Influence of sea surface temperature variability on global temperature and precipitation extremes

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 114,, doi: /2009jd012301, 2009 Influence of sea surface temperature variability on global temperature and precipitation extremes Lisa V. Alexander, 1,2 Petteri Uotila, 1 and Neville Nicholls 1 Received 22 April 2009; revised 1 June 2009; accepted 13 July 2009; published 24 September [1] The HadISST1 data set was used to categorize seasonal patterns of observed global sea surface temperature (SST) variability between 1870 and 2006 using the method of Self-Organizing Maps (SOM). Eight patterns represented the majority of global SST variations associated with the El Niño Southern Oscillation (ENSO). Time series of the eight patterns exhibited periods with preferred SST states since the late 19th century, i.e., when one or more patterns occurred more frequently than in other periods. The eight patterns were used to investigate the global land-based response of observed extreme temperature and precipitation indices from the HadEX data set to different nodes of SST variability between 1951 and Results showed very strong statistically significant opposite temperature and precipitation extremes associated with the first pattern (strong La Niña) and the last pattern (strong El Niño). Extreme maximum temperatures were significantly cooler during strong La Niña events than strong El Niño events over Australia, southern Africa, India, and Canada while the converse was true for United States and northeastern Siberia. These responses were larger when global warming was retained. Even intermediate patterns representing a shift from a weak El Niño to a weak La Niña with associated variability in the North Atlantic were linked with statistically significant increases in warm nights and warm days particularly across Scandinavia and northwest Russia. While the link between precipitation extremes and global SST patterns was less spatially coherent, there were large areas across North America and central Europe, which showed statistically significant differences in the response to opposite phases of the El Niño Southern Oscillation. These results confirm that the variability of global SST anomaly patterns is important for the modulation of extreme temperature and precipitation globally. Citation: Alexander, L. V., P. Uotila, and N. Nicholls (2009), Influence of sea surface temperature variability on global temperature and precipitation extremes, J. Geophys. Res., 114,, doi: /2009jd Introduction [2] Sea surface temperature (SST) variability has a significant influence on the variability of atmospheric climate [Ropelewski and Halpert, 1987; Halpert and Ropelewski, 1992]. For instance, the El Niño Southern Oscillation (ENSO) brings drier/warmer conditions to much of Australia, southeast Asia, India, northwest USA and Canada during positive phases (El Niño) and wetter/cooler conditions in most of these regions during its negative phase (La Niña), e.g., McBride and Nicholls [1983], Power et al. [1998], Kiladis and Diaz [1989], Rasmusson and Carpenter [1982], Deser and Wallace [1987, 1990], and Ropelewski and Halpert [1986]. While it is important to understand how large scale variability affects mean climate, much more understanding is required on how this variability influences 1 School of Geography and Environmental Science, Monash University, Clayton, Victoria, Australia. 2 Now at Climate Change Research Centre, University of New South Wales, Sydney, New South Wales, Australia. Copyright 2009 by the American Geophysical Union /09/2009JD climate extremes since these can have a much more significant impact on communities and ecosystems [Easterling et al., 2000; Nicholls and Alexander, 2007]. [3] Previous studies have shown that natural modes of climate variability have a significant influence on the regional response of climate extremes on multidecadal and longer timescales, e.g., Kunkel [2003], Nicholls et al. [2005], Scaife et al. [2008], WASA Group [1998], and Wang et al. [2009]. Perhaps the first study which looked at the effect of a number of large scale climate variability measures on global temperature extremes was Kenyon and Hegerl [2008]. Using indices of large scale climate variation, they showed that different phases of ENSO appeared to influence temperature extremes globally, often affecting cold and warm extremes differently. They found areas of much stronger response around the Pacific Rim and throughout all of North America. They conclude that modes of variability need to be taken into account if we are to provide for reliable attribution of changes in extremes as well as prediction of future changes. Kenyon and Hegerl [2008] also investigated the influence of other large scale climate indices, such as the North Atlantic Oscillation 1of13

