Many studies attempt to predict species’ responses to climate change [1] and most focus on changes in geographic distributions [1–3] and potential in situ evolutionary adaptation [4–5]. However, many animal species will need to make ecological adjustments within geographic range interiors, such as altering habitat selection, to deal with increased environmental stochasticity. These responses should have consequences for species persistence and may be affected by human alterations of the landscape. To understand how populations will respond to predicted climate-change scenarios, a necessary step is to investigate temporal variation in species occurrences relative to a range of weather conditions.

Climate change is increasing the variability of precipitation and frequency of extreme flooding and drought events [6–7]. Species obligately associated with wetland and stream habitats are particularly at risk due to extreme fluctuations in water levels. As these climate-sensitive habitats become less stable, species dispersal [8], recruitment [9], and survival [9–12] could be compromised. Semiaquatic species might need to move to other suitable habitat patches to persist during times of environmental stress, but moving across terrestrial areas can be costly [13–14], especially in regions where agriculture and urbanization have destroyed linkages and reduced connectivity.

American mink (hereafter mink; Neovison vison) and muskrats (Ondatra zibethicus) are semiaquatic mammals that may be sensitive to increased variation in precipitation events. Both species are widely distributed throughout North America and are obligately associated with aquatic habitats, although the degree of this association differs between species. Muskrats are chiefly herbivores and most of their diet consists of wetland vegetation. Space use by muskrats is mostly restricted to the stream edge and movements >3 m away from water are rare [15]. Mortality from predation is high during drought conditions and likely due to the limited mobility of muskrats away from water and reluctance to leave established home ranges [16–17]. Additionally, increased flooding can reduce survival of young [18]. Mink are generalist predators that forage in aquatic and terrestrial habitats [19]. When aquatic prey (e.g., fish, amphibians, and crayfish) are unavailable, mink will forage more frequently in terrestrial areas [20], which could expose them to elevated risks. However, mortality risk for mink in terrestrial versus stream habitats is unknown.

We used 6 years of presence-absence data spanning years of record-breaking floods and drought to assess how mink and muskrats respond to conditions predicted to increase under climate-change scenarios. Specifically, we tracked annual changes in site occupancy for mink and muskrats in response to variable summer precipitation. We also radiomarked individuals to assess mortality risk for both species in terrestrial habitats as activity in these areas may become more common with increasing environmental variability. In our study system, > 90% of wetlands have been drained to accommodate agricultural production [21], thus limiting both species’ distributions primarily to flashy streams and rivers. Species occurring in these human-dominated landscapes may be at an increased risk owing to the synergistic effects of habitat loss and climate change [22–23].

We hypothesized that mink and muskrat populations in our region would be sensitive to summer precipitation because droughts reduce habitat quality for semiaquatic mammals in streams. Droughts lower water levels and persistence of flow, thus reducing the protection from predation, and availability of aquatic prey, afforded by deeper water. Thus, we predicted habitat occupancy for both species would be correlated positively with summer precipitation across years. We assumed differences in annual occupancy rates for species reflected underlying patterns of abundance [24–25]. Because muskrats are more tightly associated with streams than are mink, we predicted negative effects of droughts would depress muskrat abundance more than mink abundance. We also assessed the extent of terrestrial habitat use by mink and muskrats and predicted mortality risk would be greater in terrestrial habitat than in stream habitat.

Based on a 123-year record of precipitation, the average summer precipitation for our region was 32.3 cm (SD = 10.2; Fig 2a). During our study, summer precipitation (cm) was extremely variable among years (Fig 2a): 2007 = 27.3; 2008 = 51.7; 2009 = 41.7; 2010 = 39.3; 2011 = 27.2; 2012 = 15.2. Summer precipitation was high during a year of record-breaking floods (2008 = 51.8 cm) and low during a year with widespread drought (2012 = 15.2 cm).

Our ability to detect muskrat sign was negatively affected by the amount of rain 7 days prior to surveys (β = -0.0897, SE = 0.05; Table 1). Models including ‘Rain’ had the most support among all models of muskrat detectability (Σω = 0.67). Although four other models were competitive, none had a substantially better model fit than the top-ranked model (Table 1). Thus, we considered the variables ‘Sandbar’, ‘Date’, and ‘Debris’ non-informative [44–45]. The intercept-only model also was among the top-ranked models but it had a reduced model fit compared to our best model.

