Seasonal Prediction

This is a past E-mail discussion. Please use the BBS Discussion to join discussion.

---------- Forwarded message ----------
From: <>
Date: Mon, Jun 22, 2009 at 2:07 PM
Subject: RE: Seasonal forecasts from 1 June 2009 (monthly mean maps)

Thanks Jing-Jia. That helps me explain. I hope you get this sorted out.

Best regards, Peter.


From: Jing-Jia Luo []
Sent: Monday, 22 June 2009 2:35 PM
To: McIntosh, Peter (CMAR, Hobart)
Cc: Toshio Yamagata
Subject: Re: Seasonal forecasts from 1 June 2009 (monthly mean maps)

Dear Peter,

Nothing except the computer has changed since 1 April 2009; the forecast model is the same as before. We repeated the forecasts initiated from 1 March 2009 (with the same model and initial conditions), 9-ensemble mean did show certain differences as I mentioned before. I am still not quite sure what the actual reasons for this difference are. One possible factor can be due to the different FORTRAN compiler. This means the executable codes of the coupled model are different now though the source code itself has no any change.

I asked NEC system engineering. The answer is that it is basically no way to get the same results on the new Earth Simulator (like chaos). Theoretically, if we have infinite ensembles, the results may be equal if the new compiler does not change the code systematically. But who knows (sometimes, bug fix in the compiler can induce big changes in the model results).

We are planing to redo the hindcast step by step (we are facing another technical problem. Our model speed become slower despite the much faster new machine).

Bets regards, Jing-Jia


On Mon, Jun 22, 2009 at 1:08 PM, <> wrote:

Dear Jing-Jia

I regularly talk to wheat farmers in NW Victoria, Australia, at a place called Birchip. The Birchip Cropping Group are the most active farmer group in Australia, and they hold their annual Grains Expo in early July each year. This year, Australia's Governor-General will be attending. Over the years I have given talks about the various climate models, including SINTEX, and they have come to trust SINTEX forecasts. As you know, SINTEX has been successful at predicting the three positive IOD events recently, and the IOD seems to be the most important effect on rainfall at Birchip.

I will certainly get questions regarding the change of forecast in SINTEX this year, and I would like to be able to answer as clearly as possible. Can you explain to me why the SINTEX forecasts changed so much? I don't understand why changing computers would make such a big difference. Normally one would expect very minor changes going from one computer to another. Were software changes required in order to change computers? Did data sets change? Any information you can give me will be helpful.

Regards, Peter.

Dr Peter McIntosh
Centre for Australian Weather and Climate Research (CAWCR)
CSIRO Marine Laboratories


From: Jing-Jia Luo <>
Date: Wed, Jun 17, 2009 at 10:06 PM
Subject: Re: no skill for predicting the IOD before June [SEC=UNCLASSIFIED]
To: Harry Hendon
Cc: David Jones, Toshio Yamagata, Grant Beard, Oscar Alves

Dear Harry,

So we are reaching some agreements. The averaged hindcast skill just gives you a rough reference. If you do carefully for the real time forecasts, I believe you should go beyond this (even without the increase of ensemble members); you have much more information/analysis than the mean hindcast skill tells.

Concerning the smoothing issue: When we look at the monthly prediction plumes of 12 target months, we will focus on the signal beyond the intraseasonal disturbance. And we will look at the consecutive forecasts performed during several months. In this sense, we are also doing the smoothing. Or like IRI, we can directly do 3-month average to remove the noise in the prediction plumes.

Because of the uncertainty caused by the new Earth Simulator, I do not know how much we can still trust the SINTEX-F model forecast, particularly for the current IOD prediction. I hope POAMA model forecast would be correct. Let's see the forecasts in following months and see what will happen in the real ocean.

Best regards,


On Wed, Jun 17, 2009 at 4:58 PM, Harry Hendon <> wrote:

Hi Jing-Jia, I think the point I am trying to make is that you can't have it both ways: yes you sometimes get a few forecasts right, but based on a long record, the skill is low. This means there are lots of times that you are wrong (as are forecasts from our model). The problem is that you don't know in advance whether you are right or wrong. This is what you are not communicating. I agree that it is an interesting scientific question to sort out why some events are more predictable than others, and this may lead to improved forecast systems, but this doesn't change the fact that the current level of skill is low. This isn't an issue of whether we or you issued a right or wrong forecast. It is an issue of communicating the level of skill of your forecast system.

And yes, the skill in the east and the west is better separately than the skill of the IOD index. I included these plots. However, the story is unchanged: skill in the east is low before June for the unsmoothed forecasts as are provided in real time. Your system (and our system) can't make reliable forecasts of the IOD before June.

You indicate that you might know in advance that you have a skilful forecast because of low ensemble spread. However, I doubt that there is a spread/skill relationship in your forecasts. If so, this would be an interesting result.

