Predation by wolves interacts with the North Pacific Oscillation (NPO) on a western North American elk population



    Corresponding author
    1. Department of Biological Sciences, University of Alberta, Edmonton, Alberta, Canada T6G 2E9
      Mark Hebblewhite, Department of Biological Sciences, University of Alberta, Edmonton, Alberta, Canada T6G 2E9. Tel: 1 (780) 492 0083; Fax: 1 (780) 492 9234; E-mail:
    Search for more papers by this author

Mark Hebblewhite, Department of Biological Sciences, University of Alberta, Edmonton, Alberta, Canada T6G 2E9. Tel: 1 (780) 492 0083; Fax: 1 (780) 492 9234; E-mail:


  • 1Recent research reveals the widespread influence of the North Atlantic Oscillation (NAO), a large-scale climatic variation, on northern ungulate populations. Little is known, however, about the influence of the North Pacific Oscillation (NPO), a similar climatic index, on ungulates.
  • 2The influence of the NPO on elk population dynamics in Banff National Park (BNP) was examined using a 15-year time-series of three elk subpopulations exposed to different levels of predation by wolves. NPO was strongly related to local climate data including snow depth (r =+0·61) and winter temperatures (r = −0·51).
  • 3Higher NPO values reflected increasing winter severity, reducing elk population-growth rate irrespective of wolf-predation pressure. Elk population-growth rate, however, declined more strongly in areas with wolf predation through the interaction with winter severity, indexed by NPO. Effects of NPO were weaker in the absence of wolf predation.
  • 4Differences between the effects of snow and NPO were revealed that depended on wolf-predation pressure. Without wolf predation, NPO reduced elk population-growth rate, suggesting overall climate was important. With wolf predation, snow depth was more important than NPO, suggesting a mechanism of increased wolf-predation rates in deeper snow.
  • 5This study is the first to demonstrate the influential role of Pacific climate on western North American ungulate population dynamics and provides further evidence of the role of large-scale climatic variation in terrestrial ecosystem dynamics.


Winter climate influences population dynamics of ungulates through effects on locomotion (Telfer & Kelsall 1984), thermoregulation (Cook et al. 1998), reproduction (Portier et al. 1998), predator–prey processes (Post et al. 1999; Hebblewhite, Pletscher & Paquet 2002) and access to food resources (Turner et al. 1994). These effects often interact with population density in populations of northern ungulates (Portier et al. 1998; Milner, Elston, & Albon 1999; Coulson et al. 2001). Early studies used snow depth (Messier 1991), snow-water equivalents (Turner et al. 1994) or local climatic indices (Picton 1984) to measure the effects of climate on population dynamics. Recent studies reveal widespread effects of large-scale climatic variations such as the North Atlantic Oscillation (NAO; Hurrell, Kushnir & Visbeck 2001; Stenseth et al. 2003) on ungulate population dynamics (Post & Stenseth 1999; Patterson & Power 2001).

The NAO climatic phenomenon is measured by the NAO index, a sea-surface pressure-difference index measured between the Azores and Iceland. The effects of the NAO have been demonstrated on population dynamics of soay sheep (Ovis aries L.) in the north sea (Coulson et al. 2001), red deer (Cervus elaphus L.) in the north sea and Norway (Post & Stenseth 1999), white-tailed deer (Odocoileus virginianus L.) in New Brunswick and Minnesota (Post & Stenseth 1998; Patterson & Power 2001) and moose (Alces alces L.) on Isle Royale (Post et al. 1999). The NAO also links predator–prey dynamics in lynx (Lynx canadensis L.) – snowshoe hare (Lepus americanus L.) cycles (Stenseth et al. 1999) and wolf (Canis lupus L.) – moose population dynamics (Post et al. 1999). The mechanisms linking the NAO to local climate may be spatially idiosyncratic. For example, negative NAO values were related to high snowfall in inland areas of Norway, yet warmer, moister winters in coastal areas (Mysterud et al. 2000). In eastern North America, positive NAO values were related to cold, snowier winters (Hurrell et al. 2001). The ecological effects of NOA are pervasive across terrestrial ecosystems throughout the northern hemishphere (Ottersen et al. 2001; Stenseth et al. 2002; Stenseth et al. 2003).