2 and the North Pacific Index, on extremes of temperature in different regions of the globe. However, the aim of our study is to look at the influence of global patterns of SST variability on temperature extremes and for the first time to look at the global response of precipitation extremes to these patterns. This could determine whether patterns of variability associated with the El Niño Southern Oscillation, but other than those defined by positive or negative phases of indices such as NINO3.4, NINO4 etc. had any impact on global extremes. The method of Self-Organizing Maps (SOM) was used to categorize observed seasonal SST anomalies between 1870 and 2006 from the HadISST1 data set [Rayner et al., 2003] into common patterns. The SOM algorithm objectively categorizes patterns from large data sets and therefore has an advantage over some of the subjectively derived large-scale climate indicators. The results were used to show how modes of SST variability influence, or are related to, global temperature and precipitation extremes. 2. Data and Methods [4] Two observational data sets were used in this analysis: HadISST1 [Rayner et al., 2003] and HadEX [Alexander et al., 2006]. HadISST1 is a 1-degree latitude by 1-degree longitude grid of globally complete monthly SST and sea ice concentration (from 1870 onward) derived from in situ measurements, digitized sea ice charts and passive microwave retrievals. For this study HadISST1 data were obtained from 1870 to Because our interest is in sea surface temperature patterns, areas of 100% sea ice cover were considered to be land. HadEX is a 2.5-degree latitude by 3.75-degree longitude gridded data set containing 27 indices derived from daily in situ temperature and precipitation observations over much of the global land area from 1951 to Therefore while our analysis of SST variability covers the entire period from 1870 to 2006, the comparison of this variability with land-based climate extremes is limited to the period when HadEX data were available. To investigate how temperature and precipitation extremes are affected by SST variability, we focused on six of the HadEX indices (four temperature and two precipitation). These indices are as follows: [5] 1. Percent of days below the 10th percentile (of the climatological reference period) of seasonal minimum temperature (TN10p); [6] 2. Percent of days above the 90th percentile of seasonal minimum temperature (TN90p); [7] 3. Percent of days below the 10th percentile of seasonal maximum temperature (TX10p); [8] 4. Percent of days above the 90th percentile of seasonal maximum temperature (TX90p); [9] 5. Maximum 1-day seasonal rainfall event (RX1day); [10] 6. Maximum consecutive 5-day seasonal rainfall event (RX5day). [11] These indices were chosen because they were available seasonally and in addition the temperature indices represent the warm and cold extremes of both maximum and minimum temperature. In addition, the monthly NINO4 index was also obtained from NOAA s National Weather Service Climate Prediction Center. This index is calculated from SST anomalies in the tropical Pacific averaged in the box 6N 6S, 160E 150W and was available from 1950 to provide an independent measure for comparison with the results presented here. [12] Firstly, seasonal SST anomalies (SSTA) were created from HadISST1 for December to February (DJF), March to May (MAM), June to August (JJA) and September to November (SON) between 1870 and 2006 using as the climatological reference period. Next, these seasonal SSTA were detrended by fitting a linear ordinary least squares regression to each grid box with data and calculating the residuals of the seasonal SSTA time series from this trend. The reason for removing trends from the data is so that changes observed from one state to another reflect the actual variability of the SSTA patterns rather than long-term trends toward warming oceans. The detrended SSTA were then categorized into most common patterns of variability using a clustering technique called Self- Organizing Maps (SOM [Kohonen, 2001]). The SOM technique has been found to be useful in reducing the dimensions of large data sets to enable the visualization and interpretation of the data characteristics. Details of the approach which is closely followed here are given by Cassano et al. [2006]. In addition, Leloup et al. [2007a] give a detailed description in their appendix of the construction of a SOM of SSTA in the equatorial Pacific between 1950 and Our technique differs in that we create a SOM from all available global SSTA over a longer period, i.e., 1870 to Also, we construct global SSTA patterns directly from the SOM algorithm rather than through a combination of this with a Hierarchical Agglomerative Clustering method as in the case of the Leloup et al. [2007a] study. In our case, the SOM algorithm produced a one-dimensional set of patterns or nodes of global SSTA. After testing several different numbers of nodes, and other variables which are required by the SOM algorithm (such as learning rate parameter), it was found that eight nodes provided enough patterns to cover most aspects of global seasonal SST variability while ensuring that each node is frequent enough to provide robust statistics for the comparison with the global climate extremes. By comparison, Leloup et al. [2007a] found that 12 patterns best described SSTA variability in the equatorial Pacific region. The SOM algorithm determines which node of the SOM best matches the observed SSTA for that season (that is the SOM node that gives the smallest Euclidean