WPCV{ 2BJ4urierRomanTimes Roman Bold\  PCXP3|omanHPLAIIPO.PRSx  @hhhh GX@292 : Z>3|o45MONGBHPLAIIPO.PRSXN\  Phhhh GXPDefault Para66 O2=(2*2W: ]$E%$ A\  P  XN\  P 2k pkAA0]Page Number668 O2=(2 ]B*2W: ]$E$ A\  P ,JR Z bj  XN\  P Footer11W:6N3, ]]O21O2kc21]6N]t216NL1`  XN\  P  R Z bj  XN\  P HeaderЁ]]c2L]iW:RF ]  L]`  XN\  P"  R Z bj  XN\  P# ѫXN\  PXP(9 Z 6Times New Roman RegularXA\  PP\  `$Times NewRomanXN\  PXP\  `$Times NewRomanXA\  PP\  `$Times NewRomanXN\  PXP\  `$Times NewRomanXXN\  PXP\  `$Times NewRomanXXN\  PXP\  `$Times NewRomanXXN\  PXP\  `$Times NewRomanXXN\  PXP\  `$Times NewRomanXXN\  P XP\  `$Times NewRomanXA\  P P\  `$Times NewRomanXN\  P XP\  `$Times NewRomanXXN\  P XP\  `$Times NewRomanXA\  P P(9 Z 6Times New Roman RegularXN\  PXP\  `$Times NewRomanXA\  PP\  `$Times NewRomanXN\  PXP\  `$Times NewRomanXXN\  PXP\  `$Times NewRomanXA\  PP\  `$Times NewRomanXN\  PXP\  `$Times NewRomanXXN\  PXP\  `$Times NewRomanXA\  PP\  `$Times NewRomanXN\  PXP\  `$Times NewRomanXA\  PP\  `$Times NewRomanT\  PP\  `$Times NewRomanXN\  PXP\  `$Times NewRomanXT\  PP\  `$Times NewRomanXN\  PXP\  `$Times NewRomanX[\  PP\  `$Times NewRoman h\  P P\  `$Times NewRoman XN\  PXP\  `$Times NewRomanXT\  PP\  `$Times NewRoman[\  P P\  `$Times NewRomanXN\  P!XP\  `$Times NewRomanXXN\  P"XP\  `$Times NewRomanXXN\  P#XP\  `$Times NewRomanX2)a ##a!%("m^2CRddCCCdq2C28dddddddddd88qqqYzoCNzoozzC8C^dCYdYdYCdd88d8ddddCN8ddddY`(`lC2CC!CCCCCCCCCCd8YYYYYYzYzYzYzYC8C8C8C8ddddddddddYdddddodYYYYYYYdzYzYzYzYddddddddC8C8C8C8Ndz8z8z8z8z8ddddddCCCoNoNoNoNz8z8z8dddddddzYzYzYdz8dCoNz8dddddNF2[dCYddddd7>d<d<$YYdCCddooCYpppppppppp>>~~~cΡ|JWǡ||ӡJ>JipJcpcpcJpp>>p>ppppJW>ppppck-kyJ8JJ%JJJJJJJJJJp>cccccǕcccccJ>J>J>J>ppppppppppcppppp|pcccccccpccccppppppppJ>J>J>J>Wp>>>>>ppppppǡJJJ|W|W|W|W>>>ppppppӡpcccp>pJ|W>pppppNN8epJcppppp>EpCpC(ccpJJpp||JcC~~{{>%~II{{{~܇Y~=n7\܇bn{{{ϡ{\nbntJI{{ptn{{|{bbt{bII{bbbbbbbIIIIIIIIIIII{{{{{{{J08/09/97 15:40 Climatic Extremes: How well can these be estimated Julie Burgess (&230F)Xv)a,a80Times RomanTimes Roman BoldTimes Roman Italic7oC2*o\  PCXP=|J8*U|\  PCP=J8 U4  p(AC<;{J8ڴU{*f9 xCX|>|c|WJpp|p|NN8ypJppppppCJpCpC>pppJJpp||JpC{{>%II{{{܇Y=n7\܇bn{{{ϡ{\nbntJI{{|tn{{|{bbt{bII{bbbbbbbIIIIIIIIIIII{{{{{{{J"m^8J|ppߺJJJp8J8>ppppppppppJJpСWpӡ|ߡJ>JpJp|c|cJp|>J|>|p||cWJ|pppcX1XtJ8JJ%JJJJJJJJJJ|>pppppߡcccccW>W>W>W>|pppp||||pp|pppp|pppcccc|ccccpppppp||W>W>W>W>p|>>>>>||||ppߡccc|W|W|W|WJJJ||||||ߡpccc|>|c|WJpp|p|NN8ypJppppppCJpCpC>pppJJpp||JpC{{>%II{{{܇Y=n7\܇bn{{{ϡ{\nbntJI{{|tn{{|{bbt{bII{bbbbbbbIIIIIIIIIIII{{{{{{{J"m^8J^ppJJJp8J8>ppppppppppJJp͈Jc|p|||W>W^pJppcpc>pp>>c>ppppWW>pcccWY=YyJ8JJ%JJJJJJJJJJp>pppppǕcccccJ>J>J>J>ppppppppp|cpppp|cpppppccccpccccppppppppJ>J>J>J>cc|>|>|>|>|>ppppppӕWWWpWpWpWpW|>|>|>pppppp|c|W|W|Wp|>pWpW|>|c|cpppNN8upWpppppp>EpCpC0||pJJppppJ|Cї{{>%II{{{܇Y=n7\܇bn{{{ϡ{\nbntJI{{ptn{{|{bbt{bII{bbbbbbbIIIIIIIIIIII{{{{{{{J2Da3Z,7a:a="m^!,6CCoh,,,CK!,!%CCCCCCCCCC%%KKK;{`YY`QJ``,4`Qw``J`YJQ``~``Q,%,?C,;C;C;,CC%%C%hCCCC,4%CC`CC;@@H,!,,,,,,,,,,,,C%`;`;`;`;`;wYY;Q;Q;Q;Q;,%,%,%,%`C`C`C`C`C`C`C`C`C`C`;`C`C`C`C`CJC`;`;`;Y;Y;Y;Y;`CQ;Q;Q;Q;`C`C`C`C`C`C`C`C,%,%,%,%4`CQ%Q%Q%Q%Q%`C`C`C`C`C`Cw`Y,Y,Y,J4J4J4J4Q%Q%Q%`C`C`C`C`C`C~``CQ;Q;Q;`CQ%`CY,J4Q%`C`C`C`C`CN/!tbttYtkYbttttb5,5KP5GPGPG5PP,,P,|PPPP5>,PPtPPGM MW5(555555555555P,tGtGtGtGtGkkGbGbGbGbG5,5,5,5,tPtPtPtPtPtPtPtPtPtPtGtPtPtPtPtPYPtGtGtGkGkGkGkGtPbGbGbGbGtPtPtPtPtPtPtPtP5,5,5,5,>tPb,b,b,b,b,tPtPtPtPtPtPtk5k5k5Y>Y>Y>Y>b,b,b,tPtPtPtPtPtPttPbGbGbGtPb,tPk5Y>b,tPtPtPtPtPN8(HP5GPPPPP,2xxP0zzPx0GGP55PPYY5G0ZZXXr,Z55XXXr{rrZ``@Z,rO(Bn``{rrrrF{{{``iOXXXrrrtekX`BbObFbOt`wS55tXnXPtSgOtX{XYX_`FbFn\zStXn{neF`55X\\nec`reeeeeeeeeeeeeeeeeeeFFFFFFF````````````````````555555555555XXXXXXX\\\\\\\\\\\\nnnnnnnnnnnnnnnnnnnntbt5tn{tQcK\  PCP QcK 4  p(ACk(Q1%Q\  PCP"X^Kc*cccKcKScct*cScccScStcvBvcKcc*2ccccccccccS*tStStStStStStStSSSSSS*ttttccc*StcNhKcZbZZS*cc**c*ZS2bb&&7wRJ{&&&7{cb bbbbbbbbbbbbbbc"m^(1<90th (<10th) percentile counts are strongly positively (negatively) correlated with mean annual temperature (not shown). 