‹header›
‹date/time›
マスタ テキストの書式設定
2 レベル
3 レベル
4 レベル
5 レベル
‹footer›
‹#›
OK, my talk is about seasonal predictions. As you may know, many centers in the world are now doing ENSO forecast regularly. But for the IO, there are still very few activities up to now. Here, I am going to introduce the IO predictions mainly based on JAMSTEC-Frontier coupled model.
Based on this model, we have designed a kind of semi-multi-model ensemble prediction scheme. We perturb the model coupling physics in three different ways. And for each model we use three initial conditions generated by a simple nudging scheme assimilating SST information only. This scheme works well for ENSO prediction because ENSO is mainly determined by air-sea coupling.  But for the Indian Ocean prediction,  assimilating subsurface important can be very important because of strong stochastic and influential intraseasonal disturbances there.
This shows correlation skill of global SST anomaly at 3, 6, 9 and 12 month lead. You can see very low predictability in the equatorial Atlantic; a common challenge to current coupled models. For ENSO prediction, the model has high skill above 0.6 up to 1 year lead.
In fact, ENSO can now be predicted up to 2 years ahead. This shows ACC started from each calendar month, May 1st, June 1st,.. until May 1st next year. You can see the two spring prediction barriers. The model predictions can successfully across the first spring time. But after the second spring time, the skill become less skillful. If averaging the skill over all calendar month, it shows high skill above 0.6 up to 16 months lead. At 2 years lead, it is about 0.5.  RMSE.. This is index prediction. The model also shows success in predicting ENSO-related global climate anomalies. Here, I just show you one example…. The 2-year lead predictability of ENSO is very close to its theoretical limit.
In the IO, SST anomaly can be predicted up to about 2 seasons ahead. The highest skill are found in the central IO, southwestern IO, and WA. In general, source of predictability in the IO can be divided into  two parts. One is from external force like ENSO, the other is due to internal climate variability such as IOD.
This shows SST and rainfall anomalies in the IO associated with ENSO in boreal winter. During El Nino, warm SST appears in the tropical IO basin, SW IO, and cold anomaly in WA. This pattern was predicted very well. For rainfall prediction, the model has systematic bias, particularly in Indonesia region. The drought in Australia is too strong in the model. The  dipole pattern in the IO (less rainfall in the east and floods in the west) can be predicted basically.
IO has its own climate variability, that is the IOD. IOD usually start to develop in boreal summer when the Asian summer monsoon starts, it peaks in fall and decay rapidly in early winter when the monsoon winds reverse. IOD is strongly modulated by Asian monsoon. Generally, IOD signal is weaker than ENSO and difficult to predict. But in the IO the SST is very warm, above 28C, and the convection is strong. Even small SSTA may induce large rainfall variations not only in local areas but also can affect the climate in far region through atmospheric bridge. So IOD prediction can be very important and beneficial to the society.
This shows probably the best skill for IOD prediction which current coupled models may achieve. Starting from each month January 1st, February 1st, March 1st, the skill decrease gradually and reach minimum values in December. This is related to the strong phase locking with monsoon. The skill then rebound and decrease quickly in next spring. This part is associated with ENSO influences. So you can see both winter and spring barrier exist in eastern pole prediction .In general, the signal can be predicted up to about 2 seasons ahead. 
In 1994 boreal fall (Sep-Nov), a strong positive IOD happened. Very cold SST is observed in the EIO and weak warming in the west. This was predicted from 1 June 1994. This IOD caused severe drought in EIO, Indonesia, East Asia, and Australia, floods in the WIO, South India, and East Africa. The climate impacts were also predicted basically despite some biases in the model. 
This is another example in 1997. You can see the similar climate anomalies caused by this IOD event. Again the model shows good skill in predicting this event. 
Another strong pIOD happened in 1997 boreal fall. Again, you can see the drought in EIO, Indonesia, East Asia, and parts of Australia, floods in the WIO, South India, and East Africa.  Both the IOD signal and its-related rainfall anomalies can be predicted but the intensity of the IOD was underestimated. 