distance error when compared to the seasonal SSTA pattern). [13] Seasonal values were also calculated for global temperature and precipitation extremes between 1951 and 2003 when data are available from the HadEX data set. Alexander et al. [2006] showed that there has been a significant warming of both maximum and minimum temperature extremes globally during the second half of the 20th century. In order that the results here are not simply a reflection of this warming climate, each of the extremes indices was detrended in a similar way to the SSTA. In addition, because of the varying distribution globally of the rainfall indices, RX1day and RX5day were normalized (i.e., each value of the detrended index was calculated by subtracting the mean and dividing by the standard deviation) so that changes can be compared consistently between different regions. For each season between 1951 and 2003 it was possible to relate the temperature and precipitation extremes to a node of the SOM. 2of13

3 [14] Note that seasonal results are used so that the sample size for analysis is sufficiently large to perform robust statistical analysis. In the following results section, while the SOM algorithm is calculated seasonally, the results will not be considered for individual seasons as the samples would be too small to make sound conclusions. In any case, it will be shown in the next section that the distribution of the SOM nodes is reasonably equiprobable between seasons, i.e., there is not a tendency for any pattern to occur only during one or two seasons. 3. Results 3.1. Sea Surface Temperature Patterns [15] The SSTA patterns produced by the SOM algorithm are shown in Figure 1. Clearly the SOM is able to distinguish the modulating influence of ENSO moving from a strong La Niña phase through to a strong El Niño phase with weaker or neutral phases in between. The first node (strong La Niña) appears most frequently within the SOM although all nodes are approximately equally distributed. Note also that the SOM nodes are reasonably well distributed throughout the seasons, e.g., SOM node 1 occurs 21, 17, 22 and 24 times between 1870 and 2006 in DJF, MAM, JJA and SON respectively. As we move through the nodes of the SOM, the similarity of the patterns to a typical La Niña pattern weakens with the middle nodes (4 and 5) indicating more neutral global SST conditions. The nodes were compared to the independent NINO4 index during the time period when this index is available (Figure 1). For each occurrence of a node, the average seasonal value of the NINO4 index was obtained. It is clear from the probability distribution functions (PDFs) of these values (see Figure 1) that the strong La Niña pattern (node 1) is almost always associated with a negative NINO4 index and conversely the strong El Niño pattern (node 8) has almost always positive NINO4 index values. The intermediate patterns (nodes 3 6) are less strongly associated with one or other of La Niña or El Niño and appear to have greater variability in ocean basins other than the tropical Pacific, e.g., the North Atlantic in nodes 3 and node 4. Interestingly nodes 3 and 4 have a distinctive bimodal distribution of NINO4 values (Figure 1). Each seasonal SSTA pattern was compared visually with the node assigned to it by the SOM algorithm. In all cases, it was found that the key features of the SSTA pattern, for example, the presence or absence of an ENSO signal, were adequately reflected by the assigned SOM node. [16] Because it was possible to link each seasonal SSTA pattern to a node, we were able to determine whether the differences between the nodes were statistically significant. To do this, seasonal SSTA from each one degree grid box assigned to each node were compared to every other node by means of a Kolmogorov-Smirnoff (K-S) test. This test can determine whether two samples are likely to come from the same overall distribution. The null hypothesis that the two samples come from the same distribution was tested at the 5% level. Figure 2 shows a selection of the results when comparing different nodes. The results show large coherent areas of statistically significant differences between the majority of nodes. Not surprisingly the tropical Pacific emerges as the most common region with significant differences between the nodes, most strongly evidenced by the difference between nodes 1 and 8 (see Figure 2a) although there is also a strong signature across most of the Pacific basin and the Indian Ocean. What is also apparent from Figure 2a is that SST anomalies during the strong El Niño are larger in absolute terms than the SST anomalies during the strong La Niña node. However, comparison of some other nodes does not exhibit such strong statistically significant differences. Figure 2b shows that there are few regions where there are coherent differences between node 4 and node 5. In addition, there are regions other than the tropical Pacific where significant differences occur. Figure 2c shows that while there is an ENSO-like signature in the difference between nodes 3 and 6, the North Atlantic, North Sea and Yellow Sea also exhibit large variability between these nodes. While this is possibly an artifact