Timing outweighs amount of rainfall in shaping population dynamics

Climate variability has been widely documented to have bottom-up effects on the population dynamics of animals 1,2 , but the mechanisms underlying these effects have been rarely investigated through �eld manipulative experiments that control for confounding factors 3 . Here, we examined the effects of different rainfall patterns (i.e. timing and amount) on the population size of Brandt’s voles Lasiopodomys brandtii in semi-arid steppe grassland in Inner-Mongolia by conducting a 10-year (2010-2019) rainfall manipulation experiment in twelve 0.48 ha �eld enclosures. We found that moderate rainfall increase during the early rather than late growing season drove marked increases in population size through increasing the biomass of preferred plant species, whereas heavily increased rainfall produced no further increase in vole population growth. The increase in vole population size was more coupled with increased reproduction of overwintered voles and increased body mass of young-of-year than with better survival. Our results provide the �rst experimental evidence for the bottom-up effects of changing rainfall on the population growth of small mammals, and highlight the importance of rainfall timing on the population dynamics of wildlife in the steppe grassland environment.


Main Text
Changes in rainfall regime are occurring worldwide 4,5 and have substantial consequences for vertebrate population dynamics, the species composition of communities and ecosystem functions and services 6,7 .
Both timing and amount of rainfall are recognized as major components that contribute synergistically to the population dynamics of wildlife through shaping plant primary productivity 1,2 .However, it remains unclear whether changes in timing or amount of rainfall play the more dominant role in bottom-up regulation processes, despite their distinct and different effects on aboveground annual net primary productivity 8 .It is important to disentangle the independent effects of rainfall timing and amount to predict responses of species' populations and ecosystems to global climate change scenarios.
Three main factors interact to in uence the dynamics of species' populations: bottom-up regulation by rainfall 1 , top-down control by predators 9 , and interspeci c competition 10 .Among rodents, for example, rainfall is well recognized to induce a bottom-up increase in abundance via increasing food availability, as observed in Phyllotis darwini and Octodon degus in South America 11,12 , Pseudomys hermannsburgensis and Mus domesticus in Australia 13,14 , Spermophilus dauricus 15 and Cricetulus barabensis 16 in East Asia, Dipodomys merriami in North America 17 and Mastomys natalensis in Africa 18 .However, these observations are all based on the correlation between rodent abundance and precipitation; the mechanism underlying the bottom-up effects of precipitation on rodents through plant productivity is often assumed but has been rarely investigated (but see ref. 19 ).In addition, some studies have found neutral [19][20][21] or even negative [22][23][24] effects of rainfall on rodent abundance.These contradictory ndings may be ascribed to the confounding effects of biotic factors (e.g.predation and interspeci c competition) and abiotic factors (e.g. the ooding of burrows) in natural environments 2 .To fully understand the bottom-up effects of rainfall on the population dynamics of target species, including the effects of rainfall amount and timing, it is therefore necessary to exclude or control for confounding factors.
Manipulative experiments provide a means to do this, but are very challenging even for species such as small rodents owing to the need for large-scale enclosures that prevent immigration/emigration of individuals and disturbance by predators.
We conducted a 10-year, large-scale, manipulative experiment to examine the bottom-up effects of changes in rainfall regime (including timing and amount; Fig. S1) on the population dynamics of Brandt's voles Lasiopodomys brandtii and the relative contribution of vital rates (i.e., survival, reproduction, and body growth) to population growth.In our study region in Inner Mongolia, an increase in annual rainfall, especially during the early growing season, can markedly enhance annual net primary productivity 25 , with more rain in particular increasing the biomass of rye grass Leymus chinensis 26,27 , a major and favored food source for Brandt's vole 28 .Additional rainfall in the early growing season can provide a match between peak food resources and peak food requirements of young voles.Therefore, we hypothesized that rainfall would increase the population density of voles by enhancing the biomass of their preferred food species, and that rainfall increase in the early growing season would be of vital importance in triggering population increases, or outbreaks, of voles in arid steppe grassland.
We captured and marked a total of 18,452 Brandt's voles from 2010 to 2019.The population density of voles uctuated dramatically between years (ranging from 59 to 667 voles/ha).The rainfall increase treatments showed various effects on the population density of Brandt's voles, depending upon the timing and amount of increased rainfall (Fig. 1a).In Phase 1 (2010-2015), the experimentally increased amount of rainfall was 50 mm and 100 mm (averaging 18% and 36% increase compared to the natural rainfall amount; termed the R50 and R100 treatments respectively; Table S1) and was evenly distributed throughout the growing season (Table S1).We found that allocating 50 mm and 100 mm evenly through the growing season to enclosures had no signi cant effect on either the biomass of preferred plant species or the population abundance of voles (Fig. 1a, 2a, Table S2).The high evaporation rate in the semi-arid study region could reduce the water-use e ciency of plants and weaken the positive effect of rainfall on food resources for voles, especially when the amount of rainfall increase is relatively low 29 .This nding is similar to those of Carrier and Krebs (2002) who conducted a 5-yr rainfall manipulation in the Yukon Territory without varying the timing and magnitude of rainfall manipulation and not excluding predators and interspeci c competition.They found little effect of rainfall increase on food resources for the boreal red-backed vole Clethrionomys rutilus 19 .The lack of a signi cant response of food availability to low rainfall increases may be the reason for the observed neutral effect of rainfall on the population density of L. brandtii and other small rodent.
