Beyond climate change attribution in conservation and ecological research


  • Camille Parmesan,

    Corresponding author
    1. Integrative Biology, University of Texas, Austin, Texas, USA
    • Marine Institute, Level 3 Marine Bldg., Plymouth University, Drakes Circus Plymouth, Devon, UK
    Search for more papers by this author
  • Michael T. Burrows,

    1. Scottish Association for Marine Science, Scottish Marine Institute, Oban, Argyll, UK
    Search for more papers by this author
  • Carlos M. Duarte,

    1. Department of Global Change Research, IMEDEA (CSIC-UIB), Instituto Mediterráneo de Estudios Avanzados, Esporles, Spain
    2. The UWA Oceans Institute and School of Plant Biology, University of Western Australia, Crawley, WA, Australia
    3. Faculty of Marine Sciences, King Abdulaziz University, Jeddah, Saudi Arabia
    Search for more papers by this author
  • Elvira S. Poloczanska,

    1. Climate Adaptation Flagship, CSIRO Marine and Atmospheric Research, Ecosciences Precinct, Dutton Park, QLD, Australia
    Search for more papers by this author
  • Anthony J. Richardson,

    1. Climate Adaptation Flagship, CSIRO Marine and Atmospheric Research, Ecosciences Precinct, Dutton Park, QLD, Australia
    2. Centre for Applications in Natural Resource Mathematics (CARM), School of Mathematics and Physics, University of Queensland, St Lucia, QLD, Australia
    Search for more papers by this author
  • David S. Schoeman,

    1. Faculty of Science, Health, Education and Engineering, University of the Sunshine Coast, Maroochydore, QLD, Australia
    2. Department of Zoology, Nelson Mandela Metropolitan University, Port Elizabeth, South Africa
    Search for more papers by this author
  • Michael C. Singer

    1. Integrative Biology, University of Texas, Austin, Texas, USA
    Search for more papers by this author

Correspondence: E-mail:


There is increasing pressure from policymakers for ecologists to generate more detailed ‘attribution’ analyses aimed at quantitatively estimating relative contributions of different driving forces, including anthropogenic climate change (ACC), to observed biological changes. Here, we argue that this approach is not productive for ecological studies. Global meta-analyses of diverse species, regions and ecosystems have already given us ‘very high confidence’ [sensu Intergovernmental Panel on Climate Change (IPCC)] that ACC has impacted wild species in a general sense. Further, for well-studied species or systems, synthesis of experiments and models with long-term observations has given us similarly high confidence that they have been impacted by regional climate change (regardless of its cause). However, the role of greenhouse gases in driving these impacts has not been estimated quantitatively. Should this be an ecological research priority? We argue that development of quantitative ecological models for this purpose faces several impediments, particularly the existence of strong, non-additive interactions among different external factors. However, even with current understanding of impacts of global warming, there are myriad climate change adaptation options already developed in the literature that could be, and in fact are being, implemented now.


Detailed understanding of the mechanisms driving the global climate system has resulted in a set of modelling and analytical approaches that is widely accepted by the climate science community. The result is a series of quantitative assessments of the relative roles of natural and anthropogenic drivers of climate trends (Fig. 1, IPCC 2007a). Climate scientists have successfully provided analyses that yield ‘very high’ confidence [sensu Intergovernmental Panel on Climate Change (IPCC)] in attributing the bulk of the past 50 years’ rise in global mean temperatures to rises in human-caused greenhouse gases (GHG), (Fig. 1, see Box 1 for IPCC definitions of ‘confidence levels’ and ‘attribution’, IPCC 2007a). The policy sector drove that research, both in terms of funding priorities and in terms of IPCC mandates. The result has been an ever-increasing call for worldwide reductions in GHG emissions.

Figure 1.

Estimates (and ranges) of warming effect in 2005, in terms of radiative forcing, stemming from different natural and anthropogenic sources. (Source: Figure 2.20 in: IPCC 2007a)

With consensus that human activities are leading to dangerous interference in Earth's climate (Rockström et al. 2009), there has been growing policy pressure for clear quantification and attribution of the resulting biological impacts. Encouraged by successful attribution of global warming to GHG, policymakers have advocated extension of this approach to generate quantitative attribution of biological events not merely to changing climate (CC), but specifically to anthropogenic climate change (ACC), the component of change that has been driven by increase in GHG. When particular biological systems are dramatically altered, such as when a population goes extinct, policymakers (and the public) often demand a level of ‘attribution’ that clearly delineates the role of ACC from those of other potential drivers. A recent IPCC Guidance document for the Fifth Assessment Report attempted to satisfy this demand by developing a framework for detecting biological impacts and attributing them to anthropogenic GHG forcing (Box 1 and Hegerl et al. 2010).

Biologists have used multiple approaches to link long-term observational changes in wild species or systems to ACC. One approach focuses on documentation of ‘fingerprints’ of ACC in systematic global trends and in specific patterns of observed changes in species' traits that are uniquely predicted as responses to global warming. Such studies have focused on globally consistent, systematic poleward and upward shifts of range boundaries and advancement of spring phenologies (Root et al. 2003; Parmesan & Yohe 2003; Fig. 2 ‘meta-analyses’). Another approach relies on scientific inference drawn from independent, multiple lines of evidence that, taken together, support a coherent and parsimonious conclusion that ACC has driven particular observed biological changes (see discussions under Best Practices and Table 1). A third set of approaches has analysed observed biological changes against climate models to link biological responses to greenhouse gas (GHG) forcing (Root et al. 2005; Rosenzweig et al. 2008).

Table 1. Attribution and lines of evidence. The stronger the evidence within each line, and the greater the number of different lines of evidence that support a climate change interpretation, the higher the confidence in attributing an observed change in a particular population, species or system to observed climate change. Coherence in responses across taxa and geographical regions, via global meta-analyses of observed changes, implicates anthropogenic climate change as a primary causal agent. Effects of anthropogenic climate change on coral reefs and on a butterfly provide examples of this approach to attribution. Full references in Supplemental Material
Lines of evidenceTropical coral reefsEuphydryas editha butterflyReferences
Paleo data: document associations between historical climate change and ecological responsesOver the past 490 My, coral reef die-off coincided with increases in CO2, methane, and/or warm temperatures1NA1Vernon (2008)
Experiments: document a significant role of climate in species' biologyLaboratory experiments show corals bleach under stresses such as warm temperatures, extreme salinities and high rates of sedimentation2Laboratory, greenhouse and field experiments of temperature manipulations show detailed responses to temperature change3

2Lesser (1997), Jones et al. (1998), Glynn & D'Croz (1990), Anthony et al. (2007)

3Weiss et al. (1988), Hellmann (2002)

