VALIDATION OF ECHAM GCM IN TERMS OF CIRCULATION PATTERNS AND THEIR INTERDIURNAL VARIABILITY

Martin Dubrovsky* and Radan Huth

Institute of Atmospheric Physics, Czech Republic

*** presented at 78th Annual Meeting of the American Meteorological Society, Phoenix, 11-16 January 1998 ***

(to be published in the proceedings)

*corresponding author:
Martin Dubrovsky, Institute of Atmospheric Physics, Husova 456, 50008 Hradec Kralove, Czech Republic

visit my home page home      mail

1. INTRODUCTION

Estimating impacts of climate change often require high resolution (in space and time) data to be used as an input to the simulation models [crop growth models (Dubrovsky, 1998); rainfall-runoff models]. A possible method how to obtain such data - typically required on a daily scale - is to downscale GCM-based upper-air circulation patterns to finer spatial resolution. In this approach, the surface weather characteristics are estimated from circulation with use of statistical methods (Hewitson, 1994) or stochastically synthesised by a weather generator with parameters conditioned on circulation (Bardossy and Plate, 1992). Three assumptions must be satisfied for the series of surface weather characteristics generated in this way to have realistic structure:

  1. GCM reliably simulates spatial and temporal structure of upper-air circulation patterns;
  2. surface weather characteristics are satisfactorily explained by circulation characteristics;
  3. the links between the circulation and surface weather will be preserved even in changed climate conditions.

The present contribution focuses on the ability of GCM to simulate upper-air circulation. The GCM is validated in terms of leading circulation modes and their interdiurnal variability. The motivation for this study is twofold: Apart from being a necessary condition for circulation-based weather generators, it is assumed that preserving interdiurnal variability of circulation is an important, but not yet discussed in literature, feature of GCMs' performance.

2. DATA

The circulation is characterised by 500 hPa geopotential (63 grid points over Europe, 5.6deg x 5.6deg spacing) taken from three sources:

  1. 47year (1949-1995) series of observed data,
  2. 30y GCM (ECHAM3/T42) control simulation of a present climate,
  3. 30y GCM simulation of 2xCO2 climate.

To extract circulation variations on different time scales, two temporal filters described by Blackmon and Lau (1980) were applied pointwise. In result, data from each of the three sources were available in three variants:

Since the raw data are dominated by low frequency variations (Huth, 1997a) and the results obtained are therefore very similar for the two types of data, only results related to LOW and SYN data will be discussed in this contribution.

3. MODES OF CIRCULATION VARIABILITY

Principal component (PC) analysis (S-mode, obliquely rotated) was applied to geopotential patterns to identify modes of variability (Huth, 1997b). This approach to characterising circulation has an advantage compared to discrete classification schemes (e.g., schemes by Hess-Brezowski or Lamb): time series of circulation patterns may be effectively parametrised by an autoregressive model which may be used in developing circulation-linked weather generator or in validating GCM simulations.

The PCs were determined (a) separately for each season (from seasonally stratified data), (b) for a whole year ( "all-year" PCs) from unstratified data. Since the variances of all-year PCs are about three times greater in winter than in summer (showing greater variability of circulation), the winter circulation dominates in defining all-year circulation modes.

The spatial patterns of PC loadings may be interpreted as leading modes of geopotential variability, or the circulation modes. For example, the five all-year PCs (Fig. 1a) determined from band-passed observed data may be interpreted as two wave-trains shifted by a half wavelength: the first two circulation modes (PC1+PC2) represent moving pressure systems from NW to SE over the North Sea; the other three PCs represent pressure systems moving eastwards over the Mediterranean.

4. CIRCULATION MODES IN OBSERVED DATA vs. GCM SIMULATIONS

The variances explained by PCs in individual datasets (Tab. I) show:

  1. The highest/lowest percentage of variability explained by PCs relates to winter/summer. These seasonal differences are quite well preserved in GCM simulations.
  2. The percentage of variance explained by the leading circulation modes is higher for GCM/control simulations compared to observations (in both frequency bands), and yet higher for GCM/2xCO2. This may indicate that the spectrum of processes in GCM simulations is reduced compared to reality.

For both low-passed and band-passed data, the spatial patterns of PC loadings derived from observed data and GCM simulations were compared with each other. Figure 1 demonstrates a good one-to-one correspondence between the band-passed circulation modes determined from the three sources of data. The mutual correspondence of the circulation modes is also documented by spatial correlations (Tab. II). Apparently, the five circulation modes determined from GCM simulations (both control and 2xCO2 runs) correspond in turn to circulation modes 1, 2, 5, 3, 4 determined from observed data. The correlations between corresponding circulation modes (GCM vs. observation) are always higher than 0.90. Note that the correspondence between GCM/control and GCM/2xCO2 circulation modes is yet better - the correlations between corresponding circulation modes are 0.99 in this case.

