Climate Dynamics

, Volume 39, Issue 1, pp 95–112

Assessment of atmosphere-ocean general circulation model simulations of winter northern hemisphere atmospheric blocking

Authors

    • Climatic Research Unit, School of Environmental SciencesUniversity of East Anglia
  • Tim J. Osborn
    • Climatic Research Unit, School of Environmental SciencesUniversity of East Anglia
Article

DOI: 10.1007/s00382-011-1177-z

Cite this article as:
Vial, J. & Osborn, T.J. Clim Dyn (2012) 39: 95. doi:10.1007/s00382-011-1177-z

Abstract

An assessment of six coupled Atmosphere-Ocean General Circulation Models (AOGCMs) is undertaken in order to evaluate their ability in simulating winter atmospheric blocking highs in the northern hemisphere. The poor representation of atmospheric blocking in climate models is a long-standing problem (e.g. D’Andrea et al. in Clim Dyn 4:385–407, 1998), and despite considerable effort in model development, there is only a moderate improvement in blocking simulation. A modified version of the Tibaldi and Molteni (in Tellus A 42:343–365, 1990) blocking index is applied to daily averaged 500 hPa geopotential fields, from the ERA-40 reanalysis and as simulated by the climate models, during the winter periods from 1957 to 1999. The two preferred regions of blocking development, in the Euro-Atlantic and North Pacific, are relatively well captured by most of the models. However, the prominent error in blocking simulations consists of an underestimation of the total frequency of blocking episodes over both regions. A more detailed analysis revealed that this error was due to an insufficient number of medium spells and long-lasting episodes, and a shift in blocking lifetime distributions towards shorter blocks in the Euro-Atlantic sector. In the Pacific, results are more diverse; the models are equally likely to overestimate or underestimate the frequency at different spell lengths. Blocking spatial signatures are relatively well simulated in the Euro-Atlantic sector, while errors in the intensity and geographical location of the blocks emerge in the Pacific. The impact of models’ systematic errors on blocking simulation has also been analysed. The time-mean atmospheric circulation biases affect the frequency of blocking episodes, and the maximum event duration in the Euro-Atlantic region, while they sometimes cause geographical mislocations in the Pacific sector. The analysis of the systematic error in time-variability has revealed a negative relationship between the high-frequency variability of the transient eddies in the areas affected by blocking and blocking frequency. The blocking responses to errors in the low-frequency variability are different according to the region considered; the amplitude of the low-frequency variability is positively related to the blocking frequency and persistence in the Euro-Atlantic sector, while no such consistency is observed in the Pacific.

Keywords

Atmospheric blockingAOGCMsModel evaluationNorthern hemisphereWinter

1 Introduction

Characterized by its persistence and quasi-stationary features, large-scale atmospheric blocking is often responsible for extreme weather events, which in turn can sometimes have enormous impacts on human life, economy and environment. Examples of such episodes are the cold spell that hit Europe in February 1986 (Hoskins and Sardeshmukh 1987), or the extremely dry and hot weather over western Europe in July 1976 (Green 1977). The adverse impacts of blocking on society has solicited the interest of many scientists since the middle of the twentieth century. Synoptic analyses of blocking episodes and their impact on regional climate have been documented in several papers, examples being Namias (1947), Rex (1950b), Treidl et al. (1981) and more recently Trigo et al. (2004).

The first well-accepted definition of blocking was given by Rex (1950a), and was then revised in later studies. Several theories explaining different aspects of blocking including the mechanisms responsible for its onset, persistence and decay have been developed, and can be found in Rex (1950a), Green (1977), Grose and Hoskins (1979), Tung and Lindzen (1979), Hoskins and Sardeshmukh (1987) and Ferranti et al. (1994a). This large number of theories indicates that mid-latitude blocking anticyclones are complex phenomena involving the interaction of several components of the atmospheric circulation at different scales and in different regions of the world. As a result, the model simulation of blocking is a difficult and challenging task for both operational forecast and climate studies.

Blocking highs have mostly been investigated in forecast models, as for example in Tibaldi and Molteni (1990, hereafter TM90), Anderson (1993), Ferranti et al. (1994b), Jung (2005) and Brankovic and Molteni (1997), and Atmosphere-only General Circulation Models (AGCMs) in Tibaldi et al. (1997, hereafter TEA97), D’Andrea et al. (1998, hereafter AEA98), Doblas-Reyes et al. (1998), Doblas-Reyes et al (2002) and Matsueda et al. (2009). It is only recently that studies have started to deal with blocking in coupled climate models. Ringer et al. (2006) compared blocking frequency in the northern hemisphere in the atmosphere-only HadGAM1 and the coupled HadGEM1 climate models, and found that HadGAM1 has a tendency to produce more blocking events than HadGEM1, so consistent with the work done by Hinton et al. (2009) in the case of Pacific blocking, which suggests that the atmosphere-only model performs better than the equivalent coupled model. In their assessment of the AOGCM MPEH5/MPI-OM, Sillmann and Croci-Maspoli (2009) found that the frequency distribution and the location of atmospheric blocking occurring in the Euro-Atlantic sector were relatively well simulated.

The objective of the present study is to compare the ability of several climate models to simulate atmospheric blocking in the northern hemisphere. The use of an ensemble of climate models is useful in order to estimate the uncertainty due to model formulation, which is essential when considering future projections. As context for this work, the main conclusions drawn from relevant model studies will be reviewed, emphasizing AEA98’s work in particular. A feature common to all AGCMs was to underestimate the frequency of large-scale blocking events. This model deficiency was also observed by TM90 in the ECMWF operational medium-range forecast model, by TEA97 in their assessment of the ECHAM3 model at different horizontal resolutions, by Brankovic and Molteni (1997) in four different versions of the ECMWF NWP model, and by Doblas-Reyes et al. (2002) for the ARPEGE GCM with two different methods of blocking identification. In AEA98’s study, a few cases of blocking overestimation were found in the western Atlantic region, which was also prone to large decadal variability in the frequency of blocking.

AEA98 also noted that grid-point models with high resolution tend to perform better than grid-point models with lower resolution, while for spectral models the influence of resolution was less apparent on their performance. On the other hand, TEA97 found that increasing the horizontal resolution of ECHAM3 was sufficient to enhance the number of blocking events in the European-Atlantic sector, while in the Pacific higher resolution integrations using observed variations in sea surface temperature (SST) as boundary conditions simulated blocking frequency better than when climatological SST was used. As mentioned in AEA98, this fact suggests that blocking anticyclones in the European-Atlantic and Pacific regions might be the product of different processes. Ferranti et al. (1994a) investigated the impact of tropical diabatic heating anomalies on the mid-latitude circulation, and found that blocking occurring in the European-Atlantic and Pacific regions are both enhanced with a positive SST anomaly in the Indonesian region, while a cold anomaly in the tropical Caribbean sector results in an increase of Euro-Atlantic blocking only. The latter is consistent with Hoskins and Sardeshmukh (1987)’s findings in their diagnostic study of the 1986 winter blocking event.

Another feature common to all models in AEA98’s intercomparison project was an underestimation of blocking duration; none of the models were able to produce blocks as long as some that are observed. This feature was also observed by TEA97, especially for the lower resolution integration in the European-Atlantic sector.

