Formal attribution, as defined by the IPCC, requires three steps to be fulfilled: extreme event attribution, impact detection, and impact attribution (Fig. 1). At each step, three essential components are required22. First, the relationship between cause and effect must be demonstrated. Then the detected change must be shown to be inconsistent with changes due to alternative possible drivers. And finally, a quantification of the strength of the attribution statement is required to acknowledge the uncertainty and limitations of the available data and analyses. The challenges associated with these steps are clearly illustrated in the floodplain grassland community of the River Elbe.
Recent attribution studies have evaluated the extent to which human-induced climate change has affected extreme heatwaves, drought and floods in the Elbe River region. Stott et al.38 estimated the likelihood of a heatwave of the magnitude of the 2003 European one was at least doubled under human induced climate change (confidence level > 90%). Similarly, anthropogenic forcing was found to have played a substantial role in the hot, dry summer of 2013, both in terms of the high temperatures observed and the northward shift of the North Atlantic summer storm track which led to reduced rainfall over western Europe39. In contrast, a large simulation ensemble and observation-based analysis concluded that climate change had not made the extreme rainfall of 2013 in the upper Danube and Elbe basins more likely40. The attribution of rainfall events is substantially more difficult than temperature events because event attribution relies on the model’s ability to simulate the climate conditions generating the weather event. This remains challenging for rainfall, which is naturally highly variable, and generated by processes that are not captured well at the scale of current-generation climate models41,42. Flood time series are similarly highly variable in response to natural variability and factors such as urbanization, deforestation and dike construction. These factors occur simultaneously across a catchment and often interact at multiple temporal and spatial scales, limiting attribution of extreme floods to climate change42,43.
Understanding natural modes of variability plays a crucial role in attribution studies, particularly for rainfall and associated flooding events. For example, the apparent increase in intense rainfall that we identified in the current period, particularly in summer, could have been caused, in part or fully, by natural variability. The 20-year time periods used here are sufficient to capture interannual drivers of variability such as the El Nino–Southern Oscillation and decadal modes such as the North Atlantic Oscillation and the Pacific Decadal Oscillation. However, a significantly longer time period would be needed to capture multidecadal patterns such as the AMO, which was predominantly negative during the historical period and positive during the recent period44. The lack of long-term observations means there are few studies of the influence of the AMO on the study region, although increased summer rainfall has been associated with a positive AMO in modelling over North Western Europe45.
It was possible to detect a community response to the extreme events. The composition and abundance of all taxa displayed considerable inter-annual variability, but changes in community turnover, species abundance and dominance were detected in years following the extreme events as previously shown by31. Decreased carabid species richness was found in years following drought and heatwave events, while plant species richness increased or remained stable in years after heatwaves and/or drought. Extreme floods reduced the species richness of plants and beetles in the short term. In contrast, and consistent with other studies, molluscs showed higher species richness following flooding and lower richness after heatwave and drought years46,47,48. Many of the changes detected were consistent with expectations based on understanding of life cycle biology, such as timing of reproduction and traits enabling inundation and drought tolerance (e.g. molluscs with diaphragms or lids that close shells), although not all changes had simple explanations (described in more detail in Supplementary Information).
While “detection” does not depend on an explanation of the causes of the observed change20, it does require a demonstration that the likelihood of the response is significantly different from that due to natural variability. This is challenging with biological data, for several reasons. First, extreme events occur infrequently and are difficult to predict, so it is rare to have baseline data to characterise the community prior to an event. Second, biological data are generally highly variable in space and time. As illustrated in the floodplain community, species abundance, composition and dominance commonly vary over time, with responses differing both within and between sites. In some cases, the same species responded differently under different conditions.
Further, a community response to an extreme event may be sudden or gradual, periodic or episodic, and the effects may be short-term or permanent. Without continuous observations, it is impossible to determine whether changes in community indices occurred gradually over the years from 1999 to 2003, or suddenly in response to the 2002 extreme flooding event. It is difficult to quantify the extent to which the community is altered permanently by one extreme event. In the dynamic floodplain system there may be some capacity to return to a previous state, since many organisms are adapted to cope with regular flooding events. However, species interactions and feedbacks could lead to lagged responses due to changes in resource availability or competitive interactions49.