1779 has the highest number of >90th percentile days (87) while 1814 has the highest number of days below the 10th percentile value (117). The increase in mean temperature in recent decades has not been accompanied by an increase in >90th percentile days (solid curve). In contrast, a marked decline in the number of days <10th percentile value has occurred since the late 19th century. However some caution is needed before 1880; the use of different types of screen before then (or no screen at all before the mid 19th century) may influence these results, probably by increasing the likelihood of large extremes. Bearing this problem in mind, addition of the two series in Figure3 provides an extremes index (Figure4). This can be used to assess whether extremes as a whole have changed in frequency relative to the 1961-90 period. Overall, the frequency of extremes reduces through time. Years with high*p-&+&+ values in Figure4 may be due to warmth (coldness) throughout the year as in 1779 (1814) or years when extremes of both sign occurred but cancelled each other in the annual mean temperature value. For example, in 1947 there were 127 days with extreme warm or cold temperatures, but the annual mean temperature anomaly was only 0.14C. The higher numbers of extremes, relative to 1961-90, before the 1930s is principally due to cooler temperatures in these years and hence greater numbers of very cool days (see Figure3). The results in Figures3 and 4 are moderately insensitive to the choice of base period. Fairly similar results are achieved using the 1931-60 base period. This is not surprising as the mean annual temperatures for these two periods are very similar, 9.6 (1931-60) and 9.5 (1961-90). When the extreme index was recalculated relative to the much colder period of 18811910 there was, by contrast, little change in the average number of days with extreme temperatures with time. 2.2Circulation Typing and Gale Frequency Other important extreme phenomena include very strong winds, which in the extratropical North Atlantic are generally related to deep cyclonic depressions in the westerly winds. Estimates of changes in the occurrence of such events from anemometer records suffer from numerous site and instrumentalchange problems. This makes the development of long homogeneous wind time series difficult. An alternative method of calculating time series of gale statistics is to use gridpoint meansealevel pressure (MSLP) data. The sources of the daily 5 latitude by 10 longitude MSLP gridpoint data used here are described in Jones (1987). The data have been adjusted so that the monthly average MSLP values agree with a recent version of the Global  ^' MSLP data set, updated from that in Allan etal. (1996). We use the adjusted daily MSLP to calculate the strength of the resultant wind flow (F) and the(p-&+&+ vorticity (Z) (see Jenkinson and Collison, 1977, and Hulme and Jones, 1991), yielding a gale index of   ^ 9 9  "AA'G = (F2 + (0.5Z)2)=BQQH!!M(1) While the relatively coarse grid of 5 latitude by 10 longitude and a daily sampling interval might not resolve all small intense cyclonic systems, such features are not very typical in these latitudes, the storm of 16 October 1987 being a notable exception. Figure5 shows the annual number of severe gales for each year over the United Kingdom for the period 1881 to 1996. Asevere gale day is defined for most years as having G=>40. However, owing to the varying methods of chart construction, the threshold value of G is reduced to 37 for 18991939 and 194959 and to 32 for 1960-65 to maintain longterm homogeneity (see Hulme and Jones, 1991, for details). The frequency of severe gales reached a level around 1990, averaged over a decade, higher than that recorded previously, though individual years at earlier times give the maximum values of G in the time series (e.g. 1887, 1916). The high recent level of gale activity over the United Kingdom, and in adjacent regions of similar latitude, is related to the very strong westerly phase of the North Atlantic Oscillation during  ^ the late 1980s and in the early 1990s (Jones etal., 1997b). The use of a gridbased method of estimating changes in gale frequency enables equivalent indices to be derived from the control runs of GCMs. These  ^ are rarely validated at this level of regional detail, Hulme etal. (1993) being an exception. The latter analysis showed that both the GCMs analyzed underestimated gale frequency by up to 50%, particularly during the winter season. Jenkinson and Collison (1977)also used daily mean MSLP data to classify  ^' daily atmospheric circulation objectively (see also Jones etal., 1993) to match the subjectivelyderived Lamb weather types (Lamb, 1972). In the subjective Lamb typing scheme, a strong decline in the number of westerly days (the *p-&+&+ dominant wind direction) was highlighted by Lamb (e.g.Lamb, 1972). With the objective classification scheme of Jenkinson and Collison (1977), however, there is no evidence of a longterm decline (see Figure6a) though multidecadal fluctuations are readily apparent. The objective classification scheme has higher  ^ numbers of southerly days, particularly in autumn and winter (see Jones etal., 1993). Keeping these uncertainties in mind, circulation classification schemes can be used together with temperature and precipitation series to assess whether overall trends, for example to warmer/drier conditions, are due to changes in the frequency of certain circulation types or to changes of temperature/precipitation for specified weather types. For example, a cooling in April CET between 19611987 appears to be due to an increase in blocking and thus more ENE winds. Conversely, on days throughout the year with generally westerly flow  ^ types over UK, temperatures in the Scilly Isles (near 50oN, 6oW) are correlated well with SST anomalies in the North Atlantic (Fig 6b) over five year averages (adapted from Parker and Folland, 1988). This diagram is striking in showing the existence of appreciable multidecadal variations in the mean temperature of days with specific circulation types over the UK. It is the kind of within circulation type diagnostic that should be extended to studies of extremes. The two air flow characteristics, F and Z also have considerable potential  ^ for use in GCM down scaling (Conway etal., 1996). However, calculations over diverse regions of Europe (Conway and Jones, 1996) for a run of the Hadley