Previous studies have shown that increased rainfall can induce a sharp increase in the biomass of L. chinensis (main summer food for Brandt's voles) 27,28 .Due to the high content of fructose and oligosaccharides in L. chinensis, increasing the intake of L. chinensis would alter the composition of the voles' gut microbiota, which could in turn increase the production of short-chain fatty acids, improve body growth and ultimately facilitate increases in population density 28 .In addition, increased rainfall in the early growing season can promote vegetative sprouting.In the family Gramineae, sprouting produces a plant secondary metabolite (i.e.6-methoxybenzoxazolinone; 6-MBOA) that can trigger the onset of reproduction in rodents such as the montane vole (Microtus montanus) 30 and the African multimammate rat (Mastomys natalensis) 31 .6-MBOA has been detected in seedling L. chinensis; intraperitoneal injection of 6-MBOA into Brandt's voles also hastens the onset of reproduction by increasing serum levels of luteinizing hormone and testosterone and relative mRNA expression levels of reproduction-related genes (e.g.StAR and CYP11a1) in the testes of voles maintained under a short photoperiod 32 .Such regulatory pathways could explain the positive bottom-up effect of increased rainfall on Brandt's vole populations in the early growing season.The maximum population density of voles in the ER130 and ER260 treatment groups averaged 1.7 times and 1.5 times that of the control enclosures; although when the amount of rainfall was doubled, we found no further increase in the population density achieved by the ER260 compared to the ER130 groups.This result may be explained by the nonlinear response of net primary productivity to changes in annual rainfall 33 .For example, in Phase 2, we found a signi cant increase in the biomass of all plant species in the ER130 treatment compared with that in the control, while there was no further increase in total plant biomass in the ER260 treatment compared with that in the ER130 treatment.In the extreme case, very high rainfall could cause ooding and drown voles in their shallow burrows (usually 30 cm below ground), leading to population collapse 2,34 .The population density was lowest in 2015, a trough that was most likely caused by two successive intense thunderstorms that deposited over 30 mm rainfall in less than 2 hours on 22 June and 24 mm rainfall in less than an hour on 9 July during the breeding season.After the intense thunderstorms, we found many drowned young voles in the enclosures (Fig. S2b), con rming the harmful effect of extreme rainfall events on this vole population 2,35 .The bene cial effect of rainfall through indirectly increasing food resources may be offset by the harmful in uence of rainfall through direct burrow ooding, which may lead to a nonlinear effect of experimental rainfall increase on vole populations.The variation in annual natural rainfall was large at our study site (e.g.512 mm in 2012, only 169 mm in 2017), and in turn in uenced the effect size of our experimental rainfall increase on the vole populations.For example, in Phase 2, adding 260 mm rainfall during the growing season (ER260) signi cantly increased the vole population in 2017 when the natural rainfall during the growing season was relatively low (130.7 mm), but not in the years 2016 and 2018 that received 215 mm and 230 mm of rainfall during the growing season respectively.
Total rainfall in the early growing season (May and June) for plants has been increasing since 1970 in our study region (Fig. S3).In order to understand better the biological consequence of altered rainfall timing, in Phase 3 we xed the total amount of added rainfall to 130 mm and allocated most water (61.5%) to the early growing season (ER130 treatment) and to the late growing season (LR130 treatment).This rainfall increase in ER130 treatment and LR130 treatment signi cantly increased both the biomass of the vole's preferred plant species and vole population density, compared with the control group (Table S2).Furthermore, the biomass of preferred plant species and population density of voles in the ER130 treatment were signi cantly higher than those in the LR130 treatment (both P < 0.05; Fig. 1a&2a).These results further con rm that rainfall increase in the early, not late, growing season can drive marked increases in the population density of voles.In the ER130 treatment, the enclosure received 120 mm of rainfall in total, including 40 mm of natural rainfall and 80 mm of arti cial rainfall, during May and June 2019, and the mean population density of Brandt's vole went as high as 296 individuals per hectare, which is very close to the population density in years of population outbreaks (about 300-400 voles/ha).By reviewing meteorological records and literature, we found that there were 7 years when the amount of rainfall in May and June exceeded 120mm after 1970.Five of the 7 years with wet seasons in May and June were reported to have population outbreaks of Brandt's vole (> 300 individuals per hectare, Fig. S3), which is consistent with our manipulative results.
The response of reproduction and body growth of voles to increased rainfall depended on the rainfall treatment schedule (Fig. 1b, c).In Phase 1, there was no difference in either the total number of recruits or body mass of young-of-year (YOY) between the control, R50 and R100 treatment groups.In Phase 2, the total number of recruits in 2016, 2017 and 2018 increased signi cantly, by 71.8%, 87% and 167%, respectively, in the ER130 treatment compared with the control group (2016: t = 3.68, P = 0.005; 2017: t = 4.3, P = 0.002; 2018: t = 3.5, P = 0.006), while numbers in the ER260 treatment increased signi cantly only in 2017 (t = 3.79, P = 0.004).Voles also showed greater body mass in the ER130 group compared with the control group in October over all three years (2016: t = 3.26, P = 0.01; 2017: t = 3.63, P = 0.006; 2018: t = 3.3, P = 0.009), while those in the ER260 treatment showed increased body mass only in 2018 (2018: t = 2.81, P = 0.02).In Phase 3, the total number of vole recruited and their mean body mass were signi cantly greater in both the ER130 and LR130 groups compared with the control group (Fig. 1b, c, all P < 0.05).There was no difference in the total number of voles recruited or body mass between the ER130 and LR130 treatment groups (both P > 0.05).
Population dynamics arise as a consequence of variations in population growth rates, which are determined by underlying demographic rates (survival, growth, reproduction).The relative contributions of different demographic rates to change in population growth rate vary among populations 36 .Due to limited food resources in the natural environment, individuals must allocate resources in a balanced way to growth, reproduction and survival if tness is to be maximized 37 .At the population level, the trade-offs between survival versus growth and reproduction shape population uctuations and species coexistence, but are constrained also by physiological limitations and ecological modes-of-life 37 .According to the pace-of-life syndrome (POLS) hypothesis, individuals with a fast pace (fast growth rate, early reproduction and short lifespans) can be expected to allocate more investment in reproduction than in survival, while the opposite pertains for the slow-pace-of-life individuals 38 .Small rodents are typically fast lifestyle species with relatively short lifespans, and may adopt a high-fertility/low-survival strategy to maximize tness, especially in environments where resources arrive in short pulses.In our study, Brandt's vole responded consistently to the rainfall-induced increase in plant food resource, with populations rising due to increased fertility and individual growth rate and less to changes in survival rate, in accord with the POLS hypothesis.
In summary, we provide the rst experimental evidence that rainfall can induce bottom-up effects on plant biomass that drive the population dynamics of small mammals, and in particular that the timing of rainfall is more important than the amount of rainfall in shaping these dynamics.Wet years, when > 120 mm rainfall during the growing season of May and June, can be used to predict population outbreaks of Brandt's voles in our study region.Globally, climate change is causing large variations in rainfall patterns.
Understanding the response of animal populations to these changes is becoming more urgent for biodiversity conservation, wildlife management and pest or zoonotic disease control.Although our study was restricted to Brandt's vole in a steppe grassland ecosystem, changes in the rainfall pattern clearly drove large changes in the population dynamics of this rodent.Further experimental studies are now required on other species to reveal how the amount and timing of rainfall determine demographic processes, population uctuations and changes in species distributions more generally.