Long-term observations: significant and consistent associations between a climate variable and a species' responseCoral bleaching events consistently follow warm sea surface temperature events (e.g. El Niño)4>50 years of regular (often yearly) censuses across the species' range document multiple population declines and extinctions following drought and increased variability in precipitation5, early snowmelt (false springs) and unseasonal frost6

4Hoegh-Guldberg (1999)

5Singer (1971), Singer & Ehrlich (1979), Ehrlich et al. (1980), McLaughlin et al. (2002a,b), Singer & Parmesan (2010)

6Thomas et al. (1996)

Fingerprints: responses that uniquely implicate climate change as causal factorFirst observations of mass tropical coral bleaching in 1979, concurrent with accelerating SST warming7Highest levels of population extinctions along the southern range boundary and lowest levels of population extinctions along northern and high-elevation range boundaries, uniquely consistent with regional warming and not with local or regional habitat degradation and destruction, urbanisation, agricultural expansion, or plant invasions.8 There is a significant downward step-jump in both proportion of population extinctions and in snowpack trends at 2400 m in the Sierra Nevada mountain range.9

7Hoegh-Guldberg (1999)

8Parmesan (1996), Parmesan & Yohe (2003), Parmesan (2005)

9Parmesan (1996), Johnson et al. (1999)

Change in climate variable at relevant scale has been linked to GHG forcingOcean warming has been linked to GHG forcing with some GHG projections indicating the Pacific will move towards a more ‘El Niño-like’ state10Warming and lifting snowlines across western North America has been linked to GHG forcing.11

10Hansen et al. (2006), Latif & Keenlyside (2009), Fedorov & Philander (2000)

11Karl et al. (1996), Kapnick & Hall (2012)

Meta-analyses: global coherence of responses across taxa and regions16% of tropical coral reefs lost globally in 1997/98 El Niño event12NA12Hoegh-Guldberg (1999), Wilkinson (2000)
Figure 2.

Applying climate system approaches to ACC attribution of ecological impacts. This cartoon depicts possible approaches to estimating the contribution of ACC to observed biological changes. From Figure 5, Stone et al. 2009 ‘The sequential approach differs from the end-to-end approach in having a discontinuity between the attributed climate change and the observed weather driving the ecological model. The meta-analysis approach takes results from studies of many ecological systems and takes consistency among all of these results as support for the individual results. The meta-analysis is shown here operating on results from sequential analyses but could also operate on results from end-to-end analyses. The synthesis approach compares the pattern of changes in many ecological systems to what would be expected given historical weather, and then brings the result into a sequential approach’.

Lastly, development of reliable process-based models has been possible for a few species for which we have both long observational time series and experimental studies. The expected responses to climate change can then be estimated from these ecological models, and those results compared to either observed climate changes or to General Circulation Model (GCM) outputs that can estimate the component of CC due to GHG. Ecological models are often based on physiological and developmental processes related to temperature and water availability. Such models have successfully documented significant causal relationships between observed long-term changes in a species and regional climatic changes. For example, Battisti et al. (2005) causally related the northward and upward range shifts of the pine processionary moth (Thaumetopoea pityocampa) in France and Italy to warming winters (an ACC signal) that enhanced larval survival and feeding rates. Likewise, Kearney et al. (2010) causally related advancement of spring emergence of adults of the common brown butterfly (Heteronympha merope) to warming temperatures near Melbourne (Australia) during key phases of the life cycle. Chuine et al. (2000) found that models incorporating species-specific dormancy and bud growth temperature cues explained between 57 and 94% of variance in phenology since 1884 for three North American trees.

While hundreds of process-based ecological models have been developed, particularly for agricultural crops and insect pests, they predominantly focus on climate metrics and do not incorporate the range of extrinsic factors, especially other anthropogenic stressors, that interact with climate to affect wild species (see discussions below under Impediments -Interactions and Emerging Insights).

Policymakers drive the priorities and issues addressed in reports by the IPCC; therefore, they are key players in this discussion, together with ecologists and climate scientists. Here, we argue that while all three sectors operate with high levels of internal consistency, transference of priorities and approaches among these three communities is not necessarily straightforward. An example of such transference is the application to ecological data of some of the methodologies used in the climate system and listed in Fig 2, namely sequential analysis, end-to-end analysis and synthesis analysis (Stone et al. 2009). Several of the approaches highlighted by Hegerl et al. (2010) also assume that mechanistic ecological models quantifying relationships between climate and its biological impacts can be developed from experimental results. If this were possible, such ecological models could then be coupled with climate models to derive quantitative estimates of the separate effects of different potential drivers of biological impacts. However, the literature we review below, including results from multiple large meta-analyses, points to a plethora of problems in developing ecological models aimed at clearly separating effects of different drivers, including ACC, on biological systems.

We expand an earlier study (Parmesan et al. 2011) to develop an alternative view as to the most feasible, scientifically sound and, we hope, the most useful avenues for approaching the ‘attribution’ question from an ecological perspective. We do so by examining evidence from studies performed at a range of spatial scales, and at a diversity of taxonomic levels, from individual species to communities and ecosystems. We highlight aspects of biological systems that operate differently from climate systems, specifically aspects arising from strong interaction effects among multiple external drivers and from complex and diverse responses within and among species. We then suggest ways in which biological attribution research can be refined and directed into effective and productive lines of inquiry. In so doing, we recommend a way forward that we consider more feasible and useful than attempting to dissect the roles of ACC in separate, diverse biological events. We argue for a need to move beyond narrowly defined attribution approaches focused on individual responses towards understanding the complex interactions between ecological impacts of climate change and those of other human-caused stresses.

Box 1. Definitions Detection and attribution

There are three concepts that shape the field of research in attribution of impacts to greenhouse gas emissions. Hegerl et al. (2010) encapsulate the current scientific consensus with the following definitions:

Detection of change is defined as the process of demonstrating that climate or a system affected by climate has changed in some defined statistical sense, without providing a reason for that change. An identified change is detected in observations if its likelihood of occurrence by chance due to internal variability alone is determined to be small, for example, < 10%’.

Attribution is defined as the process of evaluating the relative contributions of multiple causal factors to a change or event with an assignment of statistical confidence. The process of attribution requires the detection of a change in the observed variable or closely associated variables’.

External Forcing and Drivers. When external drivers are explicitly included in detection and attribution studies, their influence on an observed change can be estimated…. To the extent that the response to greenhouse gas forcing can be separated from the responses to other external forcing and drivers, the change attributable to greenhouse gas forcing can be assessed and further used to produce probabilistic projections of future change’.

