El Niño/Southern Oscillation-driven rainfall pulse amplifies predation by owls on seabirds via apparent competition with mice

Abstract

Most approaches for assessing species vulnerability to climate change have focused on direct impacts via abiotic changes rather than indirect impacts mediated by changes in species interactions. Changes in rainfall regimes may influence species interactions from the bottom-up by increasing primary productivity in arid environments, but subsequently lead to less predictable top-down effects. Our study demonstrates how the effects of an EL Niño/Southern Oscillation (ENSO)-driven rainfall pulse ricochets along a chain of interactions between marine and terrestrial food webs, leading to enhanced predation of a vulnerable marine predator on its island breeding grounds. On Santa Barbara Island, barn owls (Tyto alba) are the main predator of a nocturnal seabird, the Scripps's murrelet (Synthliboramphus scrippsi), as well as an endemic deer mouse. We followed the links between rainfall, normalized difference vegetation index and subsequent peaks in mouse and owl abundance. After the mouse population declined steeply, there was approximately 15-fold increase in the number of murrelets killed by owls. We also simulated these dynamics with a mathematical model and demonstrate that bottom-up resource pulses can lead to subsequent declines in alternative prey. Our study highlights the need for understanding how species interactions will change with shifting rainfall patterns through the effects of ENSO under global change.

1. Introduction

Fluctuations in climate can alter species interactions, and through these perturbations we can gain insight into shifting trophic controls in food webs [1–4]. For example, the El Niño/Southern Oscillation (ENSO) is the most prominent of the global climate fluctuations, occurring at 2–7 year intervals and impacting productivity in both marine and terrestrial environments via physical forcing that affects the base of both food webs [5]. Specifically, ENSO causes weakening trade winds and increased sea surface temperature that drives primary production in the ocean [5], while also inducing drought or extreme rainfall in different parts of the globe that influences the primary productivity of terrestrial plants [6]. These bottom-up changes have effects that end up propagating throughout food webs, often through complex interactions [6,7], that can possibly circle back in ways that connect both marine and terrestrial food webs [8].

Islands are part of the global coastal ecotone where terrestrial and marine food webs connect [9], and this connection means island communities may respond to ENSO events driven by changes to either food web. For example, islands with seabirds can experience high nutrient input from the marine environment via guano deposition [10]. These subsidies can then be disrupted during ENSO events when seabirds fail to return to the colony when foraging conditions are poor [11], which subsequently impacts island vegetation [8,12]. Islands can also be affected by ENSO-associated rainfall, and these impacts are particularly spectacular on islands that are arid or semi-arid because rainfall limits primary productivity in these areas [13]. After years with heavy rainfall, the vegetation flourishes, and arthropod and small mammal populations can all increase dramatically [14–17]. In both cases, the bottom-up effects of ENSO are well described, but very little attention has been paid to potential subsequent top-down effects.

These top-down effects may be particularly strong on islands, as they often have relatively simple systems across which cascades can easily transmit [18]. For example, rodents are a key mesopredator species on many islands, and increases in their abundance can trigger increases in top predator populations, which can lead to indirect interactions that could impact alternative prey species through ‘apparent competition’ [19,20]. This type of indirect interaction, where two species share a predator, is a fundamental community interaction module common to many ecosystems [21]. In some cases, once the resource pulse subsides and primary prey densities decline, prey switching by generalist predators can lead to a further increase in predation on alternative prey [20]. If the decline is severe enough, models that have been developed for apparent competition predict that such enhanced predation on alternative endangered prey could lead to extinction through hyperpredation [19,22]. However, often these predator-mediated indirect effects tend to be far less appreciated than direct effects [23], and only a few studies have addressed how indirect interactions are influenced by rainfall variability (e.g. [24]).

