Explaining extreme events of 2013 from a climate perspective.
SARAH E. PERKINS, SOPHIE. C. LEWIS, ANDREW D. KING, AND LISA V. ALEXANDER
Human activity has increased the risk of experiencing the hot Australian summer of 2012113, as measured by simulated heat wave frequency and intensity, by two- and three-fold, respectively.
Introduction. The Australian summer of 2012/13 was the warmest since records began in 1910 (Bureau of Meteorology 2013a). The season was characterized by the hottest month on record (January), where the continental mean temperature reached 36.9[degrees]C. Averaged nationally, the last four months of 2012 were 1.61[degrees]C higher than the long-term mean. Rainfall was below average for much of the country since July 2012. Along with the late onset of the Australian monsoon, such conditions primed the continent for extremely hot summer weather, including heat waves. Heat waves require detailed focus due to their large impacts (Karoly 2009; Coumou and Rahmstorf 2012), particularly on human health and morbidity (Nitschke et al. 2007). Much of inland Australia experienced extreme temperatures for over three consecutive weeks (Bureau of Meteorology 2013a).
By employing the fraction of attributable risk (FAR) framework (Allen 2003), Lewis and Karoly (2013) demonstrated that the likelihood of the extreme Australian heat during the 2012/13 summer had increased by between 2.5 and 5 times due to human activity. However, this assessment was on the seasonal average temperature anomaly and did not include specific heat wave measures. Here we also undertake an analysis of the summer of 2012/13 but with a metric of two heat wave characteristics (Perkins and Alexander 2013). While focusing specifically on seasonal heat wave measures, such an analysis also allows for the assessment of whether changes in risk are consistent for heat wave magnitude and frequency, thus providing important information for adaptation and impacts groups.
Data and methods. We calculate heat waves using the Excess Heat Factor (EHF) definition (Nairn and Fawcett 2013; Perkins and Alexander 2013) for November-March, where the daily average of minimum and maximum temperature must exceed a separate climatological and monthly threshold for at least three consecutive days. Here the climatological threshold is the calendar day 90th percentile, calculated from a 15-day moving window for 1961-90. Note that EHF units are [degrees]C2 (see Nairn and Fawcett 2013).
In order to investigate the effects of human activity on heat waves, the preindustrial control (289 years long), historical, and RCP8.5 experiments (Taylor et al. 2012) from the Community Earth System Model (CESM; Fischer et al. 2013) were employed. Here we use a 21-member ensemble of CESM (1.875[degrees] x 2.5[degrees] resolution; for further model details, see Fischer et al. 2013). The ensemble is generated through perturbations on the order of 10-13 applied to atmospheric temperature initial conditions. We use 1955-2005 of the historical period and merge it with 2006-13 from RCP8.5. A caveat to this study is its dependence on a single model (CESM). However, collectively, the CESM ensemble simulates reasonable changes and variability in observed heat waves over Australia (Perkins and Fischer 2013). Observed heat wave metrics were calculated for austral summers commencing in 1955-2012 using the Australian Water Availability Project (AWAP) temperature dataset (Jones et al. 2009) interpolated onto a two-degree grid. Using AWAP, the observed 2012/13 anomalies of heat wave measures were calculated.
In this study, we investigate the extreme heat of the 2012/13 summer by analyzing two heat wave characteristics (Perkins and Alexander 2013). These are the total number of heat waves and the peak amplitude (hottest heat wave day). The number of heat waves represents the frequency, and peak heat wave amplitude represents the intensity of the 2012/13 extended summer season (November-March). For each model run and the observations, the characteristics were calculated at the grid box level and expressed as anomalies against the relative 1961-90 average, with the control simulation relative to model years 111-140. The control base period was chosen to be the same length as the historical; however, the start year (111) was selected at random since no significant difference between 30-year windows from the control was detected. All anomalies were area-averaged, and all 21 CESM members were concatenated to form a longer single sample.
We employ the FAR framework to analyze changes in the risk of heat wave attributes due to increases in anthropogenic greenhouse gases. This requires conditions where no anthropogenic emissions are present (the control run) and where greenhouse gas concentrations are prescribed to observed levels to 2005 and projected to 2013, accounting for anthropogenic emissions (the historical/RCP8.5 runs). Three periods are analyzed for summers commencing in 1955-83, 1955-2012, and 1984-2012 to investigate how the risk of each characteristic changes with increasing anthropogenic forcings during the observational period. Per period, we generate 1000 bootstrapped samples, consisting of 50% of the control and historical/RCP8.5 data per heat wave characteristic (i.e., the bootstrapped sample sizes of the control and forced runs are half of the original). Selected years are in two-year blocks to account for time dependence, and using 50% of data accounts for sample size sensitivity. Bootstrapping is employed since a true estimate of FAR cannot be obtained from the original control and forced simulations. Our bootstrapping technique allows the uncertainty in FAR to be estimated. The probability of the respective observed anomaly is calculated, and 1000 FAR values are calculated by:
FAR = 1 - ([P.sub.cntrl]/[P.sub.forced]) (1)
We also compute the corresponding changes in risk of the characteristics by:
Risk = 1 / (1 - FAR) (2)
Lastly, using the FAR values for each heat wave characteristic, we compare the waiting time of the frequency and intensity of the 2012/13 summer in each of the three periods and the control. This deter mines how historical return intervals of the hot 2012/13 summer compare to a world without anthropogenic influence.
Results. Figures 10.1a and 10.1b present the area-averaged probability density functions (PDFs) of anomalies of seasonal heat wave frequency and intensity, respectively. The summer of 2012/13 experienced an unprecedented number of heat waves; however, the peak intensity was not particularly unusual (see Table 10.1). Throughout the periods of the simulations, the right tail of the PDF increases--with greater anthropogenic forcing, more extreme summers, as characterized by heat wave frequency and intensity, are expected (relative to 1961-90).
Figures 10.1c and 10.Id present PDFs of the FAR values per period for heat wave frequency and intensity, respectively. FAR values for intensity (Fig. 10.1b) and frequency (Fig. 10.1a) are very similar. This includes negative median FAR values for 1955-83 (-0.22 and -0.32 for frequency and intensity, respectively). A Kolmogorov-Smirnov test at the 5% level indicates that the 1955-83 and control simulations are not significantly different, indicating that these values hold little meaning and that the impact of human activity on Australian summer intensity and frequency had not yet emerged from natural variability. In the case of this study, this result occurs only when the first 30 years of the historical period (1955-83) is included, that is, when anthropogenic forcings were considerably lower than 2012/13.
It is very likely (>90%) that FAR values are greater than 0.26 and 0.1, respectively, during 1955-2012 (medians 0.49 and 0.27) and very likely that FAR values are greater than 0.55 and 0.37 (medians 0.69 and 0.53), respectively, for 1984-2012. The all-positive FAR values in 1984-2012 indicate that the risk of the intensity and frequency of the 2012/13 summer is always larger due to human activity during the latter decades of the 20th century. Kolmogorov-Smirnov tests on 1955-2012 and 1984-2012 against the control indicate statistical significance for both heat wave frequency and intensity.
Based on the median FAR values, Table 10.1 presents the best estimate changes in the risk of the 2012/13 summer heat wave frequency and intensity during the three time intervals and the corresponding return periods. The risk due to human activity is similar for both characteristics during 1955-1983 and is less than zero. However as discussed above, 1955-83 summer heat wave frequencies and intensities are not distinguishable from the control (i.e., cannot be separated from no human influence).
A striking result is that during 1955-2012 and 1984-2012, the risk of the summer of 2012/13 having such a high heat wave frequency anomaly increases faster than heat wave intensity. During the latter period, the risk of experiencing a summer heat wave number (intensity) greater than that of 2012/13 increases by almost three-fold (two-fold) compared to a world with no anthropogenic forcing. This corresponds to a reduction in return periods to ~33 and 3 years, respectively, compared to 1955-2012. It is also an interesting and important result that even though the 2012/13 summer heat wave intensity was much less "extreme" than heat wave frequency (see corresponding return periods in Table 10.1), human activity has clearly increased the risk of both characteristics occurring. Thus, there is a calculable human influence on the hot Australian summer of 2012/13.
Conclusions. Using a 21-member ensemble of the CESM model, we analyzed changes in the risk of the hot Australian 2012/13 summer with respect to heat wave frequency and intensity. Our study found that the risk of both simulated heat wave characteristics has increased due to human activity. The risk of summer heat wave frequency increases faster than heat wave intensity. When isolating 1984-2012, the 2012/13 heat wave frequency increased three-fold due to human activity, while heat wave intensity increased two-fold, compared to a climate with no anthropogenic forcings.
This infers a reduction in return periods when comparing 1955-2012 to 1984-2012--from 58 years to 33 years for frequency and from 4 years to 3 years for intensity. Lastly, even though heat wave intensity of 2012/13 was not the most severe Australia experienced, there is still a calculable influence on this heat wave characteristic on a seasonal scale. Overall, our study shows that the risk of the hot 2012/13 Australian summer with respect to simulated heat wave frequency and intensity increased due to human influences on climate.
Table 10.1. Changes in the risk of Australian 2012/13 heat wave frequency (number of heat waves) and intensity (peak magnitude) anomalies due to anthropogenic forcings throughout summers commencing in 1955-83, 1955-2012, and 1984-2012, as well as return periods relative to observations for 1955-2012. Note that 1955-83 values are calculated from non-significant FARs. Characteristic 1955-83 1955-2012 Return Return Risk period Risk period Frequency 0.78 145.22 1.94 58.00 Intensity 0.73 8.45 1.37 4.46 Characteristic 1984-2012 Control Return Return Risk period Risk period Frequency 2.94 32.94 NA 112.64 Intensity 2.31 2.97 NA 6.12
11. UNDERSTANDING AUSTRALIA'S HOTTEST SEPTEMBER ON RECORD
JULIE M. ARBLASTER, EUN-PA LIM, HARRY H. HENDON, BLAIR C. TREWIN, MATTHEW C. WHEELER, GUO LIU, AND KARL BRAGANZA
Record high September maximum temperatures over Australia arose from a combination of a strongly anomalous atmospheric circulation pattern, background warming, and dry and warm antecedent land-surface conditions.
Introduction. September 2013 was Australia's warmest September since records began in 1910, with anomalous heat across most of the country (Fig. 11.1a). Maximum temperatures, averaged nationally, were 3.32[degrees]C above the 1961-90 average--the highest anomaly for any month on record and almost a full degree ahead of the previous September record set in 1980 (Bureau of Meteorology 2013b). September marked the peak of a record warm period for Australia, which commenced in mid-2012. The most unusual heat began from the last week of August 2013 and continued into the first half of September. Temperatures moderated from 10 September before extreme heat returned to northern and eastern Australia in the final week of the month. Lewis and Karoly ("The role of anthropogenic forcing in the record 2013 Australia-wide annual and spring temperatures" in this report) determine that the attributable risk of such extreme heat in September has increased five-fold due to anthropogenic climate change. Here we take a different attribution approach and use multiple linear regression and experiments with a seasonal forecast system to explain and understand the magnitude of the September 2013 temperatures.
Models to understand the record Australian temperatures. A multiple linear regression model using least squares and assuming normally distributed random errors was built from observed predictors that have historically been used to explain Australia's seasonal climate (e.g., Hendon et al. 2014). These include ENSO, the Indian Ocean dipole (IOD), and the southern annular mode (SAM). Other predictors were global mean temperature, as an indication of the large-scale warming of the climate system, and Australian upper-layer soil moisture from the preceding month, which may be a source of persistence (e.g., Lorenz et al. 2010). All regression calculations were developed using predictors over 1982-2011 using monthly anomalies from the 1982-2011 base period. See the Supplementary Material for details on the predictors and datasets used.
Forecast sensitivity experiments were also performed with the 30-member Predictive Ocean Atmosphere Model for Australia (POAMA) seasonal forecast system (Hudson et al. 2013; see Supplementary Material) to investigate the importance of initial conditions in the ocean, land, and atmosphere for predicting the September record heat. Table 11.1 lists the various sensitivity experiments, all of which were initialized on 21 August 2013, i.e., 10 days prior to the month of interest. Sensitivity experiments consisted of scrambling the atmosphere, land, and ocean initial conditions by sampling the initial conditions for 21 August from the previous 30 years.