In terms of blocking signatures (difference between the average field of blocked days and that of non-blocked days), which is a useful tool to evaluate the intensity, shape and location of blocking, some models tended to simulate blocking in the European-Atlantic sector with an apparent eastward shift, while in the Pacific, models showed more varied behavior. The intensity of the simulated blocking varied across the models, however a few models simulated block intensity quite well (AEA98). Apart from a slight westward displacement and a weaker low associated to the block, TEA97 noted that the main problems of the models are their inability to simulate localized signatures, and their tendency to overestimate the longitudinal extension of Pacific blocking.

In addition to assessing the ability of a new generation of coupled climate models to simulate atmospheric blocking in the northern hemisphere, this paper also examines the relationship between model systematic errors (SE) and the simulated blocking events. The datasets and models analysed, as well as the method used to identify blocking episodes are described in Sect. 2. Section 3 is devoted to the main results of this model assessment, focusing on the frequency, duration and signature of blocking episodes. The impact of the time-mean and time-variability SE on blocking statistics is analysed in Sect. 4 Finally, Sect. 5 contains some concluding remarks and identifies possible sources of model uncertainty that could impact on the ability of the models to reproduce blocking. Additional results are included in the electronic supplementary material, and are referred as “Online Resource (OR)”.

2 Datasets, models and blocking index

The data analysed are daily fields of 500 hPa geopotential height, averaged from the 6-hourly data outputs spanning 42 years from September 1957 to August 1999. Only the winter months (December, January, February, DJF) are considered in this analysis.

Data from the models was provided by the CERA database, run by the Model and Data group at the Max-Planck Institute for Meteorology, Hamburg, Germany. The models involved in this study are part of the ENSEMBLES (Ensembles-Based Predictions of Climate Changes and Their Impacts) European project, and contributed to the Intergovernmental Panel on Climate Change fourth assessment report (IPCC AR4). The ENSEMBLES project includes eight models; only the six models listed in Table 1 have been chosen because of data availability. The models are coupled Atmosphere-Ocean General Circulation Models (AOGCMs). The first run of the 20th century (20C3M) simulation is used as a representation of the “present climate”. ERA-40 reanalysis dataset, provided by the European Centre for Medium-Range Weather Forecasts (ECMWF), was used for the verification of climate model simulations. The NCEP/NCAR reanalysis (Kalnay et al. 1996) was also used to compare blocking climatology as it emerges from both reanalyses. All analyses have been done using model output regridded (using bilinear interpolation) to match the 2.5° resolution of the reanalyses.
Table 1

Summary of AOGCMs used

Model acronym

 

Institution

 

Horizontal and vertical resolution

Atmosphere and Ocean

CNRM-CM3 (CNCM3)

Centre National de Recherches

T42 (2.8° × 2.8°) L45 and 2° × 0.5° − 2°L31

Meteorologiques (CNRM), France

BCCR-BCM2 (BCM2)

Bjerknes Centre for Climate

T42 (2.8° × 2.8°) L31 and 1.5° × 0.5° − 1.5°L35

Research (BCCR), Norway

ECHAM5/MPI-OM (MPEH5)

Max Plank Institute for Meteorology (MPIM), Germany

T63 (1.9° × 1.9°) L31 and 1.5° × 1.5°L40

MA/ECHAM4 (EGMAM)

Institute for Meteorology, Freie

T30 (3.8° × 3.8°) L39 andT42

Universitaet Berlin (FUB), Germany

IPSL-CM4 (IPCM4)

Institute Pierre Simon Laplace

3.8° × 2.5°L19 and 2° × 2.5°L31

(IPSL), France

HadGEM1 (HADGEM)

Hadley Centre for Climate Prediction and Research/Met Office, United Kingdom

1.9° × 1.3°L38 and 1° × 0.3° − 1° L40

The blocking detection methodology is based on the TM90 index, which identifies atmospheric blocking highs when easterlies are identified in the region of the mid-latitude jet streams and storm tracks. Recent studies have emphasized that the choice of the reference latitude, used to define the location to search for easterly flow, is a crucial parameter in the blocking detection method (Pelly and Hoskins 2003; Barriopedro et al. 2010). The most common reference latitude used is a fixed latitude at \(60^{\circ}\hbox{N}\,\pm\,\Updelta\), where \(\Updelta\) is usually taken as one or two grid latitudes (TM90, TEA97, AEA98). Because of their large-scale and stationary properties, atmospheric blocking are known to block the progression of the transient synoptic-scale cyclones and anticyclones. Therefore, Pelly and Hoskins (2003) suggested using the latitude of the maximum storm track intensity to search for the central latitude of a blocking structure. Scherrer et al. (2006) used all grid latitudes between 35°N and 70°N, which include the fixed reference latitudes of TM90 and the storm track latitudes of Pelly and Hoskins (2003).

In this study, the same blocking index has been tested with two different sets of latitudes. One index uses the latitude of the storm track (hereafter, BIstlat), calculated for each longitude as the latitude with maximum temporal standard deviation of the 2–8 day band-pass filtered geopotential field at 500 hPa (using a finite impulse response filter based on Kaiser’s window). The storm track latitude is climatology-dependent, so in order to take into account possible model biases, the latitude of the storm track is calculated for each reanalysis and model dataset used in this study. The other set of latitudes includes all grid latitudes between 45°N and  65°N (hereafter, BI).

The blocking criteria used in this study are then defined as follows (see Fig. 1 for a schematic representation):
$$\frac{1}{\Updelta\phi} \left(\overline{Z}_{central} -\overline{Z}_{south} \right) > 0 $$
(1)
$$ \frac{1}{\Updelta\phi} \left(\overline{Z}_{north} -\overline{Z}_{central} \right) < - 10m/^{\circ} lat $$
(2)
$$ \overline{Z}_{central}^{\prime} > 0.8 \times \sigma\left(\overline{Z}_{central}^{\prime}\right) $$
(3)
Where,
$$ \begin{aligned} \overline{Z}_{central}^{\prime} &= \overline{Z}_{central} - \mu\left(\overline{Z}_{central}\right)\\ \overline{Z}_{central} &= \frac{1}{5} \sum_{j=\phi_0}^{\phi_0+\Updelta\phi} Z(\lambda_0,j)\\ \overline{Z}_{south} &= \frac{1}{5} \sum_{j=\phi_0-\Updelta\phi}^{\phi_0} Z(\lambda_0,j)\\ \overline{Z}_{north} &= \frac{1}{5} \sum_{j=\phi_0+\Updelta\phi}^{\phi_0+2\Updelta\phi} Z(\lambda_0,j) \end{aligned}$$
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig1_HTML.gif
Fig. 1

Schematic representation of a blocking candidate centered at the reference latitude ϕ0 and longitude λ0. Figure adapted from Fig. 2 in Pelly and Hoskins (2003)

