, increasing frequency and intensity of droughts, and less consistent pulses of rainfall. Therefore, understanding the link between patterns of precipitation, temperature, and population performance of species occupying these environments will continue to increase in importance as climatic shifts occur within these natural ecosystems. We sought to evaluate how individual, maternal, population, and environmental, particularly temperature and precipitation, level factors influence population performance of a large herbivore in an arid environment. We used mule deer (Odocoileus hemionus) as a representative species and quantified juvenile survival to test hypotheses about effects of environmental factors on population performance. Precipitation events occurring in mid‐ to late‐pregnancy (January–April) leading to spring green‐up, as indexed by normalized difference in vegetation index, had the strongest positive effect on juvenile survival and recruitment. In addition, larger neonates had an increased probability of survival. Our findings indicate that timing and amount of precipitation prior to parturition have strong influences on maternal nutritional condition, which was passed on to young. These results have important implications for understanding how animal populations may benefit from timing of precipitation during spring and prior to parturition, especially in arid environments.


| 3355
HEFFELFINGER Et aL. Gregory, & Mitchell, 1997). Temporal variation in precipitation from the historical norm also is expected to occur in annual rainfall patterns (Karl & Knight, 1998). Concurrently, temperatures are expected to rise beyond the typical maximums already exhibited in these hot, dry environments (Mahlman, 1997;Meehl et al., 2000). Shifts in temporal variation of precipitation events coupled with higher temperatures are predicted to increase frequency and intensity of droughts in these ecosystems, which may have substantial effects on biota inhabiting, and already well-adapted to, these arid regions (Seager et al., 2007).
Arid environments frequently exhibit high ambient daytime temperatures, sporadic precipitation, and spatially or temporally sparse nutrients. Animals cope with high ambient temperatures through many anatomical, physiological, and behavioral processes, such as increased peripheral blood flow, long appendages, and various methods of evaporative cooling while minimizing water loss in arid environments (Cain, Krausman, Rosenstock, & Turner, 2006;Hales, 1973).
The limited amount of annual precipitation in arid environments likely has a strong influence on life history characteristics, physiology, and species composition of both plants and animals inhabiting those ecosystems (Chesson et al., 2004). Precipitation pulses not only provide animals with temporary, yet essential, sources of free-standing water, but also drive forage abundance and forage quality in these ecosystems (Marshal, Krausman, & Bleich, 2005). Vegetation in arid environments, especially in the Mojave Desert, has adapted to respond rapidly to rainfall with increased productivity (Ackerman & Bamberg, 1974).
Concurrently, animals have also adapted to high temporal variation of short-lived, rapidly responding nutritional resources and the availability of water from precipitation.
Large, herbivorous mammals can be found in almost all ecosystems ranging from the arctic tundra, to deserts, to the tropical rainforests (Nowak, 1999). In arid environments with widely distributed resources and lower rates of plant productivity, large herbivores require large home ranges to meet nutritional and, thus, energetic demands to maintain viable populations (Heffelfinger, 2006). In some areas, large herbivores expand their home ranges during the hot-dry seasons to include permanent water sources because of the lack of preformed water in biomass of forage plants (Marshal et al., 2006). Quality of forage plants in arid environments increases as a function of timing and amount of precipitation. Desert plants respond to pulses of precipitation by exhibiting increased growth, biomass, water content, crude protein content, and digestibility (Marshal et al., 2005). Thus, when plant production is high in arid environments, home range sizes may decrease with greater abundance of resources, resulting in decreased expenditure of energy because of the overall greater quality and quantity of available forage (Heffelfinger, 2006). Therefore, quantifying and understanding how environmental factors affect large mammal populations in an already arid ecosystem provide insight regarding the potential effects of a changing climate.
We used mule deer (Odocoileus hemionus) as a representative species to explore how environmental, population-level, and individual factors affect demographic rates of a large herbivore in an arid ecosystem. Mule deer occur throughout most of North America, including the arid ecosystems of the southwestern United States (Heffelfinger, 2006), illustrating a high degree of adaptability to widely disparate climatic conditions. This variation in habitat use allows investigators to evaluate population responses to many different environmental factors. Similar to other large herbivores, adult mule deer exhibit high and stable survival rates throughout most of their geographical range (Bender, Hoenes, & Rodden, 2012;Bishop et al., 2009;Hurley et al., 2011;Monteith et al., 2014). Given the importance of the influence of juvenile survival on population performance (Gaillard et al., 1998), our objectives were to evaluate and quantify factors influencing juvenile survival in an arid environment. We hypothesized that neonatal survival is strongly affected by fluctuating environmental variables frequently exhibited in desert ecosystems. Therefore, we focused on assessing the effects of amount and timing of precipitation in conjunction with vegetation green-up on juvenile survival.  Thorne et al., 1981).