In 2006 and 2007 fall, two pIOD happened during two consecutive years. This is unprecedented. The 2006 IOD is stronger, it caused severe drought in EIO and Indonesia, and floods in South Indian, WIO, and East Africa. Warm and dry anomalies appeared in East Asia and Australia. The 2007 IOD is weak and peculiar in that it co-occurred with a La Nina in the Pacific. Even though the IOD signal is weak, you can see the similar atmospheric response: dry in the EIO and Indonesia, and floods in the western IO and East Africa. Australia again suffered from warm and dry climate. It actually caused a big loss in the SE Australia. 
Since 2005, we have performed experimental real time seasonal forecast based on 27 members.  This shows real time model forecasts for these two IOD events. The dipole SST pattern, the drought in the eastern IO and floods in the western IO, East Africa were predicted. The dry and warm anomalies in Australia are also predicted well. This suggests the potential societal benefits of IOD prediction. The 2006 IOD is co-occurred with a weak El Nino, and the 2007 pIOD is happened with a La Nina . So the IOD can happen independently from the ENSO. This suggests that IOD and its predictability are originate from the internal processes in the IO. To clarify this issue
The co-occurrence of both pIOD and La Nina is very rare; a similar case was in 1967. The 1967 pIOD also caused severe drought in the Southeast Australia.   
we did two additional hindcast experiments by suppressing the air-sea coupling the tropical Pacific and IO, separately. The initial conditions do not change. 
This shows the predictions of the three strong pIOD events with and without ENSO influence. For the 2006 and 1994 events, the results are very similar at all lead times. For the 1997 event, El Nino signal does help to reduce the false alarms in 1997 fall and is important for the prediction of the warm signal in 1998. But the long-lead time predictions of this IOD event are very similar even without the El Nino.
To understand the possible reason, we checked model initial subsurface conditions in the winters before those pIOD events. One common feature is that strong cold subsurface signals appeared in the tropical southwestern IO. The structure is similar to Rossby wave characteristics. The cold signal will propagate westward and reflect as equatorial eastward-propagating Kelvin wave, providing favorable conditions for the pIOD development. This is consistent with IOD dynamics. So the strong cold subsurface signal in the southwestern IO may provide important preconditions for the long-lead forecasts of pIOD.  
This shows El Nino predictions with and without IO air-sea coupling. One common feature is that the predicted El Nino would persist for a longer period if without IO signals. The transition from El Nino to La Nina would be delayed. In 1994, strong warming appeared in the central Pacific, Modoki-type El Nino.  The 1994 IOD actually tends to enhance the El Nino signal. For the 1997/98 El Nino onset prediction (the green lines), you may also see slight improvement  if with IOD signal.
This suggest that Indian Ocean observations will be very important for the IOD prediction. In the future, we plan to assimilate both ocean and atmosphere information. The key point is that the assimilated data should be well-balanced and consistent with model physics. We will also pursue MME and societal application of the prediction. JAMSTEC is going to open an outreach lab for this. Here, I would like to talk a little more about the MME.
Observations: La Nina and IOD. Predicted by … and a few other centers. But most of them only focus on ENSO forecasts. Prediction based on a single model sometimes may be not reliable. It will be very worthwhile to perform multi-model forecasts through international collaborations. Here I show you one example
13 CGCMs including our model. This suggests that MME could be a very useful approach to improve IOD predictions at current stage.
This shows SST and surface air tem anom. during 1999/2000 winter. A strong La Nina occurred in the EP with warm signals in North and South Pacific. USA experienced long-lasting warm winter. This can be predicted at 18 and 24 months ahead. Besides, the cold signals in SA, AU, Asia, large parts of Africa, and the warm signal in Northern Eurasia are also predicted. This year, another strong La Nina appeared. The model predicted that it would be long-lived. And you may expect the similar pattern this winter.
 
For the Indian Ocean Dipole predictions, however, we are still facing large difficulties.  One problem is that the signal is not as strong and regular as ENSO. And IOD can be strongly influenced by many other physical processes like MJO, monsoon, and ENSO.  Besides, current CGCMs have large deficiencies in…. and observations in the Indian Ocean are too sparse to provide good initial conditions.