Study area and experimental enclosures
This study was conducted at the Research Station of Animal Ecology (44°11′N, 116°27′E) in Inner Mongolia, China.This region is a steppe grassland that experiences a continental arid temperate climate, with a mean annual temperature of 3.4 ºC, and mean annual precipitation of 259.2 mm (1980-2009 climate data from China Meteorological Administration).The small mammal community in this region consists mainly of Brandt's vole (Lasiopodomys brandtii), Daurian ground squirrels (Spermophilus dauricus), striped hamsters (Cricetulus barabensis), Mongolian gerbil (Meriones unguiculatus) and house mouse (Mus musculus).
In 2008, we constructed a set of 24 large outdoor enclosures (60 × 80 m each), each surrounded by a galvanized sheet metal wall extending 1 m underground and 1.4 m above ground.Wire mesh netting (mesh size: 1 cm) extending 50 cm above the sheet metal wall was welded onto the metal wall, and nylon netting (mesh size: 10 cm) was placed on the top of the enclosure (for details, see ref. 39 ).The design of the enclosure was intended to prevent voles from emigrating or immigrating, as well as to exclude all mammalian and avian predators.To manipulate rainfall, a spraying irrigation system was built in each rainfall increase enclosure (see below) using steel pipe, and consisting of 30 two-meter-high rotating sprinklers arranged in a 5 × 6 array (Fig. S2a).Each sprinkler was pressure-regulated, allowing even application over each enclosure at a rate of approximately 5 mm h -1 .Water was delivered from a well into a storage tank (6 m × 3 m × 1.5 m) before transfer to the irrigation system with a pump.