Four models of mink detectability were competitive (ΔAIC_c ≤ 2; Table 1). Each model contained the variables ‘Observer’ and ‘Sandbar’. Our second- and third-ranked model also included the effects of ‘Date’ and ‘Rain’, respectively, and the fourth-ranked model included the additive effects of ‘Observer’, ‘Sandbar’, ‘Date’, and ‘Rain’ (Table 1). In concordance with past research [36], our ability to detect mink sign was affected by observer variability (range of βs = -1.8288–0.8105), positively related to amount of trackable surface (β = 0.0031, SE = 0.0014) and survey date (β = 0.0036, SE = 0.0006), and negatively related to amount of rain 7 days prior to surveys (β = -0.0417, SE = 0.0466).

Model-averaged estimates of site occupancy by muskrats varied substantially among years (Fig 2b): 2007 = 0.58 (SE = 0.05), 2008 = 0.69 (SE = 0.05), 2009 = 0.88 (SE = 0.04), 2010 = 0.91 (SE = 0.04), 2011 = 0.36 (SE = 0.07), 2012 = 0.09 (SE = 0.04). Occupancy rates were higher in years when summer precipitation was above the 123-year mean (2008, 2009 and 2010), and occupancy rates were lower when summer precipitation was below the 123-year mean (2007, 2011 and 2012). The estimated proportion of sites occupied by muskrats each year was positively associated with summer precipitation (β = 0.1045; P = 0.0016; pseudo R² = 0.62; Fig 3a).

Our model-averaged estimates of annual site occupancy by mink also varied among years (Fig 2b): 2007 = 0.44 (SE = 0.06), 2008 = 0.90 (SE = 0.03), 2009 = 0.89 (SE = 0.04), 2010 = 0.82 (SE = 0.05), 2011 = 0.67 (SE = 0.06), 2012 = 0.56 (SE = 0.06). Mink were widely distributed during years with above-average precipitation (2008, 2009 and 2010). Estimated occupancy rates were lower during years with below-average precipitation (2007, 2011 and 2012), but remained moderately high in 2011 following three relatively wet years and did not decline to the extent observed for muskrats during the extreme drought of 2012 (Fig 2). As predicted, the proportion of sites occupied by mink each year also was positively related to summer precipitation (β = 0.0681; P < 0.0001; pseudo R² = 0.77; Fig 3b).

For the 20 mink for which we had sufficient movement data to assess extent of habitat use in terrestrial areas, each individual was relocated an average of 102 times (SE = 10.27; range = 25–192) for a total of 2035 locations (Movebank.org Data Repository, dx.doi: 10.5441/001/1.gd686078). The distribution of mink mortalities differed from the distribution of telemetry locations in relation to distance from the stream edge (D = 0.62; P < 0.0002). On average, mink were relocated >100 m from the stream edge only 14% of the time (Fig 4b; S2 Table). In contrast, of 17 known-fate mortalities (seven road kill, six predation, three poisoning and one disease), 76% (n = 13) occurred when mink were >100 m from the stream edge (Fig 4b; S2 and S3 Tables). Our previous studies of muskrat space use and survival found that muskrats rarely used upland habitat and were never relocated > 3 m from the stream edge (Fig 4a; 15). Of 15 known-fate mortalities, 80% (n = 12) occurred along the stream edge and were attributed to mink predation (Fig 4a; 28). We recovered the other 3 muskrat carcasses in or around coyote (Canis latrans) and red fox (Vulpes vulpes) burrows > 50 m from the stream edge [28]. One canid-related mortality occurred while the muskrat was displaced into a corn field during a flooding event. Because we did not detect muskrat movements > 3 m away from the water’s edge, all mortalities attributed to canid mortality likely occurred along the stream edge and carcasses were transported back to active canid burrows [28].

Annual occupancy of stream segments by mink and muskrats was strongly related to summer precipitation. Occupancy rates for both species were higher during years with above-average precipitation than years with below-average precipitation (Fig 2). This contrast was especially clear for muskrats; estimated annual occupancy rates were > 10 times lower (from 0.91 to 0.09) during the severe drought of 2012. Increased frequency of summer droughts is predicted by climate models for the Midwestern USA [29], and the patterns observed in 2012 may therefore be a harbinger for semiaquatic mammals.