Here, we describe a chain of interactions triggered by an ENSO-associated rainfall pulse and the resulting top-down cascade on an oceanic island in Southern California. In this semi-arid region, winters with increased rainfall are strongly correlated with ENSO events, which is a pattern that has persisted for at least hundreds of years [25,26]. On Santa Barbara Island, there is an endemic subspecies of deer mouse (Peromyscus maniculatus elusus) and a native population of barn owls (Tyto alba) that are known to reach extremely high densities that are associated with rainfall patterns [27]. Barn owls are generalist predators [28], and there is increasing concern about the impact they have on a declining population of nocturnal seabirds, the Scripps's murrelet (Synthliboramphus scrippsi; [29]). Therefore, we hypothesized that owl predation on murrelets would depend on the density of mice, which would in turn be influenced by variations in rainfall and food availability. To test this, first, we confirmed the links between an ENSO-driven rainfall pulse, terrestrial productivity and the subsequent increase and then sharp decline in mouse and owl numbers on the island. Next, we evaluated evidence for prey switching in barn owls and demonstrate that an ENSO-driven resource pulse mediated the top-down cascade on this threatened seabird. Finally, we simulated these dynamics with a mathematical model and demonstrate that bottom-up resource pulses can lead to subsequent declines in the murrelet population.

2. Material and methods

(a). Study location and system

Santa Barbara Island (33°29′ N, 119°02′ W) is the smallest of the California Channel Islands and is managed by the Channel Islands National Park. This 2.6 km² island is located about 63 km offshore and is 39 km and 45 km from its closest neighbours, Santa Catalina Island and San Nicolas Island, respectively. The island receives an estimated average of 21.76 cm of rainfall annually [30]. Steep cliffs rise abruptly up from the sea on nearly all sides of the island, above which is a gently sloping terrace covered mostly by non-native grassland (e.g. Avena spp, Bromus spp, etc.) and patches of low growing native shrubs and cacti (Leptosyne gigantea, Eriogonum giganteum var. compactum and Opuntia spp.; [31]). The two tallest peaks are 193 m and 171 m high and five small canyons cut into the south and east sides of the island.

Scripps's murrelets (hereafter ‘murrelets’) are small (approx. 165 g) pursuit diving alcids that have a breeding distribution of approximately 10 island groups off the coast of Southern California and Mexico [29,32,33]. Murrelets are listed as Vulnerable by the International Union for Conservation of Nature and believed to be declining [34]. On Santa Barbara Island, there are estimated to be 475–650 breeding pairs (D. Whitworth 2017, personal communication) and their breeding season typically extends from March to July [32]. Murrelets and barn owls nest primarily along the coastal areas of the island within crevices in the rocky cliffs, in sea caves and to a smaller extent underneath shrubs, while mice are found throughout all the island's habitats. The island also has a population of island night lizards (Xantusia riversiana), as well as several species of breeding and migrant landbirds [35]. Barn owls on Santa Barbara Island are demographically isolated from the mainland [36], although the degree to which dispersal occurs among islands has yet to be quantified. Evidence of owl breeding activity could be found during all months of the year, at least in some years, and up to eight eggs were found in nests (S.K.Thomsen 2012, unpublished data).

(b). Temporal trends

(i). Normalized difference vegetation index and rainfall

Rainfall in this region occurs primarily during the winter months (November–March), so we summed the daily precipitation amounts from a weather station on nearby Santa Catalina Island (Avalon Airport 33°24'18.0″ N, 118°24'57.6″ W) into rainfall year totals beginning on 1 April each year from 2005 to 2013. Next, to quantify differences in spring primary productivity from 2006 to 2013, we calculated the normalized difference vegetation index (NDVI) for the island with 30 m resolution Landsat satellite images [37]. Values of NDVI are correlated with primary productivity [38], and small mammal abundance [39], and it has been used to predict mouse densities (Peromyscus spp.) at approximately 1 year time lags [40]. We selected cloud-free images captured during March (mean ordinal date = 73; range: 59–90), which coincided with mark–recapture studies of deer mice. We used ArcGIS v. 10.1 [41] to calculate the NDVI values from the satellite images and then determined an average island-wide value for each year.

(ii). Deer mouse monitoring

We calculated deer mouse densities using mark–recapture data collected annually by the National Park Service on two plots during the month of March as well as once during the autumn/winter season (September–December) from 2007 to 2013. We focus on this time period because there are no missing observations and the timing of spring captures was consistent among years. Each plot has 100 permanent trap stations that are arranged in a 10 × 10 pattern spaced 7 m apart, at which one small Sherman live-trap is placed during trapping sessions that are conducted over three consecutive nights [42]. Abundance was estimated using Huggins [43] closed population capture–recapture models that were implemented in program MARK. Density was then calculated by dividing abundance by the estimated area trapped [44] (see the electronic supplementary material, appendix S1). Mouse captures in 2013 were too low to reliably estimate abundance, so we excluded that year from statistical calculations but display the minimum number known alive per hectare in figures.