While the NAO has been related to population dynamics of mammals in western North America, relationships were not always as strong as in eastern North America (Stenseth et al. 1999). This difference may be because of the dominance of Pacific climate effects, such as El Niño, or the North Pacific Oscillation (Trenberth & Hurrell 1994) in western North America. South Pacific climate indices such as the El Niño Oscillation (ENSO) influenced mammalian predator–prey systems in Chile (Lima, Stenseth, & Jaksic 2002). Despite growing interest in the ecological effects of the NAO (Stenseth et al. 2002, 2003), few studies have investigated the effects of Pacific climate indices on population dynamics in North America (Stenseth et al. 2003). In North America, Pacific climate variation has been related to mountain hemlock (Tsuga mertensiana) growth (Peterson & Peterson 2001), phenology of flowering plants (Beaubien & Freeland 2000), seabird survival (Jones, Hunter & Robertson 2002) and reproduction (Gjerdrum et al. 2003) and Pacific salmon (Onchorhyncus spp.) populations (Beamish & Bouillon 1993; Mantua et al. 1997). Few, if any, studies have investigated Pacific climate indices impacts on mammalian population dynamics.

I investigated the effects of the North Pacific Oscillation (NPO hereafter) in a wolf–elk time-series from 1985 to 2000 in Banff National Park (BNP) (Hebblewhite et al. 2002) to test the hypothesis that elk population-growth rates (rt) were affected by the NPO. Wolves recolonized BNP through dispersal in the mid 1980s. Similar to density–climate interactions, wolf predation interacts with climate through increased predation rates with increasing snow depths (Nelson & Mech 1986; Huggard 1993; Post et al. 1999). Therefore, I tested for effects of the NPO on elk in three different subpopulations that differed in the amount or timing of wolf predation in order to tease apart predation and climate impacts (Hebblewhite et al. 2002). I examined relationships between the NPO, NAO and local climatic indices to investigate mechanisms of any emergent relationship between wolf–elk dynamics and the NPO. Finally, I compared the strength of relationships between the NPO and local climate measures on wolf–elk dynamics.

Materials and methods

study area

Banff National Park (BNP), 6641 km2 in area, is on the eastern slope of the continental divide of the Canadian Rocky Mountains (1400–3400 m, 51°15′/116°00′, Fig. 1). The climate is continental with long, cold winters and infrequent warm weather caused by Pacific-influenced Chinook winds and short, relatively dry summers. Holland & Coen (1983) describe vegetation in the study area. Elk are the most abundant ungulate of a diverse ungulate community and comprise 40–70% of the diet of wolves (Hebblewhite et al. 2004). Mean winter snow-pack depth in the valley-bottom was 43 cm at Banff (Hebblewhite et al. 2002) but increased at higher elevations. I divided the Bow Valley of BNP into three zones that followed elk subpopulation dynamics and that differed in wolf-predation pressure; the eastern zone (67 km2, the central zone (43 km2), and the western zone (181 km2) (see Hebblewhite et al. 2002 for more details).

Figure 1.

Location of Banff National Park, Alberta, Canada.

climatic data − north pacific oscillation (npo)