| Study system
Mule deer in the Mojave Desert are similar in their reproductive biology to those studied in Round Valley, CA by Monteith et al. (2013).
The mating season occurs in the autumn, usually early November and trails off through late December (Heffelfinger, 2006;Wallmo, 1981).
Gestation in mule deer is about 200 days (Heffelfinger, 2006;Wallmo, 1981), and the period of parturition in our study area generally extends from mid-to-late May through June. Because resources are generally available during winter, and little to no snow is observed, overwinter mortality is generally low. Therefore, in this ecosystem, we consider those young that have survived to 120 days to be recruited into the population.

| Data collection
From 2013 to 2016, we captured adult female deer in late February or early March via a net gun fired from a helicopter (Krausman, Hervert, & Ordway, 1985). Captured deer were flown to a central processing station established for each study area where they were fitted with GPS (Global Positioning System) collars (Advanced Telemetry Systems, G2110D, Isanti, MN USA) and uniquely marked with ear tags. Collars were equipped with a mortality sensor and a Very High Frequency (VHF) transmitter. Additionally, collars were programmed to collect one GPS location every 90 min and to drop off approximately 1 year following deployment (McKee et al., 2015). We determined nutritional condition of adult females using standard protocols developed and validated for mule deer (Monteith et al., 2013). Those methods used measurement of subcutaneous fat on the rump using ultrasonography, as well as palpation at rump, ribs, and withers to determine a body condition score (Cook, Stephenson, Myers, Cook, & Shipley, 2007;Monteith et al., 2013;Stephenson, Bleich, Pierce, & Mulcahy, 2002). Pregnancy status also was determined using ultrasonography (Stephenson et al., 1995). Pregnant individuals were then fitted with a vaginal implant transmitter (VIT) that was equipped with both a temperature and a photo sensor (Advanced Telemetry Systems, M3930L; Bishop et al., 2007). Individuals captured nearby were then released from the central processing station or flown back to the original capture location if they were captured over 8 km from the processing station.
Vaginal implant transmitters used during this study were similar to those described by (Bishop et al., 2011). In addition to the temperature sensor traditionally equipped in VITs used in previous studies (Bishop et al., 2007;Carstensen, DelGiudice, & Sampson, 2003;Johnstone-Yellin, Shipley, & Myers, 2006), VITs in this study also had a programmed photo sensor. This design allowed for VIT expulsion to be detected at night or during the day despite high ambient temperatures in our study site. When a VIT was expelled, a preprogrammed Precise Event Timing (PET) coding was emitted once per minute within the VHF pulses to aid in the identification of expulsion time. The PET coding indicated the time since expulsion in 30-min increments up to 5 days (Advanced Telemetry Systems, 2017; Bush, 2015), thereby allowing age of the neonate to be calculated precisely.
For clarification, we use the term neonate to describe an offspring in the first week of life and the term young or juvenile for all ages thereafter. Ground crews monitored females outfitted with VITs from 1 May until parturition. Monitoring occasions were <3 days apart for each female (usually every day) to capture neonates as close to parturition as possible or to prevent neonate mortality occurring prior to capture. If PET coding (i.e., the neonate's age) was less than 3 hr, technicians would allow time for critical bonding, and colostrum transfers between neonate and mother before approaching (White, Knowlton, & Glazener, 1972). Technicians would systematically search the area using the location of the VIT or the female's location at initial sighting as the beginning of their search radius (Quintana et al., 2016). Search times were restricted to 30 min to reduce the chance of abandonment of the neonate or stress to the female (Livezey, 1990). When neonates were not located during an initial search, ground crews returned the next day to search for the neonate using the same method (Bush, 2015).
In 2013, two of our collared adult females were pregnant but did not receive a VIT because the birth canal was too narrow to successfully insert a VIT. Additionally, we had one adult each year thereafter that exhibited similar morphology. Females that did not receive a VIT were located and checked for parturition by systematically searching for a neonate every 1-3 days (Bush, 2015). In 2013, we monitored 13 adults that had been active collars from 2012 but did not receive a VIT in 2013, using the same method. Throughout the study we also captured neonates from un-collared individuals displaying signs of having a neonate nearby or opportunistically finding neonates while tracking or monitoring collared adults (Carstensen et al., 2003).
When a neonate was captured, it was immediately blindfolded, placed in a clean cloth bag and weighed to the nearest 0.1 kg using a spring scale (Pesola Scales, Baar, Switzerland). We measured new hoof growth to the nearest 0.1 mm with a digital caliper and measured chest girth and metatarsus to the nearest 0.1 cm. Sex, state of umbilicus, prominent vegetation type at the birth site, handling time, and a GPS location also were recorded. Finally, each neonate was fitted with an expandable VHF radio-collar (Advanced Telemetry Systems) with the mortality switch set to 6 hr of no movement, and released.
We used several different methods to reduce abandonment caused by handling. All technicians wore nitrile gloves during handling to reduce scent transfer, and all blindfolds and weigh bags were washed in scent-free detergent after each use and stored in scent-free bags during transport. Additionally, technicians were required to wash all clothes in scent-free detergent and to change clothes between capturing neonates from different mothers. We also stored the expandable collars in scent-free bags containing native vegetation to reduce the amount of non-native odors being transferred to the neonate (Livezey, 1990).
Age at capture for each neonate was determined using the PET coding recorded from the VIT. Nevertheless, age at capture had to be estimated for individuals caught from unmarked females or in instances where the VIT or PET coding malfunctioned. When monitoring collared females that did not receive a VIT or had a malfunctioning VIT, time and date were recorded on each occasion we encountered those females. We then used behavior of the neonate, condition of umbilicus, hoof appearance, and body size to estimate maximum possible age of those neonates (Haskell et al., 2007;Haugen & Speake, 1958;Monteith et al., 2014).
Survival of each neonate was monitored daily for the first week of life. Following the first week, each young was checked every 1-3 days until the month of August and weekly thereafter until they reached 120 days of age (4 months). We used 120 days as our measure of survival of young, because at 120 days young were completely weaned from mother and fully dependent on water sources and forage (Heffelfinger, 2006;Sadleir, 1980). All animal capture and handling procedures were approved by the Institutional Animal Care and Use Committee at the University of Nevada, Reno Protocol # 00058 and were within guidelines established by the American Society of Mammalogists for research on wild mammals (Sikes, 2016). We also complied with capture and handling procedures developed by California Department of Fish and Wildlife.

