OIKOS 59: 290-302.

Copenhagen 1990

Temporal changes in a Chihuahuan Desert rodent community

James H. Brown and Edward J. Heske

Brown, J. H. and Heske, E. J . 1990. Temporal changes in a Chihuahuan Desert rodent community. - Oikos 59: 290-302. We used time series analysis of ten years of monthly census data to assess the responses of both individual species and an entire community of rodents to a fluctuating desert environment. Autocorrelation analysis revealed different patterns of intra-annual fluctuation among the It species: Dipodomys spectabilis and Perognathus Jarus had pronounced annual cycles: D. ordii, D. merriaini, Chaetodipus penicillutu~,Onvchomys torridus, 0 . leucogusrer, and Neotoma alhigula exhibited annual cycles modified by interannual variation; and Feromyscus eremicus, Pm. maniculatus, and Reithrodontornys megalofis showed little evidence of annual periodicity. The timing af annual cycles and the pattern of inter-annual fluctuations also differed among species. However. two results suggest that several species responded similarly to long-term environmental variation: 1)population densities of four species and total rodent biomass and numbers were positively correlated with the densities of annual plants; and 2) many pairs of species exhibited positively correlated population dynamics over the ten years, Clustering of painvise cross-correlation coefficients was used to identify sets of species with similar population dynamics. These clusters did not necessarily contain closely related or ecologically similar species. Detrended Correspondence Analysis identified three independent patterns of variation in species composition: a long-term trend; a four- to five-year repeated pattern that appeared to correspond to the climatic effects of the El Nifio Southern Oscillation: and an annual cycle. In general. species appeared to respond individualistically to environmental variation. There was no evidence of an equilibrium community composition or of alternative stable configurations, Some competing species had negatively correlated population dynamics, but the majority of competitors exhibited positive correlations that apparently reflected similar responses to fluctuating resources.
J . H. Brown and E. J Heske, Dept of Bicdogy, Urrlv. of New Mextco, Albuquerque, NM 87131, C'SA

Introduction
The nature of communities has been a central issue in ecology since at least the times of GIeason (1917, 1926) and Clements (1916). Despite this attention, the processes that determine the identity. abundance, and distribution of coexisting species remain poorly understood and much debated. The questions of today, while phrased somewhat differently than by Gleason and Clements, remain essentially the same. How does the composition of assemblages vary in space and time? what are the roles of abiotic factors and interspecific interactions in limiting abundance and distribution of species? To what extent is community composition deAccepted 26 June 1990

termined by processes operating within local areas, and to what extent is it affected by physical variables and movement of organisms occurring at larger spatial scales? On what spatial and temporal scales and between what kinds of organisms do coevolutionary processes influence community organization? The organization of a community and its patterns of temporal change reflect the fluctuations of the local species populations that comprise the community. There is a rich theoretical literature on populations dynamics. Many recent papers emphasize that wide, seemingly chaotic fluctuations can be expected, not only when populations are regulated primarily by densityindependent abiotic conditions, but also when they are

OJKOS

limited largely by density-dependent intra- and inter- techniques to ten years of monthly censuses of the 11 specific interactions (May 1976, 1980, Gilpin 1979, Kot most common rodent species inhabiting a 20-ha site of and Schaffer 1984, Schaffer 1985, Schaffer et al. 1986). relatively homogeneous shrub habitat in the ChihuaThere is also a large literature of empirical studies of huan Desert of southeastern Arizona. Time series autopopulations dynamics, many of t h e m on small mam- correlation and spectral analyses of monthly census data mals. Many different patterns have been documented, permit detection of seasonal and inter-annual fluctuincluding relatively stable populations, annual cycles, ations in each species. Cross-correlation analyses bemultiannual cycles, and irregular multiannual fluctu- tween abundances of species and densities of annual ations (e.g., Krebs and Myers 1974, Hansson and Hent- plants indicate responses of populations to temporal tonen 1985, Henttonen et al. 1985, Taitt and Krebs variation in productivity and food availability. Cross1985). Studies of the dynamics of multiple, coexisting correlations between pairs of species identify subsets of species are fewer in number, confined primarily to mi- species whose temporal fluctuations are positively or crotine rodents (but see O'Connell 1989), and have negatively correlated. Ordination on the basis of relausually emphasized that the multiannual cycles tend to tive abundances of species in each sample period ilbe synchronous (e.g., Henttonen et al. 1984, Henttonen lustrates how composition of the entire community changes over time. Results suggest that this rodent com1985). Insights into the relationship between changes in munity has a "Gleasonian" nature; although our analycommunity composition and the dynamics of the com- ses reveal sets of species that exhibit generally similar ponent species populations must come from long-term temporal fluctuations and previous work has demonstudies. The responses of individual species and the strated strong interspecific competition among some community as a whole to fluctuations in abiotic condi- species, the overall picture is one of individualistic, tions, availability of resources, and abundances of other species-specific responses to the pronounced temporal species can provide information on the factors affecting variation in this desert environment. the local distribution and abundance of each species, on the nature of interactions among species, and on the extent to which the species respond to environmental variation individualistically or as an integrated unit. Methods Desert rodents have served as a model system for The system studying community organization (e.g., ~ o s e n z w e i ~ and Winakur 1969, Rosenzweig and Sterner 1970, Ro- We analyze ten years of rodent censuses from Brown's senzweig 1973, Rosenzweig et al. 1975, Brown 1975, 20-ha experimental study site 6.5 km east and 2 km 1984, 1987, 1989, Brown et al. 1979, 1986, Price and north of Portal. Cochise Countv. Arizona. For detailed Brown 1983, Price 1986, Kotler and Brown 1988, description of the site, experimental manipulations, and Brown and Harney, in press). Experimental and com- rodent trapping protocol see Brown and Munger parative studies have repeatedly demonstrated the im- (1985). With a few exceptions (see below) rodents have portance of food limitation and interspecific competi- been trapped since July 1977 at monthly intervals corretion in determining the abundance of species and their spondingly approximately to the time of the new moon. foraging behaviors (Brown 1973, 1989, Reichman 1975, Within each of the 24 experimental plots one trap was 1977, Munger and Brown 1981, Brown and Munger set for one night at each of 49 permanent grid stakes 1985, Frye 1983, Bowers et al. 1987). Since plant growth placed 6.25 m apart. The rodents captured were individand seed production in desert environments depend on ually marked, identified to species, weighed and sexed, precipitation (reviewed in Brown et al. 1979), popula- and released. tion fluctuations in desert rodents should be affected by The present analyses are based on the total number of seasonal and year-to-year variation in climate, primary individuals of each of the 11 most common species production, and other rodent species. O n the one hand, captured in each monthly census on all experimental because rodent populations appear to be limited by plots during the 10.5-yr period, July 1977 to January food availability, which depends primarily on a single 1988. By combining the data for all plots we ignore the variable, precipitation, all coexisting species might be differential responses of species to the experimental expected to respond similarly to temporal variation in treatments, which include removal of different combithe environment. O n the other hand, because coexisting nations of rodent and ant species, addition of supplespecies differ in taxonomic relatedness, body size, food mental millet seeds, and an attempt to remove annual habits, and life history, they might be expected to exhib- plants (see Brown and Munger 1985, Brown 1987). On it different patterns of fluctuation in the same tempo- the other hand, by combining data in this way we rally variable environment. greatly increase the number of rodents captured, and Time series analysis and multivariate ordination tech- hence the power of our statistical analyses. We feel niques provide potentially powerful methods for assess- justified in summing captures over all plots, because the ing the patterns of temporal variation within species same rodent species had access to the same plots and populations and entire communities. We apply-these most of the experimental treatments were continued for

