1988 NOTES AND CORRESPONDENCE 109A Cyclone Climatology of the Canadian Climate Centre General Circulation Model STEVEN J. LAMBERTCanadian Climate Centre/CCRN, Downsview, Ontario, Canada10 December 1986 and 29 July 1987 ABSTRACT The cyclone event climatology is presented for a five-year simulation by the Canadian Climate Centre generalcirculation model. Winter season results are given for the extratropical regions of both the Northern and theSouthern hemispheres and the results are compared to an event climatology based on five years of observed data. In the Northern Hemisphere, classification of the simulated and the observed 1000 mb lows as a function ofcentral height shows that the model has a deficiency of weak cyclones and an excess of intense cyclones Whilethe number of cyclones of medium intensity is well simulated. The geographical distribution of the cycloneevents shows that those features of the observed climatology which are thermal in nature, such as a preferencefor lows to form and move over relatively warm underlying surfaces, are reasonably well simulated, but thetopographical features such as lee cyclogenesis and the avoidance of high terrain by lows are poorly simulated. The Southern Hemisphere also shows a deficiency of simulated cyclones, although less pronounced than theNorthern Hemisphere. The number of weak and strong cyclones are well simulated, and the paucity of SouthernHemisphere lows results from a deficiency of cyclones of moderate intensity. The observed climatology, whichis nearly free of mountain influences, is well simulated and only a few minor problems are evident, such as apoor simulation of the winter cyclogenesis in the lee of the Andes and a tendency for model cyclones to bepresent over high terrain.1. Introduction Cyclones are important features of the observed climate. In the extratropics, much cloud and precipitationis associated with low pressure systems. Consequently,it is important for general circulation models (GCM)to simulate their behavior correctly, especially to assessthe results of climate change experimems on a regionalbasis. Observational studies of cyclones show that they arenot uniformly distributed over the earth's surface buttend to occur in preferred geographical regions. Petterssen (1956) used once-daily analyses for the period1899 to 1939 to determine the frequency of cyclonecenters in 100 000 km2 areas during winter and summer for the extratropical Northern Hemisphere. Klein(1957) used data for the same 40-yr period to determinethe number of days with cyclones for all 12 months infive degree by five degree latitude-longitude quadrangles. These studies also showed that cyclones tend tooccur in highly elongated zones or "storm tracks." Itwas also observed that lows tended to move over relatively warm underlying surfaces and that cyclones hada life span of four to five days. The effect of topographywas also evident since lows were observed to avoid high Corresponding author address.' Dr. Steven J. Lambert, CanadianClimate Center, 4905 Dufferin St., City of North York, Downsview,Ontario, Canada M3H 5T4.terrain and that the lee of mountains was a favoredarea of cyclogenesis. A Southern Hemisphere cyclone climatology wasproduced by Taljaard (1967) who used the data fromthe International Geophysical Year (July 1957 to December 1958). These results were somewhat differentin character from those of the Northern Hemisphere.The main feature of the Southern Hemisphere climatology is a circumpolar zone of high cyclone frequenciessurrounding Antarctica during winter, summer, andthe intermediate seasons. During the winter two principal storm tracks originating in midiatitudes mergewith the circumpolar maximum. One of these originates between Australia and New Zealand and theother over South America. Subsequent cyclone studies have tended to be regional in scope. Zishka and Smith (1980) and Whittaker and Horn (1981), for example, showed that cyclone frequencies over North America decreased significantly during the period 1950-77; and in theSouthern Hemisphere, an investigation ofcyclogenesisin the lee of the Andes by Chung (1977), based on thesea level pressure analyses of the Argentinian Meteorological Service for the year 1973, revealed three favored areas of lee cyclogenesis at 25-, 31- and 55-S. The behavior of cyclones in numerical forecastmodels has also been studied in some detail. Leary(1971) examined the 36-h sea level pressure forecastsproduced by the six-layer primitive equation model ofc 1988 American Meteorological SocietyI10 JOURNAL OF CLIMATE VOLUMEthe National Meteorological Center (NMC) for the1969/70 winter season. Her results showed that cyclones over the oceans were not forecast deep enoughand that those forming to the lee of the Rocky Mountains were forecast too deep. Model cyclones were alsoobserved to intensify less rapidly find dissipate morequickly than those of the real atmosphere. Also evidentfrom this study was the fact that forecast lows tendedto move too slowly and that rapidly deepening stormstended to move to the left of the forecast track. Silberberg and Bosart (1982) studied systematic cycloneerrors in