of the SOM analysis, there may be plausible reasons for a physical link between the Pacific SSTA and extratropical SST variability. For example, observational studies such as Allan et al. [2009] found links between a Rossby-like wave train in MSLP, a mid-atlantic SST anomaly and an ENSO-like signal when looking at the effect of natural variability on extremes (in this case severe storms) over the UK. Some modeling studies suggest that SSTs in the tropics and extratropics are linked via atmospheric teleconnections, e.g., Lau and Nath [1996] suggest an atmospheric bridge with perturbations in temperature, humidity and wind fields inducing changes in the latent and sensible heat fluxes across the air-sea interface while Liu and Alexander [2007] suggest that tropical Pacific SSTA and extratropical climate variability are related through the interactions of stationary atmospheric waves with midlatitude storm tracks along with other oceanic teleconnections. [17] Figure 3 shows how the occurrence of each SOM node has changed through time. Each SSTA pattern has occurred throughout most of the 137 year period analyzed although some patterns are much more bunched in some periods than are some other patterns. In the latter part of the 19th century the preferred SST state reflected weak La Niña conditions but with also quite a few episodes of strong El Niño events. Interestingly there is at least a 35-year period in the early 1900s when nodes 3 and 4 did not occur at all in HadISST1 whereas these modes have been the predominant SSTA variability modes in the last decade or so. The first half of the 20th century was classified by weaker El Niño regimes which switched to a predominance of either neutral SSTA or strong La Niña regimes between about 1950 and The results over the second half of the 20th century appear to agree broadly with those of Leloup et al. [2007a]. Figure 1. Seasonal SSTA patterns ( nodes ) from HadISST1.1 derived using Self-Organizing Maps (SOM). Each node is shown alongside the probability distribution function of the associated NINO4 index value for the seasons where this node occurs. The number of times n that each node appears within the SOM between 1951 and 2006 when the NINO4 index is also available is shown above each map. Color bar units are anomalies from the reference period in C. 3of13

4 Figure 1 4of13

5 Figure 2. Colors show SST regions from HadISST1 where differences in SSTA s ( C) between 1951 and 2006 for selected SOM nodes are statistically significant at 5% level using a Kolmogorov-Smirnoff (K-S) test. Differences are plotted as absolute values such that positive (negative) values represent where the anomaly from the first named node are larger (smaller) than the second named node for (a) node 1 and node 8, (b) node 4 and node 5, and (c) node 3 and node 6. Note that color bar values change between the diagrams. 5of13

6 Figure 3. Time series of seasonal SSTA from HadISST1 as categorized by the eight SOM nodes defined in Figure 1. Each symbol represents a different season. Lines are shaded blue where at least three consecutive seasons most closely resembled node 1 (strong La Niña) and orange where they most closely resembled node 8 (strong El Niño). [18] Now that the dominant patterns of global SSTA variability have been classified this information can be used to determine how the various nodes are associated with global patterns of extreme temperature and precipitation Linking SSTA Patterns to Global Extremes [19] Table 1 shows globally averaged deviations from the expected values of the various indices used in this study during each of the SOM nodes. Because the expected value of the globally averaged temperature indices is 10% (approximately 9 days per season), a value of 1 in the table represents an increase of approximately 1 day per season. From Table 1 we can see that there is an increase (decrease) of warm nights (TN90p) of around 1 day per season during strong El Niño (La Niña) events represented by node 8 (node 1). A similar increase in this index is also seen in node 4, a node that is also associated with a 1 day per season increase in warm days (TX90p) and a decrease in cool nights (TN10p). These three nodes, associated with the largest deviations in the minimum temperature indices (TN10p and TN90p), are associated with smaller deviations in the maximum temperature indices (TX10p and TX90p). The warm nights index (TN90p) is more sensitive to differences between nodes than cool nights (TN10p), as evidenced by the relative magnitudes of the deviations for the two indices. Similarly, in nearly all cases warm days (TX90p) are more sensitive than cool days (TX10p). No nodes are linked with large global deviations in the precipitation indices (RX1day and RX5day) although there are more wet extremes during La Niña (node 1) and more dry extremes during El Niño (node 8), similar to the well known effect of ENSO on average rainfall. [20] Table 1 shows that there is not a smooth transition between the nodes and the globally averaged deviations of extremes associated with them. This is because global averages mask some of the strong differing regional influences of each of the nodes. Figure 4 presents maps of how