the entire 10-yr period. The only exceptions were one Dipodomys spectabilis removal plot that was added in February 1980 and eight plots where annual millet seed addition treatments were terminated in July 1985 and replaced by ineffective attempts to remove annual plants through local applications of herbicide. Since the addition of millet seed affected only a single rodent species (causing an increse in D. spectabilis; Brown and Munger 1985) and the attempt to remove annual plants had no apparent effects o n rodent populations, we are confident that most of the population dynamics reported here reflect responses of species to temporal environmental variation over the entire study site, rather than differential responses of species to experimental perturbations on certain plots. We will return to this point in the discussion. Annual plant densities were censused twice each year at the end of the winter and summer growing seasons. O n each of the 24 experimental plots, plants were counted by species on 8 (1978-1982) or 16 (1983-present) permanent subplots, each measuring 0.5x0.5 m (see Davidson et al. 1985 for details). There was a weather station on the study site, and analysis of the relationship between climatic variables and annual plant density, biomass, and species composition will be reported elsewhere.

Analysis
Single species patterns: To detect patterns of intra-annual variation in rodent abundance, we calculated autocorrelations for time series of monthly censuses from July 1977 through January 1988 for each of the 11 most common rodent species. Autocorrelation analysis calculates a parametric correlation coefficient for all pairs of data points separated by a given time lag (Davis 1976). It is a powerful method of detecting the extent of cyclical variation. A n annual cycle is indicated in our analyses by a sinusoidal autocorrelogram with significant positive values at a time lag of 12 months. During the 127-month study period, censuses were missed on 9 months due to extended periods of bad weather. Because autocorrelation analysis requires a complete set of equally spaced data points, we.filled in missing values by interpolation between the preceding and following censuses. Because there was a high degree of temporal autocorrelation in censuses of all species for the time lags of one month, and we never missed more than one month in succession, the error introduced by this interpolation procedure was very small. For proper analysis of intra-annual patterns, there is the additional requirement that the data be stationary; that is, there must be no significant inter-annual trends in numbers. To remove the effects of inter-annual density variation on the autocorrelation analysis, we first normalized the data within each year separately so that all

years had the same mean and variance, and then performed the autocorrelation analysis. Spectral analysis using the Fast Fourier Transform (FFT) procedure was conducted on the time series of monthly censuses to examine the population dynamics of each species for multi-annual patterns. This technique quantifies the relative contributions of sine waves of different frequencies and amplitudes to explaining the pattern of temporal variation (Bloomfield 1976). It identifies the dominant periodicities that are present in the dynamics. We also inspected the time series of each species for long-term trends in abundance. The relationship between variation in rodent numbers and fluctuations in food availability was investigated by cross-correlating the time series of rodent censuses and annual plant censuses. This technique computes parametric correlations between two dependent variables (in this case the abundance of plants and of a rodent species) separated by time lags of varying duration. Significant values in the correlogram indicate the magnitude and time lags of covariation between two time series. This analysis required two manipulations of the annual plant data. First, since we know that some of the experimental treatments (rodent and ant removals) produced changes in vegetation on some plots (Davidson et al. 1985, D . Samson, D . Davidson, and T. Phillipi, pers. comm.), we used only data from the unmanipulated control plots as an index of inter-annual variation in plant production. Second, because winter and summer annual plants differed greatly in absolute abundance, plant size, and other factors affecting seed production, we normalized censuses for winter and summer annuals separately so that both time series had the same mean and variance, and then combined them. The resulting time series thus represents an index of relative goodness or badness of each season in terms of mean number of annual plants per m2 over the 10-yr period. No winter annual census was made in 1980. We replaced this missing value with the mean value for the winter censuses rather than using interpolation, because the plant censuses showed no evidence of being positively correlated from year to year. Because annual plant censuses were conducted twice each year, we had to construct a comparable semiannual time series of rodent abundances in order to calculate the cross-correlations between the two. Since plant censuses were made at the end of the growing seasons, usually in March and September of each year, we made comparable winter (October- arch). and summer (April-September) estimates of rodent abundance by calculating the mean number of each species captured per census for each 6-month interval

Between-species comparisons: Comparisons between the time series for pairs of rodent species were made in two ways. First, cross-correlations were calculated for all pairs of species (N = 55) using the time series of

LAG

Fig. 1. Autocorrelograms of normalized time series (see text) for the 11 common rodent species at our Chihuahuan Desert study site. Lags are in periods of months. Correlation coefficients greater than 0.18 (or less than -0.18) are significantly different from zero at P < 0.05. Low correlations at six-month lag and high values at 12 months indicate annual cycles. Species are indicated by first letter of genus and species; see text for full species names.

WSWsWSWSWSWSWSWSWSWsW 1977 1988

YEAR

monthly censuses. Because these data include both intra- and inter-annual density variation, we additionally made pairwise cross-correlations between all species pairs using time series of the mean annual abundances of each species calculated for each of the 10 yr. This second analysis allowed us to focus on how similarly rodent populations of the different species fluctuate from year to year, independent of any seasonal patterns of abundance within species. The onset of the monsoon rains in early July after a late spring - early summer drought is the most predictable seasonal event at our study site. Therefore, we based our analyses on a biological year from July through June instead of a standard calendar year beginning in January. Cross-correlation coefficients at 0 lag for both sets of painvise analyses (monthly and annual averages) were clustered using a standard UPGMA (unweighted pairgroup method using arithmetic averages) algorithm to group species according to their similarity in population dynamics. This UPGMA algorithm clusters values sequentially, beginning with low ones. For this reason and also because some coefficients were negative, and clustering algorithms cannot accept negative distance values, we used the absolute value of 1 minus the coeffi-

YEAR
Fig. 2. Population dynamics of the 11 common rodent species at our Chihuahuan Desert study site over 10.5 yr. Numbers are given as 6 month averages (see text) rather than monthly census values for clarity in presentation. Above: Heteromyid rodents. Below: Murid rodents. See text for full species names.