the NMC LFM-II (limited-area fine mesh)model during the period I October 1978 to I May1979. Their results also showed that cyclones were not. forecast deep enough over the oceans and too deep tothe lee of the Rocky Mountains. The phase speeds ofthe lows in October and November were underforecast,in March and April they were overforecast, and nosystematic speed errors were evident in December,January and February. A recent study by Akyildiz (1985) examined the behavior of cyclones in the European Centre for MediumRange Weather Forecasts (ECMWF) 1.875 by 1.875degree grid-point model and the triangular 63 wavespectral model during the 1983/84 winter. In the gridpoint model, lows moved too slowly and their deepening and filling rates were underestimated. In thespectral model, the systematic slowness was evidentonly for rapidly moving cyclones. The rate of deepeningwas reasonably well forecast but the rate of filling wasstill underestimated. For both forecast models, thehorizontal scale of cyclones in the early stages of development was too small and for mature cyclones itwas too large. Manabe and Terpstra (1974) presented NorthernHemisphere cyclone trajectories for a 45-day winterperiod from a simulation by the GFDL GCM describedin Holloway and Manabe (1971). Their results showeda major storm track originating south of Japan andextending to the Gulf of Alaska. There was a secondmajor storm track along the east coast of North America to Greenland, western Europe, and northeast of theCaspian and Aral Seas where the lows dissipated. Thecyclogenetic areas off the east coast of Asia and to thelee of the Rocky Mountains appeared tO be reasonablywell simulated, but the strong cyclogenetic area in theMediterranean was absent. There appear to be no global studies of the behaviorof cyclones in GCMs in the literature, and this paperhopes to fill this void by presenting a winter cycloneclimatology for both the Northern and the Southernhemispheres for a five-year simulation from the Canadian Climate Centre (CCC) GCM.2. Analysis There are two methods used to obtain cyclone frequencies or "event" climatologies. The first of thesecounts the number of cyclone tracks which cross unitareas, or in-many cases latitude-longitude quadrangles,and the second records the number of low centers foundin a unit area in a given time period. The method which counts tracks enumerates a lowonly once in a given unit area, hence slowly movingor quasi-stationary lows are not multiply counted andfast moving lows will be counted in each unit area alongits path. This method is very labor intensive becausethe trajectories of every low over a period of time mustbe extracted before proceeding with the counting ofthe tracks. However, in doing so, one has the advantageof being able to. eliminate those lows which are notrelevant, such as thermal lows and lows caused by reduction to sea level over high terrain. The second method of producing cyclone frequencies counts the number of lows in each unit area. Asa result, slow moving lows can be enumerated manytimes in a given unit area and rapidly moving cycloneswill not necessarily be counted in all unit areas alongits path. It is very straightforward to extract such anevent climatology using a computer, but it must beremembered that all lows including spurious ones willbe included.' Most cyclone studies have used latitude-longitudequadrangles as the "unit areas" or analysis boxes. Sincethe area of a quadrangle is a function of latitude, theextracted raw cyclone frequencies were usually areanormalized when the results were presented. This pro~cedure has been criticized by Zishka and Smith (1980)and Hayden (198 I) who recommend that the presentation of cyclone frequency results be done in terms ofthe raw values even though this would bias the results.The authors felt that the area-normalized results werealso biased and that it would be preferable to displaythe raw frequencies. An additional bias in the cyclonetrack counting method is described in Taylor (1986)who showed that the number of cyclone events dependson the orientation of the cyclone trajectories in theanalysis boxes. For this study, the second approach is used mainlybecause it can easily be done on a computer. The problem of lows "skipping" analysis boxes is a samplingproblem which is controlled by using a relatively longperiod of five years to smooth the frequency maps.Another advantage of this method is that it does multiply count slow moving cyclones and thereby providesa better measure of cyclonic activity and, hopefully,cloudiness and precipitation in each grid box. The climatology is produced.using a polar stereographic pro.jection and since its grid boxes have a relatively weakdependence on latitude, area-normalized results willbe given. The cyclone climatology is extracted for the NorthernHemisphere winter (December, January, and February)and the Southern Hemisphere winter (June, July, andAugust) from five years of simulated data from the CCCGCM (Boer et al., 1984) operating in an annual cycleJANUARY 1988 NOTES AND CORRESPONDENCE 111mode sampled once per day. The model's 1000 mbheight fields, which had been archived