Extreme weather events are occurring with increasing severity and frequency in recent decades, a trend linked to ongoing anthropogenic climate change and shifts in atmospheric circulation [1, 2]. While many studies of species and ecosystems have looked at climate impacts by using the average change in historical or projected climatic conditions [3, 4], we know far less about the ecological impacts of extreme weather events on wild organisms [5–7]. In part this is due to the regional geography in which extreme weather events occur, which requires baseline biological data as well as appropriately-scaled climatological data for a particular region prior to an event [8]. Droughts have been occurring more frequently, and across greater spatial scales, yet there are few studies addressing the effects of extreme drought on insects [9]. In general, warmer and drier conditions are known to have both positive and negative effects on insect populations. Positive effects can be associated with more rapid development and less time spent in vulnerable juvenile stages [10], while negative effects include reduced nectar availability and early senescence of host plants [11]. Here we take advantage of decades of data on 163 butterfly species across an elevational gradient in Northern California (Fig. 1a) to investigate the effects of a recent, extreme drought on butterfly populations. The years 2011 to 2015 were characterized by both record-setting high temperatures in California and low levels of precipitation, which combined to produce water deficits that were without precedent in more than 1000 years of reconstructed climatic records [12].

Investigations of butterflies at our focal sites have reported that a majority of populations at the lowest elevations have been in decline since at least the mid-1990s [13], which has been attributed to changes in land use and warming temperatures [14]. Populations at higher elevations, in contrast, have shown relatively less directional change over time, with the exception of a decline in more dispersive, disturbance-associated species that rely on demographic connections with lower-elevation source populations [15]. Previous analyses of abiotic effects have noted responses to weather that were heterogeneous and idiosyncratic among sites and species [16, 17]. For example, associations with climatic variables often differ in sign among congeneric species [18], and even among populations of a single species [17].

The extremity of the 2011–2015 drought provides a unique opportunity to study abiotic effects on butterfly populations that have already been well characterized in terms of climatic relations and long-term demographic trends. Specifically, we asked the following: 1) does an extraordinary, multi-year drought elicit a faunal response that is extreme relative to faunal behavior in previous dry years? 2) Are impacts on the fauna consistent or divergent across elevations? A theoretical expectation is that organisms living in more heterogeneous environments should be more resilient to extremes of temporal variation [19, 20]. We predicted that butterflies at montane sites would be robust relative to populations at lower elevations in landscapes that are both less spatially variable and already impacted by a history of human activity. 3) Finally, we asked if population-level responses to drought are mediated by phenological shifts. Species that are able to begin activity earlier in the spring might reach higher population densities [21, 22], potentially offsetting detrimental drought effects. Another possibility is that ectotherms exposed to longer growing seasons could fall into a developmental trap by which extra generations fail because of insufficient time [23].

A prolonged and consistent shift towards earlier spring flight during the recent drought years can be seen in Fig. 1b. While the shift in phenology is evident across elevations, the dynamics of the flight window diverge later in the season: at higher elevations, the date of last flight shifted to an earlier time during the drought, while at lower elevations the last flight dates from 2011 to 2015 are closer to the long-term average (Fig. 1c; also see Additional file 1: Figure S1 for the same results without sampling filters as described above). In other words, the total flight window expanded at lower elevations, while in the mountains the flight window shifted and compressed towards the start of the season, a change that is reflected in fewer overall flight days at higher sites (Fig. 1d).

Along with the recent reduction in the average number of days that butterflies were observed flying at higher elevations during the drought years, there have been fewer butterfly species observed per year at the same sites (Fig. 2a-e). In some cases, the millennium

drought was associated with a discrete downturn (e.g., Fig. 2b and c), while at other montane sites the recent drought years contributed to ongoing, negative trends (Fig. 2d and e). A downward trend in species richness is less evident at the highest site (CP, Fig. 2a), which previous analyses have found to be receiving immigrants from lower elevations as populations shift upslope in warmer years [15]. In a dramatic reversal of fortunes, the lowest elevation sites during the millennium drought experienced some of their most productive years in nearly two decades, as reflected both in numbers of species (Fig. 2g-j) and numbers of individuals (Fig. 2l-o). Results shown in Fig. 2 are for simple richness (the number of species observed per year). We repeated analyses using alpha diversity Hill numbers that down weight the importance of rare species (Additional file 1: Figures S2, S3), and found similar results for all sites except for GC, where a long-term decline in the number of species becomes evident when rare, or transient species have less influence.

