Model Simulations of the Presentday Arctic Climate

One measure of the level of confidence in results generated by GCMs is the degree to which they reproduce the cunent global climate. A review of the first three IPCC reports (Houghton et al. 1990, 1992, 1996) and other works (e.g. Gates et al 1996, 1999) shows that the largest disagreement between coupled climate model simulations of prcscnt-day climate is in the Polar regions. It is worth adding, however, that in more rcccnt models this disagreement is still observable, though it is lower (Boer et al. 2000; Flato et al. 2000; Giorgi and Francisko 2000a, b; Houghton et al. 2001; Lambert and Boer 2001). Lambert and Boer (2001) also found that the intermodel scatter of the simulated climate variables is the largest here and over mountains.

According to Randall et al. (1998) the large degree of disagreement among the models reflects both the weakness of our current understanding of Arctic climate dynamics and the sensitivity of the Arctic climate to different formulations of various physical processes.

Recently a number of papers have been published in which researchers have concentrated on simulating the Arctic climate with the GC'Ms (see e.g. Walsh and Crane 1992; Cattle and Crossley 1995; Tao et al 1996; Walsh et al. 1998; Randall et al. 1998; Weatherly et at. 1998; Zhang and Hunke 2001) and with limited-area models (Walsh et al. 1993; Lynch et al. 1995; Dethloff et al. 1996, 2001; Rinke et al. 1997, 1999a, b, 2000; Dom et al. 2000; Rinke and Dethloff 2000; Gorgen et al. 2001).

What arc the results presented in these papers? Figure 11.1 shows the annual mean fields of SLP simulated by the five most widely known atmospheric GCMs (GFDL, GISS, NCAR, OSU, and UKMO), and the "observed" field based on the NCAR sea level pressure analyses for 1952-1990.

Significant differences between modelled and observed fields may be clearly seen. The best simulation is given by the GFDL model. Pattern correlations computed by Walsh and Crane (1992) between simulated and observed fields based on data for the zone 70-90°N entirely confirm this conclusion. The highest correlations were noted for winter and spring (0.909 and 0.908, respectively) and the lowest for summer (0.568). Other models have significantly tower correlations. It is worth noting that even such marked baric centres as the Icelandic and Aleutian lows vary widely from model-to-model. An analysis conducted using a new model (the NCAR Climate System Model) confirms the existence of significant differences between the annual mean field of SLP simulated by the model and the "observed" field obtained from the ECMWF (European Centre for Medium-Range Weather Forecasts) analyses (Weatherly et al. 1998). On the other hand, as recent works show, the N AO and the AO, which significantly influence the Arctic climate in the wintertime, are simulated quite well by the majority of the coupled climate models (for details see e.g. Delworth 1996; Broccoli et al. 1998; Laurent et al. 1998; Saravanan 1998; Fyfe et al. 1999; Osborn et a!. 1999; Shindell et al. 1999; Houghton et al. 2001).

Figure 11.1. Annual mean sea level pressure fields produced by the five models (a-e) and the observational field based on the NCAR sea-level pressure analyses for 1952-1990 (f) (after Walsh and Crane 1992). Key: GFDL - Geophysical Fluid Dynamics Laboratory (U.S.), GISS - Goddard Institute for Space Studies (U.S.), NCAR - National Center for Atmospheric Research (U.S.), OSU - Oregon State University (U.S.), UKMO -United Kingdom Meteorological Office (U.K.).