Finally, for what it is worth, POAMA has been and is going for El Nino and positive IOD.

Harry Hendon
Centre for Australian Weather and Climate Research
A partnership of the Bureau of Meteorology
and the Commonwealth Scientific Industrial Research Organization


From: Jing-Jia Luo <>
Date: Wed, 17 Jun 2009 16:20:35 +0900 (JST)
Subject: Re: no skill for predicting the IOD before June [SEC=UNCLASSIFIED]

Dear Harry, and David,

Thank you for your comments. Personally, I know our forecasts have influence on Australian farmers but have not realized that they trust the forecasts so much. Last January in Tokyo, we had a symposium on the societal application of climate prediction ( One Australian farmer attended the meeting and gave a wonderful presentation. I asked him whether they prefer deterministic or probabilistic forecasts. He said they need deterministic forecasts, and they will take their own risk/uncertainty (even if we tell them the forecasts are 100% sure, they may believe only 50% or even lower if the forecasts have no much credit). I believe that farmers have always checked all available/updated informations and made their own decision.

I have no idea about the media, but I think they have their own responsibility and freedom for what they choose to report. May be they will be more careful when reporting our forecasts in the future.

Response to Harry's scientific questions:

Yes, IOD is much more difficult to predict compared to ENSO. The IOD index is a measure of the zonal sea surface temperature (SST) gradient in the tropical Indian Ocean. By doing this, ENSO influence is largely removed. I should clarify that we are using the eastern pole SST as an indicator for IOD. And the prediction skill of SST in the eastern Indian Ocean is much better than that of IOD index (Luo et al., J. Climate 2007). (Comment here: As Harry mentioned, we used 5-month running mean to remove intraseasonal disturbances which are very strong in the Indian Ocean. Scientifically, we believe the smoothing is reasonable since intraseasonal disturbances and interannual variations are two different phenomena with different dynamics and predictability. We should separate them properly rather than mix up everything. We are doing seasonal forecasts not weather forecasts.)

In general, prediction skill assessed from retrospective forecasts can tell us how much we may trust the real time forecasts. However, I should emphasize that it is also very important to consider the difference case by case carefully. It is not that simple: you just look at the hindcast skill and can conclude "no way to predict IOD before June". Otherwise, our model should fail to predict the last three IOD during 2006-2008 and another strong one in 1994 before June.

Concerning our forecasts of a negative IOD last months, we had some reasonable reasons for this. 1) The probability to have a fourth IOD consecutively this year could be very low. 2) Indian Ocean observations showed some evidence for the possible occurrence of a negative IOD. 3) According to my personal experience, IOD has low predictability when the model ensembles show a large spread (usually due to the large differences in our model initial conditions and strong intraseasonal disturbance influence). Our model forecasts last months actually showed small spreads; a majority of 27 ensembles predicted a negative IOD.

Concerning the sudden change of IOD forecasts this month, we think it can be due to El Nino influence. The equatorial thermocline in the Indian Ocean is deep in the east and shallow in the west. During negative IOD, internal air-sea coupling could be weak. So generally, ENSO may have larger influence on negative IOD evolution compared to positive IOD case. Scientifically, it is not surprising that the tendency of negative IOD can be weakened or even killed by El Nino.

Final words: Personally, I believe that I should learn more on how to release forecast information to the society including necessary communications with the users. It is a big lesson to me this time. I think BoM had also got many criticisms when you failed to give a correct forecast. In this sense, we may find many collaborations.

Regarding the IOD forecasts this year, let's see what will happen.

Best regards,


From: David Jones <>
Subject: RE: no skill for predicting the IOD before June [SEC=UNCLASSIFIED]
Date: Wed, 17 Jun 2009 13:24:10 +1000

It is most unhelpful having these forecasts injected into the Australian media like this. We have an extremely capable research and modelling capacity in Australia with a government agency (The Bureau of Meteorology) responsible for climate prediction for Australia.

In this instance farmers have been mislead about expected climate conditions and the long-term public confidence in climate predictions wil be eroded. Those farmers who make a loss have good reason to feel poorly served.



From: Harry Hendon
Sent: Wednesday, 17 June 2009 12:36 PM
Cc: Grant Beard; Oscar Alves; David Jones
Subject: no skill for predicting the IOD before June [SEC=UNCLASSIFIED]

Hi Luo and Yamgata-san, you have probably seen the next article in the Weekly Times where the same journalist now tries to warn everybody about El Nino and positive IOD due to the SINTEX-F forecast, whereas the previous month he advised the farmers that the drought was over based on forecasts from the same SINTEX-F system:

The problem that you are causing here by dissmeniating your forecasts in the fashion that you do is that you are not fairly representing the forecast skill of SINTEX-F. You use anectodal evidence that because you made a “skillful” forecast for an event that other systems might have missed, that you then have a skilful system that farmers ought to pay attention to. You also say that, after the fact, this event or that event was indeed predictable months or seasons in advance. But you only know this after the fact. If you score all of your forecasts for the past ~25 yrs, you wouldn’t be advising farmers to act upon your forecasts issued in April or May: you have no skill (nor does anybody else) of making IOD forecasts from before June.

