Thongchai Thailand

Event Attribution Science: A Case Study

Posted on: July 10, 2018


  2. A notable feature of the “Common but Differentiated Responsibilities” principle of the Kyoto Protocol/UNFCCC is that the rich industrialized countries (Annex-1) are required to compensate poor developing countries (non-Annex) for “adaptation cost” of climate change impacts such as rising seas and extreme weather. Initially, all such events in the nonAnnex countries were fundable under the Framework Convention but later it was argued that this funding policy is arbitrary because natural variability is known to cause extreme weather events anyway even in the absence of fossil fuel emissions and that therefore not all extreme weather events can be attributed to fossil fuel emissions and not all extreme weather events are relevant in the context of climate adaptation assistance from the Annex I countries to the nonAnnex countries (Allen, 2003). This principle was formalized in the Warsaw International Mechanism (WIM) for Loss and Damage Associated with Climate Change Impacts (UNFCCC, 2013).
  3. The Warsaw International Mechanism (WIM) has redefined climate change adaptation funding as a form of compensation for “loss and damage” suffered by nonAnnex countries because of sea level rise or extreme weather events caused by fossil fuel emissions which are thought to be mostly a product of the industrialized countries. Accordingly, the WIM requires that loss and damage suffered by the nonAnnex countries for which compensation is sought from climate adaptation funds must be attributable to fossil fuel emissions.
  4. A probabilistic methodology was devised to address the need for attribution in the WIM and It has gained widespread acceptance in both technical and policy circles as a tool for the allocation of limited climate adaptation funds among competing needs of the VNAL countries (Stott, 2013) (Otto, 2015) (Otto, 2012) (James, 2014) (Trenberth, 2015) (Peterson, 2012) (Huggel, 2013). The probabilistic event attribution methodology (PEA) uses a large number of climate model experiments with multiple models and a multiplicity of initial conditions. A large sample size is used because extreme weather events are rare and their probability small by definition. The probability of an observed extreme weather event with anthropogenic emissions and the probability without anthropogenic emissions are derived from climate model experiments as P1 and P0. If the probability with emissions (P1) exceeds the probability without emissions (P0), the results are interpreted to indicate that emissions played a role in the occurrence of the event in question. Otherwise the event is assumed to be a product of natural variation alone. The probability that fossil fuel emissions played a role in the extreme weather event is represented as P = (P1-P0)/P0. A contentious issue in PEA analysis is that of uncertainty in the values of P0 and P1 and in the model results themselves. Policy analysts fear that large uncertainties of climate models (Oreskes, 1994) (Frame, 2011) (Curry, 2011) and shortcomings of the PEA methodology (Zwiers, 2013) (Hulme, 2011) provide sufficient reason to question the reliability of PEA to serve its intended function as a criterion for access to climate adaptation funds (Hulme, 2011). Mike Hulme argues that much greater statistical confidence in the PEA test is needed to justify denial of adaptation funding for loss and damage from weather extremes that do not pass the PEA test. Yet another concern with respect to the PEA methodology, and one that is the subject of this post, is the apparent tendency in climate science to extend the interpretation of PEA results beyond their intended function of climate adaptation fund allocation and into the realm of empirical evidence.
  5. It has long been claimed that basic principles of climate science imply that fossil fuel driven AGW will increase the frequency and severity of extreme weather events such as tropical cyclones, tornadoes, heat waves, extreme cold, droughts, floods, landslides, and even forest fires (IPCCAR4, 2007) (IPCCSREX, 2012) (IPCCAR5, 2014) (Easterling-Meehl, 2000) (Easterling-Evans, 2000) (Karl, 2003) (Min, 2011) (Allen, 2002) (Rosenzweig, 2001) (Mirza, 2003) (Coumou, 2012) (VanAalst, 2006) (Burton, 1997) (Flannigan, 2000) (Stocks, 1998) (Gillett N. , 2004). It is argued that the harm from these extreme weather effects of AGW provides scientific, ecological, social, political, and economic justification for urgent and costly reductions in fossil fuel emissions as a way of attenuating AGW because the social and economic cost of adaptation later is greater than the cost of mitigation now (Stern, 2006) (Metz, 2007) (IPCC, 2014). It is this catastrophic nature of AGW that provides the rationale for the policy proposal  that requires Annex I countries to reduce emissions by changing their energy infrastructure from fossil fuels to renewables (UNFCCC, 2014) (UNFCCC, 2016). And yet, this line of reasoning is weakened by a frustrating inability of climate science to produce empirical evidence that relates extreme weather disasters to emissions (IPCCSREX, 2012) (IPCCAR4, 2007) (Sheffield, 2008) (Bouwer, 2011) (Munshi, 2015) (IPCCAR5, 2014) (NCEI, 2015) (Woollings, 2014). Of particular note in this regard is that claims made by the IPCC in 2007 with regard to the effect of AGW on the frequency and intensity of tropical cyclones, droughts, and floods have been all but retracted in their very next Assessment Report in 2014 (IPCCAR5, 2014). Thus, climate scientists, though convinced of the causal connection between AGW and extreme weather events, are nevertheless unable to provide acceptable empirical evidence to support what to them is obvious and “unequivocal” (Curry, 2011) (Zwiers, 2013) (Munshi, 2016) (Frame, 2011) (Hulme, 2014).
  6. It is likely this frustration with the absence of trends in the historical data on extreme weather that motivated climate scientists to turn to PEA analysis as an alternative to empirical evidence of the extreme weather effects of AGW (WMO, 2016) (Stott, 2013) (Trenberth, 2015) (Otto, 2015). Thus, the PEA procedure has been extrapolated and generalized well beyond the context of its narrow definition in terms of the WIM. Such extrapolation allows climate science to present positive PEA results as evidence that extreme precipitation, floods, droughts, heat waves, and cold spells are attributable to fossil fuel emissions (WMO, 2016) (Sneed, 2017). In keeping with its elevated status, the PEA methodology has been re-christened as Event Attribution Science or just Event Attribution. Here we show with the high profile example of the autumn floods of 2000 in England and Wales that this interpretation of PEA results (Stott, 2013) commits the fallacy of circular reasoning and that therefore positive PEA results by themselves do not constitute empirical evidence that AGW causes extreme weather events.