The overbar signifies a latitudinal average over 5 grid latitudes, σ and μ symbolize the temporal standard deviation and mean, and the prime is a deviation from this time average. ϕ0 is the reference latitude of the blocking structure, taken as the storm track latitude, which is allowed to oscillate up to ± 7.5° (BIstlat), or as all grid latitudes between 45°N and  65°N in turn (BI). \(\Updelta\phi=10^{\circ}\); other studies, such as Pelly and Hoskins (2003) and Barriopedro et al. (2010), have used 15°, although it was found that the exact value of this parameter is not crucial. Although in this present study the models have been interpolated to 2.5° before computing the blocking index, the applicability of this methodology should not be restricted to this specific grid resolution. Therefore, a longitude λ is represented by a centered average of three grid longitudes (7.5°), in order to broaden the applicability of the blocking index by reducing the effect of resolution biases on coarser resolution climate models. The criterion (Eq. 3) requires an anomaly threshold at the location of the blocking high, chosen at \(0.8 \times \sigma \left(Z^{\prime}\right)\) (220 gpm on average for ERA-40). Other studies have adopted subjective values ranging between 100 gpm (Carrera et al. 2004) and 300 gpm (Sausen et al. 1995). Barriopedro et al. (2010) used an anomaly threshold computed as the one standard deviation level of the daily anomaly distribution. There is no crucial value for this parameter, the higher it is, the more mature and less frequent the blocks are, so to ensure statistically meaningful results a relatively low anomaly threshold should be used. However, a subjective sensitivity analysis performed using the full and anomaly geopotential height fields averaged for the life-cyle of large-scale blocking episodes (see OR2 for more details), suggests that if the anomaly threshold is too low a large proportion of patterns identified by the blocking index are only small amplifications of the planetary ridges, which do not seem to deflect the basic current as a more intense block would do. Therefore, \(0.8 \times \sigma \left(Z^{\prime}\right)\) was found to be the more adequate value for this parameter.

The blocking index (Eqs. 13) is computed at a given longitude, specific time and for any latitude ϕ0, and an instantaneous (in space and time) blocking candidate is assigned to the reference latitude ϕ0 where the zonal wind reversal (Eq. 1) is the greatest. A large-scale blocking event is then defined when the index identifies instantaneous blocking at each grid longitude over at least 15°, within predefined blocking sectors, namely the Euro-Atlantic (EA, 90°W − 90°E) and the Pacific (PAC, 90°E − 270°E) sectors. To avoid edge problems in the attribution of a large-scale blocking event to a given sector, the central longitude of a large-scale blocking, defined here as the maximum height anomaly grid longitude (λ), is required to lie within that sector.

Finally, the definition of a large-scale blocking episode requires that a large-scale blocking candidate, having its center within a specified sector, persists for at least 5 days. However, one non-blocked day in a period of at least 4 blocked days is considered as blocked, and the entire period is then defined as a blocking episode.

Figure 2 shows the blocking frequency of all winter days belonging to large-scale blocking episodes for ERA-40, obtained with the index that uses only the storm track latitude (BIstlat, grey curve), compared with the index that scans across all grid latitudes between 45°N and  65°N (BI, black curve). Both indices represent the two main blocking sectors in the Euro-Atlantic and Pacific sectors, with more frequent Atlantic than Pacific cases, which is in agreement with observational-based climatologies (e.g. Rex 1950b). However, the latter also emphasize that the Pacific blocking maximum is located over eastern regions between 210°E and  225°E, which is in line with the results obtained using the BIstlat but not the BI index.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig2_HTML.gif
Fig. 2

Frequency of all winter days that are part of large-scale blocking episodes as a function of longitude for the BI (black curve) and BIstlat (grey curve) indices. Results for ERA-40 only are shown

The performance of each blocking index is evaluated by a subjective assessment of the full and anomaly geopotential height fields averaged for the life-cyle of large-scale blocking episodes, following a set of criteria based on well-known blocking features, as described in earlier synoptic-based blocking studies (see OR2 for more details). The results of this assessment reveal that the number of blocking patterns detected using all grid latitudes is considerably higher than if the latitude of storm track alone is used. Although the number of misdetections is lower with the BIstlat than with the BI index, the BIstlat index does miss out a large number of situations, which would be considered to be blocked. Our attempts to devise an alternative index that could capture these blocking events without also capturing a large number of non-blocked episodes, particularly in the western Pacific sector, were unsuccessful. Therefore, for consistency across all longitudes, the index with the storm-track reference latitude is being used in this study.

3 Results

3.1 Statistical significance

A Monte Carlo statistical test is performed in order to verify whether the differences in blocking frequency and duration between ERA-40 and any other reanalysis or model output are or are not significant. A 95% confidence interval (shown by the grey area in the appropriate figures below) is constructed for ERA-40, by randomly removing one year from each of the performed 500 Monte Carlo simulations. Significantly different results from those of ERA-40 should lie beyond the confidence limits.

3.2 Intercomparison of reanalyses

Blocking frequency and duration have first been analysed for the ERA-40 and NCEP/NCAR reanalyses, and results are presented in Fig. 3. There are two main regions of blocking development: the Euro-Atlantic (EA, 90°W − 90°E) and the Pacific (PAC, 90°Eto 90°W). There is possibly a third region of blocking development at the boundary between western Europe and Asia (at around 60°E, in the Siberian sector). This sector, presenting a sharper topography with the Ural Mountains, has been identified in other studies (Barriopedro et al. 2006; Doblas-Reyes et al. 2002; Pelly and Hoskins 2003; TM90), but is mostly regarded as an extension of the EA blocking. Similar amplitudes of the blocking frequency are also found in Barriopedro et al. (2006).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig3_HTML.gif
Fig. 3

a Frequency of all winter days that are part of large-scale blocking episodes as a function of longitude, b blocking duration defined as the average frequency of winter large-scale blocking episodes (at least 5 days duration) as a function of duration (in days) for the Atlantic sector and c for the Pacific sector. ERA-40 (black) and the 95% confidence interval (grey area), NCEP/NCAR (red). Note that in b and c, the frequency of blocking episodes whose rarity is below 1% is not shown; instead the duration of the event is indicated by a marker at the base of the graph. And for ease of visualisation, the x-axis is cut at 15 days in the EA and PAC sectors; the maximum duration is indicated in each panel

The two reanalyses show almost identical frequency in both sectors (Fig. 3a). However, differences arise in blocking lifetime (Fig. 3b, c), especially in the Pacific sector. These differences might be related to an increased uncertainty in the reanalysis due to a smaller number of observations in this region compared to European-Atlantic sector. It could also be due to sampling errors, because of the smaller frequency of blocking events in the Pacific sector. But overall, ERA-40 and NCEP/NCAR reanalyses show similar blocking climatologies.

3.3 The impact of model internal variability on blocking

Uncertainties in blocking climatology due to model internal variability have been examined for CNCM3. Blocking frequency and duration have been computed for an ensemble of 3 runs (including anthropogenic forcing only), and are shown in Fig. 4. For the blocking frequency diagnostic, all the three runs are very similar and the differences between them are very small compared to the difference with ERA-40. However, internal variability appears to affect more blocking duration, especially in the Pacific sector (Fig. 4c), which is consistent with a smaller number of cases in this region. The three sets of CNCM3 results span the ERA-40 frequencies for most durations (i.e., there is no systematic under or overestimation), even for the longest durations with one run producing longer blocking episodes than the longest diagnosed from ERA-40. However, in the Atlantic, the results are more uniform, and although the variability among simulations is also large, all runs produce more frequent short-lasting blocking events, and less frequent long-lasting episodes than ERA-40. In addition none of them are able to produce the longest blocks seen in the reanalysis. Therefore, overall the results show that uncertainties due to internal variability do not account significantly for the underestimation of blocking frequency and duration in the Atlantic sector, but that results should be considered with caution in the Pacific, at least for CNCM3. It must be emphasized, however, that the smaller the number of simulated blocking cases is, the larger the differences between the runs are, due to larger sampling errors. CNCM3 is the most affected by sampling errors as it is the model that underestimates the most blocking frequency in both the EA and PAC sectors (see Table 4 and Fig. 6).
Table 2