Rainfall manipulation
Twelve enclosures were used for these manipulations, and were randomly assigned to control and two levels of rainfall increase treatment, with four replicates per treatment.There were three distinct phases for the rainfall manipulation experiment (Fig. S5).Phase 1 was conducted from 2010 to 2015 and was designed to test the effect of low and moderate amount rainfall increases (averaging 18% and 36% increases as compared to the natural rainfall amount; Table S1) on vole populations.In China, the 400 mm isohyet for the mean annual rainfall is considered as the boundary between semi-humid (> 400 mm) and semi-arid regions (200~400 mm), and also corresponds to the boundary between grassland and forest or farmland.Given that the mean annual rainfall in our study site was 259.2 from 1980 to 2009, the enclosures receiving the two levels of rainfall increase were set to receive in total 50 mm or 100mm of water (termed the R50 and R100 treatment groups, respectively) during the growing season (May-September), thus ensuring that the total amount of rain in our experiment was still less than 400 mm.By contrast, the control enclosures received only natural rain.The increased rainfall was evenly distributed bi-weekly throughout the growing season (Fig. S1) for a period of one or two hours each for the R50 and R100 treatment groups, respectively.Because we did not nd signi cant differences in population abundance between the control and rainfall increase groups during Phase 1, we decided to increase the rainfall amount as well as to shift more rainfall to May and June (i.e. the early breeding season of voles) in the next phase.Phase 2 ran from 2016 to 2018 and aimed to examine the combined roles of timing shift and more rainfall (increases averaging 56% and 112% compared to the natural rainfall amount; Table S1) on vole populations.From 1980 to 2015, extreme rainfall events occurred frequently at our study site and the maximum annual rainfall was 511.7 mm in 2012.According to the maximum value of rainfall in our region, we increased the amount of added rainfall to 130 mm and 260 mm (termed the ER130 and ER260 treatment group; Fig. S1) and allocated more rainfall (61.5% of the total added rain) to the early growing season (May-June).In Phase 2, we found that rainfall increases of both 130 mm and 260 mm in May-June increased vole abundance but there was no difference between the 130 mm and 260 mm treatments, suggesting that excessive rainfall increase had no extra positive effect on vole density.Thus, we decided to further test the effects of rainfall timing on population density of voles by xing the total amount of added rainfall to 130 mm in the next experiment.Phase 3 was designed to last two years (2019-2020), but was disrupted in 2020 due to the coronavirus outbreak at the end of 2019 in China.In 2019, there were two rainfall increase treatments, with most water (61.5%)applied to enclosures in early growing season (termed the ER130 treatment group) or in late growing season (termed the LR130 treatment group; Fig. S1).