  7. Critical commentaries on the PEA methodology have been published by Hulme, Boardman, Kelman and others (Hulme, 2011) (Hulme, 2014) (Boardman, 2008) (Boardman, 2003) (Kelman, 2001). Mike Hulme’s work exposes the weaknesses and the limits of the PEA methodology, John Boardman shows that the PEA methodology had erroneously ascribed non-meteorological aspects of the floods to climate change, and Kelman exposes certain inconsistencies in the details of the flood data and what had been assumed in the Event Attribution climate model analysis. This work describes Event Attribution Science in a case study format using the high profile example of the Event Attribution analysis of the floods in England and Wales in the year 2000 by (Pall, 2011). A critical evaluation of these interpretations is made in light of the relevant precipitation data (Met Office, 2017) and non-meteorological factors that affect flood impact severity (Boardman, 2003) (Boardman, 2008). The use of the Event Attribution Science to present evidence of the effect of fossil fuel emissions on the severity of extreme weather events is evaluated in this context.
  8. In mid-September of the year 2000 an unrelenting sequence of rainstorms began to strike in various parts of England and Wales. By mid-October flooding became widespread and devastating. The rainstorms and floods continued in multiple sequential flooding events until mid-December (Kelman, 2001) (Marsh, 2001) (Marsh, 2002). The series of rainstorms taken together is considered to be a rare and extreme meteorological event in terms of the amount of precipitation, the duration, and the high runoff rates in all the rivers in the region and considered to be a consequence of climate change (Reynard, 2001) (Pall, 2011).
  9. A study of the autumn 2000 floods by Terry Marsh found that 640 mm of rain fell in the four months of rain ending December 15, 2000 and 1033 mm of rain fell in the eight months ending in April 2001. By both measures, the year 2000 ranks first in the 235-year data record going back to 1766 (Marsh, 2001). An estimate of the total volume of water delivered to the ground by the rainstorms is also reported in the Marsh study in terms of the combined flows of the Rivers Thames, Severn, Welsh Dee, and Wharfe. By this measure the year 2000 ranked second after 1947 for 10-day and 30-day outflows and first, just ahead of 1947, for 60-day and 90-day outflows (Marsh, 2001). Marsh also points out that the 2000 floods did not occur in isolation. They fall in the middle of cluster of flood years in the area prior to 2000 that includes 1989, 1993, 1994, and 1998 and since 2000 in 2007, 2013-2014, and 2015-2016 (Marsh, 2001) (Marsh, 2002) (Marsh, 2007) (Huntingford, 2014) (Schaller, 2016) (Marsh, 2016).
  10. These storm and precipitation events came to be thought of as extreme and unnatural and research interest turned to the effect of human influences in the form of anthropogenic global warming and climate change on the apparent unnatural increase in the frequency and intensity of floods in the UK (Reynard, 1996) (Reynard, 1998) (Reynard, 2001) (Macklin, 2003) (Wilby, 2008). It was in this context that the newly minted Event Attribution methodology was applied to the autumn floods of 2000 using an array of climate models and a large sample of climate model runs.
  11. The results showed that the 2000 floods were more likely in “the world as it is” (with fossil fuel emissions) than in “the world that might have been” without fossil fuel emissions. Based on this PEA result the autumn floods of 2000 in England and Wales were attributed to climate change and indirectly to fossil fuel emissions (Pall, 2011).
  12. However, the attribution of the floods to emissions remains controversial. First, floods are not purely meteorological events because important non-meteorological factors play a role in the intensity and devastation of flooding at any given level of rainfall (Kelman, 2001) (Boardman, 2003) (Boardman, 2008) (Boardman, Don’t blame the climate, 2008). Also, the known large natural variability in precipitation in England and Wales provides a simpler explanation of extreme flooding events than the effect of anthropogenic carbon dioxide emissions found in climate models (Kay, 2009) (Shackley, 1998) (Young, 1996) (Beder, 1999) (Curry, 2011) (Frame, 2011) (Hulme, 2014) (Deser, 2012).
  13. This case study examines patterns in the 251-year record of precipitation in England and Wales from 1766 to 2016 in the context of the conclusions drawn from PEA analysis about the autumn floods of 2000 in England and Wales and considers whether a simpler and more natural explanation exists for this precipitation event. Historical monthly mean precipitation data for England and Wales are provided by the Met Office of the Government of the UK (MetOffice, 2017). Precipitation data are recorded in millimeters of water equivalent at standard conditions in a continuous annual time series for a 251-year period from 1766 to 2016 for each of the twelve calendar months.
  14. The data along with their OLS linear trends are depicted graphically, month by month, in Charts1-6 below the text in the chart section of this post. We note in these figures that the calendar months differ significantly in terms of mean monthly precipitation, the variance of precipitation, and the overall trend in the study period. These differences (highlighted in Charts 7-8) show that mean monthly precipitation characteristics differ markedly among the calendar months. On average, autumn is the wettest and spring the driest. Summer and winter lie in between with winter wetter than summer. Year to year variability in precipitation is also different among the calendar months. Charts 9-14 show large differences among the calendar months in standard deviations measured in a moving 30-year window. The generational (30-year) time scale is generally used in the study of climate phenomena (Ackerman, 2006) (WMO, 2016). The red line in these charts marks the no-trend boundary between rising and declining trends.
  15. To maintain the integrity of these observed differences, monthly mean precipitation data are not combined. Instead each month is studied in isolation as a phenomenon of nature unique to that month. A benefit of this methodology is that it facilitates the interpretation of historical trend analysis in terms of the season of the floods under study.
  16. Outlier analysis is carried out by examining the ten largest values from each time series one at a time starting with the largest and moving sequentially to the smallest. At each step a hypothesis test is used to determine whether the value removed belongs to the distribution of all values that are less than the value removed. The null hypothesis is H0: testValue ≤ mean(all values less than the test value). If the null hypothesis is rejected the test value is marked as an outlier and described as an extreme year (Dixon, 1950) (Aggarwal, 2015). The procedure is carried out separately for each calendar month.