I attach a ppt which shows some skill assessment of SINTEX-F (3 members) ECSys3 (three members) and POAMA2 (3 members). Granted this is a small ensemble, but it demonstrates the point. All analyses for the three systems were done in an identical fashion using monthly data. For SINTEX-F, the hindcast climatology for each version of the model was removed, though this doesn’t have much effect.

For the Nino3 index (slide 1), all three systems are comparable, though SINTEX lags behind early probably due to a lack of data assimilation. However, POAMA and EC especially simulate a damped enso mode. Further west (Nino4; slide 2) the three systems again are similar, though SINTEX is better at longer lead. Interestingly, EC is still damped, while POAMA is overactive probably due to cold tongue bias which causes enso mode to shift west.

We note that for one season lead (ie 3 mnth lead as indicated on this plot, which means for instance forecasts initialized on 1 June, that verify for the mnth of Sep) the skill for the Nino indices is ~0.8 (or greater) for all 3 systems.

The Indian Ocean is a different story (slide 3). Here ECSys3 is clearly superior, and SINTEX-F is marginally worse than POAMA, at least for lead times 0-4 mnths. However, the critical thing is the level of skill: at lead 3 mnths EC is about 0.5, POAMA is about 0.4 and SINTEX is about 0.3: none of these 3 systems has any skill in predicting the IOD at lead times 1 season.

We haven’t done any more thorough analysis with SINTEX (but we are working on it), but we have with POAMA. I presume the results from POAMA will be similar to SINTEX because the skill assessment in slides 1-3 are all very similar.

What the further analysis of POAMA shows (slide 4) is that a correlation of 0.4 (from slide 3) at lead time one seasons (4 mnth lead on these plots in slide 4) equates to a hit rate of getting the right sign of the IOD of only about 55% (ie no better than guessing). On the other hand, a correlation of ~0.8 for Nino3 or 4 equates to a hit rate close to 80%. So, the correlations in Slide 3 for the IOD mean that all three systems are no better than guessing the sign of the IOD at 1 season lead (granted this is using all start mnths, and we know that the IOD is meaningless outside of ~June-Nov and that forecast skill is a strong fucntion of season; this is addressed below)

Further analysis of the skill of POAMA as a function of start mnth (Slide 5; I think Luo may have done something similar for SINTEX-F in one of his papers but he used some sort of multi mnth smoothing across different lead times), shows clearly that you can’t predict the dipole, east IO and west IO SST prior to June.

Slide 6 shows the overall problem of predicting SST in the Indian Ocean relative to the Pacific.

In summary: with current forecast systems, nobody can reliably predict a dipole from before June in a real time setting based on forecast model performance for the past ~25 yrs. You are doing more of a disservice rather than service by not fairly conveying the level of skill of your forecast system. Journalists are na .A Nove. They like to report sexy results, but they will only repeat what they are told.

Harry Hendon
Centre for Australian Weather and Climate Research
A partnership of the Bureau of Meteorology
and the Commonwealth Scientific Industrial Research Organization


Subject: Re: The Australian Rainfall
From: Toshio Yamagata <>
Date: Wed, 17 Jun 2009 12:49:06 +0900
To: Maree Miller <>

Dear Miller-san:
Climate change and climate variations are different. We are trying to predict climate variations such as El Nino/La Nina, IOD etc., which are mother of abnormal weather events. A long-term tendency of a drought may be due to climate change but large interannual variations of either drought or wet conditions are due to climate variations. Is this explanation sufficient?
Best regards, Yamagata


Maree Miller さんは書きました:

> Dear Prof Yamagata,
> Im am a farmer in south east Australia, and find your web site very accurate, it helps with planning.
> What I am wondering is, in your opinion, if our lack of rainfall in south east Australia, is being caused from climate change or a drought?
> I would appreciate your help, thanks,
> Kind Regards
> Maree Miller
> <>
> "Mungana"
> Temora NSW 2666
> Australia


On Tue, 16 Jun 2009 09:41:56 +0930 (Cen. Australia Standard Time), Lee-anne" <> wrote:

Dear Professor Yamagata

Thank you for forwarding the information on the forecast of both the IOD and El Nino probability. I had to laugh when Jing-Jia Luo stated that "I hope that the updated forecasts can be released to Australian farmer society in order to avoid any potential losses." A pity that this was not made known to us six months earlier as it is too late now with most of the southern crop planted.

