1 the emerging science of attributing causes to extreme eventsEvent attribution: the emerging science of attributing causes to extreme events Francis Zwiers PCIC, University of Victoria Wildfire Canada 2016, 25 October 2016 Photo: F. Zwiers (Smoke filled sunset, Aug, 2014, Winthrop, WA)
2 Outline Introduction Extreme event attribution overview ExamplesHorse River Fire (2016) Calgary Floods (2013) China’s hot summer of 2013 Record low Arctic sea ice cover (2012) Discussion Acknowledgements: Megan Kirchmeier-Young, Bernado Teufel Ying Sun, Nathan Gillett, Xuebin Zhang and many others Photo credit F Fort McMurray evacuation
3 The context for this talkExtensive reporting in the media on extreme events Google News searches of Canadian new publications for the past year find 55,300 items that refer to “extreme weather” 17,500 items that refer to “drought” 31,400 items that refer to “floods” Similar searches for 2006 yield very small numbers Public perception is that frequency and intensity is increasing Growing economic impact of extreme events Growing insurance industry concern (e.g., Munich Re)
4 The context … Media discourse tends to quickly evoke possible links to climate change As a default, we scientists tend to point to the similarity between recent events and projected change Event attribution science has been trying to find a way for science to do better than this Requires “rapid response” science Places high demands on process understanding, data, models, and statistical methods Recently assessed by US National Academies of Science
5 Event attribution Photo: F. Zwiers (Jordan River, gathering storm)
6 Event attribution The public asks: Did human influence on the climate system … Cause the event? Most studies ask: Did it … Affect its odds? Alter its magnitude? Some think we should reframe the question … Rather than “Did human influence …” (which requires comparison with a counterfactual world) Ask “How much (eg, of a given storm’s precipitation) is due to the attributed warming (eg, in the storm’s moisture source area)” (after Trenberth et al, 2015) The former allows statistical analyses that inform risk assessment The latter is much more physically based and recognizes that human influence has a role in every event given that the composition of the atmosphere is not that of the preindustrial state.
7 Most studies Compare factual and “counterfactual” climatesCounterfactual the world that might have been if we had not emitted the ~600GtC that have been emitted since preindustrial These studies almost always Define a class of events rather than a single event Use a probabilistic approach Shepherd (2016) defines this as “risk based” Contrasts it with a “storyline” based approach i.e., analysis of the specific event that occurred
8 “Framing” event attribution studiesEvent type Class vs individual Analysis approach “risk based” or “storyline” Event definition What spatial scale, duration, etc Which risk-based question Did climate change alter the odds, or the magnitude? What factors should be taken into account “Conditioning” e.g., prevailing SST anomaly pattern, circulation, etc The NAS Report (2016) struggled with these distinctions
9 Framing (i.e., how the question is asked) affects the answerPhoto: F. Zwiers (Emlyn Cove)
10 JJA temperature anomalies relative to 1961-199020 July – 20 Aug 2003 vs the same period averaged over excluding 2003 Courtesy Reto Stockli and Robert Simmon (NASA/Wikipedia) Framing … How the event is defined For example, how detailed is the definition? The first “event attribution” study (Stott et al., 2004) dealt with the 2003 European heat wave The exact definition of the evident (duration and spatial extent) is unclear, … Therefore the study focused on mean summer conditions across southern Europe JJA temperature anomalies relative to Figure 1, Stott et al., 2004
11 July 2010 mean surface temperature anomaly relative to 1880-2009Framing … July 2010 mean surface temperature anomaly relative to Choice of risk based question Two studies of the Russian 2010 heat wave came to conflicting conclusions One focused on intensity (found little human influence) The other focused on frequency (found a large human influence) Answering both questions avoids confusion, and answers questions posed by different users “Factual” and “Counterfactual” Russian (50-60°N, 35-55°E) July mean surface temperature distributions 100 10 0.01 0.1 Probability Return Time - yr °C 1
12 Framing … What factors are controlled in the analysisStatisticians call this “conditioning” Two distributions of event magnitude could be calculated taking the presence or absence of anthropogenic forcing into account Or the calculations could take additional factors into consideration as well, such as the prevailing pattern of SST anomalies “Factual” “Counterfactual” “Factual” “Counterfactual”
13 Framing … Many studies condition on SST anomaliesRestricting a source of variability may improve signal-to-noise ratios Specifying the state of the sea surface allows the use of atmospheric, rather than coupled cheapers models Cheaper Can sometimes use 1000’s or 10000’s of simulations One approach is to use personal computers volunteered by the public via Conditioning may add uncertainties Need to estimate the counterfactual SST base state Likelihood of the SSTA pattern may change
14 Two key numbers Many event attribution studies focus on the “Fraction of Attributable Risk” (Allen, 2003) Under suitable conditions Hannart et al (2016) also show that Prob of event in factual world Prob of event in “counterfactual” world
15 Horse River Fire – May through July 2016590,000 ha burnt 88,000 people displaced 2 fatalities (indirect) 2400 homes and 665 work camp units destroyed $3.6 B insured losses Mandatory evacuation. Photo, Jason Franson/CP Avian escape. Photo, Mark Blinch/Reuters Edmonton Expo Centre at Northlands. Photo, Chris Bolin Timberlea. Photo, Chris Bolin