as triangular 20wave arrays of spherical harmonic coefficients, weresynthesized on a hemispheric polar stereographic gridwith a spacing of about 440 km at 60 degrees latitude.A low event at a grid point is defined as the occurrenceof a height lower than each of the four surroundingpoints. The total number of events for the two hemispheres was determined for the five-yr simulation period.The raw event statistics were area-normalized by thesquare of the chart's map scale factor, making the unitarea approximately 200 000 km2. Because the choice of grid type, analysis box size,sampling interval and length of record influence theresults, it was decided to repeat the analysis for a fiveyear period from observations. The 1000 mb heightfield analyses from the ECMWF/WMO dataset for theyears 1980 to 1984 were interpolated to the same polarstereographic grid used for the model results in orderto make the subsequent analyses identical.3. Results ' Table 1 displays the number of lows north of 30-Nand the number of lows south of 30-S from the modeland observations as a function of central height for fivewinters. These results show that during the five seasonsthe model exhibits fewer lows in total than are observed.In the Northern Hemisphere, the results as a functionof central height show that the model had too manydeep cyclones and that the paucity of model lows isconcentrated in the weak low categories. Most of thespurious lows counted by the analysis procedure willappear in the weak low classes; it is not clear whetherthe model deficiencies in these categories result fromtoo few relevant cyclones, too few spurious lows, orboth. In the Southern Hemisphere, there is less difference between the number of simulated and observedlows than was evident in the Northern Hemisphereresults. A statistical test was made to determine if the number of simulated lows differs significantly from thenumber of observed lows. Three categories are used: aweak low category with central heights above 0 m; amedium category with central heights between -300m and 0 m; and a strong low category with centralheights below -300 m. The mean number of lows andthe standard deviation over the five seasons are calculated, and the following t-statistic is computed foreach category and the total number of lows:t = (~o - )?m) + x[.(no-1)So2+(nm- ')s,,2]'/2 no + nr~--~ (1)TABLE 1. Total number of simulated and observed cyclone events as a function of the central height for five winter seasons for theNorthern Hemisphere between 30-N and the North Pole and for the Southern Hemisphere between 30-S and the South Pole.Northern HemisphereSouthern Hemisphere Model Ratio Model RatioCentral height observations observationsOver 200 m 50 0.25 1 0.11 201 9100 to 200 m 653 0.39 264 0.74 1654 3580 to 100 m 1676 0.67 938 1.22 2510 772-100 to 0 m 1354 0.78 1178 1.07 1733 1099-200 to -100 m 1102 1.07 1534 0.86 1034 1785-300 to -200 m 530 1.45 1151 0.75 365 1538-400 to -300 m 181 2.82 439 0.81 64 541-500 to -400 m 30 2.00 67 1.29 15 52Below -500 m 2 1.00 9 1.13 2 8Total 5578 0.74 5581 0.91 7578 6162112 JOURNAL OF CLIMATE VOLUME 1 TABLE 2. Computed t-values for the difference between the numberof simulated and observed lows as a function of central height. Valuesover 3.4 are significant at the 1% level. Central height Northern Hemisphere Southern HemisphereOver 0 m 8.5 0.8-300 to 0 m 1.6 4.3Below -300 m 4.1 0.8Total 9.0 4.3where -is the mean number of lows per season in eachcategory, s its standard deviation, and n the numberof seasons. The subscript o refers to observations andthe subscript m refers to the model. The t-values aredisplayed in Table 2. For no -- nm -- 5 values over 3.4are significant at the 1% level. The results show thattotal number of simulated lows in' both hemispheres,the number of weak lows in the Northern Hemisphere,the number of medium lows in the Southern Hemisphere, and the number of strong lows in the NorthernHemisphere are significantly different from the observed number of lows. The geographical distribution of total cyclone eventsis given in the following figures. Even though thesefigures display events equatorward of 30 degrees, theywere not used in compilation of the preceding statistics. Figures 1 and 2 display the total number of NorthernHemisphere cyclone events per 200 000 km2 duringthe five-yr period from observations and the modelrespectively. Comparison of Fig. I to the previousstudies of Petterssen (1956) and Klein (1957) showsreasonable agreement indicating that the simple analysis procedure used in the present study is yielding reliable results. It is readily apparent upon comparisonof Figs. I and 2 that the model has simulated too fewlows which is also indicated by Table 1. The observations show a large area of cyclone activityin the northern Pacific with a maximum in the Gulfof Alaska, and two major storm tracks. The first originates to the west of Kamchatka and ends in the Gulfof Alaska, and the second and somev)hat stronger onebegins in southern Japan. and also ends in the Gulf ofAlaska. (The contour with a value of 20 in the middleof the shaded area in the Pacific has not been mislabeled; there is