Here, we specify two key types of attribution study relevant to the impacts of global warming on biological systems (see also Parmesan et al. 2011):

Climate Change Attribution (CC attribution) is the process of attributing some significant portion of an observed (detected) biological change to detected trends in climate. Climate here is used broadly to include not only changes in annual means but also, for example, in patterns of climate variability, in strength and timing of extreme events, and in frequency and nature of ocean-atmosphere dynamics such as the El Niño Southern Oscillation. Most notably, this type of attribution study relates biological changes to climatic changes regardless of the cause of the climatic changes.

Anthropogenic Climate Change Attribution (ACC attribution) is an assessment (ideally quantitative) of the specific role of greenhouse gas forcing in driving observed biological changes. Thus, this type of attribution study is focused on teasing apart the degree to which biological changes can be related to the component of climate change that is being driven by increasing levels of greenhouse gases. This has also been referred to as ‘double attribution’ or ‘end-to-end’ attribution (Stone et al. 2009; Hegerl et al. 2010).

Confidence in attribution

IPCC uses a defined set of ‘confidence levels’ to describe the level of uncertainty in a particular conclusion that stems from incomplete or inaccurate data, disagreement among studies, incomplete understanding of underlying processes, or inherent chaotic nature of the process. Confidence levels for particular conclusions develop from the authors’ collective expert judgement as to the state of that knowledge and describes their confidence in the correctness of a particular result or conclusion. We use confidence levels as defined in Box TS1 of IPCC 2007a (Solomon et al. 2007) as follows:

very high confidence: at least 9 of 10 chance of being correct

high confidence: ~ 8 of 10 chance of being correct

medium confidence: ~ 5 of 10 chance of being correct

low confidence: ~ 2 of 10 chance of being correct

very low confidence: less than 1 of 10 chance of being correct


IPCC: Intergovernmental Panel on Climate Change

GCM: General Circulation Models (also known as Global Climate Models)

SST: Sea Surface Temperature

GHG: Greenhouse Gas

CC: Climate Change

ACC: Anthropogenic Climate Change

Impediments to Attribution Studies

Species, communities and ecosystems are dynamic. Therefore, ecological science, by its very nature, must study the causes and consequences of biological changes and relationships between species and their dynamic biotic and abiotic environments. Thus, ‘attribution’, in the broadest sense, has long been a driving motivation in ecological research. It is not surprising, then, that many of the conceptual problems inherent in attempts to attribute ecological events to CC are familiar from prior ecological work and were anticipated long ago (Levin 1992). Here, we outline seven impediments to such attribution studies, some of which apply only to the search for the specific roles of rising GHGs (ACC), while others apply equally to CC and ACC attribution studies.

Issues of mismatch of spatial scales

The GCMs from which climate variables are routinely extracted in ACC attribution studies tend to provide outputs at coarser spatial scales than those used in biological studies. For example, high-resolution analysis of sea surface temperatures (SSTs) shows greater warming in the world's coastal regions than just offshore (Lima & Wethey 2012), a pattern not yet captured by GCMs. Yet, more than half of marine ecological records are intertidal, coastal or estuarine (E.S. Poloczanska, C.J. Brown, W.J. Sydeman, W. Kiessling, D.S. Schoeman, P.J. Moore, K. Brander, J.F. Bruno, L.B. Buckley, M.T. Burrows, C.M. Duarte, B.S. Halpern, J. Holding, C.V. Kappel, M.I. O'Connor, J.M. Pandolfi, C. Parmesan, F. Schwing, S.A. Thompson, A.J. Richardson, unpublished data). Further, species in the upper intertidal may be more strongly influenced by air, rather than water, temperatures (Helmuth et al. 2006). In open oceans, species may respond to surface warming by moving to deeper, cooler water (Perry et al. 2005). Such complexity of pattern and plasticity of response is difficult to accommodate using trends derived from large, homogeneous GCM grid squares.

Issues of mismatch of temporal scales

Global climate is characterised by ocean-atmosphere cycles of varying length. For example, in the North Pacific, the Pacific Decadal Oscillation generates periodicities of SST cycles of both 15–25 years and 50–70 years (Mantua & Hare 2002). Response of a biological system to such a cycle would confound attempts at detecting trends associated with GHG forcing, particularly when time series are short relative to cycles. This principle is demonstrated by Henson et al. (2010), who determined that existing data sets of remotely sensed ocean data are too short to distinguish trends in primary production associated with short-term (interannual to interdecadal) cycles from those associated with longer term (multi-decadal and longer) trends, such as ACC. Even though primary production integrates ecosystem function, thereby dampening ecological noise associated with individual taxa, times series of 20–30 years would be required to detect climate change in low-variability (equatorial) regions, with time series > 40 years more appropriate in most ocean basins. Because ocean-atmospheric cycles also have large continental influences, the same problems are encountered in terrestrial data sets (reviewed by Stenseth et al. 2003).

Relevant spatial and temporal scales may be unknown

Mobile species must respond to the diversity of environments that they experience as they travel. Their responses integrate environmental signals across varying temporal and spatial scales according to the organisms’ mobility and life stage. This complexity renders responses to changing climate hard to study, especially as environments for parts of the life cycle may be unknown to researchers. For example, timing of turtle breeding is predominantly driven by climate conditions on feeding grounds (Chaloupka et al. 2008; Mazaris et al. 2009) that may be many 1000s of km from nesting beaches. Nonetheless, studies of turtle nesting phenology often focus only on conditions at breeding sites (e.g. Weishampel et al. 2004). Sessile species, such as trees, contrast with this pattern, potentially integrating temperatures over long temporal scales of decades or centuries, but at very small spatial scales of metres or less that are not captured in meteorological measures.

Observed trends may obscure the true nature of species' responses

Biological processes are complex and vulnerable to misinterpretation. For example, many plants growing where regional warming has occurred have not advanced flowering times as expected and appear unresponsive to climate change. However, analysis of large data sets revealed that approximately three-quarters of these apparent non-responders were actually responding in the expected direction to spring warming, but that this trend was countered by an opposing response to warmer winter temperatures that delayed spring events (Cook et al. 2012a,b).

In a second example, many long-lived seabirds apportion their relative investment in current vs. future reproduction based on environmental conditions in the current year. When food availability is temperature-related, timing of bird reproduction may vary among years in the opposite direction to that expected from direct response to temperature. For example, a study of nine seabirds in eastern Antarctica found an overall delay in arrival and reproduction, partially related to reduced sea ice extent that has been in turn linked to reductions in krill, a primary food resource for the birds in early spring (Barbraud & Weimerskirch 2006).