Mean (μ) and maximum range of persistence corresponding to the best exponential fit of the winter large-scale blocking episodes’ lifetime in the Euro-Atlantic (ATL) sector

Models

Euro-Atlantic

μ

Max. range of persistence (max. duration)

Uncorrected

TM

TM + HFV

TM + STD

Uncorrected

TM

TM + HFV

TM + STD

ERA-40

8.4 (8.4,8.5)

   

5–15 (26)

   

CNCM3

7.3

8.4+

8.4

8.9

5–12 (13)

5–15 (28)

5–15 (21)

5–16 (32)

BCM2

7.6

7.6

7.6

8.0+

5–13 (20)

5–13 (21)

5–13 (20)

5–14 (24)

MPEH5

7.1

7.6+

7.6

7.7+

5–12 (24)

5–13 (27)

5–13 (24)

5–13 (23)

EGMAM

7.6

8.0+

8.0

8.0

5–13 (21)

5–14 (23)

5–14 (16)

5–14 (22)

IPCM4

8.0

9.2

9.2

9.1+

5–14 (23)

5–17 (27)

5–17 (25)

5–17 (31)

HADGEM

7.6

8.0+

8.0

8.4+

5–13 (33)

5–14 (33)

5–14 (33)

5–15 (27)

The lower and upper bounds of the 95% confidence interval from the Monte Carlo simulation is also indicated in brackets for ERA-40. Simulated mean values not significantly different from the analysed mean are in bold. The maximum duration of blocking episodes is indicated in brackets for ERA-40 and the models. Directions of the change toward ERA-40 results after corrections (computed with reference to the previous column) of the SEs (see in Sect. 4) are also indicated. “+” = improvement, “−” = decline in model performance, no mark = no change

Table 3

Same as Table 2 for the Pacific (PAC) sector

Models

Pacific

μ

Max. range of persistence (max. duration)

Uncorrected

TM

TM + HFV

TM + STD

Uncorrected

TM

TM + HFV

TM + STD

ERA-40

7.2 (6.9,7.2)

   

5–12 (15)

   

CNCM3

8.1

8.0+

7.2+

6.7

5–15 (17)

5–14 (17)

5–13 (12)

5–11 (13)

BCM2

7.2

7.6

7.1+

6.8

5–12 (21)

5–13 (22)

5–12 (17)

5–11 (17)

MPEH5

7.7

7.6+

7.6

6.8

5–13 (18)

5–13 (17)

5–13 (16)

5–11 (10)

EGMAM

7.2

6.8

7.2+

6.8−

5–12 (16)

5–11 (14)

5–12 (14)

5–11 (16)

IPCM4

6.8

8.0

6.8+

7.2+

5–11 (13)

5–14 (23)

5–11 (13)

5–12 (26)

HADGEM

7.7

8.0

8.0

8.0

5–13 (17)

5–14 (19)

5–14 (19)

5–14 (19)

Table 4

Number of blocked days belonging to blocking episodes for the Euro-Atlantic (EA) and Pacific (PAC) sectors independently

Models

EA

PAC

ERA-40

1,331

397

CNCM3

261

159

BCM2

676

366

MPEH5

1,116

850

EGMAM

749

285

IPCM4

610

239

HADGEM

1,144

340

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig4_HTML.gif
Fig. 4

Same as Fig. 3 but for the 3 CNCM3 runs

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig5_HTML.gif
Fig. 5

Same as Fig. 4 but for HADGEM runs. HADGEM run 2 for 3 periods of 42 years from 1873 to 1999—also shown for comparison, HADGEM run 1 from 1957 to 1999. Anthropogenic forcing only are included in those simulations

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig6_HTML.gif
Fig. 6

Frequency (in %) of winter large-scale blocking episodes (lasting at least 5 days) as a function of longitude. 95% confidence interval for ERA-40 (grey area), AOGCMs (black). The models’ blocking frequency averaged over all longitudes is indicated in the upper left corner of each plot; the frequency for ERA-40 is 3.52%

3.4 The impact of the sampling period on blocking

The estimation of blocking frequency over short time integrations can be significantly different according to the period considered because of long-term variability, driven by changing external forcing (TEA97, AEA98). Therefore, by using different periods of 42 years, the blocking responses to long-term variability have also been tested for HADGEM (Fig. 5). The differences between the runs are relatively small in amplitude, suggesting that blocking frequency evaluated on such a time sample is not significantly sensitive to the long time-scale variability in the EA and PAC sectors. Note however that a different climatological mean-state resulting from transient climates might affect the latitude of the storm tracks, potentially causing changes in blocking distributions. For the duration diagnostic, the variability among simulations is more apparent in the Pacific than in the Atlantic, similarly to CNCM3 (see the previous section). This variability increases as well with the duration of blocking episodes. There is no systematic under or overestimation of frequencies for most spells and for the maximum persistence of blocking episodes in the Pacific sector. In the Atlantic, all runs produce more frequent short-lasting blocking events (5–6-day duration) than ERA-40, and maximum duration is realistic for all three simulations of HADGEM run 2.

Overall, internal climate variability, either alone (evaluated by comparing different ensemble members, including or not natural forcing) or combined with externally-forced changes (different periods during 19th and 20th centuries), seems to have a relatively minor impact on blocking frequency compared with the magnitude of some of the apparent AOGCM errors (e.g. Fig. 4). However, the large variability among simulations for the blocking duration, of similar or higher magnitudes than the errors in blocking duration in itself, arise when the number of simulated blocking cases is low (i.e. in the Pacific) and when considering the longest blocking episodes. This problem is likely to be due to sampling errors. AEA98 computed a least-squares fit of the frequency distribution of blocking duration for events lasting less than 20 days, in order to minimize this sampling problem. Having said that, the impact of internal variability and long-time variability is different according to the persistence of events and/or diagnostics considered. Consequently, it might not be possible to explain why blocking episodes seem to last longer or not in some models, and it would probably be more appropriate to focus on the frequency of shorter blocking persistence ranges (from 11 to 17 days, depending on the model; see Tables 2, 3).