Statistically reconstructing September 2013. In calibration, the regression model explains approximately half of the variance in Australian maximum temperatures during 1982-2011 (Supplementary Fig. S11.1; correlation with observed = 0.73). The contribution from each of the predictors for September 2013 is shown in Figs. 11.1b-f. The SAM, which was the most important predictor of this hot event, contributed 20%, reflecting the second-most negative SAM in the 1982-2013 record (standardized anomaly of -1.5). A negative SAM typically results in higher-than-normal maximum temperatures across much of extratropical Australia in the spring season (Hendon et al. 2007). Note that although the SAM has been trending towards its positive phase in austral summer, there is no significant trend in the September SAM time series. Global mean temperature was the second most important predictor, accounting for 15% of the observed anomaly (using the linear trend as the predictor instead gives a similar result). Both the SAM and global temperature contributions produce similar spatial patterns to the observed. Antecedent soil moisture anomalies had a moderate positive contribution in the interior of the continent, though their contribution to the Australian average anomaly was minor. The IOD and ENSO appear to have played little role in the event. SST anomalies in the Indo-Pacific were indicative of a weak La Nina state (slightly cooler than normal in the eastern Pacific and warmer than normal in the west), which usually drives cooler temperatures over Australia. A negative IOD, which was in its decaying state in September, appears to have mitigated some of the warming in the southwest, contributing to slightly cool anomalies observed there.
The reconstructed Australian September 2013 maximum temperature anomalies are shown in Fig. 11.1g (which is simply the sum of the anomalies from the individual predictors in panels b-f). The reconstruction captures the observed pattern of a warm interior and relatively cooler southwest, though at a weaker magnitude than observed (quantified as the residual in Fig. 11.1h). When averaged over Australia, the mean reconstruction reproduces 40% of the observed anomaly (Fig. 11.11). The Australian-average anomaly for 2013 also falls outside the 95% prediction interval (which accounts for the uncertainty in the strength of the historical relationships) of the reconstruction (see Supplementary Fig. S11.1). However, 5% of the years are expected to lie outside this interval, and this appears to be, roughly, what occurred for 1982-2013. The inability to reconstruct the full magnitude of the September 2013 warmth could be due to nonlinear processes not accounted for, or it could indicate that a key predictor, not previously identified to be important for September Australian maximum temperatures, is missing. One such possibility is the MJO, which was strong at the beginning and end of the month (based on the diagnostics of Wheeler and Hendon 2004) and has recently been shown to impact weekly and seasonal Australian temperature in certain phases (Marshall et al. 2013). Preliminary analysis suggests that the MJO contributed to the warm anomaly in the first and last week of September.
Analysis of the SST anomaly for 2013 indicates that there was not much similarity with the pattern historically associated with high maximum temperatures over Australia, other than over the western Pacific where the SST warming trend is strong (compare left panels of Supplementary Figs. S11.2 and S11.3). However, the observed mean sea level pressure (MSLP) anomaly in September 2013 matches well the pattern that is associated with high maximum temperatures over Australia (compare right panels). This pattern has a region of anomalously low pressure immediately to the southwest of the continent and zonally symmetric high pressure farther to the south indicative of a negative SAM. However, while we account for the SAM in our regression model, the strong low pressure to the southwest remains in the pattern of MSLP regressed onto the residual maximum temperature time series (Supplementary Fig. S11.4). The inference is that the occurrence of this strong low pressure anomaly, unrelated to the occurrence of negative SAM or SST forcing, played a prominent role in the record hot September 2013. We further explore the importance of this inference using sensitivity experiments with the POAMA coupled model forecast system.
Dynamical predictions of the record September 2013 Australian temperature. The ensemble mean POAMA forecast for September 2013, initialized on 21 August using observed atmosphere, land, and ocean initial states (Hudson et al. 2013), produced warm anomalies across most of the continent (Fig. 11.2a). Though the ensemble mean anomaly is weaker than observed, the histogram of the 2013 Australian maximum temperature forecasts is systematically shifted towards warmer temperatures compared to the hindcast data (Supplementary Fig. S11.5), with, e.g., the likelihood of exceeding a 1.5 standard deviation warm anomaly increasing by a factor of six in 2013.
Rerunning the forecasts using atmospheric initial conditions randomly picked from the previous 30 years for 21 August, which effectively removes any predictability coming from the atmospheric initial state, reduces the ensemble mean temperature anomaly by up to 50% over the interior (Fig. 11.2c). Scrambling the atmosphere and land initial conditions (Fig. 11.2e) further reduces the temperature anomaly by an even larger amount, with stippling indicating significant differences to the original forecast across most of the continent. This result suggests that the initial land state was a dominant factor in these dynamical predictions and that SST boundary forcing contributed little to the promotion of the record maximum temperatures over Australia. The latter suggestion is confirmed by rerunning the forecasts using observed atmosphere and land initial conditions but scrambled initial ocean states (Fig. 11.2g). In fact, a larger magnitude of predicted warm anomalies is found, suggesting that the ocean had a mitigating impact on temperatures during this extreme event. This is consistent with the weak cooling contribution from tropical SST modes highlighted in the regression analysis.
Note that the low pressure anomaly to the south of Australia is evident in all POAMA experiments (Figs. 11.2b,d,f,h), but with much reduced magnitude compared to observed. The magnitude weakens further when the atmospheric and land initial conditions are scrambled, substantiating the earlier inference of the importance of this mostly unpredictable pressure pattern in the resulting extreme temperatures
Conclusions. Analysis using both multiple linear regression based on historical observational data and sensitivity experiments with a dynamical seasonal prediction system indicates that the record hot September conditions over Australia arose from the apparently random occurrence of strongly negative SAM together with an anomalously deep low pressure cell situated to the southwest of the continent, the background warming trend, and antecedent dry and warm land surface conditions. SSTs appear to have played little role in promoting the record warm anomaly and, based on the evidence presented, actually acted to mitigate the warming over Australia. The results from the regression model indicate up to 15% of the record temperature anomaly can be explained by the global temperature changes over the 1982-2013 period analyzed. This warming trend is expressed in the seasonal forecast experiments through the trend in ocean and land initial conditions, the latter of which appeared to be the dominant contributor to the September 2013 predicted anomaly over Australia. The sensitivity to the initial land conditions in the POAMA forecasts is thus consistent with the inference from the regression model concerning the substantial role of the upward trend in global temperatures. To the extent that global temperature changes have been attributed to anthropogenic climate change (Bindoff et al. 2014), a multi-step attribution process suggests that anthropogenic climate change played an important role in the record Australian maximum temperatures in September 2013.
Table 11.1. Coupled model seasonal forecast experiments described in the text and their initial conditions for atmosphere (ATM), land surface (LAND), and ocean (OCEAN). See the Supplementary Material for more details on the experimental design. Color shading indicates whether initial conditions for the 30 members are as observed (blue) or scrambled (red). Operational seasonal forecast Scrambled ATM experiment experiment ATM initial Observed for Selected from 21 conditions 21 August 2013 August 1981-2010 LAND initial Observed for Observed for 21 conditions 21 August 2013 August 2013 OCEAN initial Observed for Observed for 21 conditions 21 August 2013 August 2013 Scrambled ATM and LAND Scrambled OCEAN experiment experiment ATM initial Selected from 21 Observed for 21 conditions August 1981-2010 August 2013 LAND initial Selected from 21 Observed for 21 conditions August 1981-2010 August 2013 OCEAN initial Observed for 21 Selected from 21 conditions August 2013 August 1981-2010
12. CLIMATE CHANGE TURNS AUSTRALIA'S 2013 BIG DRY INTO A YEAR OF RECORD-BREAKING HEAT
ANDREW D. KING, DAVID J. KAROLY, MARKUS G. DONAT, AND LISA V. ALEXANDER
The record heat of 2013 across inland eastern Australia was caused by a combination of anthropogenic warming and extreme drought.
Introduction. During 2013, Australia experienced its hottest year on record (23[degrees]C on average, 0.17[degrees]C above the previous 2005 record) as well as a series of extreme heat wave events (see also "The role of anthropogenic forcing in the record 2013 Australia-wide annual and spring temperatures", "Multimodel assessment of extreme annual-mean warm anomalies during 2013 over regions of Australia and the western tropical Pacific", and "Increased risk of the hot Australian summer of 2012/13 due to anthropogenic activity as measured by heat wave frequency and intensity" in this report). Besides being the hottest year in a record dating back to 1910, a drought set in across much of the east of the country leading the federal government to announce an AUD320 million (~USD 300 million) drought assistance package for affected farmers. The severe lack of water in the region came after the exceptionally wet 2010-12 period, which brought devastating floods to Queensland and New South Wales in particular. Across almost the entirety of Australia, maximum temperatures were warmer than average in 2013 (Fig. 12.1a), and for much of the continent, it was also considerably drier than average (Fig. 12.1b). The area of greatest rainfall deficit, covering inland eastern Australia, coincided with the region where the heat anomalies were strongest. Focusing on the region of strongest 2013 maximum temperature and rainfall anomalies, an inverse relationship between maximum temperature and precipitation is found with 2013 being the hottest year and one of the driest (Fig. 12.1c). Similar temperature-rainfall relationships have previously been examined for the Murray-Darling Basin (Nicholls 2004) and multiple other regions of Australia (Karoly and Braganza 2005). A shift in the relationship between annual average daily maximum temperature and annual precipitation is observed through time (Fig. 12.1c). In the late 20th century and early 21st century, the temperature-rainfall relationship shifted towards warmer temperatures (an increase of roughly 1[degrees]C) associated with the same rainfall anomalies compared to the mid-20th century. This finding is insensitive to the choice of periods analyzed.
Given the relationship between heat and drought, this study examines first whether the risk of hot and dry years has increased due to human-induced climate change. Secondly, the role of the lack of precipitation in the 2013 record-breaking heat is analyzed and discussed.
Data and methods. Observations were obtained from the Australian Water Availability Project (AWAP; Jones et al. 2009) dataset interpolated onto a regular 0.5[degrees] grid. These were used to calculate annual anomalies of daily maximum temperature (Fig. 12.1a) and rainfall (Fig. 12.1b) relative to the 1971-2000 climatology, which encompasses both historical dry and wet periods. Anomalies were averaged over our investigation area (18[degrees]-30[degrees]S, 133[degrees]-147[degrees]E), which experienced the greatest anomalies in annual average daily maximum temperature and precipitation and covers an important area for agriculture in Australia. The temperature and rainfall anomalies were calculated at individual gridboxes relative to the gridbox mean values. The relationship between annual average daily maximum temperature and rainfall values was studied for the 1930-2013 period.
To examine possible human-induced contributions to the observed heat and drought of the 2013 calendar year, model simulations forced with anthropogenic greenhouse gases were required. Annual average daily maximum temperature and rainfall data derived from the output of single historical (1861-2005) and single RCP4.5 emissions scenario (2006-33) runs from 35 state-of-the-art Coupled Model Intercomparison Project Phase 5 (CMIP5) models were analyzed (see Supplementary Table S12.1; Taylor et al. 2012). The CMIP5 model outputs were regridded onto a common 2.5[degrees] grid. Relationships between maximum temperature and rainfall were examined for a common 1861-2033 period. Combining these model simulations, probability distribution functions (PDFs) of annual average daily maximum temperature and rainfall were compared for two 41-year periods (1861-1901 and 1993-2033), with the latter period centered on 2013, to examine for changes in the risk of very hot or very dry years related to anthropogenic activity. The earlier 41-year period represents a climate with a much smaller influence from anthropogenic factors compared to now. The risk of very hot years occurring with respect to drought conditions was analyzed by considering PDFs of temperature anomalies in wet and dry years separately (where wet and dry years are defined as 33% above and below average annual rainfall respectively for the 1971-2000 period). This is based on the assumption that rainfall and moisture availability are driving temperature variations; although, undoubtedly, there is also a feedback on precipitation from temperature. Comparisons between PDFs were made by calculating the Kolmogorov-Smirnov (KS) test statistic measuring the similarity of the PDFs. Fractional attributable risk (FAR; Allen 2003) statistics may be used to examine whether the likelihood of extreme climate events has changed in relation to a particular aspect of interannual climate variability or anthropogenic warming. FAR statistics were calculated to measure change in likelihood of:
(a) very dry years between 1861-1901 and 1993-2033 (i.e., has anthropogenic activity changed the risk of meteorological drought?),
(b) very hot years between 1861-1901 and 1993-2033 (i.e., has anthropogenic activity changed the risk of extreme hot years?),
(c) the combination of very hot and very dry years between 1861-1901 and 1993-2033 (i.e., has anthropogenic activity changed the risk of very hot dry years?), and
(d) very hot years contrasting between wet and dry years (i.e., do drought conditions change the risk of extreme hot years?).