We also established additional sites where we used track tubes to obtain an index of mouse density from 2011 to 2013 within murrelet nesting areas [45]. Each of five sites in murrelet habitat consisted of nine track tubes that were placed in a 3 × 3 grid formation with 7 m spacing. Track tubes were deployed before nightfall during the new moon phase twice during the murrelet breeding season (April–May), and the following day, we recorded the number of track tubes in each grid that had the presence of mouse tracks. The numbers for each grid were then averaged for the two months to create a metric for mouse density during each murrelet breeding season.

(iii). Barn owl abundance

We monitored the relative abundance of barn owls on the island from 2010 to 2014 by repeating the trail transect survey methods developed by Drost [46] twice each year in mid-winter (January–February) and late summer (July–August). Surveys began approximately 1 h after sunset, when two observers walked the island trails with high intensity flashlights (72 lumens) and recorded the times and locations of all observations of barn owls. The total of all detections from all observers were added together to obtain a total count. All surveys had similar conditions of winds less than 15 knots, and no precipitation or fog.

Barn owls on the island have small and extensively overlapping home ranges and tended to concentrate their activity by their roost sites [47]. Therefore, to quantify differences in owl abundance across the island, we also monitored the number and locations of owl nests and roost sites from 2010 to 2013. Surveys were conducted in mid-winter (December–February) and summer (June–September) within the canyons as well as shoreline cliff habitat accessible by non-technical climbing from the top of the island. Sites were determined to be active if there was the presence of an owl, signs of breeding activity or if pellets were observed. There was also one annual visit in April/May to one roosting site located within a sea cave (Barn Owl Cave), during all 4 years. No mid-winter visits were conducted to collect pellets in 2011 in order to avoid potentially fatal disturbance to nesting brown pelicans (Pelecanus occidentalis), so these sites were monitored for the presence of owls from a distance and were later confirmed by collecting pellets from these sites during the summer.

(iv). Barn owl diet

We assessed the diet of barn owls from 2010 to 2013 using two complementary methods. First, we looked at annual changes in owl dietary breadth by identifying faunal remains in regurgitated pellets collected from owl nest and roost sites during surveys for these sites. Second, we also collected avian prey remains from areas surrounding owl nest and roost sites as well as from murrelet nesting habitat and counted the number of individual seabirds in those remains. These areas included the murrelet nest monitoring plots and along the island's hiking trails, both of which were checked one to two times per week from March to July each year (see http://www.montroserestoration.noaa.gov/multimedia/publications). Although not all areas of the island were searched with the same intensity within a year, our methods were consistent across years, which allow us to make comparisons between years. We used a handheld GPS unit to record the location of the remains, so that we could link patterns in the number of carcasses collected in different areas to local mouse and owl abundance.

After collection, pellets were disaggregated in the laboratory and all bones were identified to species. The number of individual pellets collected varied each year owing to changes in owl abundance on the island, but we were able to analyse a minimum of 100 owl pellets per year for diet composition. When the remains of more than one individual from a species were present in each pellet, the number of skulls and/or lower mandibles was used to determine the number of prey items (minimum number of individuals; MNI) of each species per pellet. Seabird prey remains were identified to species and collected when found. All other land bird remains were identified and deposited at the Santa Barbara Museum of Natural History, Santa Barbara, CA.

(c). Data analyses

All statistical analyses were performed in R v. 3.2.1 [48]. First, we evaluated whether there were temporal relationships between regional annual rainfall, NDVI and mouse density. To do so, we used simple linear regressions to examine annual rainfall and island-wide NDVI in March, as well as NDVI in March of year t − 1 and log-transformed mouse density in March of year t. We then compared the counts of prey types (mice, island night lizards and all seabird species combined) within owl pellets among years with a contingency table χ²-test of differences. Rare prey items like arthropods and land bird species were not included in order to meet the assumptions of this test. It is possible that some of the seabird carcasses found on the island were from other predators, such as peregrine falcons (Falco peregrinus; [32]). Therefore, we also confirmed that the total number of carcasses found each year and the proportions of seabirds found in owl pellets for each year were related using correlation analyses (r = 0.89).