Centre (HADCM2) model with projected increases in greenhouse gases and changes in anthropogenic sulphate aerosols indicate no change in either F or Z, and hence no increase in gale frequency over Europe. Model projections of changes in storminess on larger scales such as the North Atlantic are beginning  ^' to be assessed (e.g. Carnell etal., 1996).'p-&+&+  ^  3.Global Temperature  ^  In this analysis, we use monthly 5x 5 gridbox temperature anomalies  ^ derived from land surface air (Jones, 1994) and sea surface (Parker etal., 1995) temperatures. The latter uses the (almost) non interpolated MOHSST6. The gamma function distribution method used for the CET analysis enables each monthly, seasonal or annual gridbox temperature to be transformed to a percentile. This allows a direct comparison of the unusualness of each gridbox value with respect to distributions based on 1961-90 averages. Figure7 shows the annual percentile map for 1996 together with the more usual anomaly form for comparison. Many lowlatitude oceanic regions, parts of Mexico and South West USA, and scattered regions elsewhere were warmer than the 98th percentile (Horton and Parker, 1997). 9.8% of the monitored area of the globe exceeded this percentile in 1996, showing that recent global warming is widespread. Figure8 shows a time series for 1951-96 of the global areas > 90% value and <10% value. These series can be calculated in several different ways of which we have done two (Figure 8): a)fitting gamma distributions to annual average temperatures or b)fitting gamma distributions to the monthly temperatures and averaging the resulting areas >90% value or <10% value for each of the 12 months. The two different methods of calculation give different results. Method a)will stress lower frequency anomalous temperatures influencing a region on the annual timescale while b)highlights higher frequency anomalies (and, in addition, any appreciable random errors or very unrepresentative, sparse data). So method b gives inherently more variable results. Thus anextremely variable year globally, but whose anomalies cancel when averaged to a whole year, will be reflected in relatively large areas covered by extreme percentiles in method (b)but a considerably smaller area covered in method a). Conversely, a year where the individual months are not( p-&+&+ extreme but most tend to be warm or cold will give a larger area of extreme percentiles covered by method a. Figure8 indicates that both methods show increases in the monitored areas of the world having warm extremes and decreases in areas with cool extremes. However, calculation of the area index using annual data (method a) leads to greater longterm variation in the areas covered by extreme percentiles than using the average of monthly calculated areas (method b). This implies that warming on the truly annual timescale is very widespread. Ina similar manner to calculations for CET (Figure 4), the sum of both warm and cold extremes provides an extremes index (Figure9). This shows slightly higher levels of extremes before 1960 and after the mid1980s. The muted variations in the extremes index relative to its components in Fig 8 reflect the fact that cold extremes are decreasing at the same time as warm extremes have increased. Thus temperatures worldwide do not show any appreciable tendency to become more variable in very recent decades but there is a marked tendency for an increased frequency of warm extremes which mainly reflects a general averaged warming.  ^  Conclusions  The 3parameter gamma distribution provides a useful method of analysing not only long daily temperature series but also monthly gridded 5x5 grid box temperatures. The method enables extremes in both types of data to be studied without resort to arbitrary thresholds and takes account of differences in variability. The distribution enables the rarity of a particular day or months temperature to be specified as a percentile value or alternatively as a return period. Use of the maximum likelihood version of the method is to be preferred because in principle biases due to the use of a rather short (30 year) fitting period are eliminated especially in the presence of appreciable skewness (e.g. as on the monthly timescale over continents in winter).* p-&+&+ԌAlthough the main aim of this paper has been exploratory, we find that  ^ the statistical structure of CET and the global scale monthly gridded 5o x 5o temperatures has changed over the last 10 to 15 years as temperatures have risen. Slight increases in warm extremes are evident in th CET series, but the larger impact on extremes has been a reduction in the number of extremely cool days. Traditional analyses of daily and monthly temperatures using thresholds are likely to miss these subtle changes. Our work also hints that linking the analysis of temperature (or other parameters) to that of regional atmospheric circulation variations may produce potentially insightful analyses of changes in the mean and extremes of climatic variables within weather types. This may be an especially good way of studying the emerging local signals of humaninduced climate change.  ^  Acknowledgements  This work has been supported by the UK Dept. of Environment (EPG1/1/16), and the United States Dept. of Energy, Atmospheric and Climate Research Division under grant No.DE-FG02-86ER60397. We would also like to acknowledge the help of Matthew O'Donnell.  ^   X` hp x (#%'0*,.8135@8:dp-++ xd++>ԯ  ^x ԇ x! 4 <DL!N9 AIQ! Measurep-++  x! N9 AIQ!4 <DL!Periodp-++  x! Correlation with temperature> p-++ x++>ԯ   4 <DL!N9 AIQ! Days below 0C p-++   N9 AIQ!4 <DL! 17721996 p-++    -0.90> p-++  ++>ԯ   4 <DL!N9 AIQ!Degree days below 0Cp-++   N9 AIQ!4 <DL!17721996p-++   -0.84>p-++ ++>ԯ  ^ ԇ  4 <DL!N9 AIQ!Winter snow index 1 p-++   N9 AIQ!4 <DL!18761995p-++   -0.65>p-++ ++>ԯ  ^ ԇ  **Number of snowy months 1 p-++   18761995p-++   -0.44>p-++ ++>ԯ  c ԇ  4 <DL!,$|, 4<D! T\  P*C  Summers XN\  P*  p-++   ,$|, 4<D!4 <DL!p-++   >p-++ ++>ԯ   4 <DL!N9 AIQ!Days greater than 20Cp-++   N9 AIQ!4 <DL!17721996p-++   0.73>p-++ ++>ԯ   4 <DL!N9 AIQ!Degree days above 20Cp-++   N9 AIQ!4 <DL!17721996p-++   0.63>p-++ ++>ԯ ^ ԇ  4 <DL!9 AIQ! :  19 9  For definition see Jones etal. 1997a. These correlations with snowiness are slightly stronger if the winter temperature definition is extended to November to April  ^g! (Jones etal., 1997a).g!p-++ p-++ >p-++ i"++>ԯ          9 AIQ!4 <DL!i$p-++    [\  PN   ru   h\  P Climatic Extremes: Approaches to creating indices from daily and monthly data  XN\  P*   c  T\  P*C by  c3 P.D. Jones1  cG E.B. Horton2  c[ C.K. Folland2  co M. Hulme1  c D.E. Parker2 and  c T.A. Basnett2 C:lC I"҇  c# 4 <DL!9 AIQ!1Climatic Research Unit  University of East Anglia  Norwich NR4 7TJ  UK 9 AIQ!4 <DL!s/_-_- 8/_-_-_-_-8ԯ 4 <DL!9 AIQ!  c! 2Hadley Centre  Meteorological Office  Bracknell RG12 2SY  UK 9 AIQ!4 <DL!&/_-_- 8/_-_-''_-_-8ԯ September 1997 [\  P N  XN\  P!*