FAR statistics were calculated based on the 2013 value of rainfall (61% of average; the fifth lowest annual rainfall in the series) and the 2002 value of annual averaged daily maximum temperature (1.52[degrees]C above average; the second hottest value). Using the 2002 temperature threshold allows us to examine whether the risk of a hotter year than 2002 (such as 2013) has changed. Since 2013 was not the driest year in the record, we use the 2013 threshold in our FAR calculations related to precipitation. The PDFs used to calculate the FAR statistics were bootstrapped 1000 times using subsamples of 50% of the models in each case. This allows a range of FAR statistics to be calculated, so uncertainty in the FAR can be assessed. The FAR statistics quoted represent the fifth percentile of the 1000 ranked FAR values calculated through bootstrapping (i.e., 95% of FAR values are above the quoted value; we may therefore write that the FAR and risk are extremely likely to be greater than these stated values).
Note that we also performed this analysis for the shorter January-March and April-December periods to examine if there was a seasonal bias in our calendar year calculations. Generally, results were very similar, although slightly weaker, when considering the shorter periods.
Results. The relationships between annual average daily maximum temperature and annual rainfall were examined in each CMIP5 model separately. All models exhibit strong, statistically significant inverse relationships between maximum temperature and rainfall over the 1861-2033 period. Additionally, there is a shift in this relationship in the models towards warmer temperatures, similar to that which is observed. Later on in the period, the same rainfall totals are associated with warmer temperatures in comparison to the earlier period. The models display a variety of different strengths in the temperature-rainfall relationship; however, the inverse nature of the relationship is captured by every model. The shifting relationship from 1861-1901 to 1993-2033 was also evident when all the model data are plotted together (Fig. 12.1d).
Has anthropogenic activity changed the risk of drought? Using these model data, the change in precipitation between the early and late 41-year periods was investigated. Using a KS test, the PDFs of rainfall in the 1861-1901 and 1993-2033 periods are statistically indistinguishable at the 5% level (Supplementary Fig. S12.1). The risk of a drought worse than the 2013 event is also not significantly different between the two periods (the range of FAR values encompasses zero). Therefore, there has been no significant change in meteorological droughts in this region related to anthropogenic climate change, as simulated by the CMIP5 models. However, in a warming climate, with increasing evaporation and reduced soil moisture, droughts may become more severe (e.g., Seneviratne et al. 2010).
Has anthropogenic activity changed the risk of extreme hot years? The change in average daily maximum temperatures was also examined for the same early and late 41-year periods. The PDFs of maximum temperature for 1861-1901 and 1993-2033 are significantly different (Fig. 12.2a). The FAR value, based on the 2002 threshold, is extremely likely to be above 0.96. Thus, the risk of maximum temperatures above the 2002 threshold is extremely likely to be 23 times greater now than in the late 19th century. Large increases in the likelihood of extremely hot seasons and years across Australia related to human-induced climate change have been documented previously (Lewis and Karoly 2013).
Has anthropogenic activity changed the combined risk of hot years and dry years? A more interesting question is to assess whether the risk of extreme heat and drought in combination has increased due to anthropogenic climate change. Therefore, a bivariate FAR analysis was conducted using the same precipitation and maximum temperature thresholds as for the previous univariate FAR analyses. We calculate that the bivariate FAR is extremely likely to be above 0.86. Therefore, between the 1861-1901 and 1993-2033 periods, the risk of extreme heat and drought in combination is extremely likely to have increased by at least seven-fold.
Do drought conditions change the risk of hot years? The difference in the risk of extreme heat related to whether a year is particularly wet or dry was analyzed. PDFs of annual average daily maximum temperatures were compiled for wet and dry years separately (Fig. 12.2b). These PDFs are also significantly different. The FAR value is extremely likely to be greater than 0.96 and the risk of extreme heat is extremely likely to be 25 times greater in dry years than in wet years. Thus, the heat of 2013 in this region of Australia was strongly related to the lack of rainfall.
Conclusions. In 2013, much of Australia experienced extreme heat and drought. Using state-of-the-art climate models, this study examines the role of climate change in the heat and drought as well as the relationship between heat and lack of rainfall. We show that the extreme heat was made much more likely by important contributions from both the anthropogenic warming of the climate and the very dry conditions over the inland eastern region of the continent. The combination of these factors increased the probability of 2013 being Australia's hottest year on record.
13. THE ROLE OF ANTHROPOGENIC CLIMATE CHANGE IN THE 2013 DROUGHT OVER NORTH ISLAND, NEW ZEALAND
LUKE HARRINGTON, SUZANNE ROSIER, SAM M. DEAN, STEPHEN STUART, AND ALICE SCAHILL
For the 2013 New Zealand drought, evidence from a number of models suggests that the meteorological drivers were more favorable for drought as a result of anthropogenic climate change.
Introduction. In the latter part of the 2012/13 austral summer season (January-March), the North Island of New Zealand endured its most severe drought in 41 years of widespread measurements of potential evapotranspiration deficit (Porteous and Mullan 2013). For the 2013 drought, 34.2% of the North Island land surface experienced its highest recorded cumulative deficits (Supplementary Fig. S13.1), significantly greater than the 14.3% recorded for the previously severest drought (1997/98). The New Zealand Treasury (2013) estimates reduced agricultural production, attributed to the drought, cost the national economy at least US$1.3 billion, with continued impacts expected for another two years (Blackham 2013).
Droughts are complex hydrologic phenomena subject to influence by numerous climatological factors, including temperature, wind speed, atmospheric humidity, and precipitation rates (Sheffield et al. 2012; Trenberth et al. 2014). The persistent dryness of the 2013 New Zealand drought has been suggested to be primarily a result of slow-moving ("blocking") anticyclones over the sector, unrelated to the El Nino-Southern Oscillation (Blackham 2013), with the increase in atmospheric subsidence suppressing precipitation. If true, the meteorological drivers of this drought might, therefore, be characterized using mean sea level pressure (MSLP) and a measure of the absence of precipitation. Using such metrics, the North Island drought was associated with an average February MSLP of 1020 hPa (90th percentile) and a record total number of dry days of 78.2 for January to March.
Was this event influenced by climate change? Previous studies concerning the attribution of individual drought events to (anthropogenic) climate change have primarily focused on precipitation departures (Rupp et al. 2013; Trigo et al. 2013) and prolonged temperature extremes (Rupp et al. 2012; Hoerling et al. 2013). For a maritime, midlatitude climate like New Zealand's, temperature is not reflective of synoptic-scale drying and, thus, does not perform well as an indicator of drought (Clark et al. 2011; Seneviratne 2012). Furthermore, analysis of precipitation rates considers atmospheric processes that operate on small spatial scales and can be poorly constrained in climate models (Trenberth 2011), thereby also failing to capture the synoptic-scale drivers of drought (Sherwood and Fu 2014).
To consider the drought in New Zealand we limit ourselves to the North Island, where extreme drought was ubiquitous, and consider only the distributions for monthly MSLP and dry days per month produced by models that participated in the Coupled Model Intercomparison Project phase 5 (CMIP5; Taylor et al. 2012). Those models for which historical simulations compare well against observations are then quantitatively compared to simulations, which excluded the impact of anthropogenic changes such as greenhouse gases and ozone depletion.
Defining circulation and dryness indices. The North Island pressure index (NIPI) is defined as the average of the monthly MSLP observed at four locations across the North Island, corresponding to weather stations at Auckland, Gisborne, New Plymouth, and Wellington (Supplementary Fig. S13.1). Because the gradients associated with high pressure systems are weak, these stations are spatially representative of the island, and their records are also of sufficient length, with the shortest starting in 1911. Analysis of the NIPI can be used to evaluate systematic changes to atmospheric circulation over the North Island.
In addition to the NIPI, changes in the intensity and frequency of broad-scale subsidence over the North Island sector are evaluated with dry day analysis, defined as less than one millimeter of rain accumulation over a 24-hour period (WMO 2010). Gridded rainfall measurements of daily resolution were available from 1960 to 2013, with individual data points assimilated across the North Island using a spline interpolation technique (Tait et al. 2006).
Can global climate models simulate NZ drought? To determine which global climate models from the CMIP5 archive are reasonable at replicating the two New Zealand drought indices, a validation methodology was applied. Only data from models that contributed to both the historical and natural-forcings only experiments and provided the variables needed for the indices were considered. Gridded data from each model were linearly interpolated onto a 1[degrees] x 1[degrees] ERA-Interim grid (Dee et al. 2011). If the median of the observed data lay outside the interquartile range of a simulation, that model was excluded from subsequent analysis.
In Fig. 13.1a, the NIPI is compared to observations for 17 different global climate models between 1952 and 2005. Figure 13.1b compares the statistical distribution for the maximum three-month accumulation of dry days (3MDD) in each extended austral summer (November-April) for 15 climate models between 1952 and 2005. This time span of 54 years is determined by the availability of daily gridded rainfall observations from 1960 to 2013 only. In total, 13 models were deemed appropriate for pressure analysis, while five models were suitable for analysis of extreme dry day totals. It has been verified that the selection of models for 3MDD is unaffected by restricting the comparison to the 45 years of actual overlap from 1960 to 2005 (Supplementary Fig. S13.2).
Is there evidence from global climate models that the North Island drought was influenced by anthropogenic climate change? Figure 13.2a shows the shift in NIPI for 13 CMIP5 members (1952-2005), comparing model simulations with all anthropogenic forcings (ALL) to each corresponding run with natural forcings only (NAT). At the observed value for the peak of the 2013 drought, 9 of the 13 models exhibit a shift towards higher NIPI values when anthropogenic emissions are included. The average human-induced shift at the 2013 observation is an increase of 0.39 hPa (significantly different from zero at the 95% confidence level of an applied Student's t-test), with a range of -0.24 hPa to 1.09 hPa. Most models in Fig. 13.2a with an NIPI change above zero remain at either a stable or increasing value towards the high-pressure distribution tail.
Figure 13.2b similarly shows the absolute change in the 3MDD summer maxima for model simulations spanning 1952-2005, again comparing the ALL forcing runs to corresponding NAT simulations. Here there are 14 simulations from five different models. The observed 3MDD total for January-March 2013 over the North Island was high enough that no simulation emulated the event over the 54-year period. This may be due in part to a persistent low bias in the dry day distribution common to all models. Regardless, models do demonstrate an anthropogenic-induced shift towards an increased frequency of three-month periods with extreme dry day totals. For example, taking the 90th percentile of observed 3MDD summer maxima as an arbitrary threshold for a drought event, there were 31 simulated extreme events for ALL ensemble members compared to only 18 for the NAT--a 72% increase. At this threshold, there is an average increase in the number of dry days over a three month period of 1.6 (significantly different from zero at the 95% confidence level).
Discussion. Our results are consistent with the broad-scale dynamical changes to the New Zealand sector expected with anthropogenic climate change. A clear poleward shift of the Southern Hemisphere jet stream has been observed for December-February in reanalyses (Swart and Fyfe 2012), with an associated increase in surface pressure over New Zealand (Thompson et al. 2011). This is attributed largely to a trend in the southern annular mode (SAM) driven by ozone depletion (Thompson et al. 2011). For the three months of January to March 2013, the station-based Marshall Index for the SAM was strongly positive, averaging 1.95. The February value of the SAM index was 2.84, the highest ever for that month. Given that the NIPI index was only at the 90th percentile for this month and lower for the other two, it seems that high pressure induced by the SAM over New Zealand is not a complete explanation for the extreme dryness that occurred. There has also been an observed trend towards less precipitation over some midlatitude regions, attributed to the combined effects of anthropogenic greenhouse gases and ozone depletion (Fyfe et al. 2012). It is plausible that these trends in circulation and precipitation conditions have made severe drought more likely.