Table 1. Frequency in months per cycle and relative heights of the major peaks (all those peaks > 25% of the height of the major peak) indicated by Fast Fourier Transform analysis of monthly censuses. Species Months per cycle

D. spectabilis D. ordii D. merriami C . penicillatus Pg. Jlavus Pm. eremicus Pm. maniculatus R. megalotis N . albigula 0. leucogaster 0. torridus
'Includes peaks at 21, 25, or 31 months

cient value instead of the actual cross-correlation coefficients in this analysis. Thus, highly positive coefficients received a low value, highly negative coefficients received a high value, and near zero coefficients received an intermediate value by this adjustment.

less distinct peaks at 6 and 12 mo indicated considerable variation in the timing of the cycles. Three species, R. megalotis, P. eremicus, and P. maniculatus, showed little or no evidence of annual periodicity. Long-term patterns of population dynamics also revealed great variation among the 11 species (Fig. 2). Some species, C . penicillatus, Onychomys spp., and N. albigula, maintained relatively constant populations throughout the 10-yr study period. Others, such as D. ordii, R. megalotis, and Peromyscus spp., which were relatively rare at the start of the study, increased substantially over the decade. In contrast, D. spectabilis and Pg. flavus, which were relatively common at the beginning of the study, decreased dramatically in 1984 and never recovered to their former densities. D. merriami remained common throughout the study, but fluctuated approximately 5-fold over the 10-yr period. Several species attained peak densities in 1982-83, 1987-88, or at both times. Spectral analysis of monthly census data using Fast Fourier Transformation gave little evidence of multiannual cycles (Table 1). Strong peaks for most species corresponded to either annual cycles or a nonrepeating

Community-level patterns: To assess temporal variation in community composition, we analyzed the proportional abundances of all 11 species in each monthly census using detrended correspondence analysis (DCA). D C A and related multivariate ordination techniques (e.g., Principal Components Analysis) decompose complex patterns of variation into a hierarchical set of independent axes; each successive axis represents a combination of variables that account for progressively smaller proportions of the total variation. D C A was chosen as the appropriate ordination technique because we expected nonlinear relationships among species. We examined values of the first three orthogonal axes for stability or a repeating pattern using time series autocorrelations with lags of 1 to 24 months. Spectral analysis (FFT) was also used to examine the first three D C A axes for repeating patterns in community composition.

DO-DM
0

CP-PF DO-DS

OT-OL
W

DO-CP

Results
Single species patterns
Autocorrelations of the normalized (adjusted year-byyear to remove long-term trends) time series revealed very different degrees of annual periodicity in the population dynamics of different species (Fig. 1). D. spectabilis and Pg. flavus exhibited regular annual cycles with strong negative autocorrelations at a lag of 6 mo and positive values at a lag of 12 mo. Several other species, C. penicillatus, N. albigula, 0 . torridus, 0. leucogaster, D. merriami, and D. ordii, had annual patterns, but the

PM-RM :*-. 2 PE-RM

.

- PE-PM

LAG
Fig. 3. Selected examples of cross-correlograms between the 11 common rodent species at our Chihuahuan Desert study site. Lags are in months. Correlation coefficients greater than 0.15 (or less than -0.15) are significantly different from zero at P < 0.05. See text for interpretation and full species names.

Table 2. Cross-correlation coefficients at lag 0 for the time series of monthly census data (N = 127) for pairwise comparisons among the 11 most common rodent species (above diagonal). Cross-correlation coefficients for the same species pairs calculated using 10 yr of annual averages (below diagonal).

D. spectabilis D. ordii D. merriami C. penicullarus Pg. flavus Pm, eremicus C. maniculatus R. megalotis N . albigula 0. leucogaster 0. rorridus

Fig. 4. Dendrograms showing the 11 common rodent species at our Chihuahuan Desert study site clustered according to similarity in their population dynamics. A: Clusters based on cross-correlation coefficients at lag 0 using monthly censuses (N = 127). B: Clusters based on cross-correlation coefficients at lag 0 using annual averages (N = 10). Species are identified by the first letter of genus and species.

B.

1977

W S W S W S W S W S w S W S W S W S v ~ S W 1988

YEAR

Fig. 5. TMal numbers and biomass of the 11 common species of rodents at our Chihuahuan Desert study site over 10.5 yr. Both axes are plotted as 6 month averages rather than monthly census values for clarity in presentation. Note peaks in biomass following the El Niiio years of 1977-78, 1982-83, and 198M7.

10-yr trend. Some species, however, showed a 5-yr peak that probably reflects peak densities in 1982-83 and 1987-88. The fact that several species peaked at similar times suggests that populations fluctuations may vary with food availability. Cross-correlations between annual plant population densities and the number of rodents were significant (P<0.05) and positive at zero time lag for four species of murids, Pm. rnaniculatus (r = 0.74), N. albigula (r = 0.59), Pm,eremictcs (r = 0.49), and R. rnegalotrs (r = 0.47). These were the same rodent species that showed strong 5-yr peaks in the spectral analysis. The heteromyids all showed positive but not statistically significant cross correlations with the peak often occurring at a lag of 6 or 12 mo. Total rodent biomass was also s~gnificantly correlated (r = 0.48; P<0.05) with annual plant densities at zero time lag; total rodent abundance was marginally significantly correlated (r = 0.41: 0.1>P>0.05) with the same variable.