The NPO is based on sea-surface temperature anomalies measured over the North Pacific Ocean (Trenberth & Hurrell 1994; Fig. 2). The NPO index is an area-weighted measure of mean sea-surface pressure over the North Pacific from 30° to 65°N, 160°E to 140°E (Trenberth & Hurrell 1994). Trenberth & Hurrell (1994) and others (Mantua et al. 1997) describe decadal length oscillations linked to the ENSO in the NPO time-series. In the Rocky Mountains, positive NPO index values are related to cooler winter temperatures, increased snowfall and increased frequency of winter storms (Trenberth & Hurrell 1994; Mantua et al. 1997). The Canadian Rocky Mountains show some of the strongest continental correlations (r = −0·50) with NPO indices of storm frequency, December–February precipitation, and November–March surface air temperatures (Trenberth & Hurrell 1994; Mantua et al. 1997). Standardized NPO anomaly values (which measure deviations) for November–March obtained from the Climate Analysis Section of the National Centre for Atmospheric Research (NCAR), USA (available from were used for analyses. Refer to Trenberth & Hurrell (1994) for more detail on the NPO.

Figure 2.

The standardized North Pacific Oscillation (NPO) Anomaly Index for November to March, 1972–2003 (data from the Climate Analysis Section of the National Centre for Atmospheric Research (NCAR), USA, Positive values relate to more severe winters in the Canadian Rocky Mountains (Trenberth & Hurrell 1994).

relationships between npo and local climate data

I tested for relationships between the NPO and local climate data between 1970 and 2001, and within the shorter 15-year time-series to confirm patterns. First, I related NPO to local snow depth, measured by Hebblewhite et al.'s (2002) mean snow-depth index consisting of averaged snow depths measured at 15-day intervals between 15 October and 30 April, 1972–2000, in the centre of the study area (1700 m). Next, I tested for relationships between NPO and mean monthly winter temperatures obtained from Environment Canada at the town of Banff (1545 m). Both local snow depth (Huggard 1993) and local temperatures (Env. Canada, unpublished data, Holland 1983) reflected conditions experienced throughout the Bow valley elk zones. For example, Huggard (1993) found strong covariation in snow depth between three sites in the central and western zones. I tested for relationships between local and large-scale climate indices using Pearson correlations in systat (Wilkinson 1998). While Post & Stenseth (1999) noted the importance of time-lag effects of snowfall on ungulate population growth and survival (but see Messier 1995), Hebblewhite (2000) found no support for such effects on elk in BNP. Therefore, I used only the previous winter's snow depths. Finally, I tested correlations between the NAO winter values obtained from the NCAR (available at, the NPO and local climate indices for BNP.

elk and wolf population data

I tested for the effects of NPO on three elk subpopulations in different zones of the Bow Valley of BNP within which wolf, human and elk densities varied. Wolves recolonized the western zone first in 1986, partially recolonized the eastern zone in 1991, and failed to recolonize the central zone due to high human use (Paquet et al. 1996; Callaghan 2002). Elk subpopulation dynamics and movements were biologically separate between these three zones (Woods 1991). Elk populations were aerially surveyed in all three zones from 1985 to 2000 during late winter by Parks Canada (Fig. 3, see Hebblewhite et al. 2002). Within each zone, intrinsic elk population-growth rate (hereafter, rt) was estimated as rt = ln(Nt+1/Nt). Within each zone I used the NPO, winter elk density (Nt), counts of elk killed by humans on highways and railways, and estimates of the number of wolf-killed prey as independent variables to explain variation in rt (variables are described in more detail in Hebblewhite et al. 2002). To account for highway fencing in the central zone in 1991 (Clevenger, Chruszcz, & Gunson 2001), I included a variable coding for fencing in this zone. Per-capita wolf kill-rates (elk/day/wolf) and wolf-pack sizes (number of wolves) were estimated following Hebblewhite et al. (2004), and their product (Messier 1994) estimates the number of elk killed per winter (181 days) in each zone. Differential recolonization provided an approximate before-after-control-impact (BACI) experimental design to evaluate the effects of wolf predation on elk (Minta, Kareiva, & Curlee 1999).

Figure 3.