For the models used in our dry days analysis, we find a strong (R = 0.95) positive correlation between the mean shift in NIPI and the corresponding shift in mean number of monthly dry days (Supplementary Fig. S13.3). This indicates that a change in the number of dry days in the models is related to higher pressure over the North Island, with increased subsidence suppressing rainfall. Thus, the models suggest that the anthropogenic contribution to the 2013 drought was mostly through changes in circulation. Since it is also clear that the causes of the 2013 North Island drought were not simply higher pressures, natural internal variability must have played a role in the extremeness of this event.
It is also important to emphasize that quantifying properties of synoptic-scale circulation change over the North Island sector provides only partial understanding of the physical mechanisms contributing to drought formation there; spatial heterogeneity in soil type and topography represent complicating factors (Dai 2011; Seneviratne 2012). There are inherent limitations with only considering 54 years of observations and a limited number of simulations. Future work using a much larger ensemble of simulations, such as the weather@home project (Massey et al. 2014), remains critical for more robust attribution statements about changing drought risk.
Conclusions. Results show the monthly pressure anomalies associated with the 2013 drought were higher (0.4 hPa on average) as a result of anthropogenic climate change. No model was able to capture the extremity of the three-month dry day index, likely a combination of bias in the models and the severity of the event relative to the size of the model ensemble. However, using the observed 90th percentile of the dry day index as an (arbitrary) threshold for drought, a 72% increase in the number of events was found, which does suggest a change in drought risk. The human-induced shift in monthly dry days and MSLP were found to be very well correlated (R = 0.95), underlining that, in these models, the human-induced contribution to drought over New Zealand occurs through increased high pressure. Since the 2013 drought was not associated with extreme high pressures, natural internal variability in the climate system must have played a role in the underlying causes of its severity.
14. ASSESSING HUMAN CONTRIBUTION TO THE SUMMER 2013 KOREAN HEAT WAVE
SEUNG-KI MIN, YEON-HEE KIM, MAENG-KI KIM, AND CHANGYONG PARK
A comparison of observations and multiple global climate model simulations indicates that extreme hot summer temperatures in Korea have become 10 times more likely due to human influence.
Introduction. East Asia, including Korea, experienced one of its hottest summers in 2013, resulting in severe damage to society and the ecosystems across the region. During summer (June-August, JJA) 2013, South Korea had its hottest summer nights and second hottest summer days since 1954 (Fig. 14.1). The JJA average daily minimum temperature (Tmin) was 22.7[degrees]C (2.2[degrees]C warmer than the 1971-2000 climatology), and the average daily maximum temperature (Tmax) reached 29.9[degrees]C (1.9[degrees]C warmer than the climatology). New high JJA temperature records have exerted adverse impacts on the country's economy, health, and infrastructure. In particular, the heat wave increased electricity consumption beyond the forecast level and forced the government to issue several power shortage warnings. In addition to these 2013 extreme events, summer temperatures have been consistently increasing during the past 60 years with statistically significant trends in Tmin and daily mean temperature (Tmean; Fig. 14.1c).
Here, we assess the Korean heat wave in the context of global warming using recent climate model datasets available from the Coupled Model Intercomparison Project Phase 5 (CMIP5; Taylor et al. 2012) experiments. Because global climate models (GCMs) like CMIP5 models generally have relatively low horizontal resolutions (typically about 100-200 km), they cannot reproduce the spatiotemporal details of local climate variability and change. Therefore, it would be inappropriate to directly compare raw GCM outputs with station-based observations without utilizing proper downscaling processes. Alternatively, large-scale patterns, very strongly associated with local changes, can be identified, and observed changes in those large-scale patterns can then be compared with model simulations. Despite unavoidable subjectivity in selecting the number of large-scale patterns, variables, and spatial domains, this up-scaling approach can enable causes of local observed changes to be assessed in view of large-scale modeled responses to different external forcing factors, such as greenhouse gas increases. Employing methods similar to multistep attribution approaches (Bindoff et al. 2014), this study analyzes the hot 2013 summer observed in Korea in the context of greenhouse warming.
Data and methods. We use Tmin, Tmax, and Tmean observations from 12 Korean weather stations for 1954-2013 (from 59 stations for 1973-2013). Monthly SST data from the Extended Reconstruction Sea Surface Temperature (ERSST) version 3 (Smith et al. 2008) dataset are used to find a large-scale SST indicator of Korean summer temperatures. The analysis domain is 15[degrees]-50[degrees]N and 100[degrees]-160[degrees]E, where SST exhibits the strongest association with the Korean temperature ([r.sup.2] > 70%). For model simulations, we use multimodel datasets available from CMIP5 experiments (see Supplementary Table S14.1) to assess human influence on the observed changes in the large-scale SST pattern. We use the "historical" experiment integrated with natural (due to changes in solar and volcanic activities) and anthropogenic forcings (due mainly to increases in greenhouse gases and aerosols). We divide it into two 60-year periods, 1860-1919 (ALL_P0) and 1954-2013 (ALL_P1), representing climate conditions without and with human influence respectively. We also use datasets from the "historicalGHG" (greenhouse gas only forcing) and "historicalNat" (natural forcing only) experiments for 1953-2012 to examine the relative contribution of individual forcings (see the Supplementary Material for more details).
To identify a large-scale SST indicator for Korean heat waves in general, we first look for a SST spatial pattern related to the Korean hot summers by regressing SST onto Korean JJA Tmin using observations from 1954-2013. Here we use Tmin, which generally better captures Korean heat waves with stronger and more spatially consistent increases (Fig. 14.1; see also Supplementary Fig. S14.1). We then project the observed SST regression field onto the observed JJA SST and also the CMIP5 simulations (referred to as "observed projection" and "modeled projection" respectively) for the entire analysis period. The time series of projections represent temporal variations of SST spatial patterns associated with the Korean heat wave (see the Supplementary Material for further details).
By comparing the observed projection with the modeled projections, an attribution analysis of the Korean heat waves can be carried out. To estimate anthropogenic influence in a quantitative manner, we employ the fraction of attributable risk (FAR; Stott et al. 2004) approach in which the probability of extreme events occurring is compared in two hypothetical worlds, without and with human influences. Here, FAR is calculated as FAR = 1-([P.sub.N]/[P.sub.A]). [P.sub.N] represents the probability that extremes will occur exceeding the observed strength (trends in SST projection or 2013 projection value) in natural unforced conditions (ALL_P0 or NAT_P1), and PA represents the same probability estimated in anthropogenic forced conditions (ALL_P1 or GHG_P1). If FAR is 0.5 and 0.67, for example, it means doubled and tripled risk of extreme events due to human influences respectively.
Results. The SST pattern associated with heat waves in Korea is anomalously warm over northern East Asia, north of 30[degrees]N, with a maximum over the East Sea (Sea of Japan; Fig. 14.2a). The time series of the observed projection is shown in Fig. 14.2b. The correlation coefficient between Korean Tmin and the SST projection is 0.79, indicating that this SST projection explains most fluctuations in Korean summer temperature. Indeed, the observed projection has a record high in 2013 with a value of 82.3 ([degrees][C.sup.2] [degrees][C.sup.-1]). Modeled projection results are illustrated in the same plot for ALL_P0 (blue), ALL_P1 (green), GHG_P1 (red), and NAT_P1 (purple). ALL_P1 reproduces the observed change with very similar trends (Fig. 14.2c). The amplitude of interannual variability is also well simulated by models (The standard deviation of observed de-trended projection series is 36.5. The multimodel mean of standard deviations of the de-trended ALL_P1 series is 39.2 with a 5th-95th percentile range of 27.9-53.0). Trends in ALL_P0 and NAT_P1 are very weak while GHG_P1 possesses stronger positive trends. This implies a dominant contribution of greenhouse warming to the ALL_P1 trend given overall long-term cooling due to other anthropogenic forcings (see Supplementary Material).
The histogram of trends in observed and modeled projections is compared in Fig. 14.2c. The distributions of ALL_P0 and NAT_P1 trends are centered near zero whereas all runs of GHG_P1 have positive trends, consistent with the time series. The observed trend is positioned near the center of the ALL_P1 distribution, indicating that the observed increasing trend in SST projection, which represents long-term warming of the ocean surface over northern East Asia and the associated intensification of the Korean heat wave, is attributable to natural and anthropogenic forcing. In the FAR analysis (Table 14.1), the trends in SST projection exceeding the observed trend (0.88 [degrees]C2 [degrees][C.sup.-1] [year.sup.-1]) are very rarely simulated in ALL_P0 (0.98% probability) and NAT_P1 (0%). However, they are simulated much more frequently in the other model experiments that include anthropogenic forcing (63.8% and 100% probability in ALL_P1 and GHG_P1, respectively). The corresponding FAR values of ALL_P1 and GHG_P1 relative to ALL_P0 (or NAT_P1) are very close to 1, confirming the qualitative comparisons.
Attribution analysis of the 2013 summer event can be done in a similar way. The distributions of all modeled projection values are compared in Fig. 14.2d together with the observed 2013 value. SST projections stronger than the observed 2013 event are extremely rare in ALL_P0 (0.54%) and NAT_P1 (0.88%). The chance of 2013-like extreme events increases to 5.28% and 51.05% in ALL_P1 and GHG_P1, respectively. The corresponding FAR values with respect to ALL_P0 are 0.90 and 0.99. Considering large variability on interannual time scales, this provides very strong evidence that the risk of extremely hot summers in Korea associated with SST warming is 10 times more likely due to human influence.
Conclusions. This study assesses the possible impacts of greenhouse gas increases on the observed long-term increases in Korean summer temperatures as well as on the 2013 extreme event. A large-scale SST indicator of the Korean summer temperature is identified from observations and then the observed occurrences of the SST patterns are compared with those from CMIP5 model simulations representing climate conditions with and without human influences. We find that a strong long-term increasing trend in the observed SST near northern East Asia during the past 60 years cannot be explained without the inclusion of recent human-induced greenhouse gas forcing. This is because other external forcings, including solar and volcanic activities (natural) and aerosols (anthropogenic), are likely to induce cooling during the latter 20th century (see Supplementary Material). Further, it is implied that extreme hot summers like the 2013 event have become 10 times more probable due to human activities. Our results indicate greenhouse warming contributes significantly to recent warming trends in Korean summer temperature. Although statistical analysis presents strong evidence for human influence, further studies are needed to better understand the physical mechanisms accounting for the large-scale SST indicator. A simple correlation analysis suggests that there may be a weak but statistically significant influence of the Pacific decadal oscillation on Korean summer temperature (r = -0.37; Kim et al. 2008) while weaker connections exist with the Pacific-Japan pattern (r = 0.26; Wakabayashi and Kawamura 2004) and the western North Pacific subtropical high (r = -0.11; Wang et al. 2013).
Table 14.1. Probability of occurrence exceeding the observed trend in SST projection and the observed 2013 projection value. The fraction of attributable risk is calculated as [FAR.sub.ALL_PI] I - [P.sub.ALL_P0]/[P.sub.ALL_PI] and [FAR.sub.GHG_PI] = I - [P.sub.ALL_P0]/[P.sub.GHG_PI]. Obs ALL_P0 Trend in 0.88 ([degrees][C.sup.2] 0.98% projection [degrees][C.sup.-1] (60 years) [yr.sup.-1]) 2013 summer 82.27 ([degrees][C.sup.2] 0.54% in SST [degrees][C.sup.-1]) projection ALL_PI GHG_PI Trend in 63.8% 100% projection FAR = 1 FAR = 1 (60 years) 2013 summer 5.28% 51.05% in SST FAR = 0.90 FAR = 0.99 projection
15. THE CONTRIBUTION OF ANTHROPOGENIC FORCING TO THE JAPANESE HEAT WAVES OF 2013
YUKIKO IMADA, HIDEO SHIOGAMA, MASAHIRO WATANABE, MASATO MORI, MASAYOSHI ISHII, AND MASAHIDE KIMOTO
Anthropogenic climate change played a significant role in increasing the probability of events such as the heat wave in Japan in 2013.