Gradual improvements

These first four impediments to attribution – mismatches of spatial and temporal scales, poor understanding of relevant scales, and mismatch between overt (observed) responses and true underlying processes – are becoming less problematic as science progresses. Spatial resolution of climate models is gradually approaching the scales of typical ecological studies. The second impediment, the need for longer time series of observational data sets, is more difficult to correct, but increasing availability of indirect indicators for both climate and biological change is helping to extend existing time series. Examples include analyses of cores of coral reefs (De'ath et al. 2009), tree-rings (Payette et al. 1989), soils (Gavin et al. 2007) and deep-sea sediments (Iglesias-Rodriguez et al. 2008). The third impediment – poor knowledge of relevant scales – will be overcome with the normal progression of science focused on this question. Likewise, the fourth impediment – that effects of warming may exist as expected, but be obscured by separate processes acting in opposition to them – becomes less obstructive as mechanisms of response in particular systems become better understood. However, the next three impediments, listed below, describe issues that will not be simply resolved with more investigation, better technologies, or better cross-discipline collaborations.

Issues of diverse metrics

While the upward trend in mean annual temperature epitomises the concept of climate change, the most biologically relevant climatic features may stem from variability itself (McLaughlin et al. 2002), or include time lags, as well as short-term extreme weather events, false springs and heating-degree days (reviewed by Buckley & Kingsolver 2012). Changing precipitation may be a key driver in terrestrial systems (VanDer Wal et al. 2012), and marine species may respond strongly to pH (Moy et al. 2009) or sea ice extent (Wassmann et al. 2011). Increasingly, climate metrics that are being added to user databases (e.g. CMIP5: are those that are most relevant for wild species. However, there are large differences among species, and even among populations of the same species, in the identities of the most relevant facets of climate. Each species or population may respond in its own way to changes in weather extremes or to complex patterns of seasonal climate that may be difficult to identify, much less model.

Ecological response variables are also diverse (Buckley & Kingsolver 2012): traits linked to climate include body size, hairiness, hair/scale colour, individual growth rate and metabolic rate, population growth rates and gene frequencies (reviewed by Parmesan 2006; Bradshaw & Holzapfel 2006; Buckley & Kingsolver 2012). Each ecological trait has been measured in a diversity of ways. Range shifts have been measured from systematic changes in population abundances at biogeographical boundaries (e.g. Sagarin et al. 1999), from geographical patterns of population extinction and persistence (e.g. Parmesan 1996), from observed new colonisations (Thomas et al. 2001; Johnson et al. 2011) and from subtle shifts in plant demography (Foden et al. 2007). Similarly, phenological data may refer to first events (first flowering, leafing, egg laying or mating call) or mean timing of the event for the entire population (see October 2010 Theme Issue on ‘The Role of Phenology in Ecology and Evolution’ in the journal Philosophical Transactions of the Royal Society B).

Experimental results from climate experiments may not reflect species' responses to climate change in nature

To construct biological models that would be useful for attribution, we would need to understand mechanistic relationships between climate and biological traits. However, recent studies indicate that our presumed understanding may still be too limited for this task. Where we have data to make appropriate comparisons, mechanisms that are important in experimental manipulations often do not appear as primary drivers under field (natural) conditions.

A recent global meta-analysis highlights the pervasiveness of the problem. Wolkovich et al. (2012) analysed phenological responses of plant species from > 50 studies with data on hundreds of species, and compared results from experimental warming studies with those from long-term observational data sets. We would hope for the warming experiments to predict observed changes accurately. In fact, experimental results underpredicted observed advances in timing of flowering by 8.5-fold and in timing of leafing by 4.0-fold, compared with long-term field observations in the same regions. Where both responses from experimental warming and from long-term field observations existed for a single species, results from the two data sets often failed to match in magnitude, and occasionally failed to match even in direction (Wolkovich et al. 2012).

For most, if not all, of the species in the Wolkovich et al. (2012) study, the level of understanding required to explain the discrepancies between observation and experiment does not exist. One factor contributing to the disparities may be that natural systems are subject to complex interactions among multiple drivers, while experimental designs purposefully control against such complexity and rarely test more than two or three potential driving factors in a single experiment (Wernberg et al. 2012). More specifically, natural populations have often been degraded from multiple anthropogenic stresses apart from climate change, whereas experimental populations are stressed in a limited way, under carefully controlled conditions.

Evolution has been shown to be an essential component of some species' responses to CC. For example, the northward expansion of the brown argus butterfly (Aricia agestis) in the United Kingdom was possible only because of an evolutionary shift from one host plant species to another (Thomas et al. 2001). There are a few model systems in which potential evolutionary responses to environmental stress have been investigated experimentally. Gonzalez & Bell (2013) performed an elegant series of experiments on yeast that documented complex genetic responses to abrupt changes in simulated range boundary conditions. However, these studies suffer from the same limitations as purely ecological experiments, in that ‘stress’ in the lab is carefully controlled, and extrapolating evolutionary response from lab trials to field conditions faces additional challenges (Gomulkiewicz & Shaw 2013). Finally, experiments on organisms with longer generation times than yeast may not capture the full range of plastic and genetic responses of the target species to the driver manipulated in the experiment.

Insight from Edith's checkerspot butterfly: A case study in complexity

The butterfly Euphydryas editha has a long history of research coupling experimental manipulations in laboratories, greenhouse and natural populations with long-term field observations and censuses. This diversity of approach has facilitated a rich understanding of the roles of climate variability and extremes in driving natural populations (Table 1). However, this understanding also provides a prime example of how diverse these roles can be, and how focusing on experimental results aimed at the target species risks false optimism about our level of understanding of the role of climate in shaping that species' responses.

Field observations over many decades have repeatedly shown that phenological mismatches between E. editha and annual host plants routinely cause young larvae to starve when hosts senesce before the insects are ready to diapause (Singer 1972; Weiss et al. 1988; Boughton 1999; Singer & McBride 2012). These mismatches existed prior to the onset of current warming and most likely result from an adaptive trade-off between fecundity of the butterflies and offspring mortality (Singer & Parmesan 2010). Whatever the cause of the mismatches, they render the insects vulnerable to climate change, since even a small phenological advance of plant senescence relative to insect development dramatically increases larval mortality. A combination of field and greenhouse experiments has assessed this vulnerability (Weiss et al. 1988; Hellmann 2002) and reinforced the idea that effects of climate on insect/host phenological synchrony should be the key factor in insect population dynamics (Singer 1972). Thus informed, modelling of observed population dynamics dissected causes of the demise of an intensively studied E. editha metapopulation in a protected reserve, attributing this extinction specifically to increasing interannual variability of rainfall (McLaughlin et al. 2002).

Up to this point, we see good agreement among results of experiments, modelling and field observations for E. editha. However, shifts in the relative timing of insect development and host senescence are not the only mechanisms related to climate change that have caused population extinctions of this species. Observed responses of different ecotypes of this butterfly living in different habitat types have been highly diverse (Ehrlich et al. 1980; Thomas et al. 1996) and extinctions have occurred in response to specific conditions and extreme events unlikely to be anticipated in controlled experiments.