Walsh and Crane (1992) also present the winter fields of surface air temperature simulated by four of the models and the "observed" climatology based on the data of Crutcher and Meserve (1970), Although all the models have temperature minima between -35°C and -45l,C, their locations vary considerably from model-to-model. It is very interesting and surprising that the GFDL model, which most successfully simulates the Arctic sea level pressure, is the least successful at simulating the Arctic air temperature. The bias in this model in mean winter and autumn air temperature is about IOl'C (the model is too "cold" in comparison with observations). The reason for this is the fact that the GFDL mode! incorporates a poor sea-ice mode! (Mysak, personal communication). The best results were obtained using the UKMO model. The average air temperature differences between model and observed values did not exceed lnC for winter, spring, and autumn. Only in summer was the model "eolder" by about 2°C. Tao et al. (1996) give the results of Arctic air temperature for the 10-year period 1979-1988 simulated by 19 numerical models participating in the AMIP (the Atmospheric Model Intercom pari son Project). For more details of this project see, for example, Gates (1992). In winter, summer, and autumn, the majority of models give the areal mean air temperature in the Arctic Ocean lower than observations (see Figure 11.2). On the other hand, in spring the models' areal means show the warm bias, with the exception of four of them. Model-to-model differences of simulated areal mean air temperature are very high (in winter and spring up to about 15°C, in autumn more than 10"C and in summer about 6"C).

Walsh et al. (1998), using 24 climate models participating in the AMIP, compared model simulations of precipitation and evaporation with observational estimates of these elements. Figure 11.3 presents the decadal (1979 1988) annual mean precipitation and annual mean values of precipitation minus evaporation from the AMIP models, as well as the corresponding observational estimates after Bryazgin (1976a), Legates and Willmott (1990), Jaeger (1983) and Vowinckel and Orvig (1970). From Figure 11.3a, it can be seen that almost all models simulate excessive precipitation. Aside from the "outlier" model of SUNYA, the wettest models are CCC, UIUC, MPI and NCAR, while the JMA and UGAMP models simulate the least precipitation. Significantly poorer simulations are evident both for shorter periods and smaller regions. For example, the monthly precipitation simulated for the Arctic Ocean differs by a factor of 2 among the various AMIP models (see Figure 3 in Walsh et al. 1998). Roughly similar results were obtained for the precipitation minus evaporation values (Figure 11.3b).

Temperature (°C)

Temperature (°C)


/ .


w c


- -1

/ '



' , s /





Gl A

T ;

^ / ✓ /J






, ,








■ , < * ?










/ /

Temperature (°C)

\ / /

j),., >i,,) * ' , , t ,




-V '/'


/ f / * / s / * , *

f * * , f / r / f / '

V // * t * * y y y /


I f 7 / * s / / /

X ';/

y / / / t / /

r , , , j , , / - ' * '


<t?tr*3, is* '


/ J /

/'■J 7-7 -, A , ,

' / t


J ' r , s / , >

>v c _:


1, - ✓ / - - ,



^ T C /


y //',', ,

t <


' / ) ' J > '

t f / y ' ' f /


Y ' , ' > , y / 1

' / X /



y >


1 / / / / t ' < '.

s / f y y J >

, , / / ' ' / ,

/ /

]' S ' ,

' /



t f * * * ' t , J


V'/ /'

/ / / ' ' / / /



'////// i / / / / r / s *

/ / / / /


' * / f / / * ' *

/ /



' / ' ' / / ' /

/ ^


/ *


/// * / S * *

' / '




\ , >, * / / s

' * / ) / ' / ' ' / / * J

, i , , , V

' , * ' s * *

' • ' , ,

1/ / f ' '


/ / / / / ' '

' ' / "1

1 ✓ / > * > \f S f J * ,

/ * ' / to

J , ,

/ t / / * J

/ ' ' / / /

—T, , , , , 7-7*7-V , / , , ' , ' ,

/ y y y / / / y y ✓ /• y .

, , , , , / ,

' y / y y y

, r , * , ,


/ , , y

t t * t * *

ft* * *


/ ' '


Figure 11.2. Seasonal mean air temperatures ("C) for the Arctic Ocean domain as computed from observed data (OBS, after Crutcher and Meserve 1970) and the 19 AM1P models. Temperatures are shown for winter (l)JF), spring (MAM), summer (JJA), and autumn (SON) (after Tao et at. 1996),

Key: AMIP models: BMRC (Australia), CCC (Canada). CNRM (France), CSIRO (Australia), CSU (U.S.), DERF/GFDL (U.S.), ECMWF (Europe), GLA (U.S.), GSFC (U.S.), JMA (Japan), LMD (France), MGO (Russia). MPI (Germany), MRI (Japan), NCAR (U.S.), NMC (U.S.) SUNYA (U.S.) UIL (U.S.), UKMO (U.K.).