Introduction. During the boreal summer of 2013, Japan experienced extraordinarily high temperatures: record-breaking daily maximum temperatures at 143 sites in the west of the country. Daily mean surface air temperature (SAT) was 1.2[degrees]C warmer than normal on average in July and August in western Japan, which was above the 90th percentile for the reference period of 1979-2012.
This heat wave was characterized by the intensification of both the Pacific high and the Tibetan high. Figure 15.1 shows observed anomalies of surface temperature, circulation, and convective activity from July to August 2013. Active convective heating, as indicated by negative anomalies in outgoing longwave radiation, was observed in the Maritime Continent and Southeast Asia (shaded areas in Fig. 15.1b) associated with La Nina-like conditions during this season (Fig. 15.1a). In the upper levels, a divergence flow from the active convective regions converged and descended north and east of the Philippines (arrows in Fig. 15.1b), which intensified the Pacific high (black contours in Fig. 15.1b). Furthermore, the active Asian monsoon resulted in the intensification of the Tibetan high (gray contours in Fig. 15.1b), which extended eastward in the upper troposphere. This double structure of high pressure systems brought a warmer SAT and SST around Japan (Fig. 15.1a).
Although most extreme heat waves might have occurred as part of stochastic atmospheric fluctuations, anthropogenic global warming has the potential to impact the probability of their occurrence (Allen 2003; Stott et al. 2004). In this study, we generated a large ensemble using an atmospheric general circulation model (AGCM) under two specific boundary conditions for July and August 2013. We prepared two scenarios for the boundary conditions: observed conditions for a real-world experiment and those without human-induced long-term changes for an idealized counterfactual experiment. This methodology was first proposed by Pall et al. (2011). We evaluated the probabilistic difference in event occurrence rates due to the contribution of anthropogenic effect through analysis of the sets of AGCM ensemble experiments.
Method. We conducted a 100-member AGCM ensemble experiment (factual run, called the ALL-run) using an atmospheric component model of the Model for Interdisciplinary Research on Climate Version 5 (MIROC5, T85L40; Watanabe et al. 2010). The model was integrated with observed SST and sea ice under the anthropogenic external conditions during the period of the 2013 heat waves. Another 100-member ensemble was constructed under anthropogenic forcing fixed at conditions of the year 1850, with modified SST and sea ice where possible human-induced components were removed (counterfactual run under natural external conditions, called the NAT-run; see Shiogama et al. 2013 for details). We removed two possible patterns of human-induced changes in SST and sea ice: a linear trend based on Hadley Centre sea ice and SST version 1 (HadISST; Rayner et al. 2003) from 1870 to 2012 (NATI; Christidis and Stott 2014) and changes detected from the multimodel mean of Coupled Model Intercomparison Project phase 5 (CMIP5) historical experiments (NAT2; Daithi 2013). To validate the model's reproducibility of interannual variability, a long-term version of the ALL-run (1949-2011) was replicated with a reduced ensemble size of 10 (ALL-LNG run).
We defined a Japanese SAT index (SATJ index) as SAT averaged from July to August over the land area of western Japan (130[degrees]-140[degrees]E, 30[degrees]-37[degrees]N) following the model geography. An ensemble-mean SATJ time series simulated by the ALL-LNG run reproduces the observed interannual variability with a correlation coefficient of 0.79 from 1979 to 2011, although the model underestimates its variance. In the following analyses of a probability density function (PDF), a Gaussian distribution is assumed and simulated variances in the ALL-run and NAT-runs are corrected with a ratio of variances between the observations (JRA25 reanalysis; Onogi et al. 2007) and the ALL-LNG run.
Results. A PDF for the SATJ index from the ALL-run is shown in Fig. 15.2a (red curve). The ensemble mean of the ALL-run indicated warmer-than-normal conditions in 2013, mainly due to the higher SST around Japan that was given to the model as a boundary condition. Of importance is the fact that the ensemble members are well spread out, covering an observed anomaly of 1.2[degrees]C (triangle in Fig. 15.2a). Several heat waves that appeared in the ALL-run (one example of such an extreme case is shown in Supplementary Fig. S15.1) represent the double structure of the intensified Pacific high and Tibetan high associated with the active convection in the western tropical Pacific and in the Asia monsoon regions, akin to observed anomalies.
When the possible anthropogenic components are removed from the boundary conditions, the PDF for the SATJ index shifts toward a normal condition (blue curves in Fig. 15.2a). The difference between the ALL-run and either NATI or NAT2 signifies the anthropogenic impact on the occurrence rate of the 2013 Japanese heat wave; that is, the anthropogenic effect tends to increase the frequency of extreme warm events. The occurrence rate of extreme warm events that exceed the observed SATJ in 2013 is 12.4% for the ALL-run, 1.73% for the NAT1-run, and 0.50% for the NAT2-run. The estimated percentiles may have some sensitivity to assumed boundary conditions for the NAT-runs, but the difference between the ALL-run and NAT-runs suggests that human activity largely contributed to the increasing probability of heat waves in Japan in 2013.
The PDFs in Fig. 15.2a show an obvious shift of their ensemble mean between the ALL-run and the NAT-runs. The human-induced effect is represented by the difference in their ensemble mean fields between the ALL-run and each of the NAT-runs. Figure 15.2b shows the differences in surface temperature, 850-hPa stream function, and 200-hPa divergence flow between the ALL-run and NAT1-runs. The figure shows a marked warming around Japan (shading in Fig. 15.2b), which is the major cause of the positive shift of the PDF from the NAT1-run to the ALL-run. This enhanced warming around Japan is also detected in the difference from the NAT2-run (Supplementary Fig. S15.2). L. Wu et al. (2012) also found a faster warming rate over the path of global subtropical western boundary currents, including areas surrounding Japan, by analyzing century-long reconstructed SST and reanalysis products. They concluded that the poleward shift of midlatitude westerlies has a role in shifting those currents that contributed to the increase in SST.
In addition to the increase in SST, enhanced upper-level divergent flow (lower-level convergence) in the central to western equatorial Pacific is visible where surface warming is evident (arrows and shading in Fig. 15.2b). The intensified upper-level divergent flow makes subsidence to the south of Japan and enhances the Pacific high (contours in Fig. 15.2b), which might also have the potential to increase the probability of high temperatures in Japan. On the other hand, there is little change in the Tibettan high (not shown). Deser et al. (2010) reported, however, that there is uncertainty in observed SST warming patterns due to the change in SST measurement technique and to different analysis procedures. They suggested that reconstructed SST datasets, including HadISST, might fail to reproduce the 20th century trends particularly over the equatorial Pacific. Trends in atmospheric responses in AGCMs, thus, may also be subject to the uncertainty in the observed SST warming patterns. Moreover, Christidis and Stott (2014) pointed to a lack of consensus among the CMIP5 models on the magnitude and spatial patterns of the anthropogenic change in SST. There is, hence, a continuing need for longer observations and for reducing model biases and uncertainty.
Conclusions. The 2013 heat wave in Japan was mainly caused by probabilistic atmospheric natural variability, but anthropogenic climate change played a significant role in raising the chance of the heat wave occurring. We emphasize that the increase of heat wave probability in East Asia is not simply a result of an increase in surrounding SST, but it is also potentially affected by atmospheric circulation changes forced remotely by western tropical Pacific warming.
16. UNDERSTANDING A HOT SUMMER IN CENTRAL EASTERN CHINA: SUMMER 2013 IN CONTEXT OF MULTIMODEL TREND ANALYSIS
TIANJUN ZHOU, SHUANGMEI MA, AND LIWEI ZOU
July-August 2013 was the warmest such period in central eastern China since 1951. Comparison based on Coupled Model Intercomparison Project Phase 5 (CMIP5) models suggest a discernible impact of anthropogenic forcing, with internal variability also being important.
The 2013 annual mean temperature in China was the fourth highest since 1961. It was 0.6[degrees]C above normal and 0.8[degrees]C higher than 2012 (CMA 2014). Using the observational data of 756 stations provided by China Meteorological Administration (CMA), we examine both the climate anomalies and the extremes (Fig. 16.1). The strongest heat wave since 1951 occurred in central eastern China (~24[degrees]-33[degrees]N, 102.5[degrees]-122.5[degrees]E), where the July-August average of daily mean temperature was as much as 3.0[degrees]C above normal (Fig. 16.1a). Extreme high temperatures broke records at many stations (Fig. 16.1b). The regional average of July-August mean surface temperature was the highest since 1950 (Fig. 16.1c).
An average maximum of 34.4[degrees]C contributed to serious summer drought and other societal consequences (CMA 2014) in nine provinces: Shanghai, Zhejiang, Jiangxi, Hunan, Chongqing, Guizhou, Hubei, Anhui, and Jiangsu. The total days with high temperature (the daily maximum temperature [greater than or equal to] 35[degrees]C) averaged in these nine provinces was 31 days, more than double the normal average of 14 days and the longest stretch of heat since 1951. Some 344 stations recorded a daily maximum equal to or higher than 40[degrees]C, and a highest temperature of 44.1[degrees]C was recorded at Xinchang station in Zhejiang province. The total station days in these nine provinces setting maximum daily temperature records was 477, another maximum since 1951. Overall, 144 stations broke records for consecutive high temperature days (CMA 2014). Great public interest necessitates understanding the reason behind the hot summer and detecting the potential effect of anthropogenic forcing.
Did anthropogenic forcing contribute to the heat? To understand the potential contribution of anthropogenic forcing to the seasonal extreme warmth, we based our inquiry on detection and attribution methods by Hegerl et al. (2009) and using the method of Knutson et al. (2013b). We compared the observed trends with model simulations of internal climate variability and model responses to both anthropogenic and natural forcing using 31 models that participated in CMIP5 (Taylor et al. 2012). The models used in the analysis are listed in Supplementary Table S16.1. A preindustrial control simulation, the 20th century historical climate simulation (hereafter "All-forcing"), and part of a climate change projection under Representative Concentration Pathways (RCP) 4.5 scenario are used. The All-forcing simulations included both anthropogenic and natural forcing agents from about 1860 to the present. Note that the preindustrial control simulations are used as an indicator of natural variability without anthropogenic impacts. Since the 20th century historical climate simulation of CMIP5 only covers 1860-2005, as Knutson et al. (2013b), data from RCP4.5 runs are used to extend the time series through 2013 where necessary. Our analysis focuses on the period of 1900-2013.
Figure 16.2a shows the July-August (JA) time series averaged over the region of central eastern China (~24[degrees]-33[degrees]N, 102.5[degrees]-122.5[degrees]E) where the JA 2013 anomalies were the warmest in the record since 1951 (red colors in Fig. 16.1b). The observations operated by CMA show almost no trend before 1980 but a gradual rising trend after the 1980s, with a distinctive warm anomaly in 2013. The regional average surface temperature in 2013 is 1.838[degrees]C warmer than normal (base period 1961-90). This observed average is within the range of the 31 CMIP5 ensemble model members but is far higher than the multimodel ensemble mean (an anomaly of 1.838[degrees]C versus the simulated value of 0.868[degrees]C). The spread among the 31 CMIP5 models is large. Some individual models do reproduce the amplitude of the observed warmth. The weak warming trend prior to 1980 in the observations is related to natural variability of the climate system (Zhou et al. 2009), and it will be discussed in the final section.
Figure 16.2b shows a trend analysis for the JA central eastern China surface air temperature time series given in Fig. 16.2a. The models are compared to the observations. We show the results of both the multimodel ensemble mean and the uncertainties. Following Knutson et al. (2013b), the uncertainties are measured by the 5th-95th percentile range of the distribution of trends, which is obtained by combing random samples from each of the 31 CMIP5 model control runs together with the corresponding model's ensemble mean forced trends. Thus, a multimodel distribution of total trends is created, and it reflects uncertainty in both the forcing responses and the influence of internal climate variability. The 31 CMIP5 models used for the analysis contribute equally to the multimodel distribution from which the percentiles are derived in the sense of "one model, one vote."