We do have one thing in our favour. Lake Eyre, Australia's inland sea, has a considerable amount of water in at the moment and the belief of the local people is that this also has a positive impact on our weather pattern. The older generation also predicted the 10 year drought and we are now in the tenth year of below average rainfall.

Like you said, we shall just have to wait and see what happens.

Most kind regards,

ML & LJ Whitehead
Brinkworth South Australia


On Tue, Jun 16, 2009 at 2:11 AM, Toshio Yamagata <> wrote:

Dear Mina-sama:

I had a TV interview a few days ago.
I explained the TV crew about possibility of cool summer after less rain rainy season in Japan, just like 1993 (nIOD + El Nino). In 1993, there was no summer in the northern part of Japan (Tohoku region). Total agricultural loss amounted to more than 10 Billion dollars! By the way, in 1994, we had pIOD and El Nino Modoki. This 1994 event led us to introduction of IOD.
There is already a symptom in Hokkaido, where farmers have started worring about outcomes of less sunshine. Also, In kennya, farmers have started complaining of less rain. The sea level hight anomaly in the Indian Ocean suggests evolution of negative IOD, together with El Nino in the Pacific.
SINTEX-F has predicted a strong El Nino and I think this has killed the evolution of negative IOD perhaps by the model bias.
Anyway, let's see. We can learn a lot from what will happen.
Best wishes from Paris, Toshio Yamagata


David Jones <> on Mon, Jun 15, 2009 at 4:51 PM


this is a serious issue for us.




From: Jing-Jia Luo <>
Date: Mon, Jun 15, 2009 at 7:57 PM
Subject: Re: Seasonal forecasts from 1 June 2009 (monthly mean maps) [SEC=UNCLASSIFIED]
To: "Saji N. Hameed" <>
Cc: Toshio Yamagata <>,

Dear Saji,

My definition of strong El Nino is that DJF Nino3.4 SSTA reaches +1.5C. I never mean that El Nino this year can be as strong as that in 1982/83 and 1997/98 (which I prefer to call extremely strong El Nino). Just looking at the observations, I had the same feeling that El Nino this year would be weak. Because of this, I told Vinson early last week that "I do not think a strong El Nino would occur this year".

I have being used clear words (may be too strong sometimes) for our forecasts. Probably, I should turn to probabilistic forecasts now concerning the machine problem, saying that "well, we may expect 30% probability of strong El Nino".

I am sure we will redo the hindcasts after careful considerations. Thank you for your kind wishes.

Best regards,


On Mon, Jun 15, 2009 at 6:32 PM, Saji N. Hameed <> wrote:

Hi Jing-Jia,

It is no more than a hunch. You said it right "No one knows the answer", despite our claim that El Nino is predictable much ahead of time. The current conditions are no way comparable to that in 1997 June or 1982 (strong El Ninos).

This is more reason to be careful, especially when forecasts have changed suddenly. So far, SINTEX has a good reputation. It is good to keep it going.

Well, it should be done somehow - maybe in a phased manner. Do 1/3rd of the runs (every third month) instead of all months and hope to complete the full hindcasts later.

Keep the good work going!

Best regards,


On Mon, Jun 15, 2009 at 4:47 PM, Jing-Jia Luo wrote:

Dear Saji,

Are you sure El Nino would not happen this year? No one knows the answer right now. I have just realized that many Australian farmers rely on our forecasts for their agricultural arrangement. I feel a kind of responsibility to raise the uncertainty issue (caused by machine) to them. Frankly, I do not know how much I can trust our model forecasts now because of the machine problem.

To repeat the hindcast is not that easy; I suppose you should know how much CPU time it will cost. According to the model speed on the new machine, the total CPU times will be:

9 member X 12 months X 27year x 10 node hours. This is more than double of our total CPU resource this year.

Best regards,


On Mon, Jun 15, 2009 at 4:12 PM, Saji N. Hameed <> wrote:

Dear Luo-san,

I think it is better to exercise some caution in making such announcements. Your new mail is now more confusing, with sudden shift to talk about a moderate El Nino. It was surprising for me to see SINTEX shift suddenly from weak El Nino to strong El Nino.

My personal intuition is that this El Nino is not going to happen. BTW, from IRI site, it is interesting that all statistical forecasts predict normal conditions, but majority of dynamical models forecasts moderate to strong El Nino (in forecasts starting May 2009, SINTEX and Cane-Zebiak model were the only two to predict near neutral conditions for the rest of this year).

p.s: your point about repeated hindcast is very relevant. I wonder why it is not done yet --- should be the first to do, before giving out a prediction.