Elk population time-series in Banff National Park, AB, winter 1986–2000, in the three different zones, east (wolves recolonized 1992), central (no wolves) and western (wolves recolonized 1986). Counts were converted to densities for analysis.

statistical analyses

I tested how elk density, winter NPO index, winter wolf kill-rate, winter human-caused elk kill-rate and central zone highway fencing affected elk rt within each of the three zones (Fig. 3) over the 15-year time-series. I also tested for two important ecological interactions between (1) NPO and wolf kill-rate (Huggard 1993; Post et al. 1999) and (2) NPO and elk density (Portier et al. 1998). I developed a set of candidate generalized linear models (GLMs) comprising different hypotheses of factors affecting elk rt (e.g. Burnham & Anderson 1998). GLMs were of the general form:


where t = 1–15 years, N is population size in year t, rt = population growth in year t, b0 is the intercept, b1bm are coefficients of independent variables x1 xm, ɛ is random error where Σ(ɛι) = 0.

GLMs were estimated using maximum likelihood estimation (Type III) in sas (proc genmod in sas 8·0, SAS Institute 1998). Models were ranked using the Akaike information criterion (AICc) (Burnham & Anderson 1998). Where model selection uncertainty arose, an approximate cut-off of ΔAICc = 4 was used to estimate the top model set (Burnham & Anderson 1998). I used the sum of Akaike weights (ωi) for each variable to rank variables by importance (Burnham & Anderson 1998: 141), following suggestions for the presentation of information-theoretic results in ecological research (Anderson et al. 2001). Autocorrelation is often problematic in time-series analysis (Royama 1992). Therefore, I corrected SEs in the b0 …ms following Post & Stenseth (1998) where a variance inflation adjusts the degrees of freedom for beta coefficients according to Bartlett (1946) and Priestly (1981); N = N′[(1 − α1α2)/(1 + α1α2)], where N is the number of observations (d.f.), α1 is the first-order autocorrelation in the dependent, and α1 is the first-order autocorrelation in the independent variable, bm. When one series is negatively autocorrelated, this and other approaches (i.e. Newey–West adjustment, McCullough & Nelder 1989) lead to increased degrees of freedom, hence we set N′ ≤N following Post & Stenseth (1998).

comparison of effects of npo and snow depth on elk

The relative strength of the effect of local snow depth vs. the NPO on elk rt was evaluated by first comparing simple correlation coefficients with elk rt across zones. Next, model fit was compared for the top model in a zone with snow and NPO effects to compare the strength of NPO vs. snow on elk population dynamics within a zone (e.g. comparing WOLF + SNOW and WOLF + NPO models). Model fit was assessed using a limited form of cross-validation (Hastie et al. 2001) where 1 year was randomly excluded and models were re-fitted. The top NPO or snow model in each zone was reformulated into a difference equation following Merrill & Boyce (1991) to predict Nt+1 to compare model predictions to observed elk density. Prediction, and hence model fit, was evaluated using modified r2 for each model (Hastie et al. 2002).


relationships between npo and local climate variables

Local snow depth was strongly positively related to the NPO anomaly index between 1970 and 2001 (Fig. 4, correlation r =+0·61, P = 0·0005). During the reduced elk population time-series (n = 15 years), the strength of this relationship diminished, but was still positive (r =+0·45, P = 0·05). Similarly, mean monthly winter temperatures declined with increasing positive values of NPO (r = −0·51, P = 0·008) in BNP. The NAO was unrelated to NPO (r = 0·14, P = 0·44) or BNP snow depths (r = 0·28, P = 0·11).

Figure 4.

Relationship between standardized North Pacific Oscillation (NPO) index and the standardized BNP snow-depth index reported in Hebblewhite et al. (2002) for the period from 1970 to 2001. (NPO data are from the Climate Analysis Section of the National Centre for Atmospheric Research (NCAR), USA,

