Multi-hazard assessment in Europe under climate change
- 8.7k Downloads
While reported losses of climate-related hazards are at historically high levels, climate change is likely to enhance the risk posed by extreme weather events. Several regions are likely to be exposed to multiple climate hazards, yet their modeling in a joint scheme is still at the early stages. A multi-hazard framework to map exposure to multiple climate extremes in Europe along the twenty-first century is hereby presented. Using an ensemble of climate projections, changes in the frequency of heat and cold waves, river and coastal flooding, streamflow droughts, wildfires and windstorms are evaluated. Corresponding variations in expected annual exposure allow for a quantitative comparison of hazards described by different process characteristics and metrics. Projected changes in exposure depict important variations in hazard scenarios, especially those linked to rising temperatures, and spatial patterns largely modulated by local climate conditions. Results show that Europe will likely face a progressive increase in overall climate hazard with a prominent spatial gradient towards south-western regions mainly driven by the rise of heat waves, droughts and wildfires. Key hotspots emerge particularly along coastlines and in floodplains, often highly populated and economically pivotal, where floods and windstorms could be critical in combination with other climate hazards. Projected increases in exposure will be larger for very extreme events due to their pronounced changes in frequency. Results of this appraisal provide useful input for forthcoming European disaster risk and adaptation policy.
KeywordsReturn Period Extreme Event Return Level Cold Wave Coastal Flood
Europe is expected to face major impacts from a changing climate over the coming decades (Kreibich et al. 2014). The hazard to society and environment will be largely connected to changes in extreme climate events due to their disproportionate rise compared to the corresponding change in climatological averages (Rummukainen 2012). Threats will be more pronounced in areas prone to multiple climate hazards. In this context, a multi-hazard assessment accounting for possible regional variations in intensity and frequency of climate extremes is essential to identify areas potentially more exposed to climate change.
A number of climate change impact studies at the European level have been achieved, usually for a single specific climate or weather hazard, such as river floods (Rojas et al. 2012; Alfieri et al. 2015), coastal floods (Nicholls and Klein 2005; Hinkel et al. 2010), heat waves (Fischer and Schär 2010; Russo et al. 2015; Christidis et al. 2015), streamflow droughts (Lehner et al. 2006; Forzieri et al. 2014), windstorms (Nikulin et al. 2011; Outten and Esau 2013) and wildfires (Bedia et al. 2013; Migliavacca et al. 2013a). The study of multiple hazards poses two major challenges: (1) hazards are not directly comparable as their processes and describing metrics differ; and (2) hazards can interact triggering cascade effects and coupled dynamics. In the existing literature, the first issue has been mainly addressed through standardization approaches, such as classification of hazard intensity and development of continuous indices (Dilley 2005; Kappes et al. 2012; Lung et al. 2013). While these approaches represent a starting point, they describe only a limited set of climate hazards and the techniques used to make different hazards comparable are largely subjective and inconsistent. The second issue has been addressed mainly qualitatively through descriptive matrices where coupled mechanisms are conceptualized based on multi-hazard dynamics observed at local scale and largely influenced by landscape figures (Kappes et al. 2012; Gill and Malamud 2014). Deeper data-driven investigations are needed before interactions between hazards can be reliably incorporated into large-scale predictive systems. Thus, in this study we mainly focus on the first above-mentioned challenge.