Best regards,


On Mon, Jun 15, 2009 at 4:01 PM, Jing-Jia Luo <> wrote:

Dear Harry and colleagues,

I am very sorry that the sudden change of our forecasts may cause some confusion, particularly to the Australian farmers. The renewing of our Earth Simulator caused certain uncertainty in our model forecasts. Without repeating our hindcast, we cannot know the exact influence of the new machine. Here, I tried to give you a few more information (see the attached PPT).

The current situation according to TAO-TRITON observations somehow is similar to those in 2001 and 1997. We can say that possible El Nino evolution this year will be between these two cases; it would be weaker than 1997 El Nino (may be much weaker) but stronger than 2001 case. If El Nino this year could not develop, then it would occur next year as in 2002 case. Our 2-year lead forecasts (see the last slide) showed that the probability of either case is equal.

These are just my personal thoughts. Hope you would find it useful.

Best regards,


On Mon, Jun 15, 2009 at 1:13 PM, Swadhin Behera <> wrote:

Hi Jon,

Yes, conditions might turn to that of 2002. I anticipate one exception though. In 2002, we saw warming in the central Pacific supporting an El Nino Modoki event. In this year, I expect a weak to moderate El Nino with warm anomalies in the eastern Pacific.

Though most of the models are predicting a pIOD, I wish those are wrong. It will be a very bad news for your part of Australia if we really get the fourth in the series of consecutive pIODs. The ocean conditions in the Indian Ocean in this month are crucial for the turnaround to a pIOD from the present nIOD and we should expect to get some indications by the end of the month.

Best regards,


On Monday, June 15, 2009 10:22 AM, Jonathon Welsh <> wrote:

Hello Jing  Jia

The discussion relating to the current season has been interesting although I do not understand fully all the technical details.
Looking back through the DMI, Nino 3.4 Indices a weak positive IOD and a Nino indice reaching 1.5 was seen in 2002. Do conditions show any similarity to that year? This was a very severe drought year across Australia.

Kind regards


On Mon, Jun 15, 2009 at 10:08 AM, Jing-Jia Luo <> wrote:

Dear Harry,

Another reason is the super-computer change. If you add 0.5C to our forecasts initiated from 1 April and May, you would find El Nino onset forecasts would be OK. We will redo the hindcast to get correct climatology for model forecast.

We had a paper (Luo et al. JClimate2008) which showed that ENSO can be predicted up to 2-year ahead, which is close to its theoretical predictability. In this sense, you may not expect too much space from ISO. Of course, ISO is important to ENSO evolution sometimes, particularly El Nino onset. We also plan to assimilate ocean-atmosphere data in near future.

One interesting but opposite example: I still remember that all other models predicted an El Nino in 2005-06 because of the strong warm intraseasonal Kelvin wave during 2005 boreal spring except our model which correctly predicted a La Nina (Luo et al. JClimate 2007). So we need to consider carefully how to properly assimilate ISO signal into forecast system.

IOD forecasts: If you check the latest Indian Ocean condition, there is some evidence for negative IOD (warm sea level height occurred along the west coast of Sumatra). One possible reason for the weak positive IOD in model forecasts can be due to the strong El Nino in model forecasts. I hope that the updated forecasts can be released to Australian farmer society in order to avoid any potential losses.

Best regards,


On Mon, Jun 15, 2009 at 8:35 AM, Harry Hendon <> wrote:

Hi, I agree with Luo that the “sudden” appearance of El Nino in the SINTEX-F forecasts maybe a result of not using observed ocean and atmos states to initial the SINTEX-F model (the initial states would underestimate intraseasonal variations associated with such things as ocean K-waves and the atmos MJO). It would be an interesting exercise to re-run forecasts from 2 mnths ago with and without atmos-ocean data assimilation. This isn’t possible with SINTEX-F but it would be possible with systems such as POAMA and ECSys3.

I thought there was also going to be some discussion about the SINTEX-F model predicting a negative Indian Ocean dipole two mnths ago, which received press here saying that this was good news for Australian farmers. The new forecasts flip the other way.

Harry Hendon
Centre for Australian Weather and Climate Research
A partnership of the Bureau of Meteorology
and the Commonwealth Scientific Industrial Research Organization


From: Jing-Jia Luo <>
Sent: Sunday, 14 June 2009 15:29

Dear Saji,

After the long-lasting La Nina during 2007 to early 2009 (this was correctly forecast up to 2-year ahead), Pacific ocean condition is ready for an El Nino. This is why I had tried to overestimate the probability of El Nino occurrence during last months even though our model forecasts did not show a clear tendency for El Nino (another reason is due to the machine change as I mentioned). The sudden change in the model forecasts can be triggered by the warm subsurface signal. The latest NCEP analysis showed a strong warming along the equatorial thermocline, a good indicator of a strong El Nino. Our model initial condition is determined by observed SST (SST forces AGCM which in turn produces wind stress to drive OGCM). The warm signal in the model subsurface is not as strong as the NCEP data and sometimes can be delayed if the warm observed SST is caused by a warm Kelvin wave. According to our hindcast during 1982-2008, El Nino intensity was often underestimated in the model prediction (as many other models). Considering all these factors and according to my personal experience, I think El Nino signal this year would reach >1.5C.