| Individual and environmental covariates
We tested several parameters that may have influenced survival of young in our study. First, we created an index of body size for each neonate at initial capture using chest girth and metatarsus in a principal components analysis (PCA; Dunteman, 1989; R 3.3.2, R Core Team).
Both morphological variables loaded similarly in the eigenvector for principal component 1 (PC1) 0.707, and PC1 had an eigenvalue of 1.549, which explained 77.4% of the variation in those morphometric measurements. Therefore, PC1 scores for each neonate were used as an index of body size. We performed a linear regression of body weight against the PC1 scores (R 2 = .45, p < .0001) with the residuals used as an index of body condition of young. Positive residuals were considered above average condition given its size, while negative residual values were indicative of poorer condition of neonates at initial capture (Bush, 2015;Jakob, Marshall, & Uetz, 1996;Schulte-Hostedde, Zinner, Millar, & Hickling, 2005).
We created a covariate representing timing of birth relative to the median parturition date for the given year of the study. We also included additional covariates that were descriptive of the neonate including sex, handling time during capture, and a binomial covariate indicating whether the neonate was one of a set of twins or triplets.
We also included covariates describing nutritional condition of the mother at capture, which included body condition score and maximum depth of rump fat from ultrasonography, described previously.
We created a series of covariates based on location data from the GPS radio-collar of the mother and location data from the birth site obtained from ground crews following capture of neonates. We considered birth site covariates including relative distance from water, distance from roads, and the amount of shrub cover at the birth site to determine if the spatial location of the birth site affected survival of young. We obtained vegetation layers including shrub cover (LANDFIRE, 2014) and used ArcGIS (Version 10.3) to extract values of relative shrub cover and distance to water that the female occupied while the young was known to be alive and with the mother. The series of extracted values were averaged to create individual covariates for shrub cover and distance to water prior to mortality or weaning of young (Table 1). All numerical covariates were converted to standard normal variables prior to model evaluation (Cooch & White, 2017;Zar, 2010).
We considered several environmental covariates relating to temperature, precipitation, and greenness of vegetation that may affect covariate was used to test how maximum temperature affected survival of young. We also directly tested how rainfall patterns influenced juvenile survival probabilities within our system, a priori. Data were retrieved from the Mid Hills Station operated by the Western Regional Climate Center, in the center of our Midhills study area, and were assumed to be representative of rainfall conditions of all our study areas.
We summed total rainfall within each calendar month. We then evaluated various combinations of monthly sums of total rainfall within and among the trimesters of pregnancy for mule deer in our study area.
We started with seasonal groupings of months identified by McKee et al. (2015). We then expanded our modeling to align with biologically meaningful time periods (i.e., pregnancy trimesters or typical rainfall patterns). Our a priori models of monthly rainfall during gestation showed that total rainfall from January through April best-explained juvenile survival over all other temporal models of biologically meaningful periods of rainfall (ΔAIC c = 1.61).