Through a unique collaborative effort of different European modeling groups, a consistent set of climate hazard modeling data has been produced for this study including heat and cold waves, river and coastal floods, droughts, wildfires and windstorms. Future climate hazards in Europe have been generated for an ensemble of regional climate simulations under a “business-as-usual” (SRES A1B) greenhouse gas (GHG) emissions trajectory (Solomon 2007) and synthesized in a coherent multi-hazard framework. The method is based on the analysis of the changes in frequency of climate-induced extreme events and the corresponding variations in expected annual exposure to these events. The latter is hereafter defined as Expected Annual Fraction Exposed (EAFE), where the fraction can relate to any variable of interest (e.g., population, cropland). For a range of hazard severities, single-hazard EAFEs and changes therein are combined into multi-hazard indices to synthesize the potential exposure to multiple climate hazards (Methods). This work provides the first comprehensive multi-hazard assessment for Europe under climate change and focuses in particular on the comparability amongst single-hazard exposures and on the degree of overlap between areas exposed to multiple hazards throughout this century. The overall goal is to identify geographic areas with the highest potential exposure to multiple climate hazards in order to better steer adaptation efforts and land planning across Europe. It is worth to stress that this contribution should not be confused with a risk assessment study. Risk assessments imply the combination of hazard, vulnerability and spatial distribution of settlements (e.g., population, assets). In this work we focus on the hazard component, the integration of local vulnerabilities of settlements to different types of hazards will be tackled in a separated research work. Results are shown in spatial maps as well as aggregated for five European regions to simplify interpretation (Figure S1, Supplementary Material): Southern, Western, Central, Eastern and Northern Europe.
2.1 Climate hazard indicators
The analysis focuses on seven critical climate hazards for Europe: heat and cold waves, river and coastal floods, droughts, wildfires and windstorms, each one described by an indicator that relates to the physical impact (magnitude of warm/cold period, flood extents, minimum discharge, burned area, wind speed, Text S1 Supplementary Material). Climate hazard indicators were derived for the baseline (1981–2010), 2020s (2011–2040), 2050s (2041–2070) and 2080s (2071–2100) for an ensemble of climate projections obtained from different Global Circulation Model-Regional Climate Model (GCM-RCM) simulations under the A1B emissions scenario (Solomon 2007) (table S1, Supplementary Material). Precipitation and temperature fields utilized in our study as climatic drivers have been bias corrected by the quantile method (Dosio and Paruolo 2011).
Heatwaves were defined by the Heat Wave Magnitude Index daily (HWMId) that is based on the daily maximum temperature anomalies (Russo et al. 2015). Cold waves were similarly calculated by referring to minimum temperatures. Return levels of heat and cold waves were retrieved by kernel density estimator with triangular kernel. Wildfires were derived from projections of the monthly percentage of burned area (Migliavacca et al. 2013a). Beta functions were selected to fit the annual fractions of burned area and to derive extreme events. Extreme windstorms were calculated using the Generalized Pareto distributions that have been derived through a peak-over-threshold analysis for daily maximum wind speeds (Outten and Esau 2013). Relative Sea Level Rise projections were combined with current extreme value distributions of total water levels obtained using a peak-over-threshold method (Pardaens et al. 2011; Cid et al. 2014). Following, a static inundation approach was applied to generate inundation maps along the coastline. For inland flooding the annual maximum discharges and flood inundation maps were derived from earlier works (Rojas et al. 2012; Rojas et al. 2013). For drought the minimum discharges and return levels were obtained from a previous study (Forzieri et al. 2014). Details in Text S1, Supplementary Material. All climate hazard indicators have been scaled to the common 1000-m grid. Figure S2 shows the spatial modeling domain for each hazard. Note that for river and coastal floods the baseline 500-yr. flood extension is used as reference modelling domain. Details in Text S2, Supplementary Material.
2.2 Frequency of extreme events in current and future climate
Climate model variability was quantified by the coefficient of variance of the future return periods retrieved for the different climate realizations. The significance of the changes in climate hazard was evaluated by the Kolmogorov-Smirnov test applied on the annual values of future time windows versus baseline, separately for each climate model.
2.3 Expected annual fraction exposed
2.4 Combining multiple hazards
To identify areas subject to large increases in exposure to multiple hazards, we define the Change Exposure Index (CEI). CEI expresses the number of hazards - of a given baseline return level - with a future relative increase in EAFE over a certain threshold (20 %, 100 % and 1000 %). The use of three different thresholds allows capturing moderate, strong and extreme changes in hazard exposure. The number of hazards with an increase in exposure over the given threshold is calculated in each grid cell and then aggregated at NUTS3 level as the 0.99 percentile of this distribution over all cells. The 99th percentile of the exposure change distribution within the NUTS3 region excludes local fitting extrapolation errors and is considered representative of the maximum degree of change in exposure. CEI allows identifying key hotspots subject to predefined levels of change in exposure (details in Text S4, Supplementary Material).