In late winter at low elevation (500–1000 m), host seeds normally germinate in response to rainfall, and diapausing larvae ‘germinate’ in response to a combination of winter chilling and rainfall. It turns out that there are levels of drought at which the larvae ‘germinate’, but the plants do not, and that this caused several insect population extinctions in the 1970s (Ehrlich et al. 1980). At higher elevation, both seeds and larvae break dormancy at snowmelt, so milder winters and reduced snowpack have advanced both plant and insect activity to the point where both are vulnerable to ‘false springs’ followed by temporary return of winter. A set of E. editha populations at 2300 m elevation went extinct due to these effects (Thomas et al. 1996). In one year, the plants were unaffected, but the insects eclosed so early that many adults were killed by heavy April snowfall and the population size was reduced by an order of magnitude. In a subsequent year, similarly early eclosion preceded normal flowering of nectar sources and the adult butterflies quickly starved. In a final year, the reverse occurred: a late frost killed plants but not insects, causing starvation of insects in populations using a frost-sensitive host, while leaving nearby populations unharmed on a less sensitive plant. The butterfly populations on the frost-sensitive host had not recovered from the first two false spring events and were rendered extinct by the third extreme climate event (Thomas et al. 1996; Boughton 1999).

In summary, three different mechanisms of climate change-caused population extinction were observed in a single butterfly species: shifts in insect/host phenological synchrony, differing dormancy responses by plants and insects to the same climatic event, and direct mortality of both plants and insects caused by a series of false springs. It is hard to believe that experiments simulating changing climate would have predicted such a diversity of mechanisms in advance. Indeed, we don't know how many more possibilities still exist and will escape our imaginations until we observe them in the field.

Role of experiments in attribution studies

Mis-matches between experimental results and natural changes in real time are disconcerting, regardless of their source. The moral from this complexity is that, while controlled experiments will always be a fundamental part of ecological science, they are best interpreted in the context of complementary approaches that incorporate manipulations of natural systems (including large-scale reciprocal transplants), long-term observations and opportunistic use of ‘natural’ experiments that capture extreme events.

Interaction effects are often large and significant

Many ecological experiments manipulate only one driver, and even when multiple drivers are included, they are often not in fully crossed designs simply because of constraints on research facilities and resources. Designs that are not fully crossed preclude accurate estimation of interaction terms, yet when interaction terms are both significant and large, it is not statistically valid to attempt to estimate main effects.

Darling & Cote (2008) performed a meta-analysis of 112 published experiments from freshwater, marine and terrestrial systems that manipulated two or more factors both in isolation and in combination. They found that in more than three-quarters of the experiments interaction terms were significant and effects of different stressors on animal mortality were non-additive (as opposed to being synergistic or antagonistic). Crain et al. (2008) found similar results in a meta-analysis of 171 studies from marine and coastal systems, with 74% of studies showing significant, non-additive interaction effects among two or more stressors. These large meta-analyses indicate that strong, non-additive interaction effects among global change factors seem to be the norm rather than the exception. Thus, the ultimate impacts of climate change are likely to be highly context-dependent, in particular depending on the presence and strengths of other rapidly changing anthropogenic stressors.

These last two impediments, problems with extrapolating from experimental results to natural systems and the presence of strong interactions among drivers, are the least recognised in the literature, yet pose the most severe difficulties for a quantitative approach to ACC attribution, since they are inherent in the nature of ecological processes and not likely to disappear with improvements in techniques or tools.

Emerging Insights

ACC is driving major system changes and already causing problems for the most sensitive species such as those that are cold-adapted, range-restricted and/or already endangered (Parmesan 2006; IPCC 2007b). Thus, studies of climate change impacts are clearly important to ‘pure’ ecology in terms of understanding drivers of observed changes, to conservation in terms of adaptive planning, and to policy in terms of motivation to reduce GHG emissions.

Multidimensionality of anthropogenic global change

As global GHG emissions have risen over the past 150 years, so too have other forms of human impacts on the Earth System (Steffen et al. 2007; Rockström et al. 2009). Among other human activities that drive ecological and evolutionary change, the usual suspects include the following: (1) habitat degradation and destruction, including urban and agricultural expansion and the footprint of seabed trawling; (2) addition of nitrogen and other nutrients from fossil fuel combustion, fertiliser applications, aquaculture and intensive livestock ranching; (3) invasive species; (4) hydrological changes resulting from building of dams, pumping from rivers and aquifers and diversion of above-ground water; (5) hunting, fishing and harvesting; (6) increased ultraviolet (UV) radiation. These drivers are correlated through time, with each other and with ACC, exhibiting nonlinear increases over the past 150 years (Steffen et al. 2007).

A corollary of the finding that strong interactions are frequent (see above) is that increasing pressures from multiple anthropogenic stressors, including ACC, are likely to also be strongly interacting in their impacts. For example, it is generally agreed that ACC is a major cause of tropical coral reef declines (Table 1), but there is also evidence that increased UV radiation has acted both independently and synergistically along with warming sea surface temperature (SST) to stress and harm corals (Llabrés et al. 2013). Briefly, tropical corals worldwide experienced widespread bleaching and mortality in 1997–98 (Wilkinson 2000), attributable to an extremely strong El Niño that raised SSTs abnormally high (Hoegh-Guldberg 1999). However, record high levels of UV radiation were also recorded in the southern hemisphere in the summer of 1998–99 (McKenzie et al. 2009), and physiological studies have indicated that increased UVB radiation may enhance the negative impacts of warming on corals due to loss of protective pigments that constitute warming-induced bleaching events (Banaszak & Lesser 2009).

In an analogous example from a terrestrial system, we already referred (above and Table 1) to attribution of the extinction of a meta-population of Edith's checkerspot butterfly to an increase in variability of rainfall (McLaughlin et al. 2002). However, even though this meta-population was ‘protected’ on Stanford University's Jasper Ridge Preserve, it suffered anthropogenic habitat degradation. Urban expansion increased nitrogen loading in the soil, that in turn favoured exotic annual grasses in competition with native plants, causing decline of Plantago erecta, the butterfly's principal host plant (Weiss 1999).

It seems clear that these drivers do not act in isolation. However, the policy community has reacted with separate responses to different human drivers. For example, the Kyoto Protocol, responding to threats from ACC, focused largely on greenhouse gas emissions. The Montreal Protocol, responding to impacts of increased UV, addressed ozone-destroying substances. Similarly, the Stockholm Convention addressed persistent organic pollutants, the Convention on Biological Diversity addressed biodiversity loss and the UN Convention to combat Desertification addressed land degradation. The strong role of policy in driving science to address these global challenges results in encouragement of scientists to respond to specific, clearly delineated problems. This approach risks detracting from considering the interactions and connections among multiple stressors. There are hints, such as the recently announced UN Oceans Compact (cf., that the need to consider these multiple stressors in concert is starting to percolate into policy forums.