The control run internal variability can be used as a surrogate for natural variability in the real world (Knutson et al. 2013b). The observed trends in Fig. 16.2b (black line) generally lie outside of the control run 5th-95th percentile range (green shaded region and the purple shaded region) after about 1990, indicating that the observed trends are inconsistent with internal climate variability alone. The observed trends lie within the pink regions for nearly all starting dates, indicating that the observed JA trends are consistent with the CMIP5 All-forcing model simulations. In particular, after the early 1990s, the observed trends lie outside of the control run 5th-95th percentile range (purple shaded region) but within the All-forcing run 5th-95th percentile range (pink shaded region), indicating the observed JA trends are inconsistent with internal climate variability alone but consistent with the CMIP5 All-forcing model simulations. Based on the comparison, we suggest that the observed warming after the early 1990s over central eastern China is very likely partly attributable to anthropogenic forcings.
Does the anthropogenic forcing contribute to the observed anomalies of JA 2013 over the central eastern China? Based on Fig. 16.2a, the 2013 observed JA surface air temperature anomaly is ~1.838[degrees]C, and the modeled value is ~0.868[degrees]C, a rough estimate of the anthropogenic contribution to the magnitude of the temperature anomaly is 47.23%. If we interpret the difference between the All-forcing and the control run distributions as the anthropogenic influence as Knutson et al. (2013b), the observed ~1.838[degrees]C warming of JA 2013 was 2.12 times stronger than the expected multimodel ensemble mean contribution of 0.868[degrees]C due to anthropogenic forcing in 2013. The difference indicates that internal variability also played a substantial role. Based on the ensemble of 31 CMIP5 models, an event this hot or hotter would occur with a probability of 1.047% in the control run and 2.518% in the forced simulation. Under the forced scenario, the fraction of risk of such an extreme warm event that is attributable to the anthropogenic forcing is (2.518-1.047) / 2.518 = 58.42%.
In addition, one may hypothesize that a study focused on maximum temperatures should be more relevant to diagnosing contributions to the anomalies listed in the previous section of the paper. We have assessed the performances of CMIP5 models in reproducing the long-term changes of maximum temperatures over central eastern China. We found that the models are poor in this regard and do not allow us to detect the potential contributions of anthropogenic forcings to maximum temperature changes over this domain by using CMIP5 models.
Concluding remarks. In this study, seasonal extreme warmth is placed in the context of long-term climate change by analyzing the time series for the region, comparing the observed trends with simulations of 31 CMIP5 models that include both internal variability and responses to anthropogenic and natural forcings. We found that long-term anthropogenic warming in central eastern China is detectable only after the early 1990s. The anthropogenic contribution to the magnitude of the extreme July-August mean warmth over the region is expected to be about 47.23%. Regarding the anthropogenic contribution to the increased probability of the extreme July-August mean warmth, the fraction of attributable risk is estimated to be 58.42%.
Central eastern China is dominated by the monsoon. The strength of the East Asian summer monsoon circulation has exhibited robust interdecadal variability in past decades and resulted in excessive precipitation along the middle and lower reaches of the Yangtze River valley (Yu et al. 2004; Zhou et al. 2009). The excessive precipitation is associated with colder temperatures (Yu and Zhou 2007). This could explain why a warming trend is not evident in Fig. 16.2a in the early decades. In addition, the aerosol emission induced cooling effect also partly slows down the anthropogenic warming trend (Qian et al. 2003, 2009; Song et al. 2014).
The Pacific decadal oscillation (PDO) or inter-decadal Pacific oscillation (IPO; Power et al. 1999) is a prominent natural internal variability mode of the coupled climate system. There is increasing evidence that the East Asian summer monsoon circulation is inversely correlated with the phase of the PDO/IPO (Li et al. 2010b; Qian and Zhou 2014); thus, both monsoon precipitation and the associated temperature changes are largely dominated by natural internal variability (Zhou et al. 2009, 2013). The large contribution of internal variability explains why the estimated fraction of attributable risk (58.42%) is lower than that over the eastern United States, which was 92% for the March-May 2012 warm anomaly (Knutson et al. 2013b). Since averaging coupled climate model simulations averages out different phases of the PDO, such variability, for example if it caused the observed cooling during 1960-70, would not be seen in the ensemble mean of CMIP5 historical climate simulations.
17. SEVERE PRECIPITATION IN NORTHERN INDIA IN JUNE 2013: CAUSES, HISTORICAL CONTEXT, AND CHANGES IN PROBABILITY
DEEPTI SINGH, DANIEL E. HORTON, MICHAEL TSIANG, MATZ HAUGEN, MOETASIM ASHFAQ, RUI MEI, DEEKSHA RASTOGI, NATHANIEL C. JOHNSON, ALLISON CHARLAND, BALA RAJARATNAM, AND NOAH S. DIFFENBAUGH
Cumulative precipitation in northern India in June 2013 was a century-scale event, and evidence for increased probability in the present climate compared to the preindustrial climate is equivocal.
The Event: June 2013 Flooding in Northern India. Parts of mountainous northern India--including Himachal Pradesh, Uttarakhand, and Uttar Pradesh--experienced extremely heavy precipitation during 14-17 June 2013 (Fig. 17.1a,b). Landslides, debris flows, and extensive flooding caused catastrophic damage to housing and infrastructure, impacted >100000 people, and resulted in >5800 deaths (Dobhal et al. 2013; Dube et al. 2013; Dubey et al. 2013; Joseph et al. 2014; Mishra and Srinivasan 2013). Subsequent heavy rains on 24-25 June hampered rescue efforts, ultimately leaving thousands without food or shelter for >10 days (Prakash 2013).
Causes of the mid-June precipitation and associated flooding have been analyzed in detail (Dobhal et al. 2013; Dube et al. 2014; Mishra and Srinivasan 2013; Prakash 2013). Anomalously early arrival of monsoon-like atmospheric circulation over India (Fig. 17.1c, Supplementary Figure S17.1a) brought heavy rains to the mountainous regions where snow cover typically melts prior to monsoon onset (Dube et al. 2014; Joseph et al. 2014). Snow cover in local river basins was -30% above normal in early June 2013 (Durga Rao et al. 2014). Heavy precipitation led to rapid snowmelt, overwhelming the regional hydrologic system, causing glacial lake outburst floods, and triggering catastrophic mass wastage events (Andermann et al. 2012; Dubey et al. 2013; Durga Rao et al. 2014; Prakash 2013; Siderius et al. 2013).
The upper- and lower-level synoptic conditions in early and mid-June supported the anomalously early monsoon-like circulation (Supplementary Fig. S17.1a) and excessive precipitation in northern India (Fig. 17.1a,b). In the upper atmosphere (200 mb), a persistent anticyclonic anomaly formed over Central Asia (Fig. 17.le). This upper-level blocking pattern guided mid-to-high-latitude troughs southward, thereby facilitating the advection of relatively cold, dry, high-potential-vorticity air to the upper levels of the atmosphere over northern India (Joseph et al. 2014). In the lower atmosphere (850 mb), low-pressure systems formed over both the northern Bay of Bengal and the northern Arabian Sea (Joseph et al. 2014), with the Bay of Bengal system moving inland over central India and persisting for the duration of the event (Fig. 17.1f). Low-level convergence associated with these systems and a stronger-than-normal Somali Jet facilitated anomalous moisture advection to the Indian subcontinent (Fig. 17.1c). These co-occurring upper- and lower-level dynamics are consistent with a convectively unstable atmosphere (Hong et al. 2011; Ullah and Shouting 2013; Wang et al. 2011), which, when combined with orographic forcing from the surrounding northwestern Himalayan terrain, create an environment ripe for intense mesoscale convection (Houze et al. 2011).
In this study, we analyze the dynamics of this event within the context of the historical and preindustrial climates.
Historical context. We contextualize June 2013 precipitation using the Indian Meteorological Department (IMD) 1951-2013 I[degrees] x I[degrees] gridded dataset (Rajeevan et al. 2010), with the caveat that the rain gauge network in the region could have changed over this period. Cumulative June precipitation exceeded the 80th percentile over much of central and northern India, and it exceeded the maximum quantile over a majority of the flood region (Fig. 17.1a). From 14 to 17 June, this domain (29[degrees]-33[degrees]N, 77.5[degrees]-80[degrees]E) received four-day total precipitation that was unprecedented in the observed record (Fig. 17.1b), with the heaviest day (16 June) exceeding the previous one-day June maximum by 105% (Supplementary Fig. S17.2). Consequently, the flood region recorded the highest total accumulated June precipitation in the 1951-2013 record, with the previous maximum June total equaled by 17 June and exceeded by 31% by the end of the month (Fig. 17.1b).
Monsoon dynamics and thermodynamics were also unusual relative to June climatological norms. The monsoon onset date is closely associated with the reversal of the zonally averaged (52[degrees]-85[degrees]E) meridional tropospheric (500-200 mb) ocean-to-continent (5[degrees]-30[degrees]N) temperature gradient (Ashfaq et al. 2009; Webster et al. 1998), and with the vertical easterly zonal wind shear between 850 mb and 200 mb averaged over 0[degrees]-30[degrees]N and 50[degrees]-90[degrees]E (Li and Yanai 1996; Webster et al. 1998; G. Wu et al. 2012; Xavier et al. 2007). The 2013 meridional temperature gradient (MTG) reversai dates were among the earliest on record (1951-2013, Fig. 17.1d) and the vertical easterly wind shear was stronger than normal during early-June (Supplementary Fig. S17.lb). The early MTG reversal resulted from anomalously high land temperatures (-2 standard deviations; Supplementary Fig. S17.1c,d), which co-occurred with record-low Eurasian snow cover (NOAA 2013). In addition, as a result of the early monsoon-like circulation, low-level atmospheric humidity exceeded 2 standard deviations above the climatological 14-17 June mean (Fig. 17.1c).
Synoptic conditions were likewise extremely rare for mid-June. We categorize the occurrence of upper-and lower-level daily June atmospheric patterns in the National Centers for Environmental Prediction (NCEP) R1 reanalysis using self-organizing map (SOM) cluster analysis (Borah et al. 2013; Chattopadhyay et al. 2008; Hewitson and Crane 2002; Johnson 2013; Kohonen 2001; see Supplemental Materials). SOM analyses reveal persistent upper-level blocking patterns from 10 to 17 June and lower-level troughing patterns from 11 to 17 June (Supplementary Fig. S17.2). Additionally, the upper- and lower-level patterns (Fig. 17.1g,h) that persisted during the core of the event (14-17 June) are each historically associated with heavy precipitation over northern India (Fig. 17.1i,j). Although occurrence of the core-event upper-level pattern is not rare for June (median frequency of occurrence), the 850-mb pattern is much less common (<6 percentile frequency of occurrence). Further, mid-June 2013 was the only instance that the core-event upper- and lower-level patterns co-occurred in June during the 1951-2013 period. The atmospheric configuration associated with the unprecedented mid-June extreme precipitation, therefore, appears to also have been unprecedented.
We note that this configuration is not necessarily unprecedented later in the monsoon season. For example, the co-occurrence of upper-level blocking with tropical moisture advection is similar to the conditions identified during the July 2010 Pakistan floods and during heavy precipitation events that occur during the core monsoon season (Hong et al. 2011; Houze et al. 2011; Lau and Kim 2011; Ullah and Shouting 2013; Webster et al. 2011).
Quantifying the probability of a 2013-magnitude event. In quantifying the probability of a 2013-magnitude event, we restrict our focus to the June 2013 total precipitation. We select the monthly scale extreme rather than the daily scale extreme because both the extreme magnitude of this event relative to the observed distribution of four-day June totals and the limited ability of climate models to accurately simulate the daily scale extremes make the problem intractable at the daily scale. Therefore, hereafter, "a 2013-magnitude event" refers to the total June rainfall, which in June 2013 was the most extreme on record (Fig. 17.1b).
Given the rarity of the event in the observed record (Fig. 17.2a), we fit a Pareto (heavy-tailed) distribution to the 1951-2012 observations of spatially averaged (area-weighted average) rainfall over the selected domain (Fig. 17.2a; Supplementary Fig. S17.3a). From the Pareto distribution, we estimate the sample quantile ([Q.sub.o]) and return period ([R.sub.o]) of the June 2013 total precipitation in the present climate (see Supplemental Materials). We find that the 2013 event exceeds the 99th percentile in the observed distribution ([Q.sub.o] = 99.1th quantile), yielding a return period of 111 years (Fig. 17.2a). Because the Pareto is a heavy-tailed distribution, extreme events are less likely to be found anomalous, and, thus, the corresponding return period can be considered a lower bound.