The Climate of the Antic Arctic Ocean: Annual Mean P

Figure 11.3. Annual mean rates of (a) precipitation, P, and (b) precipitation minus evaporation, P - E. for the Arctic Ocean as evaluated from observational estimates (bars at left) and from AMIP models (after Walsh et at. 1998).

Observational sources are: B = Bryazgin (1976a), L = Legates and Willmott (1990), J = Jaeger (1983), and O = Vowinckel and Orvig (1970) (hatched bar is for domain excluding Barents-Norwegian Seas). Other key as in Figure 11.2 plus COLA (U.S.), DNM (Russia), GFDL (U.S.), NRL (U.S.), UCLA (U.S.), UGAMP (U.K.), UIUC (U.S.).

Figure 11.3. Annual mean rates of (a) precipitation, P, and (b) precipitation minus evaporation, P - E. for the Arctic Ocean as evaluated from observational estimates (bars at left) and from AMIP models (after Walsh et at. 1998).

Observational sources are: B = Bryazgin (1976a), L = Legates and Willmott (1990), J = Jaeger (1983), and O = Vowinckel and Orvig (1970) (hatched bar is for domain excluding Barents-Norwegian Seas). Other key as in Figure 11.2 plus COLA (U.S.), DNM (Russia), GFDL (U.S.), NRL (U.S.), UCLA (U.S.), UGAMP (U.K.), UIUC (U.S.).

The simulation of the total cloudiness varies tremendously from model-to-model (Figure 11.4). During summer, for example, the simulated cloudiness over the Arctic Ocean varies from 30% to 98%, while its value, accord ing to observations, is equal to about 82-84%. An intriguing issue is the fact that majority of the models do not show the summertime increase of cloudiness. Moreover, some of them simulate even the lower cloudiness in this season (e.g. CSI, GLA, LMD, and SUN).

Arctic Ocean

Arctic Ocean






















Figure 11.4. Annua! cycle of monthly mean total cloudincss (%) over the Arctic Ocean as simulated by the 19 AMIP models. Heavy lines show the annual cycle for 75-85"N derived from observations made at drifting ice stations (Vowinckel and Orvig 1970, p. 150) (after Tao et al. 1996). Key as in Figure 11.2.



moo mpi

mr! nca


nmc sun






Figure 11.4. Annua! cycle of monthly mean total cloudincss (%) over the Arctic Ocean as simulated by the 19 AMIP models. Heavy lines show the annual cycle for 75-85"N derived from observations made at drifting ice stations (Vowinckel and Orvig 1970, p. 150) (after Tao et al. 1996). Key as in Figure 11.2.

The above review reveals that the confidence of the GCMs in simulating the present-day climate on a regional scale (here for Arctic climate) is as yet rather unsatisfactory. According to Lynch et al. (1995) the model biases can be attributed mainly to inadequate topographical representation due to low horizontal resolution and the inadequate representation of cloud and sea ice distribution. Moreover, they state that GC'Ms appear to be inadequate for Arctic climate simulation and prediction. Therefore, to overcome this problem. they propose the use of limited-area models (so-called Regional Climate Models (RCMs)). The development of RCMs was initiated by Dickinson et al. (1989) who nested the NCAR regional model MM4 with a resolution of 60 km for the western United States in a GCM. Walsh et al. (1993) and Lynch et al. (1995) presented the first RCM (called the Arctic Region Climate System Model (ARCSyM)) for the Western Arctic. This model is based on the NCAR RCM and is a significant step in the development of a fully coupled regional Arctic model of the atmosphere-ice-ocean system. Furthermore, a team of German and Danish scientists have recently applied the RCM (called HIRHAM) to the whole Arctic (Dethloff et al. 1996; Rinke et al. 1997).