Why did the low elevation sites apparently rebound during the drought years? Using the lowest sites (RC, WS, NS and SM) and a span of years starting just before the millennium drought, we discovered a potential effect of phenological plasticity. Specifically, species whose first flight shifted to an earlier day were the species that became more abundant (r = − 0.50, P < 0.001; Fig. 3). Butterflies at the lowest elevations are almost entirely multivoltine, and an earlier start for those species could have allowed populations more time to build to greater densities which might have in turn translated to overall higher levels of observed abundance and species richness (Fig. 2). However, we note that causality in the other direction is also possible (higher abundance in favorable years leading to earlier detection of first flight dates), and the two possibilities cannot be separated at this time. Nevertheless, the association between change in abundance and change in first flight date (Fig. 3) raises the potential importance of phenological flexibility, which we sought to further understand by modeling the dates of first flight as a function of maximum and minimum temperatures, precipitation, and El Niño (ENSO) conditions. Models explained 60% of the inter-annual variation at low elevations (F _6,138  = 35.17, P < 0.001), and 72% at the higher elevation sites (F _6,138  = 59.68, P < 0.001) (Additional file 1: Table S1). Minimum and maximum temperatures had negative effects on first flight dates (warmer temperatures lead to earlier flights), and the effect of the former was most noticeable at higher elevations (Fig. 4a and b). Precipitation, as reflected by local weather and ENSO conditions, had a delaying effect on phenology (positive β coefficients in Fig. 4c and d), which is expected as wetter conditions are associated with cooler, cloudy days and delayed spring emergence.

Models of species richness revealed even more pronounced variation in weather effects across elevations, including an increased importance of minimum temperatures (Fig. 4e), maximum temperatures (Fig. 4f), and precipitation at higher elevations (Fig. 4g) (see Additional file 1: Table S2 for full details). The highest elevations are most negatively affected by dry years with warmer nighttime temperatures. While daily maximum temperatures have risen everywhere (Fig. 4j) and patterns of precipitation have fluctuated in concert across sites (Fig. 4k), minimum temperatures have risen most steeply at the mountain sites (Fig. 4i). Models of species richness also included phenology (date of first flight) as an explanatory variable, and we found an overall negative association (Fig. 4h), such that earlier emergence is associated with elevated richness (consistent with the effect of phenology on abundance at low elevations; Fig. 3). However, the beneficial effects of earlier emergence at higher elevations might not be as consequential because of a lack of multivoltine species, or they may simply be outweighed by the negative effects of minimum temperatures. Negative effects of minimum temperatures at the higher elevation sites range from 0.48 fewer species seen for every degree Celsius of warming at WA, to 6.46 fewer species seen for every degree at DP (Additional file 1: Table S3). Effects of ENSO on species richness did not differ across elevations (not shown in Fig. 4, see Additional file 1: Table S4), and were generally weak at individual sites (Additional file 1: Table S2).

In contrast to both the previously documented heterogeneity in population response at our focal sites [16] and the observation from other systems that extreme weather elicits species-specific responses [32], the millennium drought in California produced a consistent response across many sites that included advanced dates of first flight with elevation-specific changes in flight windows and species richness. The resilience exhibited by the lowest elevations appears to be associated with phenological flexibility combined with multivoltine life histories and climatic associations that are less detrimental (in the context of current climatic trends) than biotic-abiotic associations at higher elevations. These results are interesting in the light of the finding that a major drought in 1995 in Great Britain shifted species assemblages towards more wide ranging, multivoltine butterfly species with generalist host associations [33]. The drought response in the British butterflies also included an increase in abundance for a subset of sites, although (in contrast to our findings), the populations that increased during the drought tended to be associated with wetter sites, while we saw positive responses at the hotter and drier low elevations.

Many researchers have hypothesized an impending mismatch between trophic levels as a result of climate change [34]. The results from low elevation butterflies in California perhaps challenge that hypothesis, or at least suggest that a shift in phenology at the consumer trophic level need not always have negative consequences. In addition to having multiple generations each year, populations at the lowest elevations have access to agricultural lands. Although association with irrigation does not appear to predict population dynamics during the drought (Fig. 3), we cannot rule out the possibility that low elevation populations were buffered during the drought by irrigated crops or agricultural margins. If true, a reliance on agriculture would be interesting in the light of a recent hypothesis that long-term declines in low elevations butterfly populations are associated with intensification of pesticide applications [35]. It is possible that the rebound of the drought years could be followed by a more severe decline following concentrated agriculture dependency and toxin exposure.

It has been known for some time that high latitude environments are warming faster than the rest of the planet, with negative consequences for many high latitude species [36] but positive or neutral effects for others [37]. It is only recently that climatologists have become aware that higher elevations may also be experiencing a disproportionate share of warming [38], which raises the issue of how cold-adapted, montane ecosystems will respond. Contrary to the expectation that mountains offer microclimatic refugia and preadapt species for climatic variation [39], we found high elevation butterfly communities to be declining and especially sensitive to dry years with warmer minimum temperatures. Warmer and drier years are associated with lower productivity of mesic-adapted plant communities [40], and shorter windows during which montane plants are optimal for nectar and herbivory [41]. We did not model snowfall because it is highly correlated with annual precipitation at our sites (see Methods), but reduced snowpack in dry, warm years would have additional negative effects including higher overwinter mortality for life history stages that typically spend the winter under a blanket of snow [42]. The daily temperature range (the difference between maximum and minimum temperatures) has been shrinking around the globe [43], but the ecological consequences of this thermal homogenization are poorly understood and not yet incorporated into theoretical expectations of global change biology [44]. The results reported here suggest that we have much yet to learn about organismal responses to extreme weather, and the extent to which different habitat types might or might not buffer populations against climate change [45, 46]. However, there is hope for progress because powerful and simple statistical models predicting faunal dynamics with annual climatic values are possible for ectotherms even in the face of unprecedented climatic variation.