Lastly, we extracted normalized difference in vegetation index
(NDVI) values to further test for the general effect of precipitation on nutritional resources available to pregnant females and thus, juvenile survival in our analysis (Climate Engine, 2017). Within each study area, we randomly selected 10 points to collect NDVI values.
We were careful to exclude and resample points to minimize bias in NDVI values from nonfluctuating greenness in certain vegetation types (i.e., Pinyon-juniper) and effects of north-or south-facing slopes. None of our NDVI images were confounded with atmospheric obstructions and thus, we did not have to remove any of them prior to analyses. When rainfall patterns and NDVI were plotted together, our data indicated a 30-day lag in green-up following a rainfall pulse.

| Survival analysis
We assessed daily survival probabilities of young by study area and among years using the nest survival module in Program MARK (Version 8.0; White & Burnham, 1999;Cooch & White, 2017 We employed an iterative approach to model selection (Burnham & Anderson, 2002) and first created models to test for annual or study site differences in survival of young. Then, using the top model from this step, we tested various age trends that allowed survival to vary for each day, each week, and all possible combinations that potentially were biologically relevant. Trend models of this type require fewer parameters to estimate and evaluate whether neonate survival varied with age (Bishop et al., 2007).
After identifying the top model including a year-or site effect and the appropriate age trend, we began evaluating individual, maternal, population, spatial (birth site characteristics), and environmental covariates (temperature, precipitation, NDVI). We evaluated collinearity among those covariates using a correlation matrix (R 3.3.2 R Core Team). None of our covariates were correlated >0.65; therefore, we considered all predictor variables in candidate models. We realize the several of the variables (e.g., body size and NDVI) were estimates used in the survival analysis, and that in a maximum-likelihood framework, our estimated covariates may result in overly precise parameter estimates, and may have introduced some additional error into our analysis. At this point, this problem is difficult to resolve in a maximum-likelihood framework, but also likely results in increased variation among individuals. Therefore, we feel that our analysis and models are appropriate even if there is some increased variation among individuals, resulting from those estimated covariates in the survival analysis.
We employed a sequential model-building procedure to evaluate hypothesized biological processes (i.e., individual, maternal, population, spatial, and environmental factors) while minimizing the total number of models considered (Blomberg, Sedinger, Nonne, & Atamian, 2013). Precipitation data were available from only the single weather station in the Midhills; as a result, only annual variation could be tested. Additionally, NDVI data were constrained to study site and year, so they were treated as a site by year effect. Covariates for birth site characteristics were constrained to the first week of life because neonates become much more mobile and adults moved away from the vicinity of the birth sites thereafter (White et al., 1972).

| RESULTS
From 2013 to 2016, we captured 119 adult female mule deer throughout all of our study sites. Average body condition score for adult females during the study was 2.22 (SD = 0.45), and mean depth of rump fat was 0.14 cm (SD = 0.17;  Figures 2 and 3).