For the spatial domain common to all hazards OEI and CEI are calculated for each return level and time slice using the ensemble median of all climate model combinations for each hazard as inputs because only one single GCM-RCM configuration is common amongst the hazards (Table S1). To understand the possible effects of climate uncertainty on multi-hazard metrics, OEI and CEI have been also calculated for the maximum and minimum hazard scenarios obtained by combining the single-hazard grid-cell ensemble maximum and minimum, respectively, as inputs.
3 Results and discussion
3.1 Single-hazard impacts
Interestingly, larger increases in EAFE can be observed at higher return levels and for long-term scenarios due to the progressive intensification of very extreme events. This occurs also in regions prevalently experiencing a reduction (or slight change) in future frequency of climate hazards, such as Central and Eastern Europe for droughts, Southern Europe for wildfires and Southern, Central and Northern Europe for floods. The apparent contradiction manifests where few localized areas experience a very large increase in frequency that outweighs the opposite tendency occurring in most of the region.
Projected changes in single-hazard exposure suggest that future hazard scenarios will considerably deviate from those observed in current climate, especially for climate hazards strongly linked to temperature rises (e.g., heat and cold waves, droughts and coastal floods). Despite the general good agreement in the direction of change in exposure amongst the minimum/maximum hazard scenarios (whiskers in Fig. 2) opposite variations in EAFE are apparent in some situations. This is evident for droughts in Central, Eastern and Northern Europe where upper and lower bounds of the range are greater and lower, respectively, than the baseline value (e.g., 0.01 for 100-yr. baseline return period). A deeper inspection of the EAFEs values originated from single GCM-RCM combinations reveals that changes of different hazards may present a dependence across models, with generally more pronounced increases in exposure in models with a larger overall warming (e.g., C4I-RCA-HadCM3, METO-HadRM3-HadCM3, Fig. S5).
3.2 Changes in overall and concurrent exposures
Relevant climate variability emerges for both OEI and CEI, more pronounced for long-term scenarios, larger return periods and higher degree of overlapping/change when compared to their median values (Fig. 3a and 4a), largely consistent to the ranges of variability observed in minimum/maximum single-hazard EAFE scenarios (Fig. 2). We argue that the uncertainty captured by the minimum and maximum scenarios tends to overestimate the one that would originate ideally from a combination of each model individually into multi-hazard metrics. Then, sampled ranges of variability should be considered as a qualitative proxy of how model uncertainties of single hazards propagate into multi-hazard metrics.
3.3 Sources of uncertainty
Despite the depth of this study, results should be viewed in light of the potential uncertainty sources and caveats of the proposed methodology. The multi-hazard maps are dependent on the chosen set of climate hazard indicators: the use of diverse input hazards (e.g., hail, landslides) might lead to different findings. We argue that the set of hazards selected includes the most relevant hazards for Europe in terms of average annual losses and deaths (Guha-Sapir et al. 2014; NatCatSERVICE 2015). Metrics used to represent the selected climate hazards are crucial for the resulting impact scenarios: changes in return periods depend on the time scale selected to characterize an event type, e.g. 1-day temperature extremes, weekly heatwaves or seasonal heat anomalies experience different changes in return periods (Perkins and Alexander 2012; Trenberth et al. 2014). In our approach we focus on hazard-specific metrics of impact relevance that have been documented in recent literature. Details on the sensitivity analysis and calibration/validation exercises for each single hazard are reported in the references (Rojas et al. 2012; Migliavacca et al. 2013b; Outten and Esau 2013; Forzieri et al. 2014; Cid et al. 2014; Russo et al. 2014). We recognize that extreme value fitting and kernel density estimators may introduce additional uncertainty in the projections of climate hazards especially at high return periods. Recent studies, though, documented its secondary role with respect to the inter-model spread (Rojas et al. 2012; Forzieri et al. 2014).