Additionally, ecosystem responses may not always occur immediately after a single weather event, but in response to the long-term stress of the changing climate, in combination with extreme weather events9 (the ‘press and pulse’ framework17). Mean temperature has increased in the River Elbe region over recent decades, so any community response may be influenced by this change in the background ‘press’, as well as the magnitude, duration, frequency and timing of the extreme ‘pulse’ events.
The severity of flood impact will also be affected by changes in timing in relation to the traits and phenological stages of species within the community50. Annual spring floods resulting from snowmelt represent natural variability to which the community is adapted. This is supported by the fact that all taxa declined in species richness following the year in which the spring flood did not occur. However, the summer floods of 2002 and 2013 were extreme in terms of severity, duration, and timing. Floods occurring in summer were associated with reduced species richness in carabids and plants. Carabids exhibit adaptations to flood such as autumn emigration, hibernation as adults or physiological adaptations such as low physiological activity or higher submergence resistance in low temperatures. These adaptations enable survival through the usual winter and spring floods, but do not confer resilience to summer floods, which occur when many species are in sensitive larval or pupal phases31.
Differences in traits across taxa mean that different responses will be shown by different taxonomic groups51. The ability to detect a response will therefore depend on what “community” is of interest. Here we found, for example, the pattern of species re-ordering over time was in the opposite direction in the carabids and molluscs after the 2002 flood and 2003 heatwave. Both aquatic and terrestrial molluscs are well adapted to floods, as even land snails can survive in water provided the water is oxygenated and not too warm. In contrast, carabid beetles range in their ability to survive inundation and dispersal ability is important for recolonising after floods52.
Multiple events also complicate detection53. Here, for example, the low mean monthly precipitation recorded in 2003 fell within the lower 25th percentile, so does not represent a climatological extreme in isolation. However, at the same time, extreme maximum temperatures were recorded for extended periods (Table 1) and water levels were significantly lower than the long-term mean (Supplementary Figure S1), with the maximum water level below the 5th percentile of the historical period (Fig. 3d). A strong biological response in all taxonomic groups was associated with the year 2003, in which extended heatwaves coincided with low water levels. However, plant data from 2010 suggest that a similar response as that found in 2003 (increased species richness and decreased turnover) is also associated with recurrent flooding in combination with heatwaves. The mechanisms driving the response are obviously quite different, associated with the added nutrients provided by the fine sediments carried by floods54.
In the current case, not only were the impacts of droughts and heatwaves superimposed on the impacts of floods (natural and extreme), but these events are likely to act in opposing directions. In the short term, floods act to homogenise the habitat and provide nutrients, while drought and heatwaves are more likely to increase heterogeneity across microhabitats with differing elevations and exposure to water50. Over longer timescales, however, increased homogeneity could be expected as the habitat dries out in the absence of regular flooding events. The impact of drought and heatwaves on floodplain communities is likely to be greater than that of floods, given the high proportion of aquatic and inundation-dependent and tolerant species.
Attribution in the climate system relies on the ability to quantitatively model the system55. The mismatch in temporal and spatial resolution between available biological data and climate observations and models42 limits the ability to apply statistical analyses and develop models, and reduces the chances that responses at fine spatial resolution will be successfully attributed to climate change. This is compounded by the high natural variability in the climate variables of interest, in addition to the variability in biological communities, as discussed above. Each extreme event is essentially not replicable16, and even where multi-event responses are available, each event has specific characteristics. Attempts to link biological responses to climate change are therefore likely only to be possible at continental to global scales, or over the timescale of decades56.
Interactions and feedbacks are important structuring factors in natural systems. Extreme events can alter species interactions by reducing populations of common species, allowing another species to increase in population sufficiently to prevent the dominant species re-establishing. In many cases, quantification of such interactions remains problematic, and cause and effect cannot be inferred from correlations between observations and events.