the regional response of TN10p (cool nights), TN90p (warm nights) and RX1day (maximum 1-day rainfall) varies during each of the eight SOM nodes. The response of cool nights and warm nights are similar in most regions (note that the color bar has been reversed between the two indices so that yellow/red indicates warming and green/blue indicates cooling). For both TN10p and TN90p, which represent cool and warm nighttime temperatures respectively, there are clear regional effects of SSTA variability. Much of the globe shows a cooling of temperature extremes during strong La Niña phases (node 1) except for the eastern United States and north east Russia. Southeastern Australia also shows a slight warming during this phase. Australian temperature and rainfall are significantly correlated with ENSO and each other but there are complex interactions and regional differences [e.g., Jones and Trewin, 2000; Nicholls, 2003; Nicholls et al., 1996; Power et al., 1998]. [21] The strong El Niño phase (node 8) is associated with warmer than usual extreme minimum temperatures and almost the opposite pattern of warming and cooling to node 1. The remaining patterns also exhibit large coherent regions of warming and cooling although generally not as large as the most extreme ENSO phases. However, perhaps what is most striking is that neutral or weak ENSO phases have almost as strong links to global climate extremes as do the very strong positive or negative phases of ENSO. For instance, a very weak La Niña like or neutral SSTA pattern (node 4) is associated with warming in most of the global land areas and globally averaged has the same global extreme minimum temperature warming as the strong Table 1. Globally Averaged Deviations From Expected Value of Each Extreme Index for Each SSTA SOM Node From Figure 1 a Node TN10p TN90p TX10p TX90p RX1day RX5day a Values for temperature indices (TN10p, TN90p, TX10p, and TX90p) are deviations from expected value (i.e., 10 %). Values for precipitation extremes are numbers of standard deviations from expected value (i.e., average value of index from 1951 to 2003). 6of13

7 Figure 4 7 of 13

8 El Niño (Table 1). The regional variations between node 4 and node 8 are quite different however; the above average extreme warming across much of Russia in node 4 is countered by quite strong below average extreme cooling across the same region in node 8. Results for node 4 are likely related to the warm North Atlantic temperatures that are associated with this weak La Niña like pattern (Figure 1) and as discussed above there may be reasons to believe that the two SST regions are related by teleconnections. In general, node 4 is not associated with as strong a warming or cooling as node 8 although parts of southeast Asia, India, northwest Australia and southern Africa represent the equivalent of over 3 days increase in the number of warm nights in these regions during this neutral/weak La Niña pattern. Similarly, nodes 1 and 6 are associated with a general global cooling of warm nighttime temperatures. In particular, node 6, which relates to a weak El Niño pattern, is associated with substantial nighttime cooling across Europe and northwest Russia. While the patterns of warming and cooling are similar between cool nights and warm nights, in general the upper end of the minimum temperature distribution (TN90p) appears to have a slightly stronger response to SST variability than the lower end of the minimum temperature (TN10p) distribution. Figure 5 shows the results for the maximum temperature indices. It is clear that maximum temperature extremes show similar patterns of regional response as minimum temperature extremes (Figure 4) for each of the SOM nodes. However, in the Southern Hemisphere maximum temperature extremes seem to be more sensitive to SST variability or vice versa than are minimum temperature extremes. [22] For the precipitation index, RX1day, the patterns are not as coherent as the temperature indices during each of the SSTA phases but there are still some large regions where the SSTA pattern has a clear effect. Increases in RX1day, which represents the wettest day in a season, are likely to indicate increased chances of flooding. The strongest increases are mostly associated with nodes 6 to 8 (representing weak to strong El Niño events), particularly over Europe during nodes 6 and 7, countries adjacent to the Caspian sea during nodes 7 and 8 and western USA, southern Brazil and southeast China during node 8 (strong El Niño). The response of RX1day is generally opposite in these regions during La Niña like patterns. The results for the other precipitation index, RX5day, which represents the five wettest consecutive days in a season (see Figure 5) are very similar to RX1day indicating the strong relationship between these two indices. [23] Are any of these differences between the nodes statistically significant? A K-S test was again used to compare the data from every node with the data from every other node. The results indicate that there are large areas where different nodes exhibit statistically significantly differences. Figure 6 shows the results when comparing node 1 with node 8 and node 3 with node 6 for the