IOD has important influence on El Nino (we have done some experiments for this; the paper will appear in J. Climate), but I do not think IOD is a necessary condition for strong El Nino. Let's see what will happen in the real ocean.

Best regards,


On Sat, Jun 13, 2009 at 11:06 PM, Saji N. Hameed <> wrote:

Dear Luo-san and colleagues,

> between this month and last months. The failure in forecasting the
> strong El Nino last months may be attributed to two reasons.

It is interesting that there is chance of El Nino all of a sudden. Any interesting explanation, why we are so sure that a strong El Nino is going to fall upon us, other than that models are showing some signals? Is the Pacific all setup and ready, just waiting for that kick to explode....? ;)

Also a comment on the ENSO + IOD forecasts.

> > ENSO forecast: A strong El Nino would occur.
> >
> > IOD forecast: A weak positive IOD would appear, probably in response
> > to the strong El Nino.

In the last 50 or so years of observed records, there is not one strong El Nino without a strong positive IOD (Saji and Yamagata, 2003a,b). In view of this limited viewpoint, the forecast of weak IOD with strong El Nino is also not consistent with the past observations.

just my 20 wons...!

Saji N. Hameed

APEC Climate Center
BUSAN 612-020, KOREA


On Mon, Jun 15, 2009 at 1:16 PM, David Jones <> wrote:

Jing-Jia Luo,

The Australian media has heavily featured your rainfall forecasts - and I imagine that many farmers have considered your forecasts of good rainfall when planting crops and planning activities for the year ahead. This is most unfortunate as the Bureau's official guidance has been for an elevated risk of El Nino and dry conditions all year.


Dr David Jones

Head of Climate Analysis
National Climate Centre
Bureau of Meteorology
GPO Box 1289K, Melbourne
Victoria 3001, Australia
email :


on Mon, Jun 15, 2009 at 3:50 PM, from Lee-Anne Whitehead:

Dear Professor Yamagata

I have just read your latest IOD forecast on the website and was wondering if you can please tell me what the rest of winter and spring will be like in southern Australia, in particular, the state of South Australia.

The last couple of weeks have seen us complete the sowing of our harvest and to date, we have received sufficient rain but I just need some indication as to what the rest of the year would be like, if this is at all possible.

Once again I thank you for your informative website and the work that you do.

Kind regards

Lee-Anne Whitehead
ML & LJ Whitehead
Agriculturalists in the Mid North of
South Australia


On Sun, Jun 14, 2009 at 5:35 PM, Craig and Anna Anderson <> wrote:

dear jing-jia,

our coordinates are -34 lat and 144 long and am wondering if our area of s/e australia will receive above or below average rainfall between july + oct '09.

we have experienced severe drought conditions here since '02 having watched el ninos come and go with positive iod and am eagerly awaiting a la nina event.

your el nino and iod forecasts i assume are for the northern hemisphere, do we expect the opposite as we are in the southern hemisphere?

kind regards,
craig anderson
(australian ricefarmer)


On Fri, Jun 12, 2009 at 10:39 PM, Toshio Yamagata <> wrote:

Dear Luo-san:

I have several questions.
The model still does not use the subsurface data assimilation. However, the prediction result has become drastically better in June sepite that the colder SST bias of the new model. The prediction in May looks so bad in the Pacific. Isn't this due to the spring barrier?

Our model prediction skill was the best so far even if the subsurface data was not used. Why was the forecast in May was so bad compared to other institutions' and our forecast in June? Was our model performance worse compared to others' for El Nino associated with strong ocean dynamics? Was our model performance better comapred to others' for El Nino with less ocean dynamics (surface mode)? Was our model performance the best because of predicting El Nino of this kind successfully?

Best regards, Yamagata


On Fri, Jun 12, 2009 at 5:12 PM, Lixin Qi <> wrote:

Dear Jing-Jia,

The rainfall patterns from the model runs in May and June look quite different, say over Australia, the mode runs in May predicted wet season while the latest runs in June pointed at dry condition (pls see the figures below). Hope you have noticed this difference.



On Fri, Jun 12, 2009 at 6:10 PM, Jing-Jia Luo <> wrote:

Dear colleagues,

Please find the monthly mean maps as attached below.