Next, we used piecewise structural equation modelling (with the R package ‘piecewiseSEM’) to conduct a confirmatory path analysis of the effects of the ENSO-associated rainfall pulse on owl predation of murrelets in spatially replicated plots. The Piecewise structural equation model (SEM) has advantages over traditional path analysis because it allows fitting several different individual linear models with random effects that together form a network of predictive relationships [49]. To be sure we had spatially linked data for this analysis, we created 250 m spatial buffers with ArcGIS around each mouse track tube grid located in murrelet nesting habitat. We used the estimated maximum number of active owl roosts observed within the 250 m buffer distance during the mid-winter period preceding each murrelet breeding season 2010–2013 as a proxy for localized owl abundance [47]. We then summed the number of murrelet remains found within that distance for each year by using the GPS locations of the remains of murrelets found killed. We fitted linear mixed effects models for each hypothesized pathway. We included site (n = 5) as a random effect as well as a temporal correlation term with the function corCAR1 from the package ‘nlme’ in all models [50]. Our a priori path model included: (i) NDVI from the year t − 1 influences the (log-transformed x + 1) mouse density in April–May of year t as well as the number of owl roost sites in the mid-winter period of year t; (ii) the number of owl roost sites in the mid-winter period influences mouse density in April–May of year t; and (iii) both mice and owls influence the (log-transformed x + 1) number of murrelets killed during their breeding season of year t. Hence, the model includes the lagged direct and indirect effects of NDVI on both mice and owl abundance the following year, capturing the bottom-up processes that in turn drive the top-down effects of owls on mice and murrelets. The d-separation test was conducted to determine if the overall model was a good fit, which is indicated by a non-significant p-value [51], and standardized beta coefficients were used to compare effect sizes and directions for each path.

(d). Mathematical model

Relatively few studies that examine the ecological effects of extreme climatic events have been able to include more than one event [52], therefore the simulation of these climate-driven predator–prey interactions is necessary to understand the potential impacts of ENSO-associated rainfall under global change. To test whether climate could influence murrelet population trends via predation by owls, we parametrized a mathematical model consisting of a set of ordinary differential equations modified for our system from the classic hyperpredation model [53]. Details of this model and others like it have been covered elsewhere [19,22] and parameters were based on empirical or derived sources for our system (electronic supplementary material, appendix S2; table 1). Differential equations were solved numerically with the R package ‘deSolve’ [54]. Briefly, all three species are characterized by intrinsic growth rates (r_i), carrying capacities (K_i) and mortality rates (μ_i). The carrying capacities of mice (M) and owls (O) are linked by their respective predation rates on their terrestrial prey, which is itself influenced by climate variability. Murrelets (S) are consumed by owls at a rate that is proportional relative to the availability of mice. Their populations then thus can be represented by

Table 1.

Proportions of prey types identified in owl pellets collected each year from Santa Barbara Island. (Numbers in parentheses are individuals identified in the pellets.)

	year
prey type	2010	2011	2012	2013
deer mice	88.2% (405)	98% (776)	78.5% (681)	40.7% (474)
Scripps's murrelets	3.3% (15)	0.1% (1)	6% (52)	1.5% (18)
night lizards	6.3% (29)	1.8% (14)	14.1% (122)	57.2% (666)
landbirds	1.1% (5)	0.1% (1)	0.9% (8)	0.4% (5)
all seabirds	4.4% (20)	0.1% (1)	6.5% (56)	1.5% (18)

With this model, we demonstrate how the number of murrelets killed by owls and murrelet population change (N_t+1/N_t) is driven by the strength of climate forcing of the previous year. To do so, we extracted a time series of the past 120 years of annual rainfall from a 4 km gridded surface (PRISM; [55]) and examined how variation in climate-driven resources influenced community dynamics in the model. To assess the accuracy of the model output, we compared the predicted versus observed population dynamics of mice and owls as well as predation on murrelets by calculating correlation coefficients and mean absolute errors (MAE; [56]). See the electronic supplementary material, appendix S2 for further details.