a minimum in this area.) The modelresults show a similar pattern in the north Pacific, butthe maximum cyclone frequency occurs near Kamchatka and the more southerly storm track is muchweaker than observed. Over North America, the observations show a maximum of cyclonic activity in the favored cyclogeneticareas in the lee of the Rocky Mountains. From this'region a broad band of high events extends eastwardacross southern Canada and the northern United Stateswith the maximum activity located, over the GreatLakes and Hudson Bay. The model results show fewcyclone events and consequently little cyclogenetic activity to the lee of the Rocky Mountains. There is asmall region in the northern United States with a weak- storm track extending to the Great Lakes. The mostprominent model storm track over North America isa continuation of the storm tracks in the Gulf of Alaskaalong the north coast of Canada to Hudson Bay. Theobserved results do not show such a track implyingthat very few cyclones move inland from the Gulf ofAlaska and along the Arctic coast to Hudson Bay. Thespurious event maximum off the coast of Baja California is the surface reflection of upper cold lows whichare frequently present in this area during winter. The major observed cyclonic activity over the Atlantic Ocean is found off the east coast of North America, both coasts of Greenland, northern Scandinavia,and the northern coast of the U.S.S.R. There is a lesserarea of cyclone events over and to the west of the BritishIsles. A similar behavior is seen in the model results.Lows in the model tend to be farther off the east coastof North America and there is no splitting of the stormtrack along the two coasts of Greenland. The cycloneareas to the north of Scandinavia and the U.S.S.R. arepoorly defined, and there is an event maximum overthe high terrain of Norway and Sweden where the observations show an expected minimum. In Europe, the major area of cyclonic activity isfound in the Mediterranean with centers of maximumevents near Italy and Cyprus and a storm track extending northeastward into the western Soviet Unionand another extending eastward towards the CaspianSea. The model fails to show any significant cyclonicactivity in the lee cyclogenetic area in the Gulf of Genoabut does simulate the storm tracks running eastwardfrom a center near Cyprus. Over the land mass of Asia, the model and the observations show little organized cyclone activity. Themost striking features occur over mountainous terrainand are assumed to result from extrapolation belowthe surface. A few individual model cyclones were followedthroughout their life spans, and the period of time between genesis and lysis was observed as typically aboutten days in contrast to the four or five days reportedby Klein (1957). This inordinately long life of simulatedcyclones was a result of their slow dissipation rate. Asmentioned previously, a similar behavior was foundin the ECMWF forecast models by Akyildiz (1985). Figures 3 and 4 show the Southern Hemisphere cyclone event climatologies for winter from observationsand the model, respectively. The most prominent feature in the observations is the ring of high events surrounding Antarctica. The number of events is not uniformly distributed in the ring, and the highest numberof cyclones is found on the Indian Ocean side of Antarctica, with a secondary maximum occurring in thequadrant between the dateline and 90-W. BetweenSouth America and the Prime Meridian, the circumpolar ring becomes quite broad, suggesting a stormOBSERVEDWINTER FIG. 1. The total number of lows per 200 000 km2 for five Northern Hemisphere winters(December-February) during the period from 1980 to 1984 using the once-daily analyses of theECMWF/WMO dataset.MODELWINTERFIG. 2. As in Fig. I except for five winters simulated by the GCM.OBSERVEDWINTERFIG. 3. As in Fig. I except for the Southern Hemisphere winter (June-August) as observed.tt !MODELWINTERFIG. 4. As in Fig. 2 except for five Southern Hemisphere winters simulated by the GCM.JANUARY 1988 NOTES AND CORRESPONDENCE 115track in the southern Atlantic Ocean which begins inSouth America and merges with the polar ring. Thereis a maximum in the number of events in the TasmanSea with a storm track extending to the east and merging with the circumpolar maximum. A similar view ofthe Southern Hemisphere was given in Taljaard (1967)except that the South American storm track beginsfarther north around 30-S where Fig. 3 has a maximumbut no obvious associated storm track. Figure 4 shows that the model has been quite successful in simulating the geographical distribution ofthe event climatology. There is a nonuniform circumpolar ring with the largest number of events on theIndian Ocean side and another maximum in the sectorbetween the dateline and 90-W. The ring is broadesteast of South America and there is evidence of an Atlantic storm track originating over South America. Astorm track extends eastward from New Zealand butthe event maximum in the Tasman Sea is absent. Inaddition, there is little cyclogenetic activity in the leeof the Andes as expected from the results of Chung(1977). Although not so pronounced