The role of models in ecological attribution

Theoretically, it should be possible to provide high confidence for ACC attribution in studies of individual species, provided that we have accurate predictive biological models, data gathered at a scale to match GHG-climate attribution studies and sufficient mechanistic (process-based) understanding of how climate variability affects key trait. Such modelling is more tractable for some questions than for others. For example, phenological processes are generally better understood at a mechanistic level than are processes that control species' range boundaries, although the two can be intertwined (Chuine 2010).

Unfortunately, the studies highlighted above call into question our presumed understanding of mechanistic relationships between climate variability and biological traits. Further, even if we did have complete understanding there would be no single common metric, either in terms of climate drivers or in terms of biological response variables, that would allow different studies to be combined across biological systems in a simple fashion into one model analogous to those used to model the climate system itself. Developing hundreds of separate ecological models for individual species is not an efficient approach for drawing general conclusions about impacts of ACC on global biodiversity.

We might hope that using metrics that quantify aggregate properties of whole ecosystems, such as NPP or carbon flux, would smooth out problems that are idiosyncratic across field sites and species. Given suitable data sets (e.g. long time series), these aggregate system properties might be expected to ‘behave’ better in a predictive modelling framework than would traits of individual species. However, recent studies (reviewed by Luo et al. 2011) conclude that the obstacles to attribution inherent in studies of individual species also apply at the ecosystem level to properties such as carbon and nitrogen flux and evapotranspiration. Luo et al. (2011) conclude that:

‘… quantifying these [ecosystem] processes is difficult because of their slowness in change, small signal to noise ratios, limited knowledge of key mechanisms, impediment in identifying generalisable properties, their interactive responses to multiple global change factors and other aspects of complexity (e.g. nonlinearity, thresholds and tipping points)’. (Luo et al. 2011)

In terms of modelling ACC impacts, the finding of strong interactions operating at all levels (from species to ecosystems), both in experiments and in natural systems, means that more and more terms would have to be modelled to approach any level of realism, with many of those extra terms being interactions. This makes it even more difficult to identify, much less quantify, the influence of individual drivers. These underlying, and typically unknown, complexities reduce confidence in estimates of attribution from such modelling exercises.

In summary, there are four key insights to gain from recent ecological studies: (1) The diversity among species in both their key climate drivers and associated responses renders it difficult to develop a single ecological model that would represent many species; (2) It is not possible to predict effects of multiple drivers acting in concert from knowledge of effects of different drivers acting alone; (3) Therefore, results from individual ecological experiments are difficult to simply place into a complex model of multiple drivers, leading to low confidence in the ability of such complex models to project a species' response to future climate change; and (4) For a high proportion of natural systems, non-additive interactions among global change factors are significant and strong, making it statistically illegitimate to estimate main effects, such as the independent influence of GHG-driven climate change.

Moving Forward: The Role of Attribution In Ecology and Conservation

Attribution is the foundation of ecological science

A large part of ecology is, and always has been, about detecting and explaining change, whether change in distribution and abundance of individual species (Andrewartha & Birch 1954), in community composition (Elton 1958), in biodiversity (MacArthur 1972), or in ecosystem dynamics (Valiela 1995). So attribution to either CC or to ACC is a familiar enterprise, a subset of ecology that in this case depends on the availability (at appropriate scales) of climatological data and of information from experimental and/or correlational studies that predicts biological responses to changing climate. Given that ACC attribution is problematic, often to the point of being unprofitable to pursue, we recommend that ecological CC and GHG attribution studies should focus on understanding of underlying ecological process, especially net effects of multiple, interacting forces (see Box 2). The understanding gained can be usefully applied to adaptation strategies without going through the complex process of attempting to quantitatively attribute some precise proportion of ecological change to ACC.

This shift in priorities should stimulate experiments designed to maximise our ability to estimate interaction terms in addition to main effects of single factors. This means conducting experiments with fully crossed designs, with the result that sample sizes required could quickly inflate beyond practicality. One possible solution is to test fewer factors in any given experiment, but to co-ordinate replicated experiments across laboratories/institutions. Better coordination within the ecological community might facilitate conducting experiments simultaneously at multiple locations, so at larger spatial scales than has historically been the norm, and including more factors, fully crossed designs and large sample sizes. Existing efforts to ‘scale-up’ ecological science include the Phenological Gardens in Europe, and the National Ecological Observatory Network in the United States.

Mechanistic understanding at the level of individual field studies using traditional ecological approaches is a vital part of ecology, and its techniques are available to be applied to the approach that we outline above. This should provide deeper understanding of complex biological systems that ultimately will help develop practical approaches to climate change adaptation. The end result of bringing this approach to the ‘attribution’ question should incrementally, and as a by-product (not a goal), eventually shed additional light on the ‘anthropogenic attribution’ question advocated by IPCC.

Best practices for ecological attribution

We argue that multiple approaches that have already been successfully implemented in biological studies have been able to provide high to very high confidence (sensu IPCC) that wild species and sensitive ecosystems have been impacted by recent climate change (Box 1). Further, as the number of observations of CC impacts has increased, our confidence in attribution of biological changes to ACC has also increased.

Inference can provide high confidence

Inferential methods have been the principal basis for generally acccepted conclusions in disciplines where direct observation and experimentation is difficult, or impossible. For example, astronomers infer the presence of invisible celestial objects (e.g. a black hole) based on the behaviour of surrounding observable objects. Biological inference of the impacts of ACC is based on the principle of parsimony (Occam's razor), as ACC allows for a parsimonious explanation of the suites of changes observed. Indeed, most conclusions of ACC impacts on biological systems are based on inference as much as on statistical analyses (Table 1).

Multiple lines of evidence further increase confidence

Combining multiple, independent lines of evidence provides some of the strongest support for inferential conclusions that ACC has affected particular species or ecosystems. Table 1 outlines this approach for two well-studied examples: Edith's checkerspot butterfly and tropical coral reefs. Even at the scale of ecosystem processes, the existence of idiosyncratic responses and strong non-additive interactions among multiple anthropogenic drivers, including ACC, may necessitate similar inferential approaches to attribution. Luo et al. (2011), in describing this ecosystem-level complexity, have also called for synthesis of diverse approaches – experimental, observational and modelling.