Cross-correlations using the monthly census data revealed 26 significant positive and 5 significant negative painvise associations at zero time lag (Table 2). There were an additional 7 species pairs that had significant positive cross-correlations and 2 that had negative crosscorrelations at time lags of 3 to 7 mo. Similar analysis using annual averages gave 10 significant positive and one significant negative cross-correlations at lag zero. This, together with the predominance of positive but nonsignificant values in the pairwise species-by-species cross-correlation matrices (Table 2) suggests that most, but not all. species have population dynamics influenced by the same environmental conditions. Cluster analysis revealed sets of species that exhibited similar population dynamics (Fig. 4). Dendrograms based on both the monthly censuses and the yearly averages were very similar, suggesting that species that cluster together exhibit both similar seasonal fluctuations and similar year-to-year dynamics. The only qualitative difference was in the linking of C. peniculla-

roo
8G
(r,

20

Between-species patterns
Cross-correlograms showed a variety of relationships between the population dynamics of different pairs of species. Since there are 55 pairs of species, we only illustrate a few examples of the patterns in Fig. 3. Some pairs of species maintained nearly constant positive, negative, or zero correlations regardless of time lag (Fig. 3, top). These were species pairs that maintained trends for several successive years. For other pairs, the value of the cross-correlations showed an annual cycle with significant peaks at time lags different from zero (Q. 3, center). indicating positive year-to-year covariation but different dynamics within years. Still other pairs exhibited cross-correlations which were significantly positive at zero time lag but declined with increasing time lags (Fig. 3, bottonl). These were species pairs that fluctuated synchronously. but s h ~ w e dhigh variation between successive seasons and years.

40

W S W S W S W S W S W ~ W S W S W S W S W 1988 1977

YEAR

Fig. 6. Proportional contribution of four functional groups (based on cluster analysis; see text) to total rodent numbers and biomass over 10.5 yr. Functional groups are: 1) Vertical stripes: D. spectabilis and Pg, flavus, 2) White: D. merriami and D.ordii, 3) Black: Pm. eremicus, Pm. municulutus, and R. megalotis, 4 ) Gray: remaining species; C. penicillatu, N. albigula, 0. leucogaster, and 0. torridus. Numbers and biomass given as 6 month averages rather than monthly census values for clarity in presentation.

AXlS 1

20000

1751%.

AXIS 2

AXlS 2

A
175 150

AXlS 3

AXlS 3

1 B77

MONTHS

Ism

Fig. 7. Time series of scores on the first three DCA axes (see text) for 127 monthly rodent censues at our Chihuahuan Desert study site. Note the long-term trend in Axis 1, the peaks corresponding to El Nido years in Axis 2, and the annual pattern in Axis 3.

tus and N. albigula only when the monthly values were

used, which probably reflects the fact that both species peak in summer (in part because the former species is an obligate hibernator) but their inter-annual fluctuations are not well correlated. Although the species in the clusters were sometimes similar in taxonomy, morphology, and life history (e.g. the medium-sized kangaroo rats, D. ordii and D. merriami, and the small murids, Pm. eremicus, Pm. maniculatus, and R. megalotis), this was not always the case (e.g., D. spectabilis clustered with Pg. flavus, and the medium-sized kangaroo rats and the small murids were more similar to each other that either was to more closely related species).

Fig. 8. Spectral analysis of the time series of monthly rodent census scores on the first three DCA axes using the Fast Fourier Transform procedure. Frequency of peaks is given as months per cycle. Results are presented as a continuous function for clarity in presentation; however, actual values were obtained only for frequencies corresponding to approximately m, 126,63,42,31,25,21, 18,16,14, 13, 11, and 10 months per cycle. Note the peaks corresponding to: the entire study period for Axis 1; the entire study period, the four- to five-year El Nido pattern, and the annual cycle for Axis 2; and the annual cycle for Axis 3.

biomass tended to follow the El Nino years of higher than normal winter precipitation in 1977-78, 1982-83,

Table 3. Species loading and eigenvalues (in parentheses) for the first three DCA axes, calculated from monthly census data. Species
D. spectabilis D. ordii D. merriami C. penicullatus Pg. flavus Pm. eremicus Pm. maniculatw R. megalotis N. albigula 0. leucogaster 0. torridus

Axis 1 Axis 2 Axis 3 (Eig=O.151) (Eig=0.050) (Eig=0.045)

Community-level patterns
Total numbers and biomass of rodents fluctuated approximately 3-fold and 2-fold, respectively, over the course of the study (Fig. 5). Because the Dipodomys spp. accounted for the vast majority of both individuals and biomass, these patterns largely reflect the combined influence of D. merriami and D. spectabilis. Peaks of

and 198687. Equally interesting are the changes in the proportional contribution of individual species and functional groups (Fig. 6). These patterns were quite similar for both numbers of individuals and biomass. The cluster of D. spectabilis and Pg. flavus decreased, the clusters of small murids and of D. merriami and D. ordii increased, and the other rodents showed little change except for a late increase in biomass owing to a few individuals of N. albigula. The plots of the D C A values as a function of time (Fig. 7) and FFT analysis of these time series (Fig. 8) showed that the composition of the community as a whole neither remained constant over time, nor exhibited pronounced multi-annual cycles. The first three D C A axes accounted for a substantial proportion of the month-by-month variation in abundances of the 11 species and were readily interpretable (Table 3). Axis 1 loaded heavily on the small murids and D. ordii, and negatively on D. spectabilis and Pg. flavus. This axis corresponded primarily to directional changes in these species over the decade as indicated by the major peak at 126 mo in the FTT analysis. Axis 2 loaded positively for Pg. flavus, the small murids, and D. spectabilis, and negatively for C. penicillatus and Onychomys spp. The FFT analysis indicated that this axis reflects fluctuations both on the scale of the entire study and with a period of approximately five years. It also indicated a substantial annual cycle. Only an annual cycle was detected by FFT analysis of Axis 3. Results of the D C A gave no evidence of alternative stable community structures (sensu MacArthur 1972). If the community composition fluctuated between alternative stable configurations, we would expect the D C A scores to shift rapidly and then to remain relatively constant for long periods. Instead, the different components of the community identified by the D C A axes appeared to vary continuously and independently over the decade of the study. Similarly, there was continuous variation over time in the extent to which the functional groups identified in the cluster analyses contributed to the total community composition (Fig. 6).