Mink and muskrats were widely distributed during years with above-average precipitation. Higher water levels due to increased precipitation likely provided more suitable habitat and increased connectivity between areas of high-quality habitat for both species. Higher water levels also may have provided escape routes from terrestrial predators and lowered predation risk for both species. Previous research demonstrated that site occupancy for mink and muskrats is correlated positively with water depth [35–36, 46]. Furthermore, the probability of vacant sites being recolonized by both species is positively related to water depth [35–36]. Conversely, mink and muskrat occupancy rates were lower during years of below-average precipitation. Low water levels can limit available resources and reduce overall body condition of muskrats resulting in increased mortality [47]. Muskrats also are susceptible to increased predation risk during drought because their locomotion is more limited on land than in water, and the openings of their burrow dens may be exposed as streams dry [17]. Despite this deterioration in habitat quality during drought, muskrats are typically reluctant to leave their home ranges to find other suitable habitat [16], and this effect may be exacerbated in areas where habitat loss has reduced spatial connectivity.

Investigations of how mink respond behaviorally to drought are lacking. However, reduced occupancy rates during years of below-average precipitation suggest mink are foraging in terrestrial habitat away from the stream edge. We think the decline in occupancy for mink partly represents increased use of alternate habitats rather than just mortality because mink are not as constrained to aquatic habitats in our region as are muskrats. All telemetry locations of muskrats occurred within 3 m of stream banks [15]. In contrast, 14% of telemetry locations of mink occurred >100 m from stream banks (Fig 4b), revealing more flexibility by mink in habitat use. The switch to terrestrial habitats may come with increased mortality costs, however, as mortality risk was disproportionately greater for mink when moving through terrestrial areas (Fig 4b). Thus, the greater mobility of mink may allow them to exploit secondary habitats during droughts, reducing their susceptibility to degradation of stream habitats relative to muskrats in the short term. However, if climate change increases the frequency of droughts [7], increased use of more risky habitats by mink should eventually reduce survival rates and affect population dynamics. Unfortunately, we have insufficient data on mortality by muskrats or mink during the drought year to evaluate this hypothesis directly.

Although the patterns are clear, we acknowledge a caveat associated with our interpretations. We cannot directly link discrete flooding and drought events during our study to changing climate. Nevertheless, these extreme events will be more common in the future [6–7, 29–30]. Contemporary climate models suggest severe and widespread drought this century [7]. Species obligately associated with drought-sensitive habitats will be most at risk. Thus, population patterns associated with observed climate-driven events should mimic those during future climate-change conditions.

In Canada, mink and muskrats represent a classic predator-prey system in which mink populations exhibit a lagged numerical response to changes in muskrat abundances [48–50]. However, there is significant geographic variation in the strength of this predator-prey relationship [48, 50]. Mink and muskrat population dynamics in Canada may be partially affected by the spatial variability in winter precipitation [51]. Additionally, differences in predator-prey interaction strength may be partially attributed to spatial variability in prey richness across Canada [52]. In our region, habitat occupancy for mink and muskrats was strongly related to summer precipitation (Fig 3). This correlation suggests environmental variability affects populations of both species similarly and possibly decouples any classic predator-prey relationship. Mink diet in this region is diverse and seasonal occurrence of mammals in mink scats were always <50% of the percentage volume of sampled scats [53]. Increased diversity of mink prey in this region may release muskrats from the specialized predation pressure necessary for cyclic population dynamics [54]. In our study, synchrony of mink and muskrat populations (without a time lag) and their sensitivity to summer precipitation suggest these populations are largely limited by external forcing.

Many semiaquatic species may be negatively affected by the synergistic effects of habitat loss and climate change. Interactions between these stressors can depress population densities and reduce species diversity [23]. Moreover, increases in the frequency and intensity of regional flooding and drought can potentially synchronize population dynamics of species at large spatial scales—especially habitat specialists occurring in homogenous agricultural landscapes [22, 55]. For instance, intensively farmed landscapes can function as habitat sinks for common frog (Rana temporaria) populations during extreme drought compared to landscapes retaining some heterogeneity [55]. Additionally, populations of platypus (Ornithorhynchus anatinus), another obligate wetland species, are threatened by both increasing thermal stress and habitat loss due to climate change and increased irrigation demands for agriculture [56].

In many ecosystems worldwide in which most wetland habitat has been converted to agriculture, the primary remaining habitats for semiaquatic species are small, flashy streams. Because habitat suitability for these species is generally linked with water availability, increased variability in precipitation should drive spatial and temporal variation in habitat quality. A more mechanistic understanding is urgently needed of how extreme weather events, like those observed in our study and predicted under climate-change models, will affect populations of semiaquatic species in human-modified environments.