Animal trapping and vegetation survey
In April each year from 2010-2019, we conducted live-trapping for 7 consecutive days to remove all overwintered individuals in the enclosures, and then introduced fteen (2010-2011) or thirteen (2013,  2015-2019) pairs of voles from the eld into each enclosure to create a new founder population.The size of the founding population was the same in all enclosures, which helped to exclude the confounding factor of initial population size.Voles were given 10-15 days to acclimatize to their enclosure before the manipulative experiments started.In 2014, the overwintered individuals were not removed and continued to be observed for another year to test the carry-over effect of rainfall increase on vole populations.Capture-Mark-Recapture (CMR) was used to investigate the demography of vole populations biweekly (2010-2013) or monthly (2014-2019) from May to October.Each newly trapped vole was marked with a numbered aluminum ear tag (2010-2017) or implanted subcutaneously with a passive integrated transponder (PIT) tag (2018-2019).Information on sex, body mass, reproductive status, and family location was recorded before individuals were released at their capture sites.A vegetation survey was also conducted each month following the census of the vole population.We randomly selected ve quadrats (1 × 1 m) in each enclosure, and measured the density (number of plants in each quadrat) and dry biomass of each plant species.For a more detailed description of the experimental procedure, see refs 40 28 41 .Based on the relative preference index for each plant species consumed by Brandt's vole 41 , all plant species within enclosures were a priori categorized into two groups: preferred plant species (i.e.Leymus chinensis, Setaria viridis, Agropyron cristatum and Medicago sativa) and less preferred plant species (i.e.Stipa krylovii, Lepidium apetalum, Corispermum declinatum, Carex tristachya, Cleistogenes squarrasa, Neopallasia pectinate, and Phlomis dentosa).

Data analysis
The population density in each enclosure was estimated using the CMR data with Program CAPTURE 42 , and expressed as numbers per 0.48 ha (i.e. the size of the enclosures).Recruits were de ned as individuals that were born in the current year.Because it was di cult to identify the exact birth date for each individual, we pooled yearlings together each year in our analysis.The total number of recruits and yearling body mass in October were compared between different rainfall treatments using analysis of variance.Differences in population density and biomass of food resources for voles between different rainfall treatments were analyzed with linear mixed models, with enclosure speci ed as random effect.All individuals in the enclosures were classi ed into two cohorts, overwintered and YOY, for analysis of survival.For overwintered voles (i.e. the population founders), survival analysis was performed using the Cox proportional hazards model to assess the effects of rainfall manipulation, sex and overwinter body mass.For YOY, logistic regression was used to test the effects of rainfall manipulation and sex on the survival rate of YOY.All statistical analyses were performed using R 43 , version 3.5.1.

Declarations
Author contributions ZZ, XW, GL designed and lead the long-term study; XW, GL, BY, WW, XH, XZ, JZ, EB, XX, SH, YS, and JL collected data; ZZ and GL conceived the ideas for the paper and its structure; GL conducted the analyses; ZZ and GL wrote the manuscript; GW, AO, CK and CD modi ed the draft and commented on the manuscript.

Figures
Figure 1 Changes (means ± se) in population density, numbers of recruits and body mass of yearling Brandt's voles in 0.48-ha enclosures between different rainfall increase scenarios from 2010 to 2019.The asterisks denote signi cant overall treatment effects (P < 0.05) in some speci c years.R50: 50 mm rainfall were added evenly throughout enclosures during the growing season; R100: 100 mm rainfall were added evenly throughout enclosures during the growing season; ER130: 130 mm rainfall were added during the growing season but with more rainfall allocated to the early growing season (May-June); ER260: 260 mm rainfall were added during the growing season but with more rainfall allocated to the early growing season.LR130: 130 mm rainfall were added during the growing season but with more rainfall allocated to the late growing season (July-September).

Figure 2 Changes
Figure 2