Discussion
Temporal variation in this rodent community
This study revealed a variety of patterns that characterize the dynamics of single species populations and the community as a whole. Rodent populations in this desert habitat fluctuated considerably, but the magnitude and periodicity of these fluctuations varied among species. The most dramatic fluctuations were exhibited by species such as Pg. flavus and the small murids, which varied from being virtually absent from the study site to having substantial populations. Not mentioned here were several rare species that were absent for most of the study, but made irregular appearances. A t the other

extreme were species such as C. penicillatus and Onychomys spp. that maintained very constant populations from year to year. The numerically dominant Dipodomys spp. all varied several-fold in density, but this was substantially less than the orders-of-magnitude fluctuations of other rodents (e.g., some microtines and Australian murids) that are the dominant small mammals in other habitats. Despite the large responses of several rodent species to experimental manipulations of food supply and presence of potentially competing rodent and ant species on some of the experimental plots, there is little evidence that these manipulations had significant effects on the patterns of temporal variation reported here. For example, although D. spectabilis increased on plots where millet seeds were added (Brown and Munger 1985), the crash in the D. spectabilis population occurred in summer of 1984, a year before t h e seed addition treatments were discontinued. D. spectabilis also declined synchronously and has subsequently failed to recover on a nearby site studied by P. Waser (pers. comm.). Similarly, the long-term increase in the abundance of D. ordii has been reported in nearby areas of similar habitat (J. Randall, pers. comm.). Finally, the densities of the small granivorous murids, which were greatest on Dipodomys removal plots (Brown and Munger 1985), showed similar temporal trends on control plots. Therefore we are confident that the population and community dynamics that we report here reflect regional trends rather than artifacts of our experimental manipulations. Correlations of rodent densities with annual plant densities, annual population cycles in some species, and the tendency for many species to show positively correlated inter-annual fluctuations all suggest that population dynamics respond to environmental variation (see also Kenagy 1973, O'Farrell 1974, Kenagy and Bartholomew 1985). O n the other hand, the fact that none of these patterns were consistent across all species has two implications. First, no one aspect of environmental variation has a pervasive coordinating effect on rodent population dynamics and community composition. This is true even though infrequent, unpredictable precipitation events determine the amount and timing of primary production in desert ecosystems (Went 1949, Tevis 1958, Hillel and Tadmor 1962, Rosenzweig 1968, Brown et al. 1979), and the reproduction of 5everal rodent species apparently is stimulated by the appearance of newly-germinated annual plants (Beatley 1969, 1974, 1976, Van D e Graff and Balda 1973, Reichman and Van D e Graff 1975). Second, the rodent community is comprised of species that respond to temporal variation in the same environment in very different ways. Some species differed strikingly in the degree and timing of their annual cycles and in their pattern of inter-annual variation. The rodent community could be clustered into functional groups of species on the basis of the extent to which their population dynamics were correlated. These

clusters did not consistently correspond to taxonomic relationships, nor did they consistently reflect similarities in morphology, behavior, and life history. Instead they appeared to reflect the way that the population dynamics respond to environmental variation. For example, the two medium-sized kangaroo rats appear so similar in many aspects of their ecology that they act almost like a single species (Schroder 1987). They exhibited moderate year-to-year variation; their fluctuations were also negatively correlated with those of D. spectabilis (Table 2), and both species increased on plots where D. spectabilis had been removed (Brown and Munger 1985). The small granivorous murids seem to be opportunists that are able to colonize and increase rapidly in density in response to temporarily favorable conditions (Whitford 1976, Conley et al. 1977, Munger et al. 1983). The almost equally tight clustering of Pg. flavus with D. spectabilis is more difficult to explain because these species are different in so many respects. This may reflect response to some common environmental factor, perhaps summer precipitation. Both of these species have biogeographical affinities with the Chihuahuan Desert, which is characterized by summer precipitation, and both begin reproducing in late winter, presumably using the productivity of the previous season to support their initial reproductive effort. Another possibility is that D. spectabilis facilitates Pg. fluvus, either by 1) reducing the local density of the other kangaroo rat species which may interfere with Pg. fluvus (Brown and Munger 1985, Bowers et al. 1987), o r 2) maintaining burrow systems and food stores that Pg. flavus exploits. The remaining species all have relatively stable population densities from year to year, and are linked at lower levels in the dendrograms. These last species include a granivore, a folivore, and two insectivores, and a wide range of body sizes as well as the only obligate hibernator. This implies that stable population dynamics cannot be attributed to any single combination of traits (see also Kenagy and Bartholomew 1985, Brown and Zeng 1989). D C A provided a method of characterizing each census in terms of the composition and relative abundances of species present. The D C A quantified the contributions of the different species to eigenvectors corresponding to patterns of community composition that varied independently over time. Three major axes were identified. Fast Fourier Transform analysis of values for these axes for each census identified three patterns: annual cycles, a non-repeating, long-term change in the relative proportion of species, and an approximately 5-yr repeated pattern that may correspond to rodent responses to climatic changes induced by the El Niiio Southern Oscillation. There was a strong El Nifio event in 1982-83, and there were moderate ones in 1977-78 and 1987-88 (Quinn et al. 1987, Molles and Dahm 1990). El Niiio years are characterized by unusually heavy winter precipitation in the desert region of the southwestern United States, so it is not surprising that

the resulting productivity and food availability stimulated increases in many rodent populations. Our community-level analyses provide no evidence either for an equilibrium community composition that remained constant or was restored after perturbation, or for shifts in community composition between alternative steady states. Rather, community composition varied continously over time, reflecting the different responses of species and groups of species to different environmental variables. The community was comprised of species whose responses to environmental variation were neither completely independent nor highly coordinated. Instead, there were functional groups of species that responded similarly, and other sets of species that showed independent or opposite trends. Species that are known or suspected to compete did not necessarily exhibit negatively correlated population fluctuations. Variations in D. ordii and D. merriami were negatively correlated with those in D. spectabilis, a known competitor (Frye 1983, Brown and Munger 1985, Bowers et al. 1987). The proportion of total rodent numbers and biomass comprising the two former species also increased when D. spectabilis declined (Fig. 6). However, D. ordii, which was most negatively correlated with D. spectabilis, had been increasing since the beginning of the study so this should be interpreted with caution. The small granivorous murids, which increased dramatically on plots where kangaroo rats were experimentally removed, had positively correlated dynamics and clustered with D. merriami and D. ordii. Although the appropriate competition experiments have not been performed, there is much circumstantial evidence to suggest that there should be substantial competition between D. ordii and D. merriami and also among the three species of small murids; the species within each of these two clusters (Fig. 4) exhibited strongly positively correlated population fluctuations (Table 2).

General implications
The results of this study imply that there is a great deal of temporal variation, both in the composition of the community and in the dynamics of the individual species populations. This is consistent with empirical observations and theoretical predictions that "stochastic" variation in the abiotic environment can result in wide fluctuations in population density (Leigh 1975, Beddington et al. 1976, Goh 1980, Grossman et al. 1982, Strong 1986). The fluctuations seen here, ranging through several orders of magnitude, are comparable to those reported for invertebrates and some vertebrates (e.g., Connell and Sousa 1983, Ostfeld 1988). The local extinctions and colonizations are also consistent with both other empirical observations (e.g., Whitford 1976, Pimm et al. 1988) and with "source-sink" models of local population dynamics (Hanski 1982, Holt 1983, Stenseth 1983, Wiens 1986, Pulliam 1988).