5. INTERDIURNAL VARIABILITY OF CIRCULATION

Interdiurnal variability of circulation may be expressed in terms of lag-1 correlations between PC scores. To make the correlation matrices comparable, PCs derived from observed data were now used to characterise circulation also in GCM data. This step is justified by a good correspondence between circulation modes derived from the three sources of data (previous section). To allow quantitative comparison of two r x r correlation matrices (denoted A and B) as a whole (Tab. III), heuristic similarity score was used:

D(A,B) = SUMi=1,..,r; j=1,..,r (aij - bij).

The better the correspondence, the lower the similarity score. Analysis of lag-1 correlation matrices shows:

a) band-passed data (the most significant lag-1 correlations related to PCs displayed in Fig. 1 are given in Tab. IV):

  1. The lag-1 autocorrelations are close to zero due to the manner of filtering and due to the velocity of synoptic disturbances.
  2. The GCM/control simulation overestimates lagged crosscorrelations given in Tab. IV (by 22% on average). This may be related to a longer life time of GCM-simulated synoptic waves moving eastward across Europe.
  3. The best fit between observed data and GCM/control simulations occurs in summer (Tab. III).
  4. The lagged crosscorrelations are yet greater in GCM/2xCO2 simulations compared to GCM/control.

b) low-passed data Since the high-frequency variations were removed by low-pass filter, the attention was also paid to correlations for longer lags (lag = 1, 2, 5, 7, 10, 14 days were considered).

  1. Apart from several exceptions, the lag-correlations smoothly converge to zero with increasing lag. For the exceptions, the lagged crosscorrelations exhibit maxima for 5 <= lag <=10 but the values are too low (about 0.3) to make any specific conclusion.
  2. The best/worst fit between observed data and GCM/control data occurs in winter/spring (Tab. III).
  3. The autocorrelations for longer lags imply that the GCM-simulated circulation in a low-frequency band is more persistent compared to observation. The only exception was found - the circulation mode governing meridional circulation appears less persistent in GCM data.
  4. The differences between GCM/control and GCM/2xCO2 are lower (apart from winter, and all-year lag-correlations in autumn) than the differences between GCM/control and observation (Tab. III)

6. CONCLUSION

The tests were made to validate the ability of the ECHAM GCM to simulate upper-air circulation on a subcontinental scale. The circulation was expressed in terms of PCs which were derived from a series of 500 hPa geopotential patterns. The comparison of GCM simulations (for present and 2xCO2 climates) with observed data was made in terms of (a) leading modes of variability of geopotential heights and (b) lag-correlations between PCs. The results may be summarised in following points:

  1. The leading circulation modes are well reproduced by GCM.
  2. GCM simulation exhibits a longer life time of synoptic waves moving eastward across Europe.
  3. The GCM- simulated circulation in a low-frequency band is more persistent compared to observation. This corresponds with greater persistence in daily series of surface weather characteristics (temperature and precipitation) found by Nemesova et al. (1998).
  4. Slight changes in a circulation regime over Europe due to doubled CO2 may be deduced from GCM simulations: spectrum of processes governing upper-air circulation will be reduced and the links between circulation on successive days will be stronger. However, since the differences between GCM/2xCO2 and GCM/control simulations are smaller than the differences between observations and GCM/control simulation, we may hardly draw any reliable conclusion concerning the circulation regime in a changed climate.

ACKNOWLEDGEMENTS This study was supported by the Grant Agency of the Czech Republic under projects 205/96/1669 and 205/96/1670 and by the Grant Agency of the Czech Academy of Sciences under project A3-060-605.

REFERENCES

Figure

Figure 1. Modes of synoptic-frequency variability (PC loadings) in the geopotential height, an all-year version.

left column: observed data
center column: GCM/control data
right column:GCM/2*CO2 data

Tables

Table I. Percentage of cumulative variance explained by leading PCs (three PCs for low-passed data and five PCs for band-passed data) in observations (OBS), and control (CTR) and 2xCO2 (SCA) model runs for individual seasons and the whole year. 

         |     low-passed data (3 PCs)       |   band-passed data (5 PCs)      
         |   OBS        CTR        SCA       |  OBS        CTR         SCA    
------------------------------------------------------------------------------
 ALL     |   66.6       70.8       72.2      |  55.4        64.1       68.3      
 DJF     |   72.0       76.3       76.8      |  60.8        68.7       72.0      
 MAM     |   66.3       70.6       73.2      |  53.2        62.5       66.4      
 JJA     |   59.4       61.5       64.4      |  52.8        62.4       66.1      
 SON     |   65.0       70.5       70.9      |  56.5        64.8       70.1