Separation of drivers is a key element of formal detection and attribution analysis21. Biodiversity responses, however, are likely to be driven by multiple factors, acting on a range of timescales. Hydrologic conditions, land use and management, for example, are important drivers of vegetation and invertebrate floodplain communities (e.g.57,58,59,60). In some cases, such as the impact of mowing, riverbed erosion or water extraction, the non-climate driver is easily identified. However, many land use changes develop over decades to centuries, and so would more likely have a long-term effect on biodiversity. Multiple-driver attribution would require the role of such non-climatic drivers to be accounted for and shown to be inconsistent with the observed community response.
The case of the floodplain community suggests that the formal joint attribution of community responses to extreme events caused by climate change will rarely be possible. The detection of responses to extreme events, however, is feasible, and important to improve understanding of the connections between climate, extreme weather events and biodiversity. Such knowledge is essential to inform conservation management attempts to mitigate the impacts of extreme events or ongoing climate change.
Monitoring is essential, ideally before and particularly after an extreme event, to improve our knowledge of the connections among climate, weather and biodiversity. Baseline data needs to be long-term and spatially extensive, due to the highly variable nature of biological data61. More intensive temporal monitoring is essential to better understand patterns and drivers of natural variability so that extreme responses may be identified. Improved spatial replication would increase the likelihood of having biological data from areas that did not experience the extreme event and could also enable other important, non-climatic, drivers to be identified16. Better spatial monitoring is also necessary to predict a response to extreme events in ecosystems other than those with long-term observations.
Long-term monitoring should be designed within a sound hypothesis testing framework16 to encourage a more thorough consideration of the important (and possibly interacting) drivers and their potential effects on biological communities and the mechanisms driving change. Although some taxa are more difficult and expensive to sample, it is important that the best taxon to identify a community response is monitored. While plants are the cheapest and easiest taxon to monitor, they might not be the most appropriate group, as differences in biology across taxonomic groups are likely to lead to a range of responses56.
To strengthen our understanding of impacts and responses, biological monitoring should be complemented with evidence from observations, remote sensing, experimental data, models and ecological theory. Experiments to identify mechanisms driving a response can support observational studies and establish causal relationships11,16,62. For example, Rothenbucher and Schaefer63 used exclosure plots on the Lower Oder floodplain after the 2002 flood to identify species responses in relation to inundation tolerance and immigration. Such experimental information, combined with observations over time, could contribute to greater understanding of how extreme events affect the distribution of species and the structure of communities.
Mesocosm experiments are particularly appropriate for testing the impacts of extreme events on aquatic and riparian invertebrate communities. Experimental manipulations of temperature and moisture can be used to test hypotheses generated by observations of community responses and determine cause and effect64. An additional advantage of mesocosm experiments is the ability to incorporate the effect of carbon dioxide on species responses, an important component that is frequently ignored in observational studies of global change. While short-term experiments can provide important biological knowledge, longer-term mesocosm experiments are essential to identify ongoing impacts of extreme events on the structure and function of communities, including potential lag effects, feedbacks and interactions65.
Meta-analyses to combine results from studies of different single extreme events are needed to consolidate observations and identify trends, similarities, divergences and exceptions. Such analyses will be more informative if studies report similar aspects in a comparable way. For instance, the magnitude of the extreme event should be defined and robust estimates of the magnitude of ecological responses and other drivers reported1,16. Global and regional trends are more likely to be identified through such syntheses.
Reanalysis products based on climate observations could help link weather patterns associated with an extreme event and an observed biological response. Such products are now available at resolutions fine enough to capture processes at biologically relevant scales. For example, the NWP model COSMO regional reanalysis data sets66,67 provide hourly atmospheric data for Europe with a resolution of 6 km for the years 1995–present. The application of reanalysis products, seasonal forecasts and high-resolution projections will strengthen the link between biological and climate knowledge.
Central Germany, in common with many regions of the world, has experienced several extreme weather events over recent decades, in addition to gradual background warming. There is increasing interest in attributing biological responses to extreme events and climate change, but there are many challenges that limit our ability to achieve formal, quantified joint attribution. Nevertheless, it is important that we detect responses and improve our understanding of the mechanisms behind change to inform conservation management and restoration. This is particularly important as the incidence of extreme events is projected to increase in the future.