3. Results

Annual rainfall totals and NDVI in March between 2007 and 2013 varied considerably among years and were positively correlated (b = 0.01 ± 0.003, r² = 0.56, p = 0.03). Rainfall amounts ranged from as low as 17.8 cm to over twice that at 48.1 cm. NDVI in March between 2007 and 2012 predicted mouse densities in both monitoring plots one year later (figure 1; b = 12.9 ± 1.5, r² = 0.85 p < 0.001) after controlling for differences between plots (b = 0.27 ± 0.37, p = 0.48). Just 20 individual mice were captured in March 2010, but their density increased substantially by March 2011, peaking in October 2011 at 468 ha⁻¹ before steadily declining to 21% of the peak by March 2012 (figure 2a; mean density of the two plots relative to peak mean density). By March 2013, the mice had declined even further and only two individual mice were captured. The numbers of owl detections on the trail surveys followed a similar pattern and were low in early 2010 at just three observations, and peaked in August 2011 at 32 owls, before a steep decline by January 2013 (figure 2a). Although mouse and owl abundances both peaked relatively close in time, the mouse population declined in 2012 before the owl population did, which coincided with major changes in owl diet.

Figure 1.

Density of mice on Santa Barbara Island in two plots in March 2007–2013 as predicted by NDVI values from the previous March (t − 1). (Online version in colour.)

Figure 2.

(a) Trends in abundance of deer mice and barn owls relative to their respective peaks in density from 2010 to 2013. Lines connect irregularly spaced monthly surveys representing trends relative to maximum peaks observed in mouse density (mean density of two monitoring plots) and owl abundance (maximum number of owl detections on trail transect surveys). (b) Total number of murrelets found killed by owls each year.

A total of 3281 vertebrate prey items were identified to species in the owl pellets. Deer mice dominated the diet of barn owls with 71% of the total prey items (n = 2331). However, there were significant differences in the proportions of different prey types in owl pellets collected each year (table 1; χ²₆ = 1075.5, p < 0.001). When mice were super abundant in 2011, deer mice were the main dietary item, comprising 98% of all prey items. In other years, this percentage varied from a low of 40.7% in 2013 to 78.5% and 88.2% in 2010 and 2012, respectively. Of the seabirds, murrelets were the most common prey (86 of 95 seabird remains in pellets). Murrelets were identified in 3.3% of prey items in 2010, but comprised only 0.1% of owl prey items in 2011. The following year, after the mouse population had crashed, owl diet changed substantially. Murrelets were far more common prey items in 2012, peaking at 6% of prey items before dropping in 2013 to 1.5%. The number of murrelet carcasses found each year followed a similar pattern as the pellets (r = 0.89), with island-wide annual totals of 53, 11, 172 and 80 individuals represented in the prey remains, respectively, over the 4 years (figure 2b).

The results from the piecewise SEM (figure 3; Fisher's C = 2.44, p = 0.296) indicated that the pathway from NDVI to owls was positive (standardized β = 0.33 ± 0.11; p = 0.008), and with greater owl abundance, the log number of murrelets killed also increased (standardized β = 0.50 ± 0.18; p = 0.02). The magnitude of this indirect effect is obtained by multiplying the two coefficients along the path model, 0.165, indicating a moderately strong effect. However, log mouse density was negatively correlated with the log number of murrelets killed (standardized β = −0.45 ± 0.19; p = 0.04). Hence, the ENSO-associated increase in rainfall indirectly increased owl predation of murrelets, but this was also mediated, in large part, by mouse density. Lastly, the bottom-up influence of NDVI on mice was positive but not significant (standardized β = 0.38 ± 0.19; p = 0.08), but there was a strong top-down negative effect of owls on log mouse density (standardized β = −0.67 ± 0.22; p = 0.02).

Figure 3.

All arrows represent hypothesized pathways included in the a priori piecewise structural equation model (SEM). Arrows indicate the direction of the relationship and the numbers shown are the standardized coefficients. Solid arrows indicate significant pathways (p < 0.05) and the dashed arrow indicates the one non-significant pathway (p = 0.08). Values of NDVI in March from the year t − 1, a proxy of mouse abundance, were positively related to the number of owl roosts in mid-winter of year t, which negatively impacted log-transformed mouse density in April and May in year t. Mouse density was negatively related to the log-transformed number of murrelet carcasses collected in year t in each site, which in turn was positively related to the number of owl roosts. (Online version in colour.)