16 Fire risk (Kirchmeier-Young et al, in prep)Annual area burned Canadian National Fire Database We ask whether human induced climate change has affected extreme fire indices We consider 90th percentile of fire index values for each fire season (MJJAS) the “Southern Prairie” Homogeneous Fire Regime zone fire indices that reflect variations in fire risk on different time scales Fire Weather Index Fine Fuels Moisture Code Duff Moisture Code Drought Code Southern Prairie HFR Zone the indices depend on temperature, relative humidity, wind speed, and precipitation
17 Models, data processingWe use the CanESM2 large ensemble simulations 50 run ensembles with historical anthropogenic and natural forcing combined (ALL) and historical natural forcing (NAT) only We downscale the model to a finer resolution using an advanced statistical downscaling scheme surface air temperature, relative humidity, wind speed and precipitation are downscaled using the Global Fire Weather Database (MERRA reanalysis based) and a new, high resolution blended precipitation dataset as the “downscaling targets” fire indices are derived using the downscaled data
18 Results for HFR zone 9 Estimated distributions of the 90th percentiles of the weather drivers for under ALL and NAT forcing Estimated distributions of the 90th percentiles of the fire indices for under ALL and NAT forcing The shift towards higher values is driven primarily by changes in temperature and wind speed as seen from the distributions of the 90th percentiles of the underlying meteorological variables Vertical lines represent Canadian Wildland Fire Information System (CWFIS) “extreme” levels
19 Has human induced climate change increased fire risk in HFR zone 9?=FAR p1 p0
20 Calgary flood, 2013 100,000 displaced, 5 deaths2nd costliest (?) disaster event in Canadian history Estimated $5.7B USD loss ($1.65B USD insured) Calgary East Village (June 25, 2013), courtesy Ryan L.C. Quan
21 Calgary floods Southern Alberta MJ max 1-day precipDistribution of annual May-June maximum 1-day southern-Alberta precipitation in CRCM5 under factual and counter-factual conditions (conditional on the prevailing global pattern of SST anomalies) Frequency doubles (~25-yr ~12 yr) Magnitude increases ~10% Observed (dashed line) is from CaPA Solid lines are CRCM-5 “ref” simulations (0.11 deg resolution, ERA-Int driven, launched at 6 hourly intervals beginning 00Z June 12, continuing to 00Z June 22) Factual and counterfactual simulated with CRCM5 variable resolution mode (0.5 degree over N. America, 2 degree elsewhere). Counterfactual uses 1850 anthropogenic forcing, observed SST patterns adjusted to 1850 using CanESM2, GFDL-ESM2M, and GISS-E2-H Factual and counterfactual ensembles contain 1000 simulations for each SST configuration (4000 runs in total). Runs are launched hourly beginning 08Z 20 Nov Initial conditions are from a longer spin-up runs that start early in 2012 (one spin up for each SST configuration) FAR=PN≈0.5 PS≈0.04 Teufel et al (2016)
22 China’s Hot Summer of 2013 Impacts included estimated $10B USD agricultural yield loss Photo: F. Zwiers (Yangtze River)
23 How rare was JJA of 2013? Estimated event frequencySun et al, Nature Climate Change, 2014 Anomaly relative to 1.1°C ≈ 3.5 SD above the mean Estimated event frequency once in 270-years in control simulations once in 29-years in “reconstructed” observations once in 4.3 years relative to the climate of 2013 Fraction of Attributable Risk in 2013: (p1 – p0)/p1≈ 0.984 Prob of “sufficient causation”: PS=1-((1-p1)/(1-p0)) ≈ 0.23 FAR=PN=1-p1/p0=0.984 PS =1-((1-p1)/(1-p0))=0.23
24 Projected event frequencyRCP4.5 RCP8.5 Mean temp Frequency + + × × 23%, 4.3-yr Figure 4 | Frequency of extreme hot summer recurrence. Time evolution of the frequency of summer temperature anomalies above 1.1 ◦ C, relative to the 1955–1984 mean, in the reconstructed observations (1955–2013) and in the observationally constrained projections (2014–2072) under RCP4.5 (plus) and RCP8.5 (cross) emission scenarios (left-hand scale). The solid smooth curves are LOESS (local regression) fitting. The dashed curves represent projected ensemble mean temperature changes under the relevant emission scenarios (right-hand scale) and are shown here for reference. Results for RCP4.5 and RCP8.5 are represented by red and green, respectively.
25 Record low Arctic sea ice cover - 2012Photo: F. Zwiers (approach to Alert, Aug., 2009)
26
27 Arctic sea-ice extent event attributionCurves of FAR,PS,RR for an event more extreme than the given threshold (x-axis) using the anomalies and for the decade (limited by CMIP5) Vertical bar at observed 2012 event Cutting off RR plot at 10**6 Resampled pool of data (from years in decade and ensemble members) to estimate uncertainty Kirchmeier-Young et al (2016; in press) All models indicate an event of a magnitude equal to or more extreme than the 2012 record minimum would be exceptionally unlikely to occur under natural forcing alone. ALL forcing is a necessary, but not sufficient cause.
28 Some unresolved issuesPhoto: F. Zwiers (Marsh Wren)
29 Some unresolved issuesEvent characterization Class vs individual, risk-based vs storyline Individual is not completely synonymous with storyline Data assimilation approach of Hannart et al (2016) Event definition Dependence on models Counterfactual state specification uncertainty when conditional approach is used Selection bias Need objective event selection criteria Communications At each stage of the media and disaster response/recovery cycle
30 Questions? https://www.pacificclimate.org/ Photo: F. Zwiers