as in the NorthernHemisphere, there is a tendency for model cyclones tobe present over high terrain as shown by the eventsover the southern Andes and the southern AntarcticPeninsula where few cyclones are observed.4. Summary and conclusions The winter season event climatologies of the CCCGCM for the Northern and Southern hemispheres havebeen determined from a five-year simulation. In theNorthern Hemisphere, the simulations have too fewweak lows and too many intense lows when comparedto observations. The model simulates well those characteristics of the event climatology for which thermalinfluences are important including oceanic storm tracksand the concentrating of cyclones over large inlandbodies of water. The model has problems simulatingthose features of the event climatology which are topographical in nature. Observed lee cyclogenesis areassuch as those east of the Rocky Mountains and in theGulf of Genoa are essentially absent in the model. Thesimulated cyclones also tend to traverse mountainranges which block observed cyclones. In the Southern Hemisphere, the model also produces too few lows in total resulting from a deficiencyof medium intensity lows. The number of strong andweak lows is well simulated. In general, the features ofthe event climatology are well reproduced by the GCM. Acknowledgments. The author wishes to thank StuartTitle for carrying out most of the calculations, LyndaSmith for typing the manuscript, and Brian Taylor fordrafting the figures.REFERENCESAkyildiz, V., 1985: Systematic errors in the behaviour of cyclones inthe ECMWF operational models. Tellus, 37A, 297-308.Boer, G. J., N. A. McFarlane, R. Laprise, J. D. Henderson and J.-P. Blanchet, 1984: The Canadian Climate Centre atmospheric general circulation model. Atmos.-Ocean, 22, 397-429.Chung, Y. S., 1977: On the orographic influence and lee cyclogenesis in the Andes, the Rockies and the east Asian mountains. Arch. Meteor. Geophys. Bioklim., Ser. A, 26, 1-12.Hayden, B. P., 1981: Cyclone occurrence mapping: Equal area or raw frequencies? Mon. Wea. Rev., 109, 168-172.Holloway, J. L., and S. Manabe, 1971: Simulation of climate by a general circulation model. Part I: Hydrologic cycle and heat bal ance. Mon. Wea. Rev., 99, 335-370.Klein, W. H., 1957: Principal tracks and mean frequencies of cy and anticyclones in the Northern Hemisphere. Res. Pap. No. 40, U.S. Weather Bureau, U.S. Govt. Printing Ot~ce, Washing ton, DC, 60 pp.Leary, C., 1971: Systematic errors in operational National Meteo rological Center primitive-equation surface prognoses. Mon. Wea. Rev., 99, 409-413.Manabe, S., and T. Terpstra, 1974: The effects of mountains on the general circulation of the atmosphere as identified by numerical experiments. J. Atmos. Sci., 31, 3-42.Petterssen, S., 1956: Weather Analysis and Forecasting, Vol. 1. McGraw-Hill, 422 pp.Silberberg, S. R., and L. F. Bosart, 1982: An analysis of systematic cyclone errors in the NMC LFM II model during the 1978-79 cool season. Mon. Wea. Rev., 110, 254-271.Taljaard, J. J., 1967: Development, distribution and movement of cyclones and anticyclones in the Southern Hemisphere during the IGY. J. Appl. Meteor., 6, 973-987.Taylor, K. E., 1986: An analysis of the biases in traditional cyclone frequency maps. Mon. Wea. Rev., 114, 1481-1490.Whittaker, L. M., and L. H. Horn, ! 981: Geographical and seasonal distribution of North American cyclogenesis, 1958-77. Mort. Wea. Rev.,Zishka, K. M., and P. J. Smith, 1980: The climatology of cyclones and anticyclones over North America and surrounding ocean environs for January and July, 1950-77. Mon. Wea. Rev., 108, 387-401.
The cyclone event climatology is presented for a five-year simulation by the Canadian Climate Centre general circulation model. Winter season results are given for the extratropical regions of both the Northern and the Southern hemispheres and the results are compared to an event climatology based on five years of observed data.
In the Northern Hemisphere, classification of the simulated and the observed 1000 mb lows as a function of central height shows that the model has a deficiency of weak cyclones and an excess of intense cyclones while the number of cyclones of medium intensity is well simulated. The geographical distribution of the cyclone events shows that those features of the observed climatology which are thermal in nature, such as a preference for lows to form and move over relatively warm underlying surfaces, are reasonably well simulated, but the topographical features such as lee cyclogenesis and the avoidance of high terrain by lows are poorly simulated.
The Southern Hemisphere also shows a deficiency of simulated cyclones although less pronounced than the Northern Hemisphere. The number of weak and strong cyclones are well simulated, and the paucity of Southern Hemisphere lows results from a deficiency of cyclones of moderate intensity. The observed climatology, which is nearly free of mountain influences, is well simulated and only a few minor problems are evident, such as a poor simulation of the winter cyclogenesis in the lee of the Andes and a tendency for model cyclones to be present over high terrain.