Finally, the mathematical model demonstrated that the climate perturbations led to marked variability in the population trends for all three species over time (electronic supplementary material, appendix S2, figures A2 and A3). Specifically, our model predicts outbreaks in both the mouse and owl populations as well as subsequent sharp increases in the number of murrelets killed by owls after an increase in rainfall (figure 4; electronic supplementary material, appendix S2). The predicted dynamics of the mouse population from the model are similar to those actually observed on the island, particularly since the year 2001 (n = 14 years, r = 0.73 p = 0.003). The model also performs fairly well in predicting owl abundance similar to the observed mean annual abundance as estimated from trail surveys (r = 0.64, p = 0.03; MAE = 5.5; n = 12 years), as well as the number of murrelets killed by owls each year (figure 4; r = 0.72; p = 0.03; MAE = 56; n = 9 years).

Figure 4.

Changes in the number of murrelets killed by owls each year as a result of rainfall from the year t − 1 as simulated by a mathematical model of community dynamics over 120 years (details in electronic supplementary material, appendix S2). Each filled circle represents the predicted number of murrelets killed by owls as calculated from the model given the effects of estimated rainfall on mice and owl abundance. Each open circle represents the observed number of murrelets killed and rainfall the previous year (data from [46,61,77]).

4. Discussion

Seabird survival and breeding parameters in many systems are typically strongly influenced by food availability and ocean climate regimes including ENSO events [57], i.e. bottom-up processes. Here, our results demonstrate the consequences of an El Niño event on breeding seabirds that propagated through the terrestrial island food web where they nest. Specifically, our results suggest that owl predation of murrelets was mediated by prey switching triggered by climate-driven changes in the density of mice and owls (figure 2). Other island systems in this region and elsewhere may be similarly affected. For example, deer mice are already known to increase by over 400% after El Niño years on islands in the Gulf of California [15], and rodent population outbreaks following increased rainfall are also commonly observed elsewhere [14–17]. Barn owls are also distributed widely around the world [29], and are known to disperse and readily establish on oceanic islands [58]. However, other predators may respond similarly to resource pulses. Similar effects of lagged hyperpredation by introduced cats and foxes on native mammals following rainfall from the La Niña phase of ENSO have been observed even in a much more complex terrestrial food web in Australia [59]. These lagged top-down effects of ENSO could therefore be more widespread in diverse systems, particularly on islands in both the Pacific and Atlantic Oceans that receive increased rainfall associated with either phase of ENSO events [6,60,61].

We observed an unprecedented level of predation by owls on murrelets 2 years after the ENSO-associated rainfall pulse. Our high count of 172 murrelets killed during the 2012 breeding season potentially represents approximately 15% of the estimated breeding population on the island (475–650 breeding pairs, D. Whitworth 2017, personal communication). This is unlikely to be an isolated case, as we note that previous owl population peaks on the island often occurred at a approximately 1 year lag from an ENSO event ([27,61]; National Park Service 2014, unpublished data, electronic supplementary material, appendix S2), and the highest numbers of murrelet carcasses have often been found in the years after an ENSO event [46,61]. Moreover, the mathematical model with climate variability also predicts such dynamics (figure 4). Our results therefore demonstrate the importance of apparent competition in communities [62,63], where impacts to alternative prey are accelerated by high predator abundance following the sharp decline of the primary prey species [64]. Here, we observed the mouse population plummet to 21% of the previous peak in just six months while the owl population had scarcely declined until after the 2012 murrelet breeding season. This timing was probably responsible for the increased predation on murrelets seen in 2012, as well as other island species such as the night lizards (table 1). For a long-lived seabird with a low reproductive effort, high mortality events such as this could have lasting consequences on population dynamics [65].

During the 2009–2010 ENSO event, the increased rainfall triggered a cascading bottom-up indirect effect leading to increased numbers of both mice and owls. During the time that mice were increasing to their peak abundance in 2011, owls consumed almost exclusively mice. Very few murrelets were found killed that year, despite the island having one of the highest known densities of barn owls documented in the literature [27]. Barn owls can double-brood and do not defend foraging ranges which means they can rapidly increase in abundance [28]. Small seabirds like murrelets are highly vulnerable to avian predators while on land owing to high wing-loading and reduced manoeuvrability [32], so these interludes of relaxed predation pressure could be very important. In several other systems there is evidence that the temporary extreme abundance of a rodent can substantially benefit the nesting success or survival of birds by distracting predators [66,67]. For example, in the Arctic, during peak lemming years, predators like foxes often largely ignore avian prey (e.g. [68,69]). For some geese, the loss of these buffer years owing to dampening lemming cycles is even suspected to be a contributing factor in declining population trends [70].