Next, we assess the influence of anthropogenic forcings on the likelihood of extreme June precipitation using the historical (20C) and preindustrial (PI) simulations from the CMIP5 climate model archive (Taylor et al. 2012). We use the Kolmogorov-Smirnov (K-S) goodness-of-fit test to identify the models that most closely simulate the observed distribution of the area-weighted average June total precipitation over the impacted region (Fig. 17.1a). (To control for the mean bias in the models, we first re-center each model's distribution so that the model mean matches the observed mean.) Because the simulated change in likelihood of extremes can be heavily influenced by biases in the simulated distribution, we restrict our analyses to 11 models whose K-S value exceeds 0.2 (Supplementary Fig. S17.3b), ensuring a comparatively good fit of the overall distribution, including in the tails. Among these 11 models that pass this goodness-of-fit criterion, 4 show greater mean and variability of June precipitation in the 20C simulations (Fig. 17.2b). However, 7 of the 11 show increased exceedance of the PI 99th percentile value (Fig. 17.2c), suggesting increased probability of extremely high June precipitation in the current climate. This result is consistent with studies that indicate an increase in extremes primarily from increased atmospheric-moisture availability (Allan and Soden 2008; O'Gorman and Schneider 2009).
Next, we use Pareto distributions to estimate the return period of the June 2013 total precipitation in the 20C and PI simulations. To control for the variability-bias in the models, we first determine the magnitude of the 111-year event (Q0= 99.1th quantile) in the fitted 20C distribution (PrH), and then determine the quantile ([Q.sub.PI]) corresponding to [Pr.sub.H] in the fitted PI distribution (see Supplemental Materials; Supplementary Fig. S17.3c). Further, we quantify the uncertainty in these likelihood estimates ([Q.sub.o]/[Q.sub.PI]) using the bootstrap (Fig. 17.2d). We find that 5 of the 11 models show >50% likelihood that the extreme June total precipitation has higher probability in the 20C climate. In addition, of the three models that have high p-values from the K-S test (> 0.8) and similar sample sizes in the 20C and PI populations (Fig. 17.2d), two suggest >50% likelihood that the extreme June total precipitation has higher probability in the 20C climate, and the third model suggests -50% likelihood. Further, the model with the largest 20C ensemble (Centre National de Recherches Meteorologiques Coupled Global Climate Model; CNRM-CM5) demonstrates a ~50% likelihood that the probability of the extreme June total precipitation has at least doubled in the 20C climate. CNRM-CM5 also has the highest skill in simulating the summer monsoon precipitation and lower-level wind climatology (Sperber et al. 2013).
Conclusions. Our statistical analysis, combined with our diagnosis of the atmospheric environment, demonstrates that the extreme June 2013 total precipitation in northern India was at least a century-scale event. Precise quantification of the likelihood of the event in the current and preindustrial climates is limited by the relatively short observational record and by the resolution and ensemble size of the small subset of models that credibly simulate the seasonal rainfall distribution over northern India. Indeed, an attempt to quantify the probability of the unprecedented four-day precipitation total would present even greater analytical challenges. However, despite these limitations, our analyses of the observed and simulated June precipitation provide evidence that anthropogenic forcing of the climate system has increased the likelihood of such an event, a result in agreement with previous studies of trends in rainfall extremes in India (Goswami et al. 2006; Krishnamurthy et al. 2009; Ghosh et al. 2012; Singh et al. 2014).
18. THE 2013 HOT, DRY, SUMMER IN WESTERN EUROPE
BUWEN DONG, ROWAN SUTTON, AND LEN SHAFFREY
Anthropogenic forcing played a substantial part in western Europe's hot, dry summer in 2013. North Atlantic sea surface temperatures were likely a factor in the large contrast with summer 2012.
Observations. Western Europe experienced sweltering high temperatures in summer 2013. On 22 July 2013, the United Kingdom recorded its hottest day since July 2006, with 33.5[degrees]C recorded at Heathrow and Northolt in west London (Met Office 2014). Averaged over western Europe (Fig. 18.1c), the seasonal mean (June-August) anomaly in surface air temperature (SAT) was 1.33[degrees]C above the mean over the period of 1964-93, which is 3.2 standard deviations of the interannual variability. [HadCRUT4 data (Morice et al. 2012) shows a similar warming of 1.28[degrees]C.] This magnitude of warming is slightly less but comparable with the previous hot summers in western Europe, such as 2003 (e.g., Schaer et al. 2004) and 2010 (e.g., Barriopedro et al. 2011) for which summer mean SAT anomalies were 1.46[degrees]C and 1.86[degrees]C respectively, corresponding to 3.5 and 4.5 standard deviations.
The atmospheric circulation in summer 2013 was characterized by anomalously high sea level pressure (SLP) extending from the United Kingdom into northern Europe and anomalously low SLP over the Arctic (Fig. 18.1a). This pattern projects strongly onto the positive phase of the summer North Atlantic Oscillation (SNAO; Folland et al. 2009). The anomalous SNAO index of 2.7 hPa in 2013 was +1.0 standard deviation of the interannual variability, in stark contrast with the previous summer of 2012 (Dong et al. 2013a) for which the index was -2.7 standard deviations (Supplementary Fig. S18.1b). The circulation pattern in 2013 was associated with a northward shift of summer North Atlantic storm track (Fig. 18.le and f). The climatology of cyclone track density (Dong et al. 2013b and Fig. 18.le) shows a split into two preferred cyclone paths at the North Atlantic jet exit region (5[degrees]W-5[degrees]E): one passing near Iceland at ~71[degrees]N and into the Nordic Seas and the other passing across the British Isles at ~56[degrees]N and into western Europe. In summer 2013, more storms than usual passed over Iceland and fewer over the United Kingdom and into Western Europe (Fig. 18.If). This led to dry conditions in the United Kingdom and most of western Europe. The area-averaged precipitation anomaly was -0.35 mm [day.sup.-1], which is -2.2 standard deviations of the interannual variability (Supplementary Fig. S18.lc). The low rainfall was also in stark contrast to the summer of 2012, which was a record wet summer in the United Kingdom and was last in a series of wet UK summers since 2007, each of which was associated with a negative SNAO index (Allan and Folland 2012; Dong et al. 2013a). [Note that the inhomogeneity of the data in E-OBS precipitation is a potential source of bias (Zolina et al. 2013), but negative precipitation anomalies in Western Europe are consistent with the northward shifted storm track.]
Global SST anomalies for summer 2013 are illustrated in Fig. 18. Id. Warm SSTs (relative to 1964-93) were present in many regions, with a prominent warm anomaly (> 1.0[degrees]C) along the Gulf Stream extension in the North Atlantic. Associated with this feature were an enhanced meridional SST gradient to the north and a reduced gradient to the south (Supplementary Fig. S18.2c). These anomalous SST gradients may have played a role in the observed northward shift of the North Atlantic storm track (e.g., Sampe et al. 2010; Ogawa et al. 2012) and influenced the related anomalies in the SNAO and western European climate (Folland et al. 2009; Sutton and Dong 2012; Dong et al. 2013b). Warm anomalies were also observed in the Arctic, consistent with the continuing low sea ice extent (SIE); these SIE anomalies might also have had an influence on the atmospheric circulation (Balmaseda et al. 2010; Petrie et al. 2014, manuscript submitted to Quart. J. Roy. Meteor. Soc.).
Relative to the climatological period of 1964-93, by 2012 there were significant increases in anthropogenic greenhouse gas (GHG) concentrations (e.g., WMO 2013) and significant changes in anthropogenic aerosols. European and North American sulphur dioxide emissions had declined while Asian emissions had increased (e.g., Lamarque et al. 2010). In this study, we investigate the roles of changes in SST, SIE, and radiative forcing in shaping the European summer of 2013, as well as possible reasons for the striking contrast between summer 2013 and summer 2012. Our focus is on seasonal mean conditions rather than on shorter-lived events that occurred within the season. We note that the sign of shorter-lived events can often differ from that of the seasonal mean, so for example, some regions of western Europe experienced floods in summer 2013 even though the seasonal mean precipitation was below average.
Climate model experiments. Climate model experiments have been carried out to identify the roles of changes in: (a) SST/SIE and (b) anthropogenic GHG and aerosol forcing in the European summer climate anomalies of 2013. In this study, we do not address the anthropogenic contribution to the SST/SIE changes, but rather consider these changes as an independent forcing factor. We use the atmosphere configuration of the Met Office Hadley Centre Global Environment Model version 3 (HadGEM3A; Hewitt et al. 2011), with a resolution of 1.875[degrees] longitude by 1.25[degrees] latitude and 85 levels in the vertical. Dong et al. (2013a) used the same model to study the 2012 summer in Europe. A series of experiments was performed, the details of which are summarized in Table 18.1. We use the same control experiment (CONTROL) for the period 1964-93 as Dong et al. (2013a) and perform two other experiments: ALL2013 and SST2013. Both of these experiments use 2013 SST and SIE boundary conditions but they differ in the specification of anthropogenic GHG and aerosol forcing: ALL2013 uses anthropogenic forcing appropriate for 2013 while SST2013 uses the same anthropogenic forcing as for the CONTROL experiment, appropriate for 1964-93. The last 25 years of each experiment are used for analysis. The CONTROL experiment reproduces realistic climatological SLP and precipitation patterns for summer (Supplemental Fig. S10.2 of Dong et al. 2013a).
The model simulates a significant warming over Europe in summer 2013 in response to changes in SST, SIE, and anthropogenic forcing (i.e., ALL2013-CONTROL, Fig. 18.2a) with an area averaged SAT change of 1.11[degrees]C over western Europe. The observed anomaly of 1.33[degrees]C is within the [+ or -]1 standard deviation range of the interannual variability of the model response (Supplementary Fig. S18.1a). Changes in SST and SIE explain 63% ([+ or -]26%) of the area-averaged western European warming response in HadGEM3 (Fig. 18.2d), with the remaining 37% ([+ or -]29%) explained by the direct impact (without forcing-induced SST and sea ice feedbacks) of changes in radiative forcings from GHG and aerosols (Fig. 18.2g and Supplementary Fig. S18.1a).
The atmospheric circulation anomalies simulated by the model (Fig. 18.2b) show notable similarities to the observed pattern over the North Atlantic and Europe (Fig. 18.1a), including low SLP anomalies over Greenland and an anomalous anticyclone over the United Kingdom. The wave train pattern of SLP anomalies suggests that changes in convection over the Caribbean Sea might be an important factor (e.g., Douville et al. 2011). However, in the model simulation the anomalous anticyclone does not extend as far eastward into central Europe as in the observations. The circulation anomalies correspond to a positive anomaly (mean = 1.2 hPa, which is only 0.5 standard deviations) in the SNAO index relative to the CONTROL, which is smaller than the observed anomaly (2.7 hPa; Supplementary Fig. S18.1b). The pattern of simulated western European precipitation anomalies (Fig. 18.2c) is consistent with the positive phase of the SNAO and is similar to the observations, with anomalously low rainfall over most of western Europe (Fig. 18.1b). As for the circulation anomaly, the magnitude of the mean precipitation anomaly is smaller than observed (Supplementary Fig. S18.1c), although there is substantial interannual variability in the model results.
The additional SST2013 experiment suggests that both SST/SIE changes and the direct impact of changes in anthropogenic radiative forcing contributed to the anomalous circulation (Fig. 18.2e and h; Fig. S18.1b) and reduced rainfall over western Europe in summer 2013 (Fig. 18.2f and i; Supplementary Fig. S18.1c). The SST change has the most impact on the low SLP anomalies simulated over Greenland, but GHG and aerosol forcing causes a substantial anticyclonic anomaly over, and north of, the United Kingdom. This anticyclonic circulation anomaly is similar to the summer mean circulation response to an increase in GHG forcing (Blade et al. 2012) and is presumably due to an increase in the frequency of the positive SNAO-like circulation regimes over the Atlantic sector (Boe et al. 2009).