3.5 Model validation

Now, we assess the ability of all six AOGCMs to simulate large-scale blocking episodes (i.e. spanning at least 15° longitude and persisting for at least 5 days) occurring in the northern hemisphere during the winter (DJF) season. For a general overview of the performance of each model to reproduce the atmospheric mean state and its variability, the 500 hPa mean state, stationary wave and high- and low-frequency variability have been computed from ERA-40 and the model outputs, and can be found in the electronic supplementary material (OR1 Figs. 18).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig7_HTML.gif
Fig. 7

Frequency of winter large-scale blocking episodes (lasting at least 5 days) as a function of duration in days, for the Euro-Atlantic (a) and Pacific (b) sectors, for the 95% confidence interval for ERA-40 (grey area) and AOGCMs

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig8_HTML.gif
Fig. 8

Scatter plot of the mean (μ) and standard deviation (σ) of the distributions of blocking lifetime (in days) for the Euro-Atlantic (a) and Pacific (b) sectors for ERA-40 (grey rectangle showing the 95% confidence limits from the Monte Carlo simulation) and AOGCMs (black squares). The straight line corresponds the relation between the mean and the standard deviation σ =  μ − 5 (see text)

3.5.1 Blocking frequency

The frequency of large-scale blocking episodes as diagnosed from ERA-40 and simulated by the models is shown in Fig. 6. The Monte Carlo significance test (as shown by the grey area in each panel) is complementary to the earlier evaluation of the internal variability influence (compare the ensemble spread in Fig. 4 with the model error in Fig. 6d). All the models simulate two main sectors of blocking occurrence in the Euro-Atlantic and Pacific sectors, although the exact longitudinal position of each sector is not always well captured. In BCM2 and HADGEM the maximum in Pacific blocking frequency is shifted westward and centered at around 180°. This model deficiency is consistent with a less prononced climatological trough between 135°E and 180°, as well as a less prononced ridge between 180° and 225°E, compared to the reanalysis (OR1 Figs. 1, 5). On the other hand, simulated Pacific blocking is shifted eastward in IPCM4, because of a more prononced trough and eastward extention of the Pacific jet stream. These features are also observed for EGMAM, although at a lesser extent.

A feature common to all models (except for MPEH5 and EGMAM in the Pacific) is an generalized underestimation of the large-scale blocking frequency, which is a well known problem that has been reported in several studies aiming at assessing climate models (AEA98; Doblas-Reyes et al. 1998; Matsueda et al. 2009; Ringer et al. 2006; Sillmann and Croci-Maspoli 2009; TEA97) and forecast models (Anderson 1993; Brankovic and Molteni 1997; Ferranti et al. 1994a, b; Jung 2005; TM90). Here this problem is more prominent for the Atlantic than the Pacific sector, although it was found that Pacific blocking frequency is also clearly underestimated by the models when the BI index is used (at the exception of MPEH5, which performs relatively well when the BI index is used - not shown). Such underestimation was suggested to be caused by several factors, such as too strong westerlies in the models, underestimation of transient wave activity, systematic errors in the tropical SST, model resolution and parameterization. In order to better understand model deficiencies, the role of errors in the time-mean atmospheric circulation and its variability on blocking simulations will be assessed in Sect. 4.

3.5.2 Blocking duration

The winter large-scale blocking duration from ERA-40 and simulated by the models are shown in Fig. 7a and b for EA and PAC sectors respectively, where the linearity of the results on the logarithmic scale used for the frequency axis suggests that the distributions of blocking lifetime follow an exponential decay. This result was verified in AEA98, and is also assessed for the AOGCMs used in this present study. The expected distribution for the models and ERA-40 is computed over different ranges of persistence from 10 days to the maximum blocking duration for ERA-40 and model simulations. The best exponential fit is then chosen at the range of persistence, such that the mean (μ) and the standard deviation (σ) of an exponential distribution of the form Y(x) = λ e−λ x, truncated at 5 days of minimum duration for the case of blocking lifetime, satisfy the best the relation σ =  μ − 5 = 1/λ (AEA98). This relation can be derived by computing (using integrations by parts) the first and second moments of the exponential distribution Y(x), to show that E(X) =  μ = 1/λ and E(X2) = σ2 = 1/λ2. Errors of the exponential fit as a function of the range of persistence are presented in Fig. 9; the best exponential fit corresponds to the minimum error values.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig9_HTML.gif
Fig. 9

Errors (in days) of the exponential fits in the Euro-Atlantic (a) and Pacific (b) sectors for ERA-40 and AOGCMs, where the error is the difference (in absolute value) between the theoretical relation μ − σ = 5, and the simulated relation. μ and σ for each simulation, plotted in Fig. 8, are taken at the maximum range of persistence where the error is minimum

As shown in Fig. 8, all the models lie close to the straight line σ =  μ − 5, at similar ranges of persistence as ERA-40. It therefore indicates a good model representation of the exponential decay of blocking lifetime. The mean (μ) and the ranges of persistence of blocking episodes giving the best exponential fit (the smallest errors) are presented in Tables 2, 3.

A common feature to all models that was reported for earlier generations (TEA97, AEA98) is the inability to reproduce the correct blocking duration for both EA and PAC regions. This problem of shorter simulated blocks seems to be alleviated in this present study, in the PAC sector. All models simulate blocking episodes at least as persistent on average as ERA-40, and are now able to produce longer blocks than in the reanalysis (see Table 3; except for IPCM4).

In the Atlantic sector, all models underestimate the average blocking lifetime, as well as the maximum duration for which the exponential distribution fits adequately the simulated distribution. The models tend to overestimate the frequency of very short-lived blocking events, and underestimate the frequency of longer-lasting episodes (Fig. 7a). It seems no longer to be the case in the Pacific, the frequency of short blocking episodes having a tendency to be underestimated, while no systematic under or overestimation is apparent for longer blocks.

TEA97 have noted that their lower resolution model reproduces more frequent short-lasting blocks, and less frequent long-lasting blocks than the analysis, which considering the straining mechanism theory by Shutts (1986), would be due to the lack of eddy activity. Their higher resolution model, on the other hand, was less affected by this problem, and could at least reproduce more longer blocks. They also emphasized that this problem was particularly prone to the Euro-Atlantic sector. In this assessment, underestimation of long-lasting blocking events and overestimation of short-lived blocks are also more prominent in the Atlantic than in the Pacific sector, but it is not clear whether this problem strictly relates to lower resolution models or not.

3.5.3 Blocking signatures

Diagnosis of blocking signatures can provide information about the average shape, intensity and location of blocking events. They are obtained by subtracting the average field of all non-blocked days from the average field of all blocked days lasting at least 5 days. They have been computed for each sector separately and are presented in Figs. 10 and 11 for the Atlantic and Pacific sectors, respectively. In addition, the average 500 hPa geopotential field during blocked days only (represented by the solid contours) can provide information about the average amplitude of the deflection of the basic flow. In the reanalysis, both the EA and PAC blocking show a clear dipole blocking structure in the 500 hPa geopotential heights, with the blocking anticyclone at high latitudes, and the associated blocking low on its equatorward side. The models mostly tend to perform well in reproducing these clear meridional dipole patterns. TEA97 found that the different versions of the ECHAM3 model were unable to produce localized signatures, and in particular wave propagation and/or larger extension of the blocking high and/or low could be identified.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig10_HTML.gif
Fig. 10

Blocking signature in the Atlantic of the 500 hPa geopotential field (color shading). Solid contours represent the average 500 hPa geopotential field during blocked days only. At least 5-day blocking events

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig11_HTML.gif
Fig. 11

Same as Fig. 10 for the Pacific sector

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig12_HTML.gif
Fig. 12

Frequency (in %) of winter large-scale blocking episodes as a function of longitude. 95% confidence interval for ERA-40 (grey area), uncorrected AOGCMs (black), time-mean corrected (green), time-mean and HFV correction (blue), time-mean and STD correction (magenta)

In the Atlantic, the intensity and location of the signatures are relatively well simulated by the models, although CNCM3, BCM2 tend to underestimate the deflection of the flow during blocking episodes. It will be shown in Sect. 4 that this deficiency is mainly related to the strong westerly bias in those models. Larger errors in the intensity and geographical location of the blocks emerge in the Pacific region. BCM2 and HADGEM shift blocking signature upstream (see also Fig. 6e, f). The opposite happens for EGMAM and IPCM4. A similar feature was also present in the frequency diagnostic (Fig. 6), and could be attributed to discrepancies in the climatological position and amplitude of the ridge and through in the Pacific region. CNCM3 and HADGEM tend to overestimate the intensity of the Pacific blocking high, while BCM2 underestimates it. Note that the robustness of those results may depends on the number of blocked days (see Table 4). In the case of CNCM3 for example, the over-intensified signature in the Pacific sector might be biased by too few blocked days (159 days). Nevertheless, the amplitude of the geopotential anomaly at the location of the blocking high, when it satisfies Eq. 3 only (not shown), is consistent with the intensity of the signatures in Figs. 10 and 11.

Model deficiencies in blocking simulations are not always straightforeward to understand. For this purpose, in the next section, SE in both the time-mean and time-variability of the atmospheric circulation are examined, and their relationship with errors in blocking simulations are evaluated.

4 The impact of the time-mean and time-variability systematic errors on blocking

The impact of the time-mean SE on blocking frequency has been analysed previously, as for example in the ARPEGE GCM (Doblas-Reyes et al. 2002), in the ECHAM3 model (TEA97), in two Hadley Centre AGCMs (HadAM3 and HadGAM1) (Scaife and Knight 2008), and in a set of 18 AOGCMs used in the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4) (Scaife et al. 2010). Those studies revealed that the frequency of blocking episodes increases substantially, although still underestimating the observed frequency, after correction of the time-mean state, which mainly corrected the too strong average zonal flow and the stationary wave pattern in the model climatology. Scaife et al. (2010) have also suggested that the time-mean bias, rather than errors in the variability of the model, is the main cause for the lack of simulated blocking episodes.

Doblas-Reyes et al. (2002) showed that blocking climatologies, as simulated by climate models, are dependent on the criterion employed to define blocking. For example, the use of the SKS index (identification of positive height anomalies) is preceded by the removal of the mean state and the annual cycle. Therefore, this method is sensitive to the model ability to simulate the variability, but not the mean. In this present study, the blocking index is based on geopotential height gradients, and is sensitive to both the mean and variability of the model. This is illustrated in Fig. 13 for the geopotential height gradient reversal (Eq. 1) at the reference latitude (ϕ0). The geopotential height gradient at ϕ0 is reversed (Eq. 1) when the curves cross (Zn) > Zs)). Given that too strong westerlies are a common feature in most models, reducing the time-mean bias (thick black arrows) will often increase (decrease) the geopotential height at ϕns), so the gradient will have more occasions to be reversed. Similarly, increasing or decreasing the amplitude of the low- and high-frequency variability (thin dashed arrows) will affect the occurrence of a positive gradient.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-011-1177-z/MediaObjects/382_2011_1177_Fig13_HTML.gif
Fig. 13

A schematic representation of the dependency between errors in the background flow and the GPH gradient reversal (Eq. 1) at ϕ0. The variation of the GPH as a function of time at a latitude 10° south (ϕs) and north (ϕn) of ϕ0 are represented in red and blue, respectively. The mean state, the low- and high-frequency variability are represented by the straight dashed, sinusoidal dotted and sinusoidal solid curves, respectively. The thick black arrows represent the time-mean bias correction, and the thin dashed arrows represent an increase or decrease in the amplitude of the low- and high-frequency variability. An occurrence of the GPH gradient reversal is indicated (Zn) > Zs)). A modification of any of the components of the flow affects the applicability of the blocking index

In order to assess systematic model errors that affect the applicability of the blocking index, blocking statistics have been re-computed for post-processed model output, in which the climatological mean state and its variability have been corrected.

4.1 Impact of the time-mean model errors

Model errors in the mean state, which affect the stationary wave pattern and the meridional GPH gradient of each model, are removed by subtracting the time-mean SE from the daily geopotential field, as follows:
$$ x_M^{\prime} = x_M - \left(\mu_M - \mu_E\right) $$
(4)
where xM and \(x_M^{\prime}\) are the daily model and daily corrected model geopotential heights respectively, μM and μE are the time-average of the model and ERA-40 outputs, respectively. For reference, the time-mean SE \(\left(\mu_M - \mu_E\right)\), is shown in OR1 Fig. 5.

Blocking frequency and duration diagnosed after removal of the time-mean bias are shown by the green curves in Fig. 12 and in the second column of Tables 2 and 3, repectively. In all models blocking distribution is now improved (Fig. 12). Correction of the strong westerly bias and stationary wave pattern enhances blocking occurrence in the Atlantic, while in the Pacific BCM2, HADGEM and IPCM4 now simulate better blocking episodes over the right regions, although for BCM2 and HADGEM some events over western Pacific are still captured. All models tend to produce more persistent blocking events after correction of the time-mean SE in the Euro-Atlantic region (See Table 2). The average blocking lifetime and maximum persistence increase in EA, and are now improved for all models; IPCM4 is the only model for which the correction is not beneficial in the EA sector. The frequency of short-lived blocking events decreases, and the frequency of long-lasting episodes increases (not shown), such that the simulated blocking lifetime distributions in the EA sector are now substantially improved.

In the Pacific, however, the time-mean correction does not necessarily contribute to improve the blocking lifetime distribution; CNCM3 and MPEH5 are the only models for which the correction improves slightly the average persistence. It must be emphasized however, that because of the low number of Pacific blocks, the results are subject to sampling problems. Note the irregularities of the Pacific blocking lifetime distributions in Fig. 7b for the models with the largest underestimation of blocked day frequency (CNCM3, IPCM4 in Table 4 and Fig. 6).

Overall, those results suggest that the climatological time-mean state is important for the initiation and maintenance of blocking episodes in the EA region, and that a reduction of the time-mean SE in climate models would significantly improve blocking diagnostics in that sector. In addition, correction of the time-mean flow does also contribute to improve the frequency distributions in the Pacific sector, and particularly it enhances the number of blocking events in the right geographical sector (i.e. eastern Pacific), but it does not improve the average persistence of blocking episodes.

Blocking signatures have also been analysed after the corrections (not shown). The time-mean corrected fields show more intense signatures and/or more prononced deflection of the flow in the Atlantic for all models. In the Pacific, correcting the time-mean geopotential field improves the geographical location of the blocking sectors for BCM2, HADGEM and IPCM4, which are now centered between 180° and 225°E. The intensity of the blocking anticyclone is modified for some models, leading to larger errors (IPCM4 and BCM2), while it is corrected for CNCM3. Therefore, time-mean circulation bias, such as too strong westerlies, underestimated or displaced stationary waves, can also induce errors in blocking intensity, longitudinal/latitudinal extension and geographic displacement of the whole structure and in the average deflection of the flow.

4.2 Impact of the time-variability model errors

To examine the impact of model errors in the amplitude of time-variability, two other analyses are considered. One consists of evaluating blocking statistics after “correction” of both the time-mean and standard deviation of the high-frequency (2-8 days) component of the variability (Fig. 12, blue line, and Tables 2, 3, third column). In the second analysis, the time-mean and total time-variability (total standard deviation, STD) are corrected (Fig. 12, magenta line, and Tables 2, 3, forth column). These two analyses will be useful to distinguish the impact of the HFV and LFV independently on blocking simulations. The calculation involved for this type of correction is as follow:
$$ x_M^{\prime} = \left[\left(x_M - \mu_M\right)\frac{\sigma_E}{\sigma_M}\right] + \mu_E $$
(5)
where \(x_M, x_M^{\prime}, \mu_M\) and μE are defined as in Eq. 4; σM and σE are the standard deviation of the model and ERA-40 outputs respectively. When the HFV is analysed, only the band-pass filtered (between 2 and 8 days) data are considered in this equation.