effects of npo on wolf–elk population dynamics

Autocorrelation.  Not surprisingly, original time-series were significantly autocorrelated at lag 1 for elk density (Fig. 3, eastern zone ACF r = 0·57, P = 0·014, central r = 0·45, P = 0·05, western r = 0·68, P = 0·001) and wolf predation (eastern ACF r = 0·56, P = 0·016, central r = −0·36, P = 0·14, west r = 0·48, P = 0·05) in all three zones. However, the dependent variables in all GLMs, elk population-growth rate rt, were negatively autocorrelated (eastern ACF r = −0·23, P = 0·33, central r =−0·57, P = 0·015, west r = −0·11, P = 0·64). Therefore, using Bartlett's (1946) adjustment led to N′ > N for all beta coefficients, suggesting that opposite signed autocorrelations acted to counteract each other (McCullough & Nelder 1989). Hence, we set N′ = N for all betas and present SEs unadjusted for autocorrelation to be conservative. Confirmatory analyses using the Newey–West estimator yielded identical results (unpublished data).

central zone

The central zone elk population was regulated by elk density, experienced higher population-growth rate after fencing and had reduced population-growth rates during positive (more severe) NPO winters (Table 1). Models of rt fitted the data well in the central zone (Table 1), and all top models explained a high proportion of the deviance (Table 1). Across all models, individual variables were ranked in order of importance as elk density (ωelk density = 0·95), fencing (ωfencing = 0·69), the interaction between elk density and NPO (ωelk* NPO = 0·28), NPO itself (ωNPO = 0·26), wolf predation (ωwolf = 0·162) and human-caused mortality (ωhuman = 0·13).

Table 1.  Elk population dynamics models for the three BNP elk populations, 1986–2000, relating elk population-growth rate (rt) to a linear function of NPO, wolf predation, human-caused elk mortality, highway fencing and elk density, with model deviance, model structure and corresponding ΔAICc and Akaike weight, ωi (Burnham & Anderson 1998)
Model structureDevianceΔAICcωib0s.e.b1s.e.b2s.e.b3s.e.
  1. Maximum likelihood estimates (MLE, type III) for coefficients in the models are presented with standard errors, coefficients with likelihood ratio χ2 probabilities < 0·05 are highlighted in black, and < 0·10 marked by *. Coefficients are reported in the order in the model, i.e. for βi, i = 1 if the model has only one parameter.

Central zone − low wolf use
1. b0 +b1(ELK) +b2(ELK NPO) +b3(FENCE)0·1350·000·32   0·970·160·120·080·0050·00080·270·08
2. b0 +b1(ELK) + β2(FENCE)0·1751·070·19   0·890·1740·110·023   0·230·09 
3. b0 +b1(ELK) +b2(NPO) +b3(FENCE)0·1452·060·11   0·9970·170·120·020·020·01 0·260·08
4. b0 +b1(ELK) +b2(FENCE) +b3(HUMAN)0·1573·280·06   1·030·190·120·02   0·270·09−0·0060·005
5. b0 +b1(ELK)0·253·500·05   0·630·170·120·02  
Eastern zone models − wolf recolonization in 1991
1. b0 +b1(WOLF) +b2(NPO) +b3(WOLF * NPO)0·5070·000·36   0·060·06−1·03*0·570·090·02 1·250·52
1. b0 +b1(NPO) +b2(WOLF NPO)0·6191·200·20   0·060·05   1·630·540·100·027 
1. b0 +b1(WOLF) +b2(ELK)0·8082·000·13   0·420·191·990·670·130·06 
1. b0 +b1(WOLF) + β2(NPO)0·7033·090·08   0·030·070·0710·028−1·580·61 
Western zone models − wolf recolonization in 1985
1. b0 + ɛ1·4590·000·250·1200·081   
2. b0 +b1(NPO)1·4570·780·17−0·120·08−0·0070·03  
3. b0 +b1(ELK)1·3640·990·15−0·0070·135−0·120·116  
5. b0 +b1(HUMAN)1·4061·440·12−0·0430·13−0·951·25   
5. b0 +b1(WOLF)1·4062·750·06−0·0070·13−0·330·84   

eastern zone

Eastern zone elk population-growth rate declined with increased wolf predation, positive NPO values and through the interaction of NPO and wolf predation (Table 1). The model fit in the top set was adequate (Table 1, all deviances < 1). Elk rt was related to variables in the following order of importance; wolf predation (ωwolf = 0·90), NPO (ωNPO = 0·85), the interaction between wolf predation and NPO (ωelk*NPO = 0·68), elk density (ωelk density = 0·25) and human-caused mortality (ωhuman = 0·11).