Changes in GHG and aerosol forcing are unlikely to be a major factor in explaining the striking contrast in circulation and precipitation between the European summers of 2012 and 2013 (Supplementary Fig. S18.2a), as the changes in these forcings between these two years were small. However, the model experiments suggest that changes in SST and SIE in the North Atlantic were a significant factor (Supplementary Fig. S18.2e). In particular, the anomalous meridional SST gradient to the north of the Gulf Stream in 2013, relative to 2012 (Supplementary Fig. S18.2c), may have favored a positive phase of the SNAO and a northward shift of North Atlantic summer storm track (e.g., Folland et al. 2009; Dong et al. 2013b), as was observed (Supplementary Figs. S18.2d and f). The model experiments show some evidence of capturing this shift, although the mean signal (Supplementary Fig. S18.2e) is again much weaker than was observed (Supplementary Fig. S18.2a).
The model results show an encouraging degree of consistency with observations, but it is difficult to assess precisely what level of consistency should be expected in view of the high level of internal variability and uncertainty about the true magnitude of forced signals in the real world. It is clear from Supplementary Fig. S18.1 that the signal-to-noise ratio for the changes in SAT is large, which permits more confident conclusions, whereas that for changes in circulation and precipitation is much lower (though it is interesting to note that the model suggests a stronger forced signal in western European summer precipitation than in the SNAO). One limitation of the current experiments, which may well influence the signal-to-noise ratio, is the use of a prescribed SST boundary condition. Active ocean-atmosphere coupling may modify the response to forcings and is an important area of future work (Sutton and Mathieu 2002; Dong et al. 2013b). Another extension not addressed here is the anthropogenic contribution to the SST/SIE changes.
Conclusions. The European summer of 2013 was marked by hot and dry conditions in western Europe associated with a northward shifted Atlantic storm track and a positive phase of the SNAO. Model results suggest that, relative to a 1964-93 reference period, changes in SST/SIE explain 63% ([+ or -] 26%) of the area-averaged warming signal over western Europe, with the remaining 37% ([+ or -] 29%) explained by the direct impact of changes in anthropogenic radiative forcings from GHG and aerosols. The results further suggest that the anomalous atmospheric circulation, and associated low rainfall, were also influenced both by changes in SST/SIE and by the direct impact of changes in radiative forcings; however, the magnitude of the forced signals in these variables is much less, relative to internal variability, than for surface air temperature. Further evidence suggests that changes in North Atlantic SST were likely an important factor in explaining the striking contrast between the European summers of 2013 and that of 2012. A major area for further work is to understand more completely the mechanisms that explain these influences.
Table 18.1. Summary of numerical experiments. Length of run Experiments Boundary Conditions 32 years CONTROL Forced with monthly mean climatological sea surface temperature (SST) and sea Ice extent (SIE) averaged over 1964-93 using HadlSST data (Rayner et al. 2003) and with anthropogenic greenhouse gases (GHG) concentrations averaged over the same period and anthropogenic aerosols emissions averaged over 1970-93. ALL2013 Forced with monthly mean SST and SIE from 27 years Dec 2012 to Nov 2013 using Had-ISST data, with anthropogenic GHG concentration in 2012 (WMO 2013) and anthropogenic aerosol emissions for 2010 (Lamarque et al. 2010), which is the most recent year for which emissions data were available. SST2013 As ALL2013, but with anthropogenic GHG 27 years concentrations and anthropogenic aerosol emissions the same as in CONTROL.
19. CONTRIBUTION OF ATMOSPHERIC CIRCULATION TO WET SOUTHERN EUROPEAN WINTER OF 2013
PASCAL YIOU AND JULIEN CATTIAUX
Winter 2013 was the second wettest since 1948 in southern Europe. This is partially explained by the atmospheric circulation. We suspect the warm Atlantic Ocean to have amplified the precipitation extreme.
Introduction. Southern Europe witnessed anomalously high precipitation amounts, associated with anomalously low temperatures, during the winter of 2013. The goal of this paper is to put this regional event into the context of long-term variability. In Europe, studies have highlighted North Atlantic atmospheric dynamics as the main driver of winter precipitation and temperatures on both intra-seasonal and inter-annual time scales (e.g., Cattiaux et al. 2010; Vautard and Yiou 2009). Here, we focus on the contribution of large-scale circulations the winter 2013 precipitation anomalies using the same flow-analogue approach as in the analysis of summer 2012 North European precipitation by Yiou and Cattiaux (2013).
Data. We use in situ measurements provided by the European Climate Assessment dataset at 5231 stations over the period 1948-2013 (Klein-Tank et al. 2002). We compute anomalies relative to a 1971-2000 climatology. The daily climatology is obtained by averaging over each calendar day in 1971-2000, then smoothing by splines. We selected 510 stations on the basis of (a) an altitude lower than 800 m, longitudes between 10[degrees]W-40[degrees]E, and latitudes between 30[degrees]-72[degrees]N; (b) the availability of more than 70% of daily values between 1 January 1971 and 31 December 2000 for a reliable estimate of the climatology; and (c) the availability of more than 90% of daily values between 1 January 2013 and 31 December 2013 for a reliable estimate of the 2013 anomalies. We averaged precipitation values over Southern Europe (10[degrees]W-20[degrees]E; 35[degrees]-50[degrees]N, see Fig. 19.1) after selecting (d) only one station per 0.5[degrees] x 0.5[degrees] latitude/longitude box for spatial homogeneity, hence leaving 45 stations in that region. Although we mainly focus on cumulative precipitation, we also computed the precipitation frequency (or fraction of wet days) for each month as in Vautard et al. (2007) and Vautard and Yiou (2009). The precipitation frequency is the empirical probability in a given month of observing a daily precipitation amount larger than 0.5 mm. It provides an indicator of the temporal continuity of precipitation.
Observed rainfall anomaly. Anomalously high precipitation amounts are observed in Southern Europe (e.g., France, Spain, and Italy) in February-March 2013 (Fig. 19.1, upper panels). These anomalously high amounts culminate in March 2013. Those winter wet conditions contrast with drier than usual conditions in Northern Europe (e.g., United Kingdom, Netherlands, Germany, and Scandinavia). The precipitation frequency during winter 2013 (from 1 January to 31 March 2013) also has positive anomalies in Southern Europe, indicating that the rain episodes lasted for prolonged periods of time (Supplementary Fig. S19.1). Therefore, the generally wet conditions mainly concern the southern part of Europe for winter 2013 (mainly February and March). Besides, cold anomalies were observed over Western Europe during this season (Supplementary Fig. S19.2).
The highest daily precipitation amounts on average over Southern Europe occurred in February and March (Fig. 19.2a), as did the longest winter spells of precipitation (32 days). Part of that precipitation was snow, which caused havoc in French public transportation, especially in the middle of March 2013, and broke local records of snow amounts in the French Pyrenees. On average over Southern Europe during the whole winter season (JFM), 2013 is the second wettest winter since the beginning of our record (1948) behind 1979. This points to the exceptional character of the Southern Europe 2013 winter. The temperatures were also anomalously low in February and March in Southern Europe, contrasting with Eastern Europe, especially in February (Supplementary Fig. S19.2), but the temperature anomaly was close to the median (e.g., 33rd coldest year on record), therefore, justifying our focus on precipitation. The temperature anomalies are not as well reconstructed by the circulation analogues in February 2013, especially over France. This is explained by the fact that the February cold temperatures were due to persisting snow cover (e.g., in France) rather than the atmospheric circulation itself.
Contribution of the atmospheric circulation. The contribution of large-scale dynamics to the precipitation anomalies of 1948-2013 is estimated from the same flow-analogue approach as used in Cattiaux and Yiou (2012) or Yiou and Cattiaux (2013). For each day, we selected the 20 days with the most correlated sea level pressure (SLP) among days of other years but within a moving window of 60 calendar days to account for seasonality (see Yiou et al. 2007 for details). SLP anomalies are derived from the NCEP-NCAR reanalyses (Kistler et al. 2001) and are considered over the period 1948-2013 and the area (80[degrees]W-20[degrees]E, 22.5[degrees]-70[degrees]N).
For each station and each day, we computed the mean of the 20 analogue composites. The mean analogue precipitation of winter 2013 (averaged over 20 analogue days) is higher than usual in Southern Europe, both in terms of precipitation amount (Fig. 19.1) and frequency (Supplementary Fig. 519.1). Spatial patterns of analogue precipitation follow the observed ones, albeit with lower amplitudes (Fig. 19.1). The spatial correlations are 0.61, 0.64, and 0.79 in January to March, respectively. On a daily time scale over Southern Europe, the precipitation analogues closely follow the observed averages (Fig. 19.2a; temporal correlation r = 0.57; p-value < 5 x [10.sup.-3]). This confirms that the large-scale atmospheric circulation influences the precipitation amounts at the intraseasonal scale.
In order to further describe the circulation patterns of the winter 2013, we used the clustering approach of Michelangeli et al. (1995), and adapted by Yiou et al. (2008), to derive the four preferential winter weather regimes over the North Atlantic region and the period 1948-2013. The rationale for this analysis, which complements the flow-analogue approach, is to visualize the atmospheric circulation temporal variability and associate episodes of high precipitation with circulation patterns (Fig. 19.2a). The weather regimes are computed from SLP anomalies during 1948-2013. These four weather regimes correspond to anomalies in the flow and affect the advection of temperature and humidity (Fig. 19.2c--f). We find that the wet spells over Southern Europe correspond with episodes of the Atlantic Ridge (AR) in February and the negative phase of the North Atlantic Oscillation (NAO-) in March, which yields a weak pressure dipole over Iceland and the Azores (Fig. 19.2d-e). When the circulation yields anticyclonic patterns over Scandinavia (blocking), daily precipitation amounts fall to low values.
Trends of precipitation. We computed the linear trends of the seasonal average precipitation over the outline region (Fig. 19.2b) for the period between 1971 and 2013. The trends for all seasons are not statistically significant (p-values > 0.1). The mean analogue precipitation for winter is well correlated (r = 0.86,p-value < 10-15) with the observed average (Fig. 19.2b), and the analogues yield a negative winter trend found in the observations, although it is not statistically significant (p-value = 0.26). The winter 2013 median analogue precipitation amount is only the 18th highest of the analogue time series, showing that even if the atmospheric conditions were favorable to wet conditions over Southern Europe, they do not fully explain the exceptional character of the precipitation anomaly. We conjecture that a potential amplifying cause could be that the oceanic air masses carried by regimes of westerly winds were moister than usual due to warmer SSTs in the Northeast Atlantic (between 0.5 and 1.5 K above normal). We performed an additional analysis by searching circulation analogues among the years of warm Northeast Atlantic SST (i.e., above the 1971-2000 average). The mean monthly European precipitation amounts reconstructed from such "filtered" analogues exceed those of "regular" analogues, picked over 1948-2012 (not shown). Although this is not a definite proof, this pleads in favor of this mechanism.
Conclusions. Our analysis suggests that the high precipitation amounts were mainly caused by the cyclonic conditions (NAO- and Atlantic Ridge) that prevailed during the late winter (February and March) over the North Atlantic. Such conditions brought moist air over Southern Europe. This conclusion is drawn from the significant correlations over Europe between the observed and the analogue precipitation, deduced from the North Atlantic atmospheric circulation. The extreme precipitation amounts, not fully explained by the atmospheric circulation, are conjectured to be due to a warmer Northeast Atlantic with more moist air (Trigo et al. 2013).
The trend in winter precipitation over Southern Europe is negative but not statistically significant. The frequency of cyclonic regimes over Scandinavia (NAO- and Atlantic Ridge) has also slightly decreased, albeit not significantly (not shown).
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||p. S34-S69|
|Author:||Herring, Stephanie C.; Hoerling, Martin P.; Peterson, Thomas C.; Stott, Peter A.|
|Publication:||Bulletin of the American Meteorological Society|
|Date:||Sep 1, 2014|
|Previous Article:||Explaining extreme events of 2013 from a climate perspective.|
|Next Article:||Explaining extreme events of 2013 from a climate perspective.|