Meta-analyses provide higher confidence than individual studies

Ecologists have long used meta-analytic approaches to tackle ‘big picture’ questions. Each meta-analysis combines many studies, which are distributed neither randomly nor uniformly across the globe. Nor do they use the same methods or study similar systems, even when they tackle the same problem. This diversity raises questions, on which ecologists have yet to reach complete agreement, about when it is legitimate to combine such disparate data, and, if so, how to do it. For example, a protracted debate on the use of meta-analyses to elucidate diversity-productivity relationships culminated in a ten-paper forum in the September 2010 issue of the journal Ecology. Similarly, during internal discussions within IPCC, some critics argue that diversity of experimental approaches and of local confounding factors clouds conclusions from existing global meta-analyses. A few critics have even called for clear (statistical) demonstration that effects of confounding factors globally sum to zero, a criterion that current data cannot fulfil.

Conversely, proponents of the meta-analysis approach argue that local confounding factors are idiosyncratic across studies, and essentially become noise at global scale. We find the proponents’ case more compelling – that if a consistent signal emerges from such messy diversity, the meaning of the signal is all the clearer. As LaJeunesse (2010) eloquently writes:

‘An important criterion for synthesis is validation through convergent confirmation of independent research using a diversity of experimental designs and measurements … when heterogeneous results that define multiple operations of the same ecological construct are combined and compared, something essential is learned about this biological effect beyond what each operation captures individually…Pooling research based on a combination of methodologies also insures (sic) that the variance of the ecological process reflects this process and not any one methodological artifact’.

Global provides higher confidence than local

We argue that higher confidence in ACC attribution is obtainable from global scale rather than regional studies. Our first basis for this opinion stems simply from global being the scale at which climate scientists have the highest confidence in attributing warming trends to GHG. An example illustrating the limitation of local studies in this sense is the long-term study of phenology in an Australian butterfly by Kearney et al. (2010). In Fig. 2a, this study is placed in the ‘high’ to ‘very high’ confidence range for attribution to CC, but lower for the ACC (Fig 2b). It is an excellent study with a very good mechanistic model of how temperature drives phenology in this butterfly species, derived from experimental manipulations coupled with a natural correlation between temperature and phenology measured across 64 years. However, because the entire study area is only one small location (Laverton – a suburb of Melbourne, Australia), the authors used a downscaling method to relate local climate change to GHG forcing. In the current state of climate science, there remains lower confidence in attributing the temperature changes in that single location, and hence the phenological changes in the butterfly, to GHG forcing (Hasselmann 2006).

Our second reason for favouring global-level analyses stems from possible synergisms between GHG and other factors. Impacts of GHG-caused ACC on local climate change may be enhanced or mitigated by other drivers operating simultaneously. Therefore, there is a risk that the effects of these non-GHG anthropogenic drivers can either mimic or mask the expected effects of warming climate. This may occur simply by chance, but urban expansion and logging are known to drive local increases in temperature, and so can act synergistically in concert with larger scale warming trends. Conversely, crop irrigation can drive local decreases in temperature, thereby acting in opposition to global warming trends.

Our best hope of escape from the risk that non-GHG anthropogenic effects might obscure ACC impacts comes with increasing the spatial scale of the study. As we scale up from local to regional scales, for example to continental Europe, the eastern USA, or, better still, to a global level, we reduce our risk of finding spurious correlations between effects of GHG-driven climate change and effects of other anthropogenic drivers that are not causally related to ACC. For example, there is no expectation that logging would drive a systematic increase in mean global temperatures over the past 50 years, as would a rise in GHGs, since estimated effects of land-use change over this period are relatively small (Fig. 1).

Combining meta-analysis with global scale provides highest confidence

We have made the case that confidence in attribution is increased both by combining many studies into a single meta-analysis and by scaling up from local to global. The logical extension of this case is that the highest confidence we can attain is from global meta-analyses. Fig. 2 is a cartoon illustrating how confidence in attribution to CC (regardless of the cause, Fig. 2a) and to the anthropogenic component of climate change (Fig. 2b) varies among studies conducted across different spatial and temporal scales. These confidence levels should apply irrespective of whether the conclusion is that ACC is involved or that it is not. To provide an example in which attribution can be made with very high confidence, but not to ACC, we included in the figure the global trend for reductions in body size of fish worldwide. This trend has occurred in parallel in many species and is well understood to have been driven by over-exploitation of fish, particularly of the largest size classes (Pauly et al. 2005). Thus, biologists have very high confidence that globally systematic declines in average body size of fish are NOT due to ACC.

Once CC impacts are detected at the global level, success in further attributing those impacts to ACC is much improved because climate scientists have convincingly attributed climate change at the global scale to GHG. Therefore, if global-scale biological patterns can be attributed to global climate warming, they can also, by inference, be attributed to ACC. This argument is not a statistical one, it is a logical argument stemming from our understanding of anthropogenic influences on ecological processes and the global pattern of human impacts. Nor is this a novel argument. Focusing on poleward/upward range shifts and advancement of spring phenologies, many authors (including those on the IPCC Third and Fourth Assessment Reports), have used some form of this argument to attribute these systematic global biological responses to CC and ACC with high to very high confidence. In developing consensus summary statements, biologists within IPCC have argued for higher confidence in either CC or ACC attribution derived from global syntheses (whether from formal meta-analyses or not) than from either studies of individual species or of many species within one region (e.g. the United Kingdom). Indeed, several authors have referred to this overall, highly consistent pattern of biological responses across the globe as a ‘fingerprint’ of ACC impacts (Table 2; Parmesan & Yohe 2003; Root et al. 2003, 2005; Rosenzweig et al. 2008).

Table 2. Global fingerprints of climate change impacts across wild species. For each data set, a response for an individual species or functional group was classified as (1) no response (no significant change in the measured trait over time), (2) if a significant change was found, the response was classified as either consistent or not consistent with expectations from local or regional temperature trends. For example, consistent responses would be poleward range shifts or advancement of spring events in areas that are warming. Probability (P) of getting the observed ratio of consistent : not consistent responses by chance is from binomial tests against P = 0.5
Studyn: species + functional groupings% changing distribution + phenology% change consistent with climate change P
  1. a

    Exact binomial probability could not be calculated because n was not stated, P taken from publication.