We apply a conservative approach without accounting explicitly for hazard interrelations that could lead to greater impacts. Regions exposed to the overlap of multiple hazards and subject to concurrent increases in single-hazard EAFEs, however, are indicative of a more likely exacerbation of the overall impacts due to inter-hazard triggering relationships. Estimation of probabilities of coincidental or cascading events would require finer time resolution of hazard metrics (here annual or monthly) and a better knowledge of the inter-hazard physical interactions and coupled processes.
The socioeconomic scenarios driving GHG emissions, the sensitivity of the climate models to GHG concentrations and the specific hazard modelling utilized are subject to uncertainty, and all are relevant in influencing the final multi-hazard assessment. The use of different climate model ensembles for each hazard may have introduced additional artifacts (Table S1, Supplementary Material). However, recent studies suggest that the reduced subsets utilized in this study for some hazards largely preserve main statistical properties of the initial 12-member ensemble (Russo et al. 2013). The use of identical - and possibly larger - ensembles could allow to better capturing climate-related uncertainties (Kharin et al. 2013; Sillmann et al. 2013). We used a different baseline and only one future time window for windstorms (see text S1, Supplementary Material). New dedicated runs for windstorms for the remaining temporal periods were not feasible within this study. We understand that such diversity may limit the comparability with the other hazards; however, changes in extreme winds seem to be lower compared to the other climate hazards, hence the potential bias is expected to play a minor role. Analyses of the multi-hazard indices are performed using the ensemble median (and minimum/maximum) of all climate model combinations for each hazard as input because only one single GCM-RCM configuration is common amongst the hazards. While the median can be considered a robust estimate of single-hazard ensembles, this inevitably hampers the analysis of how single-hazard uncertainties (Fig. 1) propagate to the combined metrics, especially in light of possible dependences of hazards across climate models (Figure S5).
Projected changes in the occurrence of the seven climate extremes depict important variations in hazard scenarios with large spatial patterns modulated by local climate conditions.
Europe will see a progressive and strong increase in overall climate hazard with a prominent spatial gradient towards south-western regions.
Key hotspots emerge particularly along coastlines and in floodplains in Southern and Western Europe, which are often highly populated and economically pivotal.
Results – interpreted in light of exposed assets and their vulnerability – provide useful input to derive future multi-hazard risk scenarios and support adaptation strategies to increasing Europe’s resilience to climate change.
G.F. and L.F. conceived and designed the study, G.F. and S.O. analyzed windstorms, G.F. analyzed streamflow droughts, G.F. and M.M. analyzed wildfires, S.R. analyzed heat and cold waves, M.V. and A.C. analyzed coastal floods, R.R. and L.A. analyzed river floods, G.F. and A.B. harmonized the climate hazards, G.F. developed the multi-hazard indices. G.F. and L.F. wrote the paper with contributions from all the authors. This research was supported by the CCMFF Project, funded by the DG Climate Action (ref. 071303/2012/630715/CLIMA.C.3) and the FP7 ENHANCE project (grant agreement number 308438) funded by the European Commission.
- Dilley M (2005) Natural disaster hotspots: A Global risk analysis. World Bank PublicationsGoogle Scholar
- Dosio A, Paruolo P (2011) Bias correction of the ENSEMBLES high-resolution climate change projections for use by impact models: Evaluation on the present climate. J Geophys Res Atmospheres 116:D16106. doi: 10.1029/2011JD015934
- Guha-Sapir, D, Hoyois, Ph, Below, R (2014) Annual disaster statistical review 2013, the numbers and trends. Centre for Research on the Epidemiology of Disasters (CRED), Brussels, BelgiumGoogle Scholar
- NatCatSERVICE | Munich Re. http://www.munichre.com/natcatservice. Accessed 26 May 2015
- NatCatSERVICE (2015) Annual statistics 2014, Loss events worldwide 2014, Percentage distribution. Munich Re, Munich, GermanyGoogle Scholar
- Solomon S (2007) Climate change 2007 - the physical Science Basis: Working Group I contribution to the Fourth assessment Report of the IPCC. Cambridge University PressGoogle Scholar
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.