As you may have already noticed the big difference of the forecasts between this month and last months. The failure in forecasting the strong El Nino last months may be attributed to two reasons. One is due to no assimilation of oceanic subsurface data in our forecast system. The other is due to the renewing of the Earth Simulator since April 2009. The same model with the same initial conditions on the new machine produces cooler SST in the Pacific as I mentioned before. So the calculated forecast anomaly (which is still relative to the climatology computed on the old Earth Simulator) is cooler (about 0.5C) than it should be.

Your comments are very appreciated.

Best regards, Jing-Jia


From: Vinson Kurian <>
Date: Sat, Jun 13, 2009 at 10:09 PM
Subject: Re: Seasonal forecasts from 1 June 2009 (seasonal mean part)
To: ashok karumuri <>
Cc:, Saji Njarackalazhikam Hameed <>, Doo-Young Lee <>, "" <>

Dear Dr Ashok,

Thanks so much for the detailed mail.

As you would see, the Indian monsoon is already running in a significant deficit after making an an early onset. Cyclone Aila seems to have been the culprit.

We would not be unduly perturbed in case of a strong El Nino (as Luo-san would have it) with a warming anomaly to the east Pacific (in 1997, we had a good monsoon despite a 'strong' El Nino with warming in the east Pacific).

But in case, as you observed, the warming gets concentrated in the central Pacific and the current La Nina Modoki gives into a El Nino Modoki by July, the monsoon is doomed (Is this transition possible?). Because, July rains are so crucial for the monsoon.

I had heard from Dr Luo-san, as also your mail to him copied to me, based on which I published a report fighting a tight deadline the same evening. Kindly see the link below:

I tried my best to condense FRCGC and APCC outlooks in the report but, on hindsight, could have done better. The time was too short; I ran a G-mail chat also with Dr Luo, gist of which too was incorprated into the report. Kindly excuse me if I've sounded hazy and non-articulate.

I'd love to hear from you all as you watch the big transition in the tropical Pacific.

Thanks and best,


2009/6/13 ashok karumuri <>

Dear Mr. Kurian,

Thanks for the kind email. I guess that Luo-san's explanation on the possible strengthening of prediction (such as the ocean assimilation) also may, at least to some extent, throw some light upon his shift in IOD signals. But this is my speculation, and Luo-san is best, and right, person to answer.

As you have mentioned, the recent SSTA conditions in the Pacific SSTA < > look like a La Nina Modoki, and may translate for better rainfall for India, when you take into account the gradient into the western tropical Pacific too. However, you have to wait and see how the SST evolves further. Our experimental MME predictions indicate the warming to be located in central Pacific, and so far it does not seem to happen in June, despite the skills for tropical Pacific SST better than those for rainfall. However, I wish that it were as simple as that. The predictions are also based on May IC, whereas the ocean signals, which are the predictive inputs, seem to have changed a lot in May, as we understand from the recent forecasts of Luo-san's explanation of possible causes. In summary, all I can say now is that the tropical Pacific needs to be watched carefully for the next month or so.

Doo-Yang, I do not have your predictions at home, and if I am wrong regarding the location of the maximum SSTA , pl. write a sentence or two.

Best regards


Date: Fri, 12 Jun 2009 15:14:13 +0530
Subject: Re: Seasonal forecasts from 1 June 2009 (seasonal mean part)

Dear all,

Thanks for the update on IOD.

Dr Luo, I'd be grateful to know why this sudden change in outlook from a negative IOD to a weak positive IOD has happened.

As of now, the SSTs over tropical eastern Pac are higher than those over central Pacific. This should actually work to the benefit of the Indian monsoon. A weak +ve IOD would further boost the monsoon prospects.

Is there a way of knowing whether the SST differential between east and central Pac would remain as such going forward? Are there any forecasts available?

I'd also like to hear from Dr Ashok as to what he makes of this.

Thanks and best,


On Fri, Jun 12, 2009 at 8:51 PM, Jonathon Welsh <> wrote:

Hi Jing Jia

Thank you for your revised forecast for our growing season. This is of great concern the developing el nino and the disappearance of the negative IOD. Do you have any monthly charts available?

Kind regards


On Fri, Jun 12, 2009 at 12:30 PM, Jing-Jia Luo <> wrote:

Dear Sensei and colleagues,

Please find the attached PPT for FRCGC/JAMSTEC seasonal forecasts initiated from 1 June 2009.

ENSO forecast: A strong El Nino would occur.

IOD forecast: A weak positive IOD would appear, probably in response to the strong El Nino.

Related with the strong El Nino and weak positive IOD; Australia would have dry ad hot boreal summer and fall. India would have poor monsoon.

Note: Certain uncertainty may arise from the renewing of Earth Simulator. El Nino prediction appears to be underestimated.

If you have any comments/suggestions/requests, please kindly inform us. Thank you very much.