Back
An interaction between juvenile size and preparturition NDVI values was not supported in our modeling procedure (ß = 0.0067 95% CI = −0.2116 to 0.2250). Back-transformed estimates, however, can be used to evaluate the effect of both parameters simultaneously.
Estimated survival to weaning for a juvenile born at the smallest size in a year from a mother that was subjected to the lowest NDVI values was 0.38% (SE = 0.01), while a juvenile born at the largest size in a year from a mother exposed to the highest NDVI had an estimated survival

| DISCUSSION
Our evaluation of juvenile survival in a population of mule deer occupying an arid environment revealed the importance of preparturition nutritional resources on population performance. Our measured NDVI values for the space occupied by pregnant females from February to May, occurred during the 2nd and 3rd trimesters of pregnancy for most individuals, and had a strong effect on survival of young. The strong evidence that mid-pregnancy NDVI values positively influenced juvenile survival likely reflects forage quality and quantity available to the mothers following precipitation events in an arid environment (Marshal et al., 2005). Development of the fetus is at a crucial point during that time period because nutritional condition of the neonate at birth is directly correlated with available nutrients and condition of the female during mid-pregnancy (Hudson & Browman, 1959;Short, 1970;Thorne, Dean, & Hepworth, 1976). Thus, higher survival was exhibited by neonates born to mothers that likely had greater access to high quality and quantity of forage in mid-to-late pregnancy. Additionally, the nutritional intake by a pregnant female is directly related to the size of a neonate at birth and is considered a form of maternal allocation (Parker et al., 2009). Therefore, maternal females that were on a higher nutritional plane and were subjected to higher quality forage during gestation generally give birth to larger young. Our data also indicated that size of a neonate at birth positively influenced survival, and thus, could potentially be used as an index of its nutritional condition and the forage quality acquired by the mother during mid to late gestation. Those young that were born larger from females that may have had access to high quality resources exhibited higher survival to recruitment and thus, had a greater effect on the population performance in that ecosystem.   (Hurley et al., 2014).
Survival of adult females within large herbivore populations has the potential to affect population growth rate, but survival of adults is generally is high and tends to vary little in most populations (Gaillard & Yoccoz, 2003;Hurley et al., 2011;Unsworth, Pac, White, & Bartmann, 1999). Conversely, survival and recruitment of young are highly variable, which can strongly affect population growth (Gaillard et al., 1998 The estimated age trend was estimated daily survival to 7 days, weekly survival to 90 days, and then constant survival from 91 to 120 days at which young were considered weaned from mothers. survival in that ecosystem and indicates the possible transition from maternal condition passed on to young in early life stages and later; the direct effects of available forage after young are weaned. In our ecosystem, however, animals were not subjected to harsh winters with extreme temperatures or deep snow and their associated constraints on access to forage. Nevertheless, our study animals were subjected to high temperatures and low precipitation that characterize hot, dry regions, during summer. Therefore, large herbivores occupying an arid environment may be more reliant on increased forage quality during gestation that is allocated to the offspring rather than relying on nutritional sources postpartum to support young to recruitment. Females also experience the most nutritionally demanding period of their life cycle during lactation (Clutton-Brock, Albon, & Guinness, 1989), which coincides with high temperatures and low precipitation in our study ecosystem. Thus, nutritional sources that female is exposed to mid-pregnancy would be better suited to allocating to development of young thereby increasing the probability of survival and increasing maternal fitness.
Arid regions of the world are expected to become hotter and dryer with less predictable pulses of rainfall (Mahlman, 1997;Meehl et al., 2000). Hence, climate change has the potential to substantially alter population dynamics of large herbivores occupying these arid ecosystems. We have identified and quantified a potential link between maternal nutritional acquisition during pregnancy and survival of young. Precipitation events that lead to increases in forage quality and availability during mid-to late-pregnancy may prove to be the optimum timeframe for acquiring and allocating nutrients to young in arid environments. As droughts become more frequent and severe, large herbivore populations and all other biota likely will have to adjust to these altered climatic conditions. If rainfall patterns shift and spring rainfall events become less common in arid regions, populations may be forced to temporally shift their reproductive cycle or develop new strategies for resource acquisition. The inability to do so could drastically affect population performance of large herbivores and potentially other species that are reliant on landscape resources in desert regions. Understanding the link between environmental patterns and population performance has vast implications for comprehending how our natural ecosystems will react to the expected shifts in climate throughout our natural world.