western zone

Model fit was poor across all top models (Table 1, all deviances > 1), and the best model was a constant rate of decline of rt = −0·12 in the western zone. Models in the top model set were restricted to simple one-parameter models, suggesting that model fit was too poor to accommodate additional parameters.

comparison of effects of npo and snow depth on elk populations

Snow depth was more strongly correlated with elk population-growth rate in zones with wolf predation (Table 2), whereas NPO was more important in zones without wolf predation (Table 2). Similarly, cross-validation revealed snow-depth models generally were better predictors of elk population-growth rates in zones with wolf predation (Table 3), and NPO was a better predictor in the zone without wolf predation (Table 3).

Table 2.  Correlation coefficients between elk population-growth rate, with NPO and BNP snow-depth indices from Hebblewhite et al. (2002)
ZoneSnow modelsNPO models
Table 3.  Cross-validation of top NPO models from this study and snow-depth models from Hebblewhite et al. (2002) for each zone showing correlations between predicted and observed population-growth rate for elk in BNP
 Cross-validation r2
Central zone
 Elk + snow + fence0·61
 Elk + NPO + fence0·69
Eastern zone
 Wolf + NPO + (WOLF * NPO)0·63
 Wolf + snow + (WOLF * SNOW)0·58
Western zone


This study demonstrates the effects of large-scale Pacific climate variation on a western North American ungulate population, confirming similar responses found in ungulate populations to the NAO in eastern North America and Europe (Post & Stenseth 1998; Post et al. 1999; Patterson & Power 2001). This research is the first to confirm similar effects on North American ungulates, revealing the influence of Pacific climate on terrestrial dynamics as far inland as the Canadian Rockies. Both local climatic comparisons to the NPO (Fig. 4), and more extensive meteorological comparisons (Trenberth & Hurrell 1994; Mantua et al. 1997) support the important role of Pacific climatic variation affecting local climate. Through these large-scale climatic effects, elk population-growth rate was affected by the NPO through interaction with predation by wolves and, to a lesser degree, interaction with elk density in the absence of wolves (Table 1).

The link between local climate and the NPO was as strong, or stronger, than connections between the NAO and local climate reported elsewhere. For example, Post & Stenseth (1999) report correlation coefficients of r =−0·33 and −0·30 between the NAO and snow in Minnesota while this study reports r =+0·60 between snow depth and NPO, and −0·50 between winter temperatures and the NPO. Trenberth & Hurrell (1994) discuss the relatively stronger Pacific influence and dominant westerly flows as a continental mechanism for these strong connections to local climate. Hence, future research in western North America, especially in the Rocky Mountains and coastal areas (Trenberth & Hurrell 1994) might usefully investigate effects of the NPO on animal population dynamics.

Differences between similar analyses of local effects of snow depth on elk population dynamics (Hebblewhite et al. 2002) and NPO suggest different climatic mechanisms operating on elk populations. Effects of snow depth were stronger in zones with wolves, suggesting the mechanism of wolves increasing kill-rates with increasing snow depths was responsible (Nelson & Mech 1986; Huggard 1993; Post et al. 1999). Without wolf predation, Hebblewhite et al. (2002) found no negative effect of snow depth on central zone elk rt. However, NPO did reduce rt in the central zone (Table 1), revealing important differences between local measures and large-scale climatic indices. Perhaps overall climate, but not snow, would still impact elk negatively due to temperature, winds and precipitation (Cook et al. 1998), particularly at the high elk densities in this zone (Portier et al. 1998). In a review of effects of NAO on ungulate populations, Post & Stenseth (1999) found NAO better-explained population dynamics in 10 of 19 ungulate time-series. This suggests, as did this study, that the influence of large-scale climate on population dynamics is varied, and may be mediated through small-scale local processes. Future analyses of ungulate population dynamics will probably benefit from the simultaneous consideration of both local and large-scale effects (e.g. Wang et al. 2002a).

