Other reports in this collection
Analysis of extreme temperature in climate model simulations has concentrated on the surface daily maximum and minimum temperature, or on the duration of hot/cold spells on the global scale (Schubert, 1998; Zwiers and Kharin, 1998; McGuffie et al., 1999; Kharin and Zwiers, 2000).
Zwiers and Kharin (1998) and Kharin and Zwiers (2000) analysed the 20-year return values for daily maximum and minimum screen temperature simulated by both CCC GCM2 and CGCM1. Comparison with the NCEP reanalyses shows that the model reproduced the return values of both maximum and minimum temperature and warm/cold spells reasonably well.
Intercomparisons among five AGCMs for the return values of extreme temperature of <-20°C and >40°C over the globe show a reasonable level of agreement between the models in terms of global scale variability (McGuffie et al., 1999).
Analysis of extreme precipitation simulated by climate models has included the daily variability of anomalous precipitation (Zwiers and Kharin, 1998; McGuffie et al., 1999; Kharin and Zwiers, 2000), patterns of heavy rainfall (Bhaskran and Mitchell, 1998; Zhao et al., 2000b), as well as wet and dry spells (Thorncroft and Rowell, 1998; McGuffie et al., 1999). The results show some agreement with the available observations but the comparatively low model resolution is an inhibiting factor.
Ideally the simulated extreme rainfall should be compared with grided data calculated from the observed station data; however, observed grided data comparable to those produced by the models are scarce. Therefore, often the NCEP/NCAR reanalysis data are used as an “observed” data set despite the fact that this data set does not appear to reproduce daily variability well (Zwiers and Kharin, 1998). Another issue is the interpretation of precipitation simulated by a climate model, some authors treat simulated precipitation as grid-box averages; others argue that it should be treated as grid-point values (Zwiers and Kharin, 1998). Hennessy et al. (1997) compared the daily precipitation by both CSIRO and UKHI (United Kingdom High-Resolution) AGCMs coupled to mixed-layer ocean models. They found that simulated frequencies of daily precipitation were close to those for grid-box average observations.
In summary, in contrast with the simulations of extreme temperature by climate models, extreme precipitation is difficult to reproduce, especially for the intensities and patterns of heavy rainfall which are heavily affected by the local scale (see Chapter 10).
Analyses of occurrences and tracks of extra-tropical storms have been performed for some climate models (Lunkeit et al., 1996; Beersma et al., 1997; Carnell and Senior, 1998; Schubert et al., 1998; Zwiers and Kharin, 1998; Kharin and Zwiers, 2000). However, very different methods are used to characterise extra-tropical storms, among the methods used are: mid-latitude storm tracks defined by 1,000 hPa wind speed (Zwiers and Kharin, 1998), maximum eddy growth rate at 350 hPa and 775 hPa (Lunkeit et al., 1996), index of storm tracks (such as 500 hPa height variability, sea level pressure, surface wind) (Beersma et al., 1997), frequency, intensity and track of 500 hPa transient eddies (Schubert et al., 1998), as well as low centres at 500 hPa (Carnell and Senior, 1998).
Kaurola (1997) compared the numbers of the observed extra-tropical storms north of 30°N for five winter seasons and from a 30-year simulation of the ECHAM3 atmospheric model for two periods during the control run. The comparisons indicated that the ratios of total numbers between the simulations and observations were 0.96 and 0.97 for two respective periods. It appears that the ECHAM3 model is able to simulate the numbers of storms north of 30°N in wintertime. The mid-latitude storm tracks over the North Pacific and Atlantic Oceans and over the southern circumpolar ocean were also well simulated by the CCC GCM2 (Zwiers and Kharin, 1998).
Kageyama et al. (1999) focused on the storm tracks of the Northern Hemisphere as simulated by several AGCMs. Intercomparisons of the nine AGCMs show that the models reproduce reasonably the storm tracks defined with high-pass second-order transient eddy quantities. These results also indicated that higher resolution models tend to be better at reproducing the storm tracks.
Schubert et al. (1998) analysed North Atlantic storms in the ECHAM3/LSG model. Their analysis indicated that the storm frequency, position and density agreed with the observations. Lunkeit et al. (1996) analysed storm activity in the ECHAM2/OPYC model. They found that the mean eddy activity and storm tracks in that simulation were in reasonable agreement with observations.
The general ability of models to simulate extra-tropical storms and storm tracks is most encouraging.
Tropical cyclones can be characterised in models by several measures such as their intensity, track, frequency and location of occurrence (Bengtsson et al., 1996; Sugi et al., 1997; Tsutsui et al., 1999). Other broad-scale fields such as maximum wind speed, maximum potential intensity (Holland, 1997) and high sea surface temperature (Hulme and Viner, 1998) are also used as indicators of tropical cyclones. Thus it is important to consider the particular characteristics that are used to describe tropical cyclones in a given analysis when results from models are compared (Henderson-Sellers et al., 1998; Krishnamurti et al., 1998; Royer et al., 1998; Walsh and Pittock, 1998).