The characteristics of temporal variation in community composition seen here are reminiscent of the temporal changes in small mammal and plant communities over the last 10,000 yr (e.g., Cole 1982, Huntley and Birks 1983, Davis 1986, Graham 1986, Van Devender 1986). The great spatial variation in the species composition of North American desert rodent assemblages (Brown and Kurzius 1987, 1989) complements the results of this study to give a relatively "Gleasonian" interpretation of the structure and dynamics of these communities. The component species have different biotic and abiotic requirements, and their abundances vary in space and time in response to variation in these environmental factors. This individualistic or Gleasonian concept of community organization must be reconciled with the abundant evidence of strong interspecific interactions. For example, for desert rodents both competition (Rosenzweig 1973, Price 1978, Munger and Brown 1981, Frye 1983, Brown and Munger 1985, Bowers et al. 1987, Brown 1987) and predation (Thompson 1982a,b, Kotler 1984a,b, 1985) have been shown to influence the combinations of species that coexist in local habiatats, and the abundance, distribution, and resource use of these species. The kinds of techniques employed here should be useful for investigating the relationships between population- and community-level dynamics. Time series analysis has often been used to examine the population dynamics of individual species, and ordination and clustering techniques have been used to characterize community structure, but rarely have the two been used in combination. Together, they can offer insights into how population- and community-level processes are coupled. However, at our present level of understanding, it is difficult to make and test unambigous a priori predictions. O n the one hand, to the extent that species require similar resources or respond similarly to abiotic conditions, their population densities will tend to covary positively. O n the other hand, direct and indirect negative interactions between species, such as competition and "apparent competition" (Holt 1977), should result in negative covariation, at least on some scale. Since both of these phenomena are known to occur, how do they interact to influence the dynamics? Results of the present study suggest that analyses that regress the abundance of one species against another (e.g., Crowell and Pimm 1976, Hallett and Pimm 1979, Dueser and Hallett 1980, Emlen 1981, Pimm 1981, Hallett 1982, Hallett et al. 1983, but see Gilpin et al. 1982, Rosenzweig et al. 1985, Schoener 1985) will often fail to detect interspecific competition when it occurs and may suggest negative interactions when they d o not exist. In a fluctuating environment, there will be periods when resources are temporarily abundant, and competing species that overlap in their requirements for these limiting resources will tend to increase synchronously. But this need not always occur. Species with quite different

patterns of population dynamics (e.g., a species that is active all year and an obligate hibernator) could still interact strongly if they overlap sufficiently in their use of limiting resources during critical time periods. Moreover, species that do not interact can have positively or negatively correlated dynamics, depending on whether they respond similarly or differently to variation in abiotic conditions (e.g., climate). A combination of nonmanipulative studies of temporal and spatial variation, and experimental manipulations to test specific mechanistic hypotheses, seems to offer the best method of clarifying the complex relationships between community organization and the abundance and distribution of the component species.
Acknowledgements - We thank all of those individuals who have helped over the years to conduct the experiments and to census the rodents at Portal. P. Nicoletto helped with the data analysis and figure preparation, J. Rotenberry and B. Milne contributed statistical advice, and L. Hawkins, B. Kotler, D. Morris, R. Pierotti, and S. Pimm made valuable suggestions for improving the manuscript.

References
Beatley, J. C. 1969. Dependence of desert rodents on winter annuals and precipation. - Ecology 50: 721-724. - 1974. Phenological events and their environmental triggers in Mohave Desert ecosystems. - Ecology 55: 856863. - 1976. Environments of kangaroo rats (Dipodomys) and effects of environmental change on populations in southern Nevada. - J. Mammal. 57: 67-93. Beddington, J . R., Free, C. A. and Lawton, J. H. 1976. Concepts of stability and resilience in predator-prey models. - J. Anim. Ecol. 45: 791-816. Bloomfield, P. 1976. Fourier analysis of time series: an introduction. - Wiley, New York. Bowers, M. A , , Thompson, D . B. and Brown, J . H. 1987. Spatial organization of a desert rodent community: food addition and species removal. - Oecologia (Berl.) 72: 77-82. Brown, J. H . 1973. Species diversity of seed-eating desert rodents in sand dune habitats. - Ecology 54: 775-787. - 1975. Geographical ecology of desert rodents. - In: Cody, M. L. and Diamond, J. M. (eds), Ecology and evolution of communities. Harvard Univ. Press, Cambridge, MA, pp. 315-341. - 1984. Desert rodents: a model system. - Acta Zool. Fennica 172: 4549. - 1987. Variation in desert rodent guilds: patterns, process, and scales. - In: Gee, J . H. R. and Giller, P. S. (eds), Organization of communities: past and present. Blackwell, Oxford, pp. 185-203. - and Harney, B. A. 198x. Population and community ecology of heteromyid rodents in temperate habitats. - In: Genoways, H . H. and Brown, J. H. (eds), The biology of the rodent family Heteromyidae. Spec. Publ. Mammal., no. 11 (in press). - and Munger, J. C. 1985. Experimental manipulation of a desert rodent community: food addition and species removal. - Ecology 66: 1545-1563. - and Kurzius, M. A. 1987. Composition of desert rodent faunas: combinations of coexisting species. - Ann. Zool. Fennici 24: 227-237. - and Kurzius, M. A. 1989. Spatial and temporal variation in guilds of North American desert rodents. - In: Morris, D .
OIKOS 59:3 (1990)