Table II. Spatial correlations (x100) between the circulation modes displayed in Fig. 1 (OBS = observations, CTR = GCM/control data, SCA = GCM/2xCO2). 

         |           CTR               
     ------------------------------
      PC |   1    2    3    4    5     
===================================
       1 | -96    8  -22  -14  -22     
  O    2 |  32   96   34   13  -38     
  B    3 |  10   -1   13   97  -39     
  S    4 |   6  -47   27  -18   90     
       5 | -32  -18  -98  -36  -19     


         |           SCA           
     ------------------------------
      PC |   1    2    3    4    5 
===================================
       1 | -94   13   22  -15  -14     
   O   2 |  36   92  -33   11  -37     
   B   3 |  18    1  -25   97  -31     
   S   4 |   3  -53  -24  -29   91     
       5 | -35  -07   98  -24  -25     


         |           SCA          
     ------------------------------
      PC |   1    2    3    4    5 
===================================
       1 |  99    7  -39   26   10 
  C    2 |  17   99  -17   15  -51 
  T    3 |  40    5  -99   22   34 
  R    4 |  34   12  -46   99  -31 
       5 |   9  -60  -22  -49   99

Table III. Comparison of lag-1 correlation matrices. The values represent similarity scores (see the text for definition) relating lag-1 correlation matrices derived from GCM/control run (CTR) vs. observation (OBS) and GCM/2xCO2 (SCA) vs. GCM/control. 

        ||               band-passed data                ||                low-passed data               
--------------------------------------------------------------------------------------------------------
        ||     all-year PCs      |     seasonal PCs      ||     all-year PCs     |      seasonal PCs     
season  ------------------------------------------------------------------------------------------------
        ||  CTRxOBS  |  SCAxCTR  |  CTRxOBS  |  SCAxCTR  ||  CTRxOBS  |  SCAxCTR |   CTRxOBS |   SCAxCTR
========================================================================================================
  DJF   ||   3.33    |   1.27    |   3.05    |   1.17    ||   0.43    |   0.78   |    0.41   |    0.69
  MAM   ||   3.44    |   1.74    |   3.10    |   1.65    ||   0.90    |   0.27   |    0.75   |    0.26
  JJA   ||   2.20    |   1.30    |   2.76    |   1.28    ||   0.74    |   0.69   |    0.63   |    0.46
  SON   ||   2.87    |   2.41    |   3.28    |   1.34    ||   0.71    |   0.83   |    0.67   |    0.44

Table IV. Significantly non-zero lag-1 correlations between PCs derived from band-passed data (SYN). 

              ||     cor[PC1(t),PC2(t-1)]       |    cor[PC3(t),PC5(t-1)]       |    cor[PC3(t),PC4(t-1)]
              ||     cor[PC2(t),PC1(t-1)]       |    cor[PC5(t),PC3(t-1)]       |    cor[PC4(t),PC3(t-1)]
   season     -------------------------------------------------------------------------------------------------
              ||   OBS    |   CTR    |   SCA    |  OBS     |  CTR     |  SCA    |   OBS    |   CTR    |   SCA
===============================================================================================================
     DJF      ||   0.71   |   0.84   |   0.85   | -0.58    | -0.70    | -0.71   |   0.42   |   0.41   |   0.51
              ||  -0.66   |  -0.79   |  -0.81   |  0.62    |  0.72    |  0.75   |  -0.47   |  -0.48   |  -0.55
---------------------------------------------------------------------------------------------------------------
     MAM      ||   0.62   |   0.80   |   0.85   | -0.49    | -0.63    | -0.69   |   0.50   |   0.50   |   0.51
              ||  -0.58   |  -0.78   |  -0.81   |  0.51    |  0.64    |  0.73   |  -0.53   |  -0.56   |  -0.55
---------------------------------------------------------------------------------------------------------------
     JJA      ||   0.56   |   0.75   |   0.81   | -0.35    | -0.43    | -0.42   |   0.49   |   0.54   |   0.49
              ||  -0.55   |  -0.74   |  -0.78   |  0.40    |  0.49    |  0.55   |  -0.48   |  -0.60   |  -0.50
---------------------------------------------------------------------------------------------------------------
     SON      ||   0.70   |   0.82   |   0.85   | -0.44    | -0.64    | -0.61   |   0.40   |   0.49   |   0.35
              ||  -0.66   |  -0.79   |  -0.83   |  0.47    |  0.69    |  0.66   |  -0.41   |  -0.53   |  -0.39