  17. A generational (30-year) moving window is used to compute a time series of 221 variability measures for each calendar month. Variability is expressed as the standard deviation and studied for trends. A statistically significant rising trend in this series is expected if climate change is causing the precipitation series to become more volatile. Statistical significance is determined using classical hypothesis testing at a maximum false positive error rate of α=0.001 consistent with “Revised Standards for Statistical Evidence” published by the National Academy of Sciences to address an unacceptably large proportion of irreproducible results in published research (Johnson, 2013) (Siegfried, 2010).
  18. The proposition that anthropogenic CO2 emissions since the Industrial Revolution have increased the amount of precipitation in England and Wales implies that the study period 1766-2016 should show a statistically significant rising trend in precipitation amounts. The simple OLS trend lines for mean monthly precipitation amounts are shown in Charts 1-6. Only three out of the twelve calendar months show a statistically significant trend. The winter months of December and January show the required rising trend in precipitation while the summer month of July shows a declining trend. No evidence of a trend in mean monthly precipitation amount is found in the other nine calendar months.
  19. To explain the autumn 2000 floods in England and Wales in terms of trends, a positive trend is necessary for the four months from September to and December. A positive result for December alone does not provide sufficient evidence that rising trends in the amount of precipitation were responsible for the floods of 2000. It is also noted that the summer floods of 2007 appear anomalous in this line of reasoning as the only evidence of a trend in the summer months is a declining trend for July. The event attribution finding that the autumn floods of 2000 were caused by climate change is consistent with historical record.
  20. Charts 9-14 are graphical displays of the standard deviation of mean monthly precipitation in a generational moving window. The results provide evidence that mean monthly precipitation in the autumn months of October and November, the winter months of January and February, and the spring month of March have become more volatile over the study period. The summer month of July shows a decreasing trend in volatility and the other six months show no evidence of a trend in volatility.
  21. The evidence of rising volatility in the months of October and November might have been consistent with the attribution of the floods of 2000 to climate change had the volatility been associated with greater precipitation. Without evidence of a rising trend in precipitation in these months it is unclear whether the greater variance relates to extreme dry years or extreme wet years. To explain floods in terms of climate change evidence of more extreme wet years is necessary.
  22. A search for extreme wet years is made by using outlier analysis. For the purpose of this analysis, a wet year is deemed to be extreme if it is an outlier in the context of all years in the time series with less precipitation. The analysis begins with the identification of the ten highest precipitation years for each calendar month. They are shown in Charts 15-20. Each of the ten wettest years is tested against all years with less rainfall to determine if it is an outlier in the sense that it does not belong in the distribution of the comparison series. These hypothesis tests are carried out at a maximum false positive error rate of α=0.001. Since twelve tests are made the overall study-wide false positive error rate is approximately 0.012 or 1.2% (Holm, 1979). Statistically significant results are identified with filled markers. These outliers are deemed to be extreme precipitation years.
  23. February, September, November, and December contain no extreme years. The other eight months contain at least one extreme wet year. April contains four, May contains three, and March and August contain two each. January, July, and October contain only one extreme year. For the autumn floods of 2000, the relevant months are September, when the rains started, and October, November, and December when they continued and intensified. Only one extreme event is found for these months and it occurred in October 1903 with 218 mm of precipitation. October of 2000 is indeed the second wettest October on record with 188 mm of precipitation but it is not an outlier as 188 mm is well within expected variability of the distribution of October precipitations at α=0.001.
  24. These results, together with the absence of a rising trend in precipitation amount or volatility show no empirical support for the claim made with Event Attribution analysis that the autumn 2000 floods were caused by extreme precipitation events attributable to anthropogenic CO2 emissions. In terms of the cluster of flood years in the UK in the period 1989 to 2015, the only extreme year that occurs in the same season as the floods is the extreme for January in the year 2014 because it coincides with the winter 2013-2014 floods. The years 2000 and 2012 are also found in the list of extremes in Figure 10 but not in the season of the floods in those years.
  25. The presentation of empirical evidence for a given theory proceeds as follows. First, a testable implication of the theory is deduced. Then, data are collected, either experimental data or field data. The data must be independent of the theory and their collection must be unbiased. In the case of classical hypothesis testing, the testable implication is then tested against the data with the null hypothesis that the theory is false. If the null hypothesis is rejected, the data constitute empirical evidence in support of the theory (Popper, 2005) (Pearl, 2009) (Kothari, 2004).
  26. In the case of Event Attribution analysis with climate models, the results serve the intended purpose of providing a non-subjective method for the allocation of climate adaptation funds in accordance with WIM guidelines. However, their further interpretation as evidence of the extreme weather effects of fossil fuel emissions involves circular reasoning because climate model results are not data independent of the theory but a mathematical expression of the theory itself; and the selection of specific events to test for event attribution contains a data collection bias (Munshi, 2016) (Koutsoyiannis, 2008) (VonStorch, 1999). A related post compares the confirmation bias in event attribution analysis with superstition. SUPERSTITION AND CONFIRMATION BIASYet another contentious issue in event attribution with climate models is the known chaotic behavior of climate that is not contained in climate models. Non-linear dynamics and chaos is discussed in a related post: IS CLIMATE CHAOTIC?