temperature indices. Statistically significant differences are apparent in how maximum and minimum temperature and precipitation extremes respond to node 1 (strong La Niña) compared to node 8 (strong El Niño). For minimum temperature extremes the most significant differences are seen in North America, northern Russia, India, southern South America, southern Africa and Australia (for warm nighttime temperatures). These regions also show a similar response for maximum temperature extremes apart from South America where the differences between node 1 and node 8 are not significant. Most of the USA and Mexico and parts of Russia have significantly warmer maximum temperature extremes during strong La Niña events than strong El Niño events but the opposite is true for most of Canada, Australia, southeast Asia, India and southern Africa. This appears to be similar to the response of average temperature in these regions, e.g., Power et al. [1998], Kiladis and Diaz [1989], Rasmusson and Carpenter [1982], and Deser and Wallace [1987, 1990]. In addition, the warmest minimum and maximum temperatures in Siberia and Australia appear to have a stronger response to varying SSTs than the coldest minimum and maximum temperatures. Although different data and seasons have been used, these results appear to agree well with the results of Kenyon and Hegerl [2008]. [24] Figure 7 shows the statistically significant differences between node 1 and node 8 for one of the precipitation indices. The results show significantly wetter 5-day precipitation events during strong El Niño events compared to strong La Niña events in southern Brazil, southwestern United States, southern New Zealand, central Asia and eastern China with an opposite signature in northern New Zealand, southern Canada and northeastern Russia. Maximum 1-day rainfall (RX1day) shows almost exactly the same response (not shown). [25] Figure 6 also shows that it is not just the opposite phases of ENSO that can exhibit significant differences. The most marked results indicate that both the warm and cold tails of maximum and minimum temperature are significantly warmer during node 3 than node 6 in Scandinavia and northwestern Russia. For warm days this signature is also seen in southern USA and Central America. Southeastern Australia shows the opposite signature, i.e., warm maximum temperature extremes are significantly cooler during node 3 than node 6. Figure 3 indicates that node 3 has been much more prevalent in the last few decades than node 6 and given that much of the differing behavior between these nodes has been seen in northern high-latitude regions this could indicate that other external influences in addition to SST variability are dominating in these regions. There are few regions where precipitation extremes indicate significant opposite responses between these two nodes (not shown). [26] It is clear from the results that maximum and minimum temperature and precipitation extremes can exhibit significant opposite behavior in certain regions, associated Figure 4. Anomalies are shown for cool nights (TN10p), warm nights (TN90p), and maximum 1-day precipitation totals (RX1day) during each SOM node. Units for temperature indices are percentage deviations from expected value, i.e., 10%, and precipitation extremes are shown as the normalized deviations (by removing the mean and dividing by the standard deviation) from average. Colors are shown such that green/blue (yellow/red) reflect a cooler/wetter (warmer/drier) climate. Anomalies are not shown for areas where each HadEX index has less than 80% available data. 8of13

9 Figure 5. As in Figure 4, but for cool days (TX10p), warm days (TX90p), and maximum consecutive 5-day precipitation totals (RX5day). 9 of 13

10 Figure 6 10 of 13

11 Figure 7. As in Figure 6, but only showing statistically significant differences between SOM node 1 and node 8 for RX5day. The color bar is presented such that green/blue (yellow/red) indicates that the node named first in the title is wetter (drier) than the node named second. Units are in mm/day. with the modulation of global SST variability by the El Niño Southern Oscillation. 4. Concluding Remarks [27] In this study we defined a set of eight SSTA patterns using the technique of Self-Organizing Maps which we found adequately reflected the variability of global seasonal SST anomalies associated with the El Niño Southern Oscillation and other large-scale circulation modes. These patterns were analyzed from 1870 to 2006 and showed that there have been periods of preferred SST states over this time, with a tendency in more recent decades toward more strong El Niño like states. This agrees with the results of other studies such as the study of Leloup et al. [2007a]. However, this trend appears more strongly here than in some other indices associated with the El Niño Southern Oscillation [Nicholls, 2008]. The occurrence of each SST pattern was compared with an independent global land data set containing extremes of temperature and precipitation between 1951 and Results showed that during SST states with strong El Niño like conditions, temperature extremes were much warmer when averaged globally and particularly for warm