W., Abramsky, Z., Fox, B. J. and Willig, M. R. (eds), Patterns in the structure of mammalian communities. Spec. Publ. No. 28 of The Museum, Texas Tech University, Lubbock, TX, pp. 71-90. - and Zeng, Z. 1989. Comparative population ecology of eleven species of rodents in the Chihuahuan Desert. Ecology 70: 1507-1525. - , Reichman, 0 . J. and Davidson, D. W. 1979. Granivory in desert ecosystems. - A n n . Rev. Ecol. Syst. 54: 201-227. - , Davidson, D . W., Munger, J. C. and Inouye, R. S. 1986. Experimental community ecology: the desert granivory system. - In: Diamond, J. and Case, T. J. (eds), Community ecology. Harper and Row, New York, pp. 41-61. Brown, J. S. 1989. Desert rodent community structure: a test of four mechanisms of coexistence. - Ecol. Monogr. 59: 1-20. Clements, F. E . 1916. Plant succession: and analysis of the development of vegetation. - Carnegie Inst. Publ. No. 242, Washington, DC. Cole, K. L. 1982. Late Quaternary zonation of vegetation in the eastern Grand Canyon. - Science 217: 1142-1145. Conley, W., Nichols, J. D . and Tipton, A. R. 1977. Reproductive strategies in desert rodents. - In: Wauer, R. H. and Riskind, D . H . (eds), Symposium on the biological resources of the Chihuahuan Desert Region. U.S. Dept Interior, Natl. Park Service, Trans. Proc., Ser., pp. 193-214. Connell, J. H . and Sousa, W. P. 1983. On the evidence needed to judge ecological stability or persistence. - Am. Nat. 121: 789-824. Crowell, K. L. and Pimm, S. L. 1976. Competition and niche shifts of mice introduced onto small islands. - Oikos 27: 251-258. Davidson, D. W., Samson, D. A. and Inouye, R . S. 1985. Granivory in the Chihuahuan Desert: interactions within and between trophic levels. - Ecology 66: 486-502. Davis, J. C. 1973. Statistics and data analysis in geology. Wiley, New York. Davis, M. B. 1976. Climatic instability, time lags, and community disequilibrium. -In: Diamond J. and Case, T. J. (eds), Community ecology. Harper and Row, New York, pp. 269-284. Dueser, R. D. and Hallett, J. G . 1980. Competition and habitat selection in a forest-floor small mammal fauna. - Oikos 35: 293-297. Emlen, J . M. 1981. Field estimation of competition intensity. Theor. Pop. Biol. 19: 275-287. Frye, R . J. 1983. Experimental field evidence of interspecific aggression between two species of kangaroo rat (Dipodomys). - Oecologia (Berl.) 59: 74-78. Gilpin, M. E. 1979. Spiral chaos in a predator-prey model. Am. Nat. 113: 306308. - , Case, T. J. and Bender, E . A. 1982. Counterintuitive oscillations in systems of competition and mutualism. Am. Nat. 119: 584-588. Gleason, H. A. 1917. The structure and development of the plant association. - Bull. Torrey Bot. Club 44: 463481. - 1926. The individualistic concept of plant association. Bull. Torrey Bot. Club 53: 7-26. Goh, B. S. 1980. Management and analysis of biological populations. - Elsevier, Amsterdam. Goodman, D. 1987. The demography of chance extinctions. In: Soule, M. E . (ed.), Viable populations. Cambridge Univ. Press, Cambridge, pp. 11-34. Graham, R . W. 1986. Responses of mammalian communities to environmental changes during the Late Quaternary. In: Diamond, J. and Case, T. J. (eds), Community ecology. Harper and Row, New York, pp. 300-313. Grossman, G . D., Moyle, P. B. and Whittaker, J. O., Jr. 1982. Stochasticity in the structural and functional characteristics of an Indiana stream fish community: a test of community theory. - Am. Nat. 119: 611-637. Hallett, J. G. 1982. Habitat selection and the community ma-

trix of a desert small-mammal fauna. - Ecology 63: 140C1410. - and Pimm, S. L. 1979. Direct estimation of competition. Am. Nat. 113: 593-600. - , O'Connell, M. A. and Honeycutt, R . L. 1983. Competition and habitat selection: test of a theory using small mammals. - Oikos 40: 175-181. Hanski, I. 1982. Dynamics of regional distributions: the core and satellite hypothesis. - Oikos 38: 21@221. Hansson, L. and Henttonen, H . 1985. Gradients in density variations of small rodents: the importance of latitude and snow cover. - Oecologia (Berl.) 67: 394-402. Henttonen, H. 1985. Predation causing extended low densities in microtine cycles: further evidence from shrew dynamics. - Oikos 45: 156157. - , Tast, J., Viitala, J. and Kaikusalo, A. 1984. Ecology of cyclic rodents in northern Finland. - Mem. Soc. Fauna Flora Fennica 60: 84-92. - , McGuire, A. D . and Hansson, L. 1985. Comparisons of amplitudes and frequencies (spectral analyses) of density variations in long-term data sets of Clethrionomys species. Ann. Zool. Fennici 22: 221-227. Hillel, D. and Tadmor, N. 1962. Water regime and vegetation in the central Negev highlands of Israel. - Ecology 43: 3341. Holt, R. D. 1977. Predation, apparent competition, and the structure of prey communities. - Theor. Popul. Biol. 12: 197-229. - 1983. Immigration and the dynamics of peripheral populations. - In: Rodin, A , , and Miyata, K. (eds), Advances in herpetology and evolutionary biology. Museum of Comparative Zoology, Harvard Univ., Cambridge, MA, pp. 68M94. Huntley, B. and Birks, H . J. B. 1983. An atlas of past and present pollen maps for Europe: C-13000 years ago. - Cambridge Univ. Press, Cambridge. Kenagy, G. J. 1973. Daily and seasonal patterns of activity and energetics in heteromyid rodent community. - Ecology 54: 1201-1219. - and Bartholomew, G. A. 1985. Seasonal reproductive patterns in five coexisting California desert rodent species. Ecol. Monogr. 55: 371-397. Kot, M. and Schaffer, W. M. 1984. The effects of seasonality on discrete models of population growth. - Theor. pop. Biol. 26: 34C-360. Kotler, B. P. 1984a. Predation risk and the structure of desert rodent communities. - Ecology 65: 689-701. - 1984b. Harvesting rates and predatory risk in desert rodents: a comparison of two communities on different continents. - J . Mammal. 65: 91-96. - 1985. Owl predation on desert rodents which differ in morphology and behavior. - J. Mammal. 66: 824-828. - and Brown, J. S. 1988. Environmental heterogeneity and the coexistence of desert rodents. - Ann. Rev. Ecol. Syst. 19: 281-307. Krebs, C. J. and Myers, J. H. 1974. Population cycles in small mammals. - Adv. Ecol. Res. 8: 267-399. Leigh, E. G . 1975. Population fluctuations, community stability and environmental variability. - In: Cody, M. L. and Diamond, J. M. (eds), Ecology and evolution of communities. Harvard Univ. Press, Cambridge, MA, pp. 51-73. MacArthur, R. H. 1972. Geographical ecology. - Harper and Row, New York. May, R. M. 1976. Simple mathematical models with very complicated dynamics. - Nature, Lond. 261: 459467. - 1980. Nonlinear phenomena in ecology and epidemiology. - Ann. New York Acad. Sci. 357: 267-281. Molles, M. C. and Dahm, C. N. 1990. El Nido, La Nina, and North American stream ecology. - J. North Am. Benth. Soc. 9: 68-76. Munger, J. C. and Brown, J. H. 1981. Competition in desert