Many analyses have been based on the physical parameters favourable for cyclogenesis as summarised by Gray (1981). Gray relates the climatological frequency of tropical cyclone genesis to six environmental factors: (1) large values of low-level relative vorticity, (2) Coriolis parameter (at least a few degrees poleward of the equator), (3) weak vertical shear of the horizontal winds, (4) high SST’s exceeding 26°C and a deep thermocline, (5) conditional instability through a deep atmospheric layer, and (6) large values of relative humidity in the lower and middle troposphere (Gray, 1981; Henderson-Sellers et al., 1998). Following the general concepts outlined by Emanuel (1987), Holland (1997) has derived an alternative thermodynamic approach to estimate maximum potential intensity of tropical cyclones. The approach requires an atmosphere sounding, SST, and surface pressure; it includes the oceanic feedback of increasing moist entropy associated with falling surface pressure over a steady SST, and explicitly incorporates a representation of the cloudy eye wall and a clear eye.
Several climate model simulations in Table 8.4 have been analysed using a variety of the above techniques to determine the frequency of tropical cyclones (Bengtsson et al., 1995; Watterson et al., 1995; Vitart et al., 1997; Royer et al., 1998). The ECHAM3 model has by far the highest horizontal resolution amongst these models. The numbers of simulated tropical cyclones are between 70 and 141 per year. The numbers of observed tropical cyclones per year are quite variable; 80 for the period 1958 to 1977 (Gray, 1979), 99 for the period 1952 to 1971 (Gray, 1975) and 86 for the period of 1970 to 1995 (Henderson-Sellers et al., 1998). Despite the differing definitions of tropical cyclones used in the different analyses, the range of tropical cyclones numbers simulated by the models are similar to the observed data.
Bengtsson et al. (1995; 1996; 1999) have analysed a five-year simulation with ECHAM3 at T106 (100 km) horizontal resolution. They conclude that the model could reproduce some aspects of the characteristic structure of tropical cyclones and some aspects of their geographical distribution and seasonal variability. They also found that in certain areas, in particular in the north-east Pacific, a realistic number of tropical cyclones was only generated by the model when the horizontal resolution was finer than 100 km. Their results showed a reasonably good agreement with the observed distribution, tracks and annual variability of tropical cyclones (Bengtsson et al., 1995, 1999). Sugi et al. (1997) and Yoshimura et al. (1999) used a 100 km version of the JMA AGCM and compared the simulated geographical distribution of tropical cyclones with observations. They obtained reasonably realistic geographical patterns. However, in contrast to the observations, they did not find a significant difference in tropical cyclone frequency when they used the SSTs representing El Niño and La Niña years.
Henderson-Sellers et al. (1998) suggested that AOGCMs could provide useful information of the frequency of tropical cyclones, but the models they studied all had coarse resolution (about 500 km), climate drift (or flux adjustment) and unproven skill for present day tropical cyclones. A first attempt of tropical cyclones simulations with a high-resolution coupled climate model was performed by Matsuura et al. (1999) with a 100 km JMA atmospheric model coupled with the GFDL modular ocean model (MOM2) (0.5°31.0°) model (but without sea ice). This model reproduced some aspects of the structure of observed tropical cyclones, although the simulated “tropical cyclones” are weaker and larger in scale than the observed. The model also shows the observed tendency of less (more) frequent tropical cyclones and an eastward (westward) shift of their locations over the northwestern equatorial Pacific during El Niño (La Niña) years. This result gives us some confidence in using a high-resolution coupled climate model in the future to explore the relationship between global warming and the frequency and intensity of tropical cyclones.
In summary, high horizontal resolution AGCMs (or AOGCMs) are able to simulate some aspects of “tropical cyclone-like vortices” with some degree of success, but it is still too computationally expensive to use such models for long experiments. The type of tropical cyclone index chosen in the analysis of low-resolution climate models is important, the use of maximum potential intensity may provide the most robust estimate, but analyses using this index remain infrequent.
Since the SAR, more attention has been paid to the analysis of extreme events in climate model simulations. Evaluations indicate that climate models are more capable of reproducing the variability in maximum and minimum temperature in the global scale than the daily precipitation variability. The ability of climate models to simulate extra-tropical storm tracks and storm frequency is encouraging. When tropical cyclones were analysed, high-resolution models generally produced better results. It is worth noting that some high-resolution operational numerical weather prediction models have demonstrated reasonable ability in forecasting tropical cyclones. This increases our confidence that they may be better reproduced by high-resolution climate models in the future.
The lack of consistent methodologies used in analyses of extreme events prevents a ready intercomparison of results between models; future IPCC assessments would be greatly assisted if common approaches were adopted.
|
Other reports in this collection |
|