minimum temperature extremes. Conversely, during SST states which exhibited strong La Niña like conditions, temperature extremes were much cooler than general when averaged globally, again most marked in warm minimum temperature extremes. Precipitation extremes did not show such marked variations but globally there were more dry (wet) extremes during strong El Niño (La Niña). The response of precipitation and temperature extremes to different SSTA patterns, while coherent across large regions, was not the same everywhere. There were statistically significant differences in how temperature and precipitation extremes responded in some regions to different SSTA patterns of variability. [28] It is important to note that the results shown here have been calculated on detrended data sets. This is important as the results only reflect the response of temperature and precipitation extremes to natural SST variability. The response is generally larger in the temperature extremes when the observed warming globally is retained. Figure 8 shows the result for warm nights when comparing node 1 with node 8 and node 3 with node 6 while retaining the trends in the land-based temperature extremes. It is clear when comparing these results with the results in Figure 6 that not only is the spatial extent of the significant differences greater but also the magnitude of the differences is larger. This is particularly true of minimum temperature extremes and particularly true for the response between the interim nodes. This would imply that minimum temperature extremes may be driven more by external anthropogenic forcing than natural climate variability (assuming that anthropogenic warming was linear). However, several studies have shown that it is the upper extremes of both maximum and minimum temperatures that appear to have a stronger response to anthropogenic forcing than the lower extremes of both variables, e.g., Kiktev et al. [2003, 2007] and Hegerl et al. [2004]. [29] Precipitation extremes in this study showed a stronger response when observed trends were removed (not shown). There were very significantly wetter 1 and 5-day Figure 6. Colors represent regions from HadEX where percentage differences (see Table 1) in the temperature indices from this study between SOM (left) node 1 and node 8 and (right) node 3 and node 6 are statistically significant at 5% level using a Kolmogorov-Smirnoff (K-S) test. Color bars are presented such that green/blue (yellow/red) indicate that the node named first in the title is cooler (warmer) than the node named second. 11 of 13

12 Figure 8. As in Figure 6 results for warm nights only without detrending land-based temperature extremes data. Note differences in color bar from Figure 6. precipitation events during strong El Niño events compared to strong La Niña events in southern Brazil, southwestern United States and to the east of the Caspian Sea. [30] A test of global climate models would be whether they could adequately simulate the significantly different responses of climate extremes to SST variability. However, Leloup et al. [2007b] recently showed that none of the global climate models in the IPCC-AR4/CMIP3 database were sufficiently good at capturing all aspects of the extent, location and timing of ENSO events. This would be a concern for our confidence in the future projections of both global, e.g., Tebaldi et al. [2006], and regional, e.g., Meehl and Tebaldi [2004], Sillmann and Roekner [2008], and Alexander and Arblaster [2009], temperature and precipitation extremes given their dependence on the modulation of SSTA and may have serious implications for the use of global climate models for impacts assessments for regions where ENSO-related variability has an important influence on climate extremes. However, perhaps the net warming estimated from global climate model projections may considerably outweigh the extent to which the models over- or underestimate ENSO-related variations and the greater uncertainty may be related to the extent of the overall warming rather than the models ability to adequately simulate SST variability. [31] Acknowledgments. This research, and N.N., were partly supported by the Australian Research Council through Discovery Project DP We are grateful to the three anonymous reviewers who helped improve this manuscript. References Alexander, L. V., and J. M. Arblaster (2009), Assessing trends in observed and modelled climate extremes over Australia in relation to future projections, Int. J. Climatol., 29, , doi: /joc Alexander, L. V., et al. (2006), Global observed changes in daily climate extremes of temperature and precipitation, J. Geophys. Res., 111, D05109, doi: /2005jd Allan, R., S. Tett, and L. Alexander (2009), Fluctuations in autumn-winter severe storms over the British Isles: 1920 to present, Int. J. Climatol., 29(3), , doi: /joc Cassano, J. J., P. Uotila, and A. H. Lynch (2006), Changes in synoptic weather patterns in the polar regions in the 20th and 21st centuries: Part 1. Arctic, Int. J. Climatol., 26(8), Deser, C., and J. M. Wallace (1987), El Niño events and their relation to the Southern Oscillation: , J. Geophys. Res., 92, Deser, C., and J. M. Wallace (1990), Large-scale