rodents: an experiment with semipermeable exclosures. Science 211: 510-512. - , Bowers, M. A. and Jones, W. T . 1983. Desert rodent populations: factors affecting abundance, distribution, and genetic structure. - Great Basin Nat. Mem. 7: 91-116. O'Connell, M. A . 1989. Population dynamics of neotropical rodents in seasonal environments. - J. Mammal. 70: 532548. O'Farrell. M. J . 1974. Seasonal activity patterns of rodents in a sagebrush community, - J. Mammal. 55: 809-823. Ostfeld, R . S. 1988. Fluctuations and constancy in populations of small rodents. - Am. Nat. 131: 445-452. Petryszyn, Y. 1982. Population dynamics of nocturnal desert rodents: a nine-year study. - Unpubl. Ph. D. Dis., Univ. Arizona, Tucson, Arizona. Pimm, S. L. 1985. Estimating competition coefficients from census data. - Oecologia (Berl.) 67: 588-590. Jones, H . 1. and Diamond, J. 1988. On the risk of extinction. - Am. Nat. 132: 757-785. Price, M. V. 1978. The role of microhabitat in structuring desert rodent communities. - Ecology 59: 910-921. Price, M. V. 1986. Structure of desert rodent communities: a critical review of questions and approaches. - Am. Zool. 26: 39-49. - and Brown, J. H . 1983. Patterns of morphology and resource use in North American desert rodent communities. Great Basin Nat. Mem. 7: 117-134. Pulliam, H . R. 1988. Sources, sinks, and population regulation. - Am. Nat. 132: 652461. Quinn, W. H . , Neal, V. T . and de Mayolo, S. E . A. 1987. El Nino occurrences over the last four and a half centuries. - J. Geophys. Res. 92: 14449-14461. Reichman, 0 . J . 1975. Relation of desert rodent diets to available resources. - J. Mammal. 56: 731-751. - 1977. Optimization of diets through food preferences by heteromyid rodents. - Ecology 58: 454-457. - and Van De Graaf, K. M. 1975. Association between ingestion of green vegetation and desert rodent reproduction. - J. Mammal. 56: 503-506. Rosenzweig, M. 1. 1968. Net primary productivity of terrestrial communities: prediction from climatological data. Am. Nat. 102: 67-74. - 1973. Habitat selection experiments with a pair of coexisting heteromyid rodent species. - Ecology 54: 111-117. - and Sterner, P. W. 1970. Population ecology of desert rodent communities: body size and seed husking as bases for heteromyid coexistence. - Ecology 51: 217-224. - and Winakur, J. 1969. Population ecology of desert rodent communities: habitats and environmental complexity. Ecology 50: 558-572. - , Smigel, B. and Kraft, A. 1975. Patterns of food, space and diversity. - I n : Prakash I. and Ghosh, P. K. (eds), Rodents in desert environments. Junk, The Hague, The Netherlands, pp. 241-267. - , Abramsky, Z., Kotler, B. P. and Mitchell, W. A. 1985. Can competition coefficients be determined from census data? - Oecologia (Berl.) 66: 194-198. Schaffer, W. M. 1985. Order and chaos in ecological systems. Ecology 66: 93-106.

, Ellner, S. and Kot, M. 1986. Effects of noise on some dynamical models in ecology. - J. Math. Biol. 24: 479-523. Schoener, T. W. 1985. On the degree of consistency expected when different methods are used to estimate competition coefficients from census data. - Oecologia (Berl.) 67: 591592. - 1986. Patterns in terrestrial vertebrate versus arthropod communities: do systematic differences exist? - In: Diamond, J. and Case, T. J . (eds), Community ecology. Harper and Row, New York, pp. 556586. Schroder, G . E . 1987. Mechanisms for coexistence among three species of Dipodomys: habitat selection and an alternative. - Ecology 68: 1071-1083. Stenseth, N. C. 1983. Causes and consequences of dispersal in small mammals. - In: Swingland, I. R. and Greenwood, P. J. (eds), The ecology of animal movement. Clarendon Press, Oxford, pp. 63-101. Strong, D . R . 1986. Density vagueness: abiding the variance in the demography of real populations. - In: Diamond, J., and Case, T. J . (eds), Community ecology. Harper and Row, New York, pp. 257-268. Taitt, M. J. and Krebs, C. J. 1985. Population dynamics and cycles. - In: Tamarin, R . H. (ed.), Biology of New World Microtus. Spec. Publ. No, 8., Am. Soc. Mammal., pp. 567-620. Tevis, 1.1958. A population of desert annuals germinated by less than one inch of rain. - Ecology 39: 688-695. Thompson, S. D. 1982a. Microhabitat utilization and foraging behavior of bipedal and quadrupedal heteromyid rodents. Ecology 63: 1303-1312. - 1982b. Structure and species composition of desert heteromyid rodent species assemblages: effects of a simple habitat manipulation. - Ecology 63: 1313-1321. Van De Graaff, K. M. and Balda, R. P. 1973. Importance of green vegetation for reproduction in the kangaroo rat Dipodomys merriami merriami. - J. Mammal. 54: 509-512. Van Devender, T. R. 1986. Climatic cadences and the composition of Chihuahuan Desert communities: the Late Pleistocene oackrat midden record. - In: Diamond. J . and Case. T. ~ . ' ( e d s ) ,Community ecology. Harper and Row, New York. DD.285-299. Wagner, ~.'1981.Population dynamices. - In: Arid land ecosystems: structure, functioning and management. Vol. 2, Int. Biol. Progr. 17. Cambridge Univ. Press, London, pp. 125-168. Went, F. W. 1949. Ecology of desert plants. 11. The effect of rain and temperature on germination and growth. - Ecology 30: 1-13. Whitford, W. G . 1976. Temporal fluctuations in density and diversity of desert rodent populations. - J. Mammal. 57: 351-369. Wiens, J. A. 1986. Spatial scale and temporal variation in studies of shrubsteppe birds. - In: Diamond, J. and Case. T. J. (eds), community ecology. Harper and Row, New York, NY, pp. 154-172. Zeng, Z. and Brown, J. H. 1987. Population ecology of a desert rodent: Dipodomys merriami in the Chihuahuan Desert. - Ecology 68: 1328-1340.
-