Comparison to the snow models of Hebblewhite et al. (2002) showed thar rt was related more strongly to snow depth with wolf predation, and to NPO without wolf predation (Table 2). Stronger partial correlation coefficients (Table 2) and higher cross-validation r2-values in wolf-zones provides further evidence for the mechanism of increasing snow depth increasing predation rates by wolves (Nelson & Mech 1986; Huggard 1993; Post et al. 1999). Stronger local climatic effects on elk population-growth rate in the presence of wolves may help to explain why local climate effects were stronger than NAO in some populations in Post & Stenseth's (1999) review. If populations were subject to predation, the local interaction between snow depth and predation may be more important than large-scale indices, potentially contributing to differences across populations.

The interaction between wolf predation and NPO was stronger (by the ratio of Akaike weights) than the interaction between elk density and NPO (Table 1). Similarly, Hebblewhite et al. (2002) found no snow–density interaction without wolves, yet a stronger snow–predation interaction with wolf predation. Therefore, density–climate interactions were weak without wolf predation. Many studies documenting strong climate–density interactions are from populations without significant predators (i.e. Portier et al. 1998). Thus, predation may reduce densities below climate–density interaction thresholds, only to be replaced by climate–predation interactions. Further, predation may reduce densities such that the effects of environmental stochasticity are reduced (Eberhardt 2002).

The cross-zone comparison suggests wolf predation can reduce elk density below the range at which density-dependent effects play an important role (for more details see Hebblewhite et al. 2002), consistent with wolf limitation and regulation of ungulates reported elsewhere (Gasaway et al. 1992; Dale, Adams & Bowyer 1994; Messier 1994; Jedrzejewski et al. 2002). Before wolf recolonization elk density regulated growth rate in the eastern zone, and after wolf predation and snow/NPO limited growth rate. Without wolves, elk density, human-caused mortality, and the NPO limited elk rt in the central zone. The western zone highlights problems with correlative approaches to the study of population dynamics because of the near constant rate of decline (Boyce & Anderson 1999). However, elk densities were similar in all three zones prior to wolf recolonization in 1986 (Fig. 3, Hebblewhite et al. 2002). Elk decline in the western zone with wolf predation, when compared to eastern and central zones, suggests that wolf predation limited elk populations in concert with the NPO.

In conclusion, this is the first report of a large-scale Pacific climate variation, the NPO, influencing ungulate population dynamics in combination with wolf predation. Future research is anticipated to reveal further the important role of the NPO on mammalian population dynamics in western North America. Anticipating the impacts of future global climate change on ecological systems is an important area of applied research (Scott, Malcolm & Lemieux 2002; Stenseth et al. 2002; Walther et al. 2002). Indices such as the NPO may be used to help forecast future impacts of climate change on ungulate population dynamics (Wang et al. 2002b). However, I agree with Lima et al. (2002) that to realize the full utility of such approaches, improved knowledge of the interactions of predation and climate must be developed.


Parks Canada, the Central Rockies Wolf Project, John/Paul and Associates, Alberta Human Resources and Employment, Human Resources Development Canada, Canadian–Pacific Foundation, Shell Environment Fund, Paquet Wildlife Fund, and World Wildlife Fund–US and Canada funded this research. The Banff National Park Warden Service and numerous research assistants helped to collect field data. I thank S. Lele, P. Biro and M.S. Lindberg for advice on statistical modelling, and J. Hurrell for help in interpreting NPO climate data. MH was supported by a Canon–National Parks Science Scholarship for the Americas during preparation of this manuscript. I thank D. Haydon for valuable editorial guidance and suggestions.