atmospheric circulation features of warm and cold episodes in the tropical pacific, J. Clim., 3, Easterling, D. R., G. A. Meehl, C. Parmesan, S. A. Changnon, T. R. Karl, and L. O. Mearns (2000), Climate extremes: Observations, modeling, and impacts, Science, 289, Halpert, M. S., and C. F. Ropelewski (1992), Surface temperature patterns associated with the Southern Oscillation, J. Clim., 5, Hegerl, G. C., F. W. Zwiers, P. A. Stott, and V. V. Kharin (2004), Detectibility of anthropogenic changes in annual temperature and precipitation extremes, J. Clim., 17, Jones, D. A., and B. C. Trewin (2000), On the relationships between the El Niño Southern Oscillation and Australian land surface temperature, Int. J. Climatol., 20, Kenyon, J., and G. C. Hegerl (2008), Influence of modes of climate variability on global temperature extremes, J. Clim., 21, Kiktev, D., D. M. H. Sexton, L. Alexander, and C. K. Folland (2003), Comparison of modelled and observed trends in indices of daily climate extremes, J. Clim., 16, Kiktev, D., J. Caesar, L. V. Alexander, H. Shiogama, and M. Collier (2007), Comparison of observed and multimodeled trends in annual extremes of temperature and precipitation, Geophys. Res. Lett., 34, L10702, doi: /2007gl Kiladis, G. N., and H. F. Diaz (1989), Global climatic anomalies associated with extremes in Southern Oscillation, J. Clim., 2, Kohonen, T. (2001), Self-organizing maps of massive databases, Eng. Intell. Syst. Electr. Eng. Commun., 9, Kunkel, K. E. (2003), North American trends in extreme precipitation, Nat. Hazards, 29, Lau, N. C., and M. J. Nath (1996), The role of the atmospheric bridge in linking tropical Pacific ENSO events to extratropical SST anomalies, J. Clim., 9, Leloup, J. A., Z. Lachkar, J. P. Boulanger, and S. Thiria (2007a), Detecting decadal changes in ENSO using neural networks, Clim. Dyn., 28, Leloup, J., M. Lengaigne, and J. P. Boulanger (2007b), Twentieth century ENSO characteristics in the IPCC database, Clim. Dyn., 30, Liu, Z. Y., and M. Alexander (2007), Atmospheric bridge, oceanic tunnel, and global climatic teleconnections, Rev. Geophys., 45, RG2005, doi: /2005rg McBride, J. L., and N. Nicholls (1983), Seasonal Relationships between Australian Rainfall and the Southern Oscillation, Mon. Weather Rev., 111, Meehl, G. A., and C. Tebaldi (2004), More intense, more frequent and longer lasting heat waves in the 21st century, Science, 305, Nicholls, N. (2003), Continued anomalous warming in Australia, Geophys. Res. Lett., 30(7), 1370, doi: /2003gl Nicholls, N. (2008), Recent trends in the seasonal and temporal behaviour of the El Niño Southern Oscillation, Geophys. Res. Lett., 35, L19703, doi: /2008gl Nicholls, N., and L. Alexander (2007), Has the climate become more variable or extreme?, Prog. Phys. Geogr., 32, Nicholls, N., et al. (2005), The El Niño Southern Oscillation and daily temperature extremes in east Asia and the west Pacific, Geophys. Res. Lett., 32, L16714, doi: /2005gl Nicholls, N., B. Lavery, C. Frederiksen, W. Drosdowsky, and S. Torok (1996), Recent apparent changes in relationships between the El Niño Southern Oscillation and Australian rainfall and temperature, Geophys. Res. Lett., 23, Power, S., F. Tseitkin, S. Torok, B. Lavery, R. Dahni, and B. McAvaney (1998), Australian temperature, Australian rainfall and the Southern Oscillation, 12 of 13

13 : Coherent variability and recent changes, Aust. Meteorol. Mag., 47, Rasmusson, E. M., and T. H. Carpenter (1982), Variations in tropical sea surface temperature and surface wind fields associated with the Southern Oscillation/El Niño, Mon. Weather Rev., 110, Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan (2003), Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century, J. Geophys. Res., 108(D14), 4407, doi: / 2002JD Ropelewski, C. F., and M. S. Halpert (1986), North American precipitation and temperature patterns associated with the El Niño/Southern Oscillation (ENSO), Mon. Weather Rev., 114, Ropelewski, C. F., and M. S. Halpert (1987), Global and regional scale precipitation patterns associated with El Niño/Southern Oscillation, Mon. Weather Rev., 115, Scaife, A. A., C. K. Folland, L. V. Alexander, A. Moberg, and J. R. Knight (2008), European climate extremes and the North Atlantic Oscillation, J. Clim., 21, Sillmann, J., and E. Roekner (2008), Indices for extreme events in projections of anthropogenic climate change, Clim. Change, 86, Tebaldi, C., K. Hayhoe, J. M. Arblaster, and G. A. Meehl (2006), Going to the extremes: An intercomparison of model-simulated historical and future changes in extreme events, Clim. Change, 79, Wang, X. L. L., V. R. Swail, F. W. Zwiers, X. B. Zhang, and Y. Feng (2009), Detection of external influence on trends of atmospheric storminess and northern oceans wave heights, Clim. Dyn., 32, WASA Group (1998), Changing waves and storms in the northeast Atlantic?, Bull. Am. Meteorol. Soc., 79, L. V. Alexander, Climate Change Research Centre, University of New South Wales, Mathews Building (4th Floor), Kensington Campus, Sydney, NSW 2052, Australia. N. Nicholls and P. Uotila, School of Geography and Environmental Science, Monash University, Building 28, Wellington Road, Clayton, Vic 3800, Australia. 13 of 13