The above models, among others, were used to simulate monthly (January and July) fields of different meteorological elements. The results obtained are better than in the case of GCMs, but are still not satisfactory. For example, in winter the simulated air temperature in some parts of the Arctic is higher than observations by up to I0°C (Western Arctic, sec Lynch et al. 1995) and !2PC (central Arctic, see Dethloff et al. 1996). In summer the differences between model and observations arc much smaller than in winter, with the largest deviations being up to 4°C over the centre of the Arctic Ocean (Dethloff el al. 1996). Here one should add that Dethloff et al/s (1996) model simulations are compared with the ECMWF analyses, which are based on the ECMWF model. The ECMWF model shows a systematic air temperature bias to be excessively low at the surface over sea ice during winter. Similar results for summer were also obtained by Lynch et al. (1995), who found a cold bias (i.e. the model is "colder"') of 1-20C in the mountainous regions and a warm bias of 3-5"C over the tundra in the Western Arctic. Significant differences in the simulated precipitation and cloud fields with the observation data were also noticed.

Rinke et al. (1997) found that model outputs greatly overestimate the incoming short-wave flux and significant differences in the net radiation over the Arctic, especially in July (up to 100 W/nr and ± 70 W/m3, respectively). It is worth adding, however, that the model experiments described in Lynch etal. (1995), Dethloff et al. (1996) and Rinke et at. (1997) have shown the potential for realistic simulations of climate processes of the Arctic troposphere and lower stratosphere meteorological fields in limited-area cliiuatc models.

The results presented here come from the first version of RCMs, which still contain many shortcomings. The improvements in the physical parameterisation packages of radiation and in the description of sea-ice thickness and sea-ice fraction introduced recently into the new version of RCMs, significantly reduce the model biases (Rinke et al 1999a, 2000; Rinke and Dethloff

2000; Dethloff et al. 2001). For example, in the case of surface air temperature, the differences between HIRHAM model simulations and "observed" values (gridded 2 m air temperahirc for the Polar Exchange at the Sea Surface (POLES), climatology of the Legates and Willmott (1990) or ECMWF analyses) generally do not exceed 5°C in winter. In June these differences are clearly lower, although locally they also reach 5"C (see Rinke et al. 2000, their Figure 11). The worst results have been obtained by applying the ARCSyM model. Rinke et al. (2000), comparing the simulated near-surface air temperature in January, found that ARCSyM temperatures arc higher than the HIRHAM simulation, to the order of 5-15"C. On the other hand, both models gave quite similar results for June.

In the case of atmospheric precipitation, the situation is opposite. Both models better simulate precipitation for January than they do for June (see Rinkc et al. 2000). Moreover, in June the differences in monthly totals between HIRHAM and ARCSyM simulations are much larger. Monthly mean surface energy balance components (sensible heat flux, latent heat flux, and net radiation) are mostly underestimated in January and June in both models (see Figures 6 and 8 in Rinke et al. 2000) in comparison with the NCEP reanalyses. The differences in the case of net radiation do not exceed 30 W/m2 in January and 40 W/m2 in June and are significantly smaller than in previous simulations (Rinke et al. 1997). Sensible heat flux differences are smaller than 10 W/m2 in both months. A similar size of the differences is noted for latent heat fluxes in January. On the other hand, in June the differences are larger, especially in case of the HIRHAM model, reaching more than 30 W/m2. Taking into account fact that the NCEP reanalyses of the energy balance components for the high latitudes arc overestimated (see Gupta et al. 1997), the real biases are probably lower.

Thus one must conclude that the RCMs, especially their new versions, more reastically simulate the present-day Arctic climate than do the GCMs and hence they should constitute a reliable tool for climate change studies in the Arctic. Up to now there have been no such studies, but probably in the near future these kinds of models will be used to simulate the Arctic climate change with the doubling of COr That is why, in the next sub-chapter, scenarios of the Arctic climate for this centuty are presented based only on the GCM outputs and the analogue method.

Was this article helpful?

0 0

Post a comment