These two particular years, 2011 and 2012, had highly contrasting impacts on murrelets. Resolving how this pattern of boom and bust years resulting from ENSO events has important long-term implications for population viability of murrelets and possibly other predator–prey interactions. However, ENSO patterns are currently shifting in ways that may already be changing the impact on murrelets (electronic supplementary material, appendix S2). Much of southern California in particular had been in a severe, record-breaking drought for most of the last decade except for the 2009–2010 ENSO [71,72], which is part of an ongoing recent trend of greater extremes in rainfall and drought cycles in this region owing to climate change [72]. Both the frequency of occurrence and intensity of ENSO events in relation to rainfall patterns are potentially important components describing the overall impact on murrelets, as variance in environmental conditions can have different effects than changes in the mean [4]. Our model, which demonstrates a connection between climate and murrelet population changes (electronic supplementary material, appendix S2, figure A4), also underscores this link.

Finally, the SEM results also suggest that the magnitude of owl predation on murrelets also strongly depended on the density of mice (figures 2 and 3). This type of indirect interaction, where an endangered prey shares a predator with a more common prey, is a dynamic that increasingly threatens many species [73]. To protect threatened species, conservation strategies have largely focused on reducing predator abundance through culling [73]. However, in this situation, there are positive indirect effects of owls on murrelets that must also be considered [45]. Mice themselves have a detrimental effect on murrelet nesting success and consume between 8 and 70% of murrelet eggs each year [32], and high owl abundance also reduces egg predation through fear-mediated changes in mouse foraging [45]. This means that both nest success and survival of adult murrelets are influenced by climate-driven indirect interactions with predators, at times both positively and negatively. Given these complexities, adaptive management strategies that strive to maintain positive indirect effects while minimizing negative direct and indirect effects may be the most effective conservation approach.

Our study demonstrates how the effects of an ENSO-associated rainfall pulse ricochets along a chain of interactions between the marine and terrestrial food webs, leading to enhanced predation of a vulnerable nocturnal seabird on its terrestrial breeding grounds. Although advances have been made in predicting the occurrence and strength of ENSO events, there is also a need for deeper understanding of the connection with global precipitation patterns [74]. Global climate change is expected to result in increased variability and changes to precipitation regimes [72,74]. Most approaches for assessing species vulnerability to climate change have focused on direct impacts via abiotic changes and less so on species interactions [75], even though the latter tends to result in stronger impacts on species than the direct abiotic effects [76]. We suggest that such lagged top-down effects of ENSO-associated rainfall are possibly widespread in systems with similar predator–prey relationships. Our study therefore highlights the urgent need for understanding how species interactions will change with shifting rainfall patterns through the effects of ENSO under global change.

Ethics

The work was conducted under the USGS Bird Banding Laboratory Permit no. 22539 and no. 20431, NPS research permit no. CHIS-2010-SCI-0007, Simon Fraser University Animal Care Protocol no. 993B-10 and a Memorandum of Understanding from CDFW.

Data accessibility

See the electronic supplementary material, appendices S1, S2 and http://researchdata.sfu.ca/islandora/object/islandora:9862 [77].

Authors' contributions

S.K.T. and D.J.G. designed the study. S.K.T. and D.M.M. coordinated collection of field data. S.K.T. conducted field and laboratory research, analysed data and developed the model. T.R.S. did the mark–recapture analysis and wrote the supplemental methods. S.K.T. wrote the rest; other authors provided editorial advice.

Competing interests

We declare we have no competing interests.

Funding

Funding was provided by the Montrose Settlements Trustee Council and the National Fish and Wildlife Foundation. Funding for T.R.S. was provided by the NPS Inventory and Monitoring Program, and the USGS, Fort Collins Science Center.

Disclaimer

Any use of trade, product or firm names is for descriptive purposes only and does not imply endorsement by the US Government.