2 Responses to "Event Attribution Science: A Case Study"

[…] EXTREME WEATHER EVENTS: The “Event Attribution Science” methodology is derived from the Probabilistic Event Attribution of PEA that was formalized as the Warsaw International Mechanism for Loss and Damage Associated with Climate Change Impacts or WIM by the UNFCCC as a fund allocation tool for the determination of whether a given weather event in a poor non-Annex country qualifies for financial compensation from rich Annex-I countries. It’s elevation to “science” and it’s use to identify extreme weather events as evidence of the harful nature of climate change is an extreme form of confirmation bias and circular reasoning. A detailed case study of PEA elevated to Event Attribution Science is provided in a related post here: EVENT ATTRIBUTION CASE STUDY. […]

[…] Such attribution serves not only to re-enforce the belief in the dangerous nature of climate change and the urgency of Climate Action to prevent the harm that it might otherwise cause. Yet, this superstition is actually presented by climate scientists as empirical evidence of human caused climate change in terms of what has come to be called “Event Attribution Science” (Munshi, 2017) (Trenberth, 2015) (Stott, 2016) (Hegerl, 2010). This aspect of human behavior, where an assumed theory of causation guides the interpretation of data in a way that re-enforces the theory of causation can be described in terms of superstition. EVENT ATTRIBUTION SCIENCE […]

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: