Seasonal Forecasting with a Simple General Circulation Model: Predictive Skill in the AO and PNA

Jacques Derome Department of Atmospheric and Oceanic Sciences, McGill University, Montreal, Quebec, Canada

Search for other papers by Jacques Derome in
Current site
Google Scholar
PubMed
Close
,
Hai Lin Department of Atmospheric and Oceanic Sciences, McGill University, Montreal, Quebec, Canada

Search for other papers by Hai Lin in
Current site
Google Scholar
PubMed
Close
, and
Gilbert Brunet Recherche en Prévision Numérique, Meteorological Service of Canada, Dorval, Quebec, Canada

Search for other papers by Gilbert Brunet in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

A primitive equation dry atmospheric model is used to perform ensemble seasonal predictions. The predictions are done for 51 winter seasons [December–January–February (DJF)] from 1948 to 1998. Ensembles of 24 forecasts are produced, with initial conditions of 1 December plus small perturbations. The model uses a forcing field that is calculated empirically from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalyses. The forcing used to forecast a given winter is the sum of its winter climatological forcing plus an anomaly. The anomalous forcing is obtained as that of the month prior to the start of the forecast (November), which is also calculated from NCEP data. The predictions are thus made without using any information about the season to be predicted.

The ensemble-mean predictions for the 51 winters are verified against the NCEP–NCAR reanalyses. Comparisons are made with the results obtained with a full GCM. It is found that the skill of the simple GCM is comparable in many ways to that of the full GCM. The skill in predicting the amplitude of the main patterns of Northern Hemisphere mean-seasonal variability, the Arctic Oscillation (AO) and the Pacific–North American (PNA) pattern is also discussed. The simple GCM has skill not only in predicting the PNA pattern during winters with strong ENSO forcing, but it also has skill in predicting the AO in winters without appreciable ENSO forcing.

Corresponding author address: Dr. Jacques Derome, Dept. of Atmospheric and Oceanic Sciences, McGill University, 805 Sherbrooke St. W., Montreal, QC, H3A 2K6 Canada. Email: jacques.derome@mcgill.ca

Abstract

A primitive equation dry atmospheric model is used to perform ensemble seasonal predictions. The predictions are done for 51 winter seasons [December–January–February (DJF)] from 1948 to 1998. Ensembles of 24 forecasts are produced, with initial conditions of 1 December plus small perturbations. The model uses a forcing field that is calculated empirically from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalyses. The forcing used to forecast a given winter is the sum of its winter climatological forcing plus an anomaly. The anomalous forcing is obtained as that of the month prior to the start of the forecast (November), which is also calculated from NCEP data. The predictions are thus made without using any information about the season to be predicted.

The ensemble-mean predictions for the 51 winters are verified against the NCEP–NCAR reanalyses. Comparisons are made with the results obtained with a full GCM. It is found that the skill of the simple GCM is comparable in many ways to that of the full GCM. The skill in predicting the amplitude of the main patterns of Northern Hemisphere mean-seasonal variability, the Arctic Oscillation (AO) and the Pacific–North American (PNA) pattern is also discussed. The simple GCM has skill not only in predicting the PNA pattern during winters with strong ENSO forcing, but it also has skill in predicting the AO in winters without appreciable ENSO forcing.

Corresponding author address: Dr. Jacques Derome, Dept. of Atmospheric and Oceanic Sciences, McGill University, 805 Sherbrooke St. W., Montreal, QC, H3A 2K6 Canada. Email: jacques.derome@mcgill.ca

1. Introduction

Forecasts of mean-seasonal atmospheric conditions, whether experimental or operational, are normally conducted with global dynamical models or with statistical techniques. The global dynamical models are either in the form of a coupled atmosphere–ocean system or are strictly atmospheric, in which case the ocean temperatures and sea ice conditions must be externally specified.

Whatever the approach used, the skill of seasonal forecasts is limited by the chaotic nature of the atmosphere, which implies that, except for the first few weeks of the forecast, the initial state of the atmosphere provides little or no information on the future state, or, conversely, that most of the predictable component of the variability is likely to come from some atmospheric forcing by the lower boundary, such as sea surface temperature (SST), sea ice, or soil anomalies. In fact, the interest in the possibility of producing skillful seasonal forecasts rose sharply with the realization that the tropical Pacific SST anomalies of the El Niño/La Niña type are correlated with atmospheric anomalies in regions as far away as the middle latitudes (Wallace and Gutzler 1981). As these SST anomalies have lifetimes of the order of seasons, they can influence the atmospheric conditions on a similar time scale, and hence can be used as predictors in statistical models or as forcing functions in dynamical models.

When a seasonal forecast is done with a numerical model it is now standard procedure to perform an ensemble of predictions whose members are started from different initial conditions and to average the predictions to filter out the chaotic component of the forecasts associated with the uncertainty in the atmospheric initial conditions. This ensemble-mean prediction can be used as a categorical forecast, or, preferably, the scatter among the ensemble members can be used to obtain a probabilistic forecast. The need to resort to an ensemble of forecasts results in a computationally intensive approach.

In the present study, we use a dynamical approach to seasonal forecasting but lower the computational cost by employing a simpler model than normally used in seasonal forecasting. We refer to the model as a “simple general circulation model” (SGCM). The SGCM uses a spectral representation of the primitive equations on the globe but is dry and uses a forcing that is kept constant throughout the forecast season. Hall (2000) has shown that the SGCM has a good boreal winter climatology, and Hall and Derome (2000) gave evidence that it produces a very creditable response to a tropical Pacific heat source or sink of the El Niño/La Niña type. Hall et al. (2001) then examined the linear and nonlinear responses of the model to midlatitude heat sources and sinks simulating sea surface temperature anomalies. The present study examines the seasonal forecasting skill of the SGCM over a period of 51 boreal winters and shows that it indeed has some statistically significant skill. The skill is then compared with that of a more complex GCM whose seasonal forecasts were available from previous work. Finally, the study shows how the skill of the SGCM and full GCM is related to the main patterns of mean-seasonal variability at 500 hPa, namely, the Arctic Oscillation (AO) and the Pacific–North American (PNA) pattern. It is now well known that a strong El Niño and La Niña can lead to a skillful seasonal forecast at least over the PNA sector, but as these are the exception rather than the rule, it is of interest to see if forecast skill can be demonstrated during seasons without SST anomalies of this type. We will show that, indeed, the SGCM has skill in predicting the AO, a skill that is not related to the presence of a strong El Niño or La Niña. To the authors’ knowledge, this is the first time that such seasonal prediction skill has been presented.

2. The model and experimental setup

Global forecasts were made for the 51 boreal winters [December–January–February (DJF)] from 1948/49 through 1998/99 with the SGCM. The latter uses a low-resolution spectral representation (T21) of the dry primitive equations, with five levels in the vertical. It is essentially the model first developed by Hoskins and Simmons (1975), to which Hall (2000) added time-independent forcing functions to the tendency equations and some linear damping of the momentum and thermodynamic equations. More details on the model and its climatology are given in Hall (2000). The forcing is obtained by computing the dynamical terms of the model, including the linear damping, with daily global analyses and setting the forcing equal to the residual term after averaging in time, as will be further discussed below. The model is then run with this time-independent forcing, which makes it computationally very economical compared to a full GCM.

The experimental protocol was designed to mimic that of an operational environment, in that the forecast system used no information whatsoever from the DJF period to be forecast. Ensembles of 70 integrations were done for each of the 51 winters starting on 1 December, but all results presented here are based on a subensemble of 24 members taken at random, in order to facilitate the comparison with the forecasts of the full GCM, for which only 24 members were available. It was found that the SGCM results, whether based on ensemble means of 24 or 70 members, were essentially the same. The initial conditions for the ensemble members were the 0000 UTC 1 December analyses from the National Centers for Environmental Predictions–National Center for Atmospheric Research (NCEP–NCAR; Kalnay et al. 1996), to which small-amplitude perturbations were added. These perturbations were in the form of scaled anomalies (deviations from winter climatology) from random winter days in the 51-yr NCEP–NCAR dataset (excluding the winter being predicted).

The daily data required to obtain the forcing were also taken from the NCEP–NCAR reanalyses. The forcing used for a given winter was the November-mean-forcing anomaly of that year added to the DJF-mean climatological forcing. Here and in the following, the climatology is defined as the average over the corresponding period (November or DJF) of the 50 remaining years, and the anomaly is the deviation of a given period from the climatology. For example, for year 1, the forcing used in the DJF forecast was the November-mean-forcing anomaly of year 1 added to the DJF-mean-forcing climatology. In short, the November forcing anomaly is assumed to persist through DJF. This approach parallels that of Derome et al. (2001) in which the November sea surface temperature anomalies were persisted through DJF in forecasts done with a full atmospheric GCM and a global numerical weather prediction model. In the latter study, the SST anomalies were transformed into heating/cooling anomalies by the models’ physical packages. This means that any deficiencies in the models’ physical packages resulted in errors in the forcing and hence presumably to errors in the forecasts. In the present study, this particular source of errors is absent, as the forcing anomaly itself (from November) is provided to the model. Naturally, the SGCM has other model errors not present, or not to the same extent, in the more complex models (e.g., dry dynamics, numerical errors due to low resolution, etc.).

3. The model forcing

Hall (2000) discussed the model forcing when based on 10 winters of data from the European Centre for Medium-Range Weather Forecasts (ECMWF). When based on the 51 DJFs of the present study, the climatological forcing of the thermodynamic equation (not illustrated here) shows the expected deep tropospheric diabatic heating in the tropical Pacific and the shallower diabatic heating over the storm tracks of the North Pacific and North Atlantic. This climatological forcing was used to run the model to an equilibrium climate, and the latter was found to be similar to that obtained by Hall (2000), that is, very reasonable compared to observations in the Northern Hemisphere, both in terms of the time-mean flow and the eddy statistics, but poorer in the Southern (summer) Hemisphere. As Hall (2000) has already discussed the model climate and our focus here is on the interannual variability when the forcing is allowed to change from winter to winter, we will not further discuss the equilibrium model climate here. In the next section, the SGCM systematic forecast error for DJF, a manifestation of the model drift toward its own climatology, will be shown to be of a similar magnitude to that of a more complex GCM.

Since our forecast approach is based on assuming that the November forcing anomalies persist through DJF, it is of interest to first check to what extent this assumption of persistence is verified in the dataset. The forcing anomalies are the synthesis of several factors, such as SST, sea ice, and land surface condition anomalies. In addition, any diabatic heating anomaly generated by the internal dynamics, such as the time-mean diabatic heating anomalies due to the synoptic-scale eddies, are present in the forcing. One might then expect that the persistence assumption would be better verified in the Tropics, where, for example, SST anomalies associated with El Niño/La Niña can last several months, than in the storm tracks off the coasts of Asia and North America. This is supported by Fig. 1, which shows a global map of the correlation between the November and DJF forcing anomalies, over the 51 winters, in the thermodynamic equation at 900 hPa. We examined the vertically averaged thermodynamic forcing field, as well as the other forcing fields. The picture is basically the same (not shown). The correlation, or persistence, is statistically significant over a considerable part of the tropical Pacific, but not over the storm-track regions mentioned above. It is also worth noting that the persistence is high over several other areas, such as the eastern part of the Asian continent, Africa, South America, and parts of the Arctic. If the SGCM is able to respond adequately to the forcing anomalies, the above results suggest that it could have some forecast skill in DJF. The fact that the forcing persistence is spatially variable suggests that it may be possible to improve on our approach of persisting the forcing anomalies everywhere, but in our first examination of the forecast skill of the model, we will not pursue this possibility.

4. The forecast skill

Figure 2a shows the temporal correlation (defined here as forecast skill), over the 51 winters, between the observed and ensemble-mean-predicted DJF-mean 500-hPa heights. Figure 2b is the corresponding figure for January–February (JF) only, that is, skipping the first month of the forecasts in order to isolate the contribution of the forcing from that of the initial conditions in the skill. We see that in DJF the SGCM has statistically significant skill over a broad tropical band and parts of the mid- and high latitudes of both hemispheres. Over the Northern Hemisphere, skill is found over the eastern part of Asia and over the Canadian Arctic, where the forcing had been observed to be persistent (Fig. 1). There is also skill over the eastern North Pacific and western and southeastern North America, in a pattern somewhat reminiscent of the PNA pattern. As expected, the skill is lower in JF (Fig. 2a) but still statistically significant at the 5% level (according to a Student’s t test) over many parts of the globe, particularly in the Tropics and those areas of the mid- and high latitudes where skill exists in DJF.

Naturally, it is of primary interest to see if the SGCM predictions have a higher skill than simple persistence forecasts. Figures 3a,b show the skill of persistence forecasts, that is, the correlation between the NCEP Z500—500-hPa height—observations in November and those in the following DJF and JF. We see that in the Tropics the persistence forecasts have a skill level that is comparable to that of the SGCM. In the extratropics of the Northern Hemisphere, on the other hand, the persistence forecasts have no statistically significant skill, with the exception of eastern Asia in DJF only, whereas the SGCM has skill in both DJF and JF in some areas, as discussed earlier (Fig. 2).

We now compare the forecast skill of the SGCM to that of a more complex GCM, namely, the second-generation general circulation model of the Canadian Centre for Climate Modelling and Analysis (GCM2; McFarlane et al. 1992). The GCM2 is a spectral model with a T32 resolution, 10 levels in the vertical, and a comprehensive suite of parameterizations of subgrid-scale processes. The model has been used in a number of studies of equilibrium and transient climate change experiments (Boer et al. 1992 and 2000a,b, respectively).

Seasonal predictions with 24-member ensembles were done with this model for the 26 DJFs during the period 1969–95 as part of the Canadian Historical Forecasting Project, and results based on the first 6 members were presented by Derome et al. (2001). Kharin and Zwiers (2001) have examined the skill of this set of forecasts as a function of time scale, and Kharin et al. (2001) have discussed the skill as a function of the ensemble size. The GCM2 results presented here are based on 24-member ensembles, as was the case for the SGCM. The SST anomalies observed in November were persisted through DJF and added to the monthly varying SST climatology during DJF. The SST data were taken from version 2.2 of the Global Sea Ice and Sea Surface Temperature dataset (GISST2.2; Rayner et al. 1996). As the model does not predict the ocean conditions, the sea ice extent was specified from climatology. The soil conditions were initialized from the model climatology, and the initial snow line in the Northern Hemisphere was taken from the NCEP–NCAR dataset for the week before the start of the forecast. As for the initial conditions for the 24 members, they were taken from the NCEP–NCAR reanalyses at lags of 6 h prior to the start of the forecast season. It is noted that the initial conditions were specified differently from those of the SGCM and that this could lead to some differences in the ensemble-mean forecasts from the two models during the first month, but this is unlikely to have a significant impact on the forecasts for months 2 and 3. We also note that, as in the case of the SGCM, no information from the forecast period was used to produce the GCM2 predictions.

Figure 4 compares the average forecast errors for the SGCM and the GCM2 over the same 26 DJF periods that were available for both models, namely, those in the period 1969–95. These average errors do not enter explicitly in the temporal correlation between the predicted and observed fields and can easily be removed from the forecasts. They nevertheless are a manifestation of a model drift over the forecast period, a drift that can influence the response to the forcing anomalies and affect the temporal correlation between the forecasts and observations. We see from Fig. 4 that the SGCM has its maximum systematic errors over the North Pacific and the polar areas, while the GCM2 has its maximum systematic errors over the North Atlantic, northeast North America, and Greenland, as well as the Tropics. These systematic errors are of comparable magnitudes, in spite of the highly simplified dynamics of the SGCM. This is consistent with the results of Hall (2000), who showed that the SGCM has a quite reasonable Northern Hemisphere climatology when run in a perpetual boreal winter mode.

Figure 5 compares the forecast skill of the SGCM and GCM2 for the 500-hPa height for the same 26 DJF periods. We see that while the GCM2 outperforms the SGCM in many areas, particularly the Tropics, the Arctic, and Southern Hemisphere, the SGCM has skill generally in the same regions as the more complex model: the Tropics, the eastern North Pacific, western North America, the southeastern United States, part of the Arctic, and the eastern part of Asia. In fact, over the eastern part of Asia, the eastern North Pacific, and western Canada, the SGCM performs somewhat better than the GCM2. The SGCM skill is poor in the Southern Hemisphere where, as noted above, the model climatology is also poor.

If we skip the first month of the forecasts to minimize the influence of the initial conditions, we see in Fig. 6 that while the skill level in JF is of course lower than in DJF, similar results are observed: namely, there is still skill in both models. It is higher in the GCM2 in many areas, particularly the Tropics, but over the eastern part of Asia, the eastern North Pacific, and western Canada, the SGCM is somewhat more skilful than the GCM2.

5. Predictive skill of the main modes of variability

The AO and PNA patterns are known to be the main modes of interannual variability in the Northern Hemisphere mean-winter 500-hPa field (Thompson and Wallace 1998; Wallace and Gutzler 1981). The AO owes part of its variability to an interaction with the midlatitude eddies (Thompson and Wallace 2000; Thompson et al. 2000)—a part that is presumably not predictable on a seasonal time scale—but part of the variability has also been shown to be linked to the tropical forcing (Lin et al. 2002; Greatbatch et al. 2003). These authors cautioned that the contemporaneous link found between the tropical forcing and the AO time series does not imply any predictability of the AO because of the need to predict the tropical forcing in the first place. As seen in Fig. 1, however, there is a significant correlation between the November and DJF forcing over large areas of the Tropics, so there would seem to be some possible predictive skill for the AO to be derived from a knowledge of the November forcing anomalies. This question was examined with the series of DJF forecasts carried out with the SGCM and GCM2.

An empirical orthogonal functions (EOFs) analysis was conducted on the 51 mean-DJF 500-hPa heights north of 20°N from the NCEP–NCAR reanalyses to obtain the main modes of variability. The spatial structures of the first two modes appear in Fig. 7. The first mode, explaining 21.9% of the variance, is the AO, while the second mode, with 16.5% of the variance, is the PNA.

The above two structures were projected onto both the observed and predicted (ensemble mean) DJF 500-hPa height fields over the 51 winters (1948–99) to obtain the corresponding observed and SGCM-predicted principal component (PC) time series. The correlation between the two series appears in Table 1, together with the associated t value of the Student’s t test. Note that in this study all temporal correlations are calculated after removing the linear trends. A correlation of 0.44 was obtained for the two AO time series, which is statistically significant at better than the 1% level. Interestingly, the correlation is higher than the value of 0.26 obtained for EOF2—the PNA mode— a value that is significant only at the 7% level. This may seem surprising, as one might expect that the PNA, with its well-known link to the tropical heating, would be the most predictable mode of variability. In fact, much of the current interest in seasonal forecasting arose from the demonstrated link between the tropical forcing and the PNA.

The above results become clearer if the data are stratified in two classes, one for the winters with a strong-amplitude El Niño or La Niña signal and one for the winters without them. This was done by conducting an EOF analysis of the mean-November Pacific SST in the region 40°S–40°N, 120°E–90°W, the first mode of which has the characteristics of the El Niño/La Niña (not shown). The 16 winters for which the PC of that EOF had an amplitude of one standard deviation or more were classified as the “extreme phase of the Southern Oscillation” (EPSO), and the remaining 35 winters were termed “nonextreme phase of the Southern Oscillation” (NEPSO). Table 1 shows that for EPSO winters, the PNA is much better forecast than the AO, with a correlation of 0.68 between the SGCM-predicted and observed PCs, while the AO correlation does not pass the 10% significance level. During the NEPSO winters, the reverse holds. The AO is skillfully predicted, but the PNA is not. As there are about twice as many NEPSO years as EPSO ones in our classification, the results for the 51 yr tend to be biased toward the NEPSO winters and show more skill in the AO than the PNA predictions.

We have diagnosed the SGCM vertical mean temperature (total) diabatic heating (the sum of the forcing and the ensemble-mean Rayleigh damping) that the model sees during the forecast period (DJF) in order to clarify the source of the SGCM predictive skill in the AO and the PNA. The results show a tendency for a cancellation between the linear damping and the forcing in the extratropics (not shown), indicating a tendency for a local relaxation of the temperature to the persisted forcing in the midlatitudes. This local response represents essentially the contribution of persistence to the predictive skill. On the contrary, in the Tropics, there is much less of a tendency for the forcing term and the linear damping to balance. The result is a stronger variability of the diabatic heating in the Tropics than in the midlatitudes.

An EOF analysis was performed on the vertically averaged diabatic heating as seen in the model forecast in the global domain. The linear trend was removed before the EOF calculation. Shown in Figs. 8 and 9 is the spatial distribution of the first and second EOFs, explaining 13% and 9% of the total interannual variance, respectively. The first EOF has a reduced heating along the equatorial middle Pacific and an enhanced heating in the Indonesian region. The second EOF, on the other hand, shows an enhanced heating in the tropical Pacific, reminiscent of an ENSO signature. Indeed, the PC of the second EOF is significantly correlated with the forecast ensemble-mean PNA index (0.56). The PC of the first EOF, on the other hand, is significantly correlated with the forecast AO index (0.47). As a result of the fact that the two EOFs are orthogonal in both space and time, their PCs reach peaks at different times. This is consistent with our forecast result that the SGCM has skill in the AO (PNA) in the NEPSO (EPSO) years.

The above correlations between observed and SGCM-predicted AO and PNA time series were also calculated for JF. Table 1 shows that while the correlations are lower than for DJF, the main results are the same as for DJF; namely, the PNA is better predicted than the AO only for the EPSO winters, and over the 51 winters, the AO is predicted with some skill.

To compare the SGCM’s skill in forecasting the AO and PNA amplitudes with that of the GCM2, we repeated the calculations of Table 1, but for only those 26 winters that are common with the available GCM2 forecasts (1969–94). The results appear in Table 2, in which the EPSO and NEPSO winters are the same as those of Table 1 for the common period. The corresponding results for the GCM2 are given in Table 3. Comparing Tables 1 and 2 for the SGCM, we note that the results for the period 1969–94 are qualitatively similar to those for the longer period 1948–99. The PNA pattern is predicted with statistically significant skill during EPSO years, both for DJF and JF, while the AO predictions have skill in NEPSO years, again both in DJF and JF. As in Table 1, there is no significant skill in the PNA predictions in NEPSO years (DJF and JF), but the skill in EPSO years is high enough (correlation of 0.66) that when all years are pooled together, the results show a skill in the PNA predictions (correlation of 0.39 for DJF and 0.43 for JF).

In Table 3, we see that the GCM2 also has statistically significant skill in predicting the PNA amplitude during EPSO years both for DJF and JF and that the skill in JF is notably higher than that of the SGCM (correlation of 0.84 versus 0.66 for the SGCM). On the other hand, the GCM2’s skill in predicting the AO amplitude in DJF and JF is lower than that of the SGCM, whether all 26 yr are considered (0.47 versus 0.66) or only the NEPSO years (0.58 versus 0.65). Monte Carlo tests were conducted to see if these differences in the correlations are statistically significant. The tests used time series generated by an autoregressive process (AR-1) with the same lag-1 correlations and variances as our data. Correlations were computed for 1000 pairs of time series. Only the above two pairs of numbers were tested, since for JF the score for GCM2 is not significant. Out of 1000 pairs of correlations, a difference as large as 0.66 versus 0.47 occurred only once. For the second pair (0.65 versus 0.58), differences as large as this occurred 30 times. Hence both differences are highly significant.

We note that the GCM2 has statistically significant skill in predicting the amplitude of the AO for DJF but not for JF. A possible explanation for this drop in skill with time could be that the model is able to derive some skill from the initial conditions, which helps the DJF results, but that it is missing an important forcing mechanism to predict the JF amplitude. In view of our discussion on the source of skill in the forcing, this suggests a difficulty for the GCM in simulating the Indonesian heating and the east Asian winter monsoon. More intriguing is the fact that the GCM2’s skill in predicting the PNA amplitude improves from DJF to JF in EPSO years, suggestive of a spinup problem. As there are only nine EPSO winters, it is also quite possible that sampling errors are responsible for this increase in skill with time.

Finally, we note that the skill of the persistence forecasts for the AO and PNA indices was also examined. For both DJF and JF, none of the correlations between the observed and predicted (persisted) indices was found to be significant at the 5% level.

6. Discussion and conclusions

Seasonal forecasts performed over 51 winters have shown that the SGCM has statistically significant skill in forecasting mean winter 500-hPa heights over the Tropics and parts of the Northern Hemisphere extratropics. When compared with the forecasts of a more complex GCM over a subgroup of 26 winters available for both models, the skill was found to be somewhat lower for the SGCM but, remarkably, was located over nearly the same areas of the Tropics and Northern Hemisphere extratropics. Over some areas of the extratropics, the SGCM outperformed the GCM. These regions tend to be associated with parts of the globe where the assumption of persistence in the model forcing anomaly, from November through DJF, is better verified in the data than in other regions of the extratropics.

The SGCM was found to have statistically significant skill in forecasting the DJF and JF AO index when initialized with data available prior to 1 December, in fact a skill that surpassed that of the PNA index forecasts. When the winters are stratified according to the strength of the main mode of tropical Pacific SST anomaly, we found that, as expected, when this mode (ENSO) has a large amplitude, the PNA index is better forecast than the AO index. In the absence of a strong ENSO-like forcing, the SGCM shows poor skill in predicting the amplitude of the PNA pattern, but significant skill in predicting that of the AO, both in DJF and JF, indicating that the skill in predicting the AO is not related to ENSO nor only to the initial conditions. Lin et al. (2002) and Greatbatch et al. (2003) have shown that part of the winter-to-winter variability in the AO index is related to atmospheric forcing anomalies in the Tropics, but the precise nature and location of this forcing remain to be elucidated.

The linear regression of November 850-hPa velocity potential anomaly to the PC of the first diabatic heating EOF shows a strong mass convergence zone over the Indonesian region (not shown). The corresponding regression for the November geopotential anomaly at 850 hPa (not shown) shows a southward extension of the Siberian high. These planetary-scale features are reminiscent of the east Asian winter monsoon (Zhang et al. 1997) where a strong (weak) Indonesian diabatic heating would be associated with a strong (weak) winter monsoon. This, in turn, could lead to a strong (weak) AO circulation (Fig. 7a). The winter monsoon is active from November through March with a complex interannual variability of cold surges on the South China Sea where the Tibetan Plateau plays an essential role as an obstacle to the cold air near the Siberian region. Zhang et al.’s (1997) statistical study on winter monsoons shows a strong persistence (1–4 months) of the Siberian highs once they are initiated, pointing out potential predictability at seasonal time scales. Their study also discusses interesting relationships between ENSO and winter monsoon features. It would be of interest in a future study to document more extensively the role of the Indonesian region in the predictive skill observed in the SGCM’s AO and its potential link to the Asian monsoon.

In future work, it would also seem highly desirable to improve on the specification of the forcing anomalies through the forecast period. In this study we have continued the November anomaly through DJF, even though the data showed that at least one component of the forcing (the 900-hPa temperature forcing) is not persistent over this period of time over many parts of the globe, for example, over the Northern Hemisphere storm tracks. It would be of interest to conduct tests in which the forcing is specified taking into account the geographical dependence of the persistence level. The challenge here will be to specify a dynamically consistent forcing, realizing that the persistence level in the forcing will vary not only in the horizontal but also in the vertical and that each of the time tendency equations of the model has a forcing function.

Another avenue for future work will be to see how the very large ensembles that this SGCM affords can be put to use in probabilistic forecasts. Relatively small ensembles can be adequate to estimate the ensemble mean, but the higher moments of the distributions, needed in probabilistic forecasts, require larger ensembles. If this SGCM has distributions that simulate those of more complex models well, its use could prove advantageous. Hall (2000) has shown that the internal variability of the model, while somewhat low compared to that of the atmosphere, is geographically well distributed; it is also of a level that one has come to expect of global atmospheric models of this resolution. The possibility also exists to run the model at a higher resolution if that proves beneficial. Tests have shown that when run at a resolution of T31 with 10 levels, the model produces ensemble-mean forecasts that are not significantly better than those done at T21 with 5 levels as in this study, but it remains to be seen if the distribution among the ensemble members would benefit from the higher resolution.

One obvious deficiency of this SGCM is that it is dry. Precipitation forecasts would therefore only be possible if a suitable statistical relationship can be found linking the precipitation to the currently available model variables. Given the relatively low level of skill in seasonal precipitation forecasts performed with full GCMs, this approach may produce competitive predictions, but at this stage, that is only a possibility that needs to be tested. Another possibility would be to add a moisture equation to the model, as was done, for example, in the GCM of intermediate complexity described by Molteni (2002).

The statistically significant skill obtained with the SGCM suggests that it may be possible to use its ensemble-mean predictions together with those of more complex models to build a superensemble (Krishnamurti et al. 1999; Palmer et al. 2000). This approach tends to filter out individual model errors and can lead, at least for some superensembles, to improved forecasts (Palmer et al. 2004). Similarly, it would be of interest to see if the very large ensembles that the SGCM allows could be used to advantage in a probabilistic, superensemble forecast system.

Whether or not the above research avenues prove fruitful, the present study has shown that it is possible to predict the AO with some skill at the seasonal time scale during winters when the ENSO signal is weak. A link was found between the predictive skill of the AO and the diabatic heating over Indonesia, but the dynamics behind that link remain to be clarified. It would be of considerable interest to clarify how the tropical influence is communicated to the middle and high latitudes.

Acknowledgments

This research was funded by the Natural Sciences and Engineering Research Council of Canada and the Canadian Foundation for Climate and Atmospheric Sciences, through grants to the Canadian Climate Variability Research network.

REFERENCES

  • Boer, G. J., N. A. McFarlane, and M. Lazare, 1992: Greenhouse gas–induced climate change simulated with the CCC second-generation general circulation model. J. Climate, 5 , 10451077.

    • Search Google Scholar
    • Export Citation
  • Boer, G. J., G. M. Flato, M. C. Reader, and D. Ramsden, 2000a: A transient climate change simulation with greenhouse gas and aerosol forcing: Experimental design and comparison with the instrumental record for the twenthieth century. Climate Dyn., 16 , 405425.

    • Search Google Scholar
    • Export Citation
  • Boer, G. J., G. M. Flato, and D. Ramsden, 2000b: A transient climate change simulation with greenhouse gas and aerosol forcing: Projected climate for the twenty first century. Climate Dyn., 16 , 427450.

    • Search Google Scholar
    • Export Citation
  • Derome, J., and Coauthors, 2001: Seasonal predictions based on two dynamical models. Atmos.–Ocean, 39 , 485501.

  • Greatbatch, R. J., H. Lin, J. Lu, K. A. Peterson, and J. Derome, 2003: Tropical/extratropical forcing of the AO/NAO: A corrigendum. Geophys. Res. Lett., 30 .1738, doi:10.1029/2003GL017406.

    • Search Google Scholar
    • Export Citation
  • Hall, N. M. J., 2000: A simple GCM based on dry dynamics and constant forcing. J. Atmos. Sci., 57 , 15571572.

  • Hall, N. M. J., and J. Derome, 2000: Transience, nonlinearity, and eddy feedback in the remote response to El Niño. J. Atmos. Sci., 57 , 39924007.

    • Search Google Scholar
    • Export Citation
  • Hall, N. M. J., H. Lin, and J. Derome, 2001: The extratropical signal generated by a midlatitude SST anomaly. Part II: Influence on seasonal forecasts. J. Climate, 14 , 26962709.

    • Search Google Scholar
    • Export Citation
  • Hoskins, B. J., and A. J. Simmons, 1975: A multi-layer spectral model and the semi-implicit method. Quart. J. Roy. Meteor. Soc., 101 , 637655.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437471.

  • Kharin, V. V., and F. W. Zwiers, 2001: Skill as a function of time scale in ensembles of seasonal hindcasts. Climate Dyn., 17 , 127141.

    • Search Google Scholar
    • Export Citation
  • Kharin, V. V., F. W. Zwiers, and N. Gagnon, 2001: Skill in seasonal hindcasts as a function of the ensemble size. Climate Dyn., 17 , 835843.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., C. M. Kishtawal, T. LaRow, D. Bachiochi, Z. Zhang, C. E. Williford, S. Gadgil, and S. Surendran, 1999: Improved skills for weather and seasonal climate forecasts from multi-model superensemble. Science, 285 , 15481550.

    • Search Google Scholar
    • Export Citation
  • Lin, H., J. Derome, R. J. Greatbatch, K. A. Peterson, and J. Lu, 2002: Tropical links of the Arctic Oscillation. Geophys. Res. Lett., 29 , 19431947.

    • Search Google Scholar
    • Export Citation
  • McFarlane, N. A., G. J. Boer, J-P. Blanchet, and M. Lazare, 1992: The Canadian Climate Centre second-generation general circulation model and its equilibrium climate. J. Climate, 5 , 10131044.

    • Search Google Scholar
    • Export Citation
  • Molteni, F., 2002: Atmospheric simulations using a GCM with simplified physical parametrizations. I: Model climatology and variability in multi-decadal experiments. Climate Dyn., 20 , 175191.

    • Search Google Scholar
    • Export Citation
  • Palmer, T. N., C. Brankovic, and D. S. Richardson, 2000: A probability and decision-model analysis of PROVOST seasonal multi-model ensemble integrations. Quart. J. Roy. Meteor. Soc., 126 , 20132034.

    • Search Google Scholar
    • Export Citation
  • Palmer, T. N., and Coauthors, 2004: Development of a European multimodel ensemble system for seasonal-to-interannual prediction (DEMETER). Bull. Amer. Meteor. Soc., 85 , 853872.

    • Search Google Scholar
    • Export Citation
  • Rayner, N. A., E. B. Horton, D. E. Parker, C. K. Folland, and R. B. Hackett, 1996: Version 2.2 of the global sea-ice and sea surface temperature data set, 1903–1994. Hadley Centre Climate Research Tech. Note CRTN 74, Met Office, Bracknell, United Kingdom, 21 pp.

  • Thompson, D. W. J., and J. M. Wallace, 1998: The Arctic Oscillation signature in the wintertime geopotential height and temperature fields. Geophys. Res. Lett., 25 , 12971300.

    • Search Google Scholar
    • Export Citation
  • Thompson, D. W. J., and J. M. Wallace, 2000: Annular modes in the extratropical circulation. Part I: Month-to-month variability. J. Climate, 13 , 10001016.

    • Search Google Scholar
    • Export Citation
  • Thompson, D. W. J., J. M. Wallace, and G. C. Hegerl, 2000: Annular modes in the extratropical circulation. Part II: Trends. J. Climate, 13 , 10181036.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., and D. S. Gutzler, 1981: Teleconnection in the geopotential height field during the Northern Hemisphere winter. Mon. Wea. Rev., 109 , 784812.

    • Search Google Scholar
    • Export Citation
  • Zhang, Y., K. R. Sperber, and J. S. Boyle, 1997: Climatology and interannual variation of the East Asian winter monsoon: Results from the 1979–95 NCEP/NCAR reanalysis. Mon. Wea. Rev., 125 , 26052619.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Correlation coefficients between temperature forcing at 900 hPa in Nov and that of the following DJF over the 51 yr. The contour interval is 0.2. The shaded areas represent significance level of 0.05 or better according to a Student’s t test.

Citation: Journal of Climate 18, 4; 10.1175/JCLI-3289.1

Fig. 2.
Fig. 2.

Same as in Fig. 1, but for correlation coefficients between the observed and the SGCM-predicted 500-hPa height anomaly in (a) DJF and (b) JF over the 51 winters.

Citation: Journal of Climate 18, 4; 10.1175/JCLI-3289.1

Fig. 3.
Fig. 3.

Same as in Fig. 1, but for correlation coefficients between the observed 500-hPa height anomaly in Nov and that in (a) DJF and (b) JF over the 51 yr.

Citation: Journal of Climate 18, 4; 10.1175/JCLI-3289.1

Fig. 4.
Fig. 4.

Systematic errors of DJF-mean 500-hPa height for (a) SGCM and (b) GCM2 over 26 winters from 1969 to 1995. The contour interval is 20 m, and the shading indicates a magnitude in excess of 40 m.

Citation: Journal of Climate 18, 4; 10.1175/JCLI-3289.1

Fig. 5.
Fig. 5.

Same as in Fig. 1, but for correlation coefficients between the observed and the predicted DJF-mean 500-hPa height anomaly by (a) SGCM and (b) GCM2 over 26 winters from 1969 to 1995.

Citation: Journal of Climate 18, 4; 10.1175/JCLI-3289.1

Fig. 6.
Fig. 6.

Same as in Fig. 5, but for JF.

Citation: Journal of Climate 18, 4; 10.1175/JCLI-3289.1

Fig. 7.
Fig. 7.

Linear regressions of 500-hPa height onto the PC of (a) EOF1 and (b) EOF2. Amplitudes are geopotential height in meters corresponding to one standard deviation of the PC. The contour interval is 10 m.

Citation: Journal of Climate 18, 4; 10.1175/JCLI-3289.1

Fig. 8.
Fig. 8.

The first EOF spatial distribution of the vertical mean diabatic heating. Arbitrary units.

Citation: Journal of Climate 18, 4; 10.1175/JCLI-3289.1

Fig. 9.
Fig. 9.

Same as in Fig. 8, but for the second EOF.

Citation: Journal of Climate 18, 4; 10.1175/JCLI-3289.1

Table 1.

Temporal correlation and associated t values between the SGCM-predicted and observed PCs of EOF1 and EOF2, for all 51 winters, for the 16 winters with strong ENSO signals, and the remaining 35 winters. The correlations that are significant at the 5% level are italicized, and those that are not significant at the 10% level have been omitted.

Table 1.
Table 2.

Same as in Table 1, but for the subset of 26 winters from 1969 to 1995.

Table 2.
Table 3.

Same as in Table 2, but for the GCM2.

Table 3.
Save
  • Boer, G. J., N. A. McFarlane, and M. Lazare, 1992: Greenhouse gas–induced climate change simulated with the CCC second-generation general circulation model. J. Climate, 5 , 10451077.

    • Search Google Scholar
    • Export Citation
  • Boer, G. J., G. M. Flato, M. C. Reader, and D. Ramsden, 2000a: A transient climate change simulation with greenhouse gas and aerosol forcing: Experimental design and comparison with the instrumental record for the twenthieth century. Climate Dyn., 16 , 405425.

    • Search Google Scholar
    • Export Citation
  • Boer, G. J., G. M. Flato, and D. Ramsden, 2000b: A transient climate change simulation with greenhouse gas and aerosol forcing: Projected climate for the twenty first century. Climate Dyn., 16 , 427450.

    • Search Google Scholar
    • Export Citation
  • Derome, J., and Coauthors, 2001: Seasonal predictions based on two dynamical models. Atmos.–Ocean, 39 , 485501.

  • Greatbatch, R. J., H. Lin, J. Lu, K. A. Peterson, and J. Derome, 2003: Tropical/extratropical forcing of the AO/NAO: A corrigendum. Geophys. Res. Lett., 30 .1738, doi:10.1029/2003GL017406.

    • Search Google Scholar
    • Export Citation
  • Hall, N. M. J., 2000: A simple GCM based on dry dynamics and constant forcing. J. Atmos. Sci., 57 , 15571572.

  • Hall, N. M. J., and J. Derome, 2000: Transience, nonlinearity, and eddy feedback in the remote response to El Niño. J. Atmos. Sci., 57 , 39924007.

    • Search Google Scholar
    • Export Citation
  • Hall, N. M. J., H. Lin, and J. Derome, 2001: The extratropical signal generated by a midlatitude SST anomaly. Part II: Influence on seasonal forecasts. J. Climate, 14 , 26962709.

    • Search Google Scholar
    • Export Citation
  • Hoskins, B. J., and A. J. Simmons, 1975: A multi-layer spectral model and the semi-implicit method. Quart. J. Roy. Meteor. Soc., 101 , 637655.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437471.

  • Kharin, V. V., and F. W. Zwiers, 2001: Skill as a function of time scale in ensembles of seasonal hindcasts. Climate Dyn., 17 , 127141.

    • Search Google Scholar
    • Export Citation
  • Kharin, V. V., F. W. Zwiers, and N. Gagnon, 2001: Skill in seasonal hindcasts as a function of the ensemble size. Climate Dyn., 17 , 835843.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., C. M. Kishtawal, T. LaRow, D. Bachiochi, Z. Zhang, C. E. Williford, S. Gadgil, and S. Surendran, 1999: Improved skills for weather and seasonal climate forecasts from multi-model superensemble. Science, 285 , 15481550.

    • Search Google Scholar
    • Export Citation
  • Lin, H., J. Derome, R. J. Greatbatch, K. A. Peterson, and J. Lu, 2002: Tropical links of the Arctic Oscillation. Geophys. Res. Lett., 29 , 19431947.

    • Search Google Scholar
    • Export Citation
  • McFarlane, N. A., G. J. Boer, J-P. Blanchet, and M. Lazare, 1992: The Canadian Climate Centre second-generation general circulation model and its equilibrium climate. J. Climate, 5 , 10131044.

    • Search Google Scholar
    • Export Citation
  • Molteni, F., 2002: Atmospheric simulations using a GCM with simplified physical parametrizations. I: Model climatology and variability in multi-decadal experiments. Climate Dyn., 20 , 175191.

    • Search Google Scholar
    • Export Citation
  • Palmer, T. N., C. Brankovic, and D. S. Richardson, 2000: A probability and decision-model analysis of PROVOST seasonal multi-model ensemble integrations. Quart. J. Roy. Meteor. Soc., 126 , 20132034.

    • Search Google Scholar
    • Export Citation
  • Palmer, T. N., and Coauthors, 2004: Development of a European multimodel ensemble system for seasonal-to-interannual prediction (DEMETER). Bull. Amer. Meteor. Soc., 85 , 853872.

    • Search Google Scholar
    • Export Citation
  • Rayner, N. A., E. B. Horton, D. E. Parker, C. K. Folland, and R. B. Hackett, 1996: Version 2.2 of the global sea-ice and sea surface temperature data set, 1903–1994. Hadley Centre Climate Research Tech. Note CRTN 74, Met Office, Bracknell, United Kingdom, 21 pp.

  • Thompson, D. W. J., and J. M. Wallace, 1998: The Arctic Oscillation signature in the wintertime geopotential height and temperature fields. Geophys. Res. Lett., 25 , 12971300.

    • Search Google Scholar
    • Export Citation
  • Thompson, D. W. J., and J. M. Wallace, 2000: Annular modes in the extratropical circulation. Part I: Month-to-month variability. J. Climate, 13 , 10001016.

    • Search Google Scholar
    • Export Citation
  • Thompson, D. W. J., J. M. Wallace, and G. C. Hegerl, 2000: Annular modes in the extratropical circulation. Part II: Trends. J. Climate, 13 , 10181036.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., and D. S. Gutzler, 1981: Teleconnection in the geopotential height field during the Northern Hemisphere winter. Mon. Wea. Rev., 109 , 784812.

    • Search Google Scholar
    • Export Citation
  • Zhang, Y., K. R. Sperber, and J. S. Boyle, 1997: Climatology and interannual variation of the East Asian winter monsoon: Results from the 1979–95 NCEP/NCAR reanalysis. Mon. Wea. Rev., 125 , 26052619.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    Correlation coefficients between temperature forcing at 900 hPa in Nov and that of the following DJF over the 51 yr. The contour interval is 0.2. The shaded areas represent significance level of 0.05 or better according to a Student’s t test.

  • Fig. 2.

    Same as in Fig. 1, but for correlation coefficients between the observed and the SGCM-predicted 500-hPa height anomaly in (a) DJF and (b) JF over the 51 winters.

  • Fig. 3.

    Same as in Fig. 1, but for correlation coefficients between the observed 500-hPa height anomaly in Nov and that in (a) DJF and (b) JF over the 51 yr.

  • Fig. 4.

    Systematic errors of DJF-mean 500-hPa height for (a) SGCM and (b) GCM2 over 26 winters from 1969 to 1995. The contour interval is 20 m, and the shading indicates a magnitude in excess of 40 m.

  • Fig. 5.

    Same as in Fig. 1, but for correlation coefficients between the observed and the predicted DJF-mean 500-hPa height anomaly by (a) SGCM and (b) GCM2 over 26 winters from 1969 to 1995.

  • Fig. 6.

    Same as in Fig. 5, but for JF.

  • Fig. 7.

    Linear regressions of 500-hPa height onto the PC of (a) EOF1 and (b) EOF2. Amplitudes are geopotential height in meters corresponding to one standard deviation of the PC. The contour interval is 10 m.

  • Fig. 8.

    The first EOF spatial distribution of the vertical mean diabatic heating. Arbitrary units.

  • Fig. 9.

    Same as in Fig. 8, but for the second EOF.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 288 86 5
PDF Downloads 156 64 3