The performance of the models at representing the time-variability of the flow is presented in OR1 Figs. 7 and 8, which show the ratio of the HFV and LFV, respectively. The main feature is an underestimation of the variability over the north Atlantic ocean and northern Europe, as well as in the north Pacific ocean, where the main storm tracks and LFV maxima are located (OR1 Figs. 3, 4). In the Pacific region, MPEH5 overestimates both the HFV and LFV, while HADGEM overestimates the intensity of the storm track with a slight southward displacement. In the north Atlantic, however, HADGEM simulates relatively well the variability.

After HFV correction blocking frequency tends to decrease in regions and models with underestimated HFV (Fig. 12, green and blue lines), although the signal seems to be weak compared to the amplitude of the model errors (e.g. CNCM3 and BCM2, OR1 Fig. 7); the opposite occurs for overestimated HFV. In the EA sector, the change in total blocking frequency seems to affect all spell durations; the average persistence is unchanged as compared to the TM correction alone (the frequency distributions before and after the correction are more or less parallel on the semi-logarithmic plots), and the maximum duration is lower (Table 2). In the Pacific, however, the average persistence and maximum duration tend to be lower (the slopes of the distributions are steeper), thus bringing an improvement towards ERA-40 as compared to the TM correction alone. Those results are consistent with an increased frequency of short-lived blocking episodes and less frequent long-lasting blocks (not shown). In the case of EGMAM, the total frequency, average and maximum persistence increase; the results are in line with less short-lasting blocking episodes and more long-lasting events, after correction of the overestimated amplitude of the HFV.

Underestimation of the LFV seems to be associated with lower blocking frequency in both sectors, and inversely (Fig. 12, blue and magenta lines). In the Atlantic, the average and maximum blocking duration tend to be increased, and as compared to TM and TM+HFV corrections, results are now closer to ERA-40 (Table 2); there is a tendency for increased frequency of long-lasting blocking episodes, and decreased frequency of short-lived events (not shown). In the Pacific, this correction tends to be associated with a decreased average persistence (Table 3) and frequency of long-lasting episodes (not shown), whatever the change in the total frequency (Fig. 12) and the type of correction (increased or deacreased LFV).

Correction of SE in the amplitude of the time-variability also impacts on blocking signatures and flow deflection (not shown). The correction of the total time-variability produces better results than the HFV correction only: the blocking events have a geographical position that is more comparable to those of ERA-40 events, but subtantial errors in the intensity of the blocking high also emerge after those corrections.

4.3 Summary and interpretation of results

The percentage change in the number of blocked days for each sector independently after each correction is presented in Table 5. It provides an estimate of the importance of each type of model errors, as well as an indication of the direction of the change (improvement or decline in model performance) in the average number of blocked days for each sector. It shows that corrections of the time-mean and LFV amplitude both contribute the most toward an increase and an improvement in the number of EA blocking cases. In the Pacific, those corrections bring also the major changes, but do not always contribute to an improvement in the number of blocked days (similar results are found for the persistence diagnostic in Table 3). It is also interesting to note the type of SEs in each model that are linked with errors in blocking diagnostics. For some models, errors in the mean state only are important (e.g. HADGEM in the Pacific). For other models, corrections including the LFV bring the best match with ERA-40 (e.g. BCM2 in the Atlantic, MPEH5 in the Pacific). Most of the models, however, need corrections in both the mean state and the variability to produce realistic blocking statistics. It should be noted however that the main source of error for the Pacific sector is blocking mislocation (e.g. EGMAM, IPCM4, BCM2 and HADGEM in Fig. 12), and rating the results by looking only at the number of blocked of days in Table 5, might be reductive as this diagnostic does not take into account model errors due to blocking mislocation. For instance, Table 5 suggests that IPCM4 best performs after the HFV correction, while from Fig. 2 TM correction is crucial for IPCM4 performance even more than what the HFV correction does. Therefore, both Table 5 and Fig. 12 should be considered to rate model performance in terms of the number or frequency of blocked days, particularly in the Pacific sector.
Table 5

Percentage change in the number of blocked days for the Euro-Atlantic and Pacific sectors after each correction (time-mean, TM; time-mean and HFV, TM + HFV; time-mean and STD, TM + STD)

Models

EA

PAC

TM

TM + HFV

TM + STD

TM

TM + HFV

TM + STD

CNCM3

211+

−25

98+

31+

0.5+

48+

BCM2

3+

−0.1

65+

26

−5+

−24

MPEH5

14+

−11

14+

−9+

−3+

−44+

EGMAM

17+

−13

43+

−20

54+

11+

IPCM4

50+

−18

32+

−13

13+

24

HADGEM

9+

−4

4+

43

−9+

1

As in Tables 2 and 3, the directions of the change toward ERA-40 results after corrections (computed with reference to the previous column) of the SEs are indicated: “+” = improvement, “−” = decline in model performance, no mark = no change

The correction exercise reported here will only change the blocking statistics if the important systematic errors occur within the blocking area itself. Systematic errors from outside the blocking area could also influence the development or removal of blocks, but the corrections applied here cannot directly compensate for those errors. This is particularly relevant for errors in the high-frequency transient eddy activity, which can act from upstream to amplify and sustain a downstream blocking high (Austin, 1980; Illari 1984; Shutts 1983). For example, a positive bias in high-frequency synoptic activity could enhance blocking activity downstream (according to current understanding of the dynamical mechanisms linking eddies and blocking). As illustrated in Fig. 13, if the HFV bias is located within the preferred area for blocking occurrence, the effect of the HFV correction would be weak compared to corrections of the LFV bias and the time-mean bias, and a small or even null impact on the total blocking frequency might be expected. The HFV bias correction would, however, induce a change in the proportion of short and long-lasting blocks: greater eddy activity inside the blocking region will increase the frequency of short-lived blocks at the expense of long-lived blocks. This is supported by observing that transient eddy activity is reduced and/or displaced in the area affected by a blocking high. The effect of the various bias corrections is likely to depend, therefore, on the relative strengths and locations of the HFV, LFV and time-mean biases, and this could explain why corrections of the HFV bias tend to improve blocking activity in PAC but do not change (or even worsen) it in ATL.

It must also be emphasized that the STD correction involves the frequencies in which blocking operates (on intra-seasonal time-scales), and hence it might be expected to see large effects or even the largest improvements on blocking statistics after its correction. However, this correction also includes other time-scales (above the seasonal one), and therefore depending on where the largest model errors are (intra-seasonal or inter-seasonal time-scales), the effect of this correction might also be different, and would explain why more diverse results are obtained according to the model and region considered.

5 Discussion

The ability of six AOGCMs to simulate winter atmospheric blocking in the northern hemisphere has been evaluated by comparison with the ERA-40 reanalysis. These six climate models all underestimate the total blocking frequency over the northern hemisphere to various degrees (except MPEH5 and EGMAM in the north Pacific). Analysis of multiple simulations and periods demonstrates that these errors are real and not the result of internal variability or changes in the external forcing.

All models underestimate the average persistence of blocking episodes in the Euro-Atlantic sector. However in the Pacific, the average blocking lifetime tends to be overestimated, and some models are even able to produce blocking episodes as long as those identified in the ERA-40 reanalyses.

In terms of blocking signatures, results are more satisfactory than reported in earlier studies (TEA97, AEA98), particularly in terms of their ability to simulate more localized signatures in both sectors. However, errors in the intensity and geographical location of blocking episodes still emerge in the Pacific region.

Another interesting feature drawn from this assessment is the difference in the ability of the models to simulate blocking in both regions, which seems to suggest that blocking highs result from different processes in each sector, as suggested in other studies (TEA97, AEA98). It must be emphasized however, that model performances in blocking simulations is also affected by the choice of the blocking detection method and of the diagnostic. When using the BIstlat index, the frequency and average lifetime of Pacific blocking episodes are not clearly underestimated by all the models, whereas it is no longer the case once the BI index is used.

The impact of models’ systematic errors (SE) on blocking simulation has also been analysed. There is a primary need to reduce the time-mean atmospheric circulation bias to improve the representation of blocking in climate models. The time-mean SEs, which are mainly too strong average westerlies and zonality of the flow, as well as discrepancies in the stationary wave pattern, are mainly responsible for the underestimation of long-lasting blocking episodes, and a reduction of the maximum event duration in the Euro-Atlantic region, as well as geographical displacements of the Pacific blocking sector. There seems to be a weak negative correlation between the amplitude of the HFV and the frequency of long-lasting blocking episodes, consistent with the observed deviation of the transient eddies’ trajectory. The amplitude of the LFV is positively linked with the average blocking lifetime and frequency of long-lasting episodes in the Euro-Atlantic sector. In the Pacific, however, there is no systematic tendency: increased or decreased amplitude of the LFV tend to be linked with decreased average persistence and frequency of long-lasting blocking episodes. In addition, the correction of the time-mean and total standard deviation does not always contribute to improve blocking simulations in the Pacific.

Overall, depending on the model and region considered, errors in blocking diagnostics can be linked with the time-mean bias only, errors in the LFV only, or both. By splitting the total flow into its mean state and time-varying part, this assessment reveals if the simulated blocks are derived from a compensation of model errors or if blocking statistics are coherent with the background flow.

Three of the models evaluated in this study (EGMAM, CNCM3 and HADGEM) are improved versions of ECHAM, METEOFR and UKMO assessed in AEA98, in terms of parameterizations, dynamics and resolution. Furthermore, in this study the models are coupled atmosphere-ocean models, while in AEA98 atmosphere-only models with prescribed observed monthly mean SST were used. It has been demonstrated in a number of studies that the impact of SST is important for the development and maintenance of both Euro-Atlantic and Pacific blocking (e.g. Ferranti et al. 1994b). Having said that, despite considerable differences between the two generations of models (about 10 years of model development), there is only a moderate improvement in blocking simulation - though different blocking indices, analysis periods and model coupling must be born in mind. Another consideration is the physics/dynamics resolution dependence. Boer et al. (1992) have mentioned that the impact of improving physical parameterizations is resolution dependent. Similar arguments hold concerning the improvement of the dynamics in the model. Ringer et al. (2006) explained that the representation of synoptic-scale eddies improves with increased resolution when climate models use a Semi-Lagrangian dynamical core. However, at lower horizontal resolutions (typical climate model resolution of 2.5° Williamson and Olson 1998; Williamson et al. 1998), Eulerian models represent transient activity better than the Semi-Lagrangian models. The AOGCMs used in this present study all use a Semi-Lagrangian scheme, except IPCM4 which uses a finite volume scheme, but it is not clear here whether or not higher resolution models represent synoptic-scale eddies better.

Ringer et al. (2006) found that the atmosphere-only model HadGAM1 has a tendency to reproduce more blocking events than the atmosphere-ocean HADGEM model, apparently a response to a cold bias in the equatorial Pacific ocean in the latter. Similar results were reported by Ferranti et al. (1994b), who found that Pacific blocking frequency was increased when a positive anomaly was superimposed on the climatological SST in the Indonesian region. So more realistic SST simulations (variability as well as mean state) in coupled climate models could enhance blocking development and maintenance. AEA98 noted that nine of their AGCMs, also analysed by Slingo et al. (1996), with low tropical variability were not able to simulate blocking well.

CNCM3, BCM2, IPCM4 and MPEH5 were also assessed in Scaife et al. (2010), using a zonal wind-based version of the TM90 blocking index applied to all grid point between 50° and 70°N, and similar model errors in blocking frequency in both the Atlantic and Pacific sectors were also found. Therefore, even though model performance in simulating blocking is sensitive to the blocking detection method, the direction of model errors (under or overestimation) is generally insensitive.

One common problem of the AOGCMs used in this study, and many other climate and forecast models is to simulate too strong westerlies (OR1 Fig. 5) and to underestimate the frequency of blocking (Fig. 6). Those two facts might in fact be linked together, and one possible explanation has been given by Palmer (2001) using the concept of non-linearity of the climate system, and specifically the way governing equations are approximated at sub-grid scale by deterministic bulk formulae - conventional parameterization schemes neglect the variability of atmospheric components that are energetically weak. Suppose the atmospheric circulation in the extratropic has a bimodal distribution with a dominant and more stable regime (zonal flow), and a sub-dominant regime (blocked flow). Neglecting even a small percentage of the variability can have an important impact on the resolved flow because of the influence of non-linearity of unresolved scales, and might in particular underestimate the synoptic-scale variability that trigger the transition between the zonal and blocked flow. As a result, the dominant and more stable regime is overpopulated and characterised by a westerly bias, and the frequency of the sub-dominant regime (atmospheric blocking) will inevitably be underestimated. This line of thoughts is supported by the results presented in this study. According to the bimodale regime of the atmospheric circulation and non-linearity described by Palmer (2001), the underestimated amplitude of the HFV itself is consistent with a downstream westerly bias (within the blocking area). In addition, the correction exercise revealed a coherent and substantial improvement in blocking simulations after removal of this time-mean bias. Some studies are currently addressing this problem (Berner et al. 2008; Palmer 2001), which can be alleviated by representing sub-grid scale processes by simple non-linear stochastic-dynamic systems coupled to the resolved flow, so that kinetic energy of the unresolved flow can be backscattered to the large-scale components of the flow. Berner et al. (2008) found a significant reduction of systematic error, and an increase in the frequency of Pacific blocking.

Acknowledgments

This work contributes toward a PhD funded by the School of Environmental Sciences at the University of East Anglia. The authors would like to acknowledge Dr. Clare Goodess and Dr. Ian Renfrew for their helpful comments and suggestions. The authors are also grateful to anonymous reviewers for their critical comments and helpful suggestions that improved the quality of this work, and the clarity of the manuscript. The ENSEMBLES data used in this work was funded by the EU FP6 Integrated Project ENSEMBLES (Contract number 505539) whose support is gratefully acknowledged. We also thank European Centre for Medium-Range Weather Forecasts and National Oceanic and Atmospheric Administration/Earth System Research Laboratory Physical Sciences Division for providing ERA-40 and NCEP/NCAR reanalyses data.

Copyright information

© Springer-Verlag 2011