Parmesan & Yohe 2003159859%84%10−13
Root et al. 2003146840%82.3%10−13
Root et al. 200514592%10−13
Rosenzweig et al. 2008not specified (~ 100–200)90%<0.001a
E.S. Poloczanska, C.J. Brown, W.J. Sydeman, W. Kiessling, D.S. Schoeman, P.J. Moore, K. Brander, J.F. Bruno, L.B. Buckley, M.T. Burrows, C.M. Duarte, B.S. Halpern, J. Holding, C.V. Kappel, M.I. O'Connor, J.M. Pandolfi, C. Parmesan, F. Schwing, S.A. Thompson, A.J. Richardson, unpublished data85776%83%10−13

Conservation planning and the role of attribution studies

A key point is that detailed attribution studies that quantify individual roles of multiple driving forces are not necessary for conservation planning. Three recent literature reviews, Mawdsley et al. (2009), Lawler (2009) and Pettorelli (2012), have merged the thoughts and insights of a diverse conservation-science community to present overviews of management and policy strategies designed to help biodiversity adapt to a rapidly shifting climate. However, on the topic of ACC attribution, all three reviews are silent. ACC attribution is simply not mentioned, either in terms of insights gained from previous studies, or in terms of need for further study. This silence sends a powerful message: ACC attribution is not perceived by these authors as essential to conservation planning. This does not mean that ACC is considered unimportant. For many conservation biologists, the main concern is actually the interaction among global change drivers, in which climate change may simply deliver the coup de grace to populations already stressed by other factors (Lawler 2009; Mawdsley et al. 2009).

Box 2 describes a recent case in which a conservation action was quickly implemented in the light of rapid deterioration of a kelp ecosystem due to combined effects of climate change, species' invasions and over-fishing. This action was taken in the absence of ACC attribution, despite the apparent stance of national and international policy makers (who shape IPCC priorities) that quantified attribution studies are essential for policy planning under uncertain future climate change. It is not yet known if this attempt at restoration will be successful. However, given the conclusions drawn here that ACC attribution is inherently difficult in ecology, it will be necessary for the foreseeable future to plan and implement adaptation strategies with the current state of our knowledge.

Box 2. Conservation case study: Kelp and climate change

Here, we present a case study of impacts of climate change on giant kelp Macrocystis pyrifera forests in eastern Tasmania (southeast Australia). This case study of climate change, invasion, interactions and adaptation illustrates many of the recommendations we make here. Recent changes in these kelp forests demonstrate how the complexity of natural systems can limit the validity of simple attribution studies and highlights the need to move beyond attribution.

Historically, there have been dense giant kelp forests in the region. These have supported a rich ecosystem, including abundant predatory rock lobsters Jasus edwardsii that have kept the abundance of sea urchins, key kelp grazers, in check (Johnson et al. 2011). Over the past several decades, two local manifestations of climate change, warming sea surface temperature and a strengthening of the East Australian Current leading to a southwards shift in ocean climatology of around three degrees latitude since 1944 (Hill et al. 2008) and range extension of the long-spined sea urchin Centrostephanus rodgersii from subtropical Australia.  Sea temperatures off Tasmania are now sufficiently warm for regular urchin invasions and for local recruitment of this climate migrant, whose larval development has been experimentally shown to require summer water temperatures > 12 °C (Ling et al. 2009). In yet another synergism between different anthropogenic factors, the local increase in urchin abundance off Tasmania is likely assisted by extensive fishing of rock lobsters, since large lobsters are the sole predators on the adult urchins.

Current abundance of long-spined sea urchins off Tasmania has led to loss of biodiversity. Urchins have over-grazed kelp and generated urchin barrens – denuded former kelp forests that support fewer resident species. This case study is a clear example in which simple attribution of kelp declines to locally warming water temperature would be misleading. Climate change and lobster-fishing have influenced a key ecological interaction – grazing on kelp by the long-spined sea urchin – and this in turn has driven kelp declines. This case study highlights how other human stresses can interact synergistically with climate change. Even more importantly, knowledge of the synergistic interaction between a local stressor (here fishing) and climate change is being used for adaptation management of an invasive species. Trials are being conducted in which large lobsters are being restocked, to assess whether this can halt the decline of giant kelp forests, despite continued warming, and continued urchin migration into the region.


Global meta-analyses are likely to remain the most reliable, scientifically defensible and robust set of approaches both for detection of long-term change in biodiversity and for determining the extent to which those changes can be attributed to anthropogenic climate change. We argue here that existing meta-analyses of global data sets provide strong evidence that major changes in natural systems have been driven by ACC (Table 2). Comparisons among global meta-analyses of biological responses show strong consistency of results across independent data sets spanning terrestrial, freshwater and marine systems and using varying statistical methods (Table 2). These observed systematic trends matched expectations unique to CC impacts – with less than 1 in a billion chance of achieving those same trends if species were changing at random (Table 2).

Individual systems with strong histories of research provide multiple lines of evidence that support conclusions from global meta-analyses and provide a rich understanding of the interactive effects of multiple drivers, including ACC, for those particular species or ecosystems. As interactions among drivers become well understood, this information may, we hope, identify those that can mitigate effects of ACC and take advantage of this information in conservation planning (e.g. Box 2). Qualitative ACC attribution studies, in which a significant impact of ACC is identified, but not necessarily quantified, are sufficient for informing future ecological research and conservation policy. We believe that further attempts at quantified deconstruction of these links are not necessary for the development of robust policies designed to minimise future negative impacts of human-driven climate change. For conservation biology in particular, limited funds will be more fruitfully directed towards implementing conservation actions already identified and advocated by conservation scientists, rather than towards pursuing detailed ACC attribution.


We thank those who responded to our initial opinion piece published in 2011: Myles Allen, Keith Brander, John Bruno, Qin Dahe, Gabi Hegerl, Alistair Hobday, Ove Hoegh-Guldberg, Pauline Midgley, Roger Pielke Jr., David Pierce, Gian-Kasper Plattner, Thomas F. Stocker, Peter Stott, Melinda Tignor, Terry Root and Francis Zwiers. The exchange sparked a lively debate on this topic, in publication and in emails, and motivated the current, more in-depth paper. This study was supported by NSF (EaSM collaborative grant #1049208 to CP). This study was conducted as a part of the Understanding Marine Biological Impacts of Climate Change Working Group supported by the National Centre for Ecological Analysis and Synthesis, a Centre funded by NSF (Grant #EF-0553768), the University of California, Santa Barbara, and the State of California.


CP conceptualised the initial manuscript and wrote the first draft; all authors contributed literature review and contributed substantially to concepts, writing and editing many versions. CP & MCS finalised manuscript; CP, CMD, ESP, MTB, AJR and DSS developed databases and associated statistics; All authors contributed to Table 1; ESP & AJR wrote Box 2; CP developed Table 2; CP & CMD developed Fig. 3.

Figure 3.

Confidence landscape in biological attribution studies. (a) Green lines depict soft boundaries for levels of confidence in attributing a detected biological change to trends in climate (CC). Area below pale green line = range of confidence levels possible for biological studies with < 20 years of continuous data. Area above dark green line = range of confidence levels possible for biological studies with > 50 years of continuous data. (b) Area below blue line represents range of confidence levels possible for studies of ‘double attribution’ (ACC). Green and blue bubbles represent the range of confidence levels for CC and ACC biological studies (respectively). References (superscripts) in Supplemental Material.