Best regards,
Jing-Jia Luo
Seasonal prediction:


--- 2007 ---

> From: Toshio Yamagata <>
> Subject: pIOD is coming back?
> Date: Sun, 24 Jun 2007 07:47:11 +0900
> Dear Mina-sama:
> It appears that IOD is coming back again in the IO just like the very rare
> case in 1967. It might be too early to say this but our CGCM prediction
> seems to be correct. Please visit the Indian Ocean observations at
> updated by Surya Rao-san. I
> am afraid that La Nina plus pIOD may cause very hot summer in East Asia
> with serious drought associated with localized heavy guerrilla rains as in 1967.
> Best regards, Yamagata

Dear Sensei, and Colleagues,

The latest observations (see attached PPT) did show some evidence for the possible occurrence of positive IOD this year. Japan had less rainfall and delayed onset of Meiyu/Baiyu season in June.

Best regards,


Dear Luo-san and colleagues,

Just for your information.
TRITON temperature at 1.5S, 90E indicates subsurface cold anomalies at the thermocline depth and deeper layer after mid-May, which is quite similar to the conditions in 2006. (This also suggests very weak spring Wyrtki jet in May this year.) Our post-doc Horii-san kindly made us the attached figure.

Best regards,


Dear Yukio and Horii-san,

Thank you for keeping us up to date. 2006 and 2007 are turning out to be most interesting years. I hope the SINTEX-F prediction is wrong because it might cut down the rainfall Australia normally receives from La Nina. But then we really don't know the rainfall pattern for La Nina/IOD+!!

And while I have your attention, would all please note my new email address and change your contact list?--gary

Prof Gary Meyers, Director
Integrated Marine Observing System
University of Tasmania


Dear Masumoto-san: Thank you for the very interesting in situ subsurface data from the TRITON buoy. The eastern IO continues to be cold since last pIOD except for the very short period of the arrival of the spring Yoshida-Wyrtki jet. The current situation is very similar to pIOD year 1967 (assocated with La Nina) after pIOD in 1966 (associated with weak El Nino; leave the precise definition alone). I think we had better go ahead about cautioning regional societies about possible floods/ droughts/hot conditions.
Best regards, Toshio Yamagata


Dear Toshio and all,

I would caution against rushing to a public announcement. Seasonal predictions can be very upsetting to the publice who live in areas at risk.
At a minimum I think any public announcement should be based on a multi-model summary. I note several groups that routinely run a seasonal forecast are on this mailing list. So I would like to know if the other groups can or will make their predictions for the Indian Ocean available. Perhaps these could be assembled and if appropriate the announcement made on the basis of a multimodel summary.--gary

Dear Colleagues

Perhaps it is a good idea to check forecasts from several models as suggested by Gary. But I am not sure if that will help to improve the confidence in model predictions. I know that MME predictions do a better job than a single model predictions. In a situation where 9 out of 10 models are giving wrong predictions, we have no way to improve a particular model forecast.

Last year, because of the debate early on during the development of IOD, we delayed the press release until August. But probably because of the delay in the announcement the information could not be utilized by people who were most affected by the 2006 IOD. More than a million people in East Africa suffered from floods and Australian farmers had to manage with severe short-supply of water and fodder. Some of the farmers who kept calling us almost on daily basis told us that they could have been benefited had the information about the IOD was passed on to them based on our model predictions.

Interestingly, several models predicted a negative IOD in 2006 until the late stage of IOD formation. So, it is not clear how soon is too soon to announce a model prediction. But I feel to save people from extreme climate conditions, at least, we need to pass some cautionary notes to the decision makers based on the models that shows consistently good skill in seasonal climate predictions. Together with the monitoring system, even a moderate prediction can be very beneficial to the society.

Best regards
Swadhin Behera

Dear All,

Following the discussions initiated by Gary and Behera I wish to bring it to your notice that many CGCMs predict a pIOD this year. Please have a look at the attached figures.

Best regards,

Dear Swadhin,

Email is not a good way to have the discussion we need to have. All I'm calling for is that we be deliberate in our judgement in this serious matter. I think it is a decision to be made by the small group of experts who generate forecasts of the Indian Ocean--gary

Hi Rao, I think you picked up the forecast for east Indian Ocean SST from POAMA. The attached figure is the prediction for the dipole index from POAMA
POAMA is going for a negative IO dipole. The other figure is a spaghetti plot of the contour of the max SST anomaly from each member that verifies in October for forecasts that were initialized this month. POAMA is obviously still going for La Niña, and warm SST in east IO. Given the weak negative SST anomalies in the east IO as of this week, and the fact that some of the members from POAMA start out going cold in the east IO, it will be interesting to see how things evolve.





I think that your comment, and those of Swadhin and Gary raise an important set of issues. It is important to communicate, along with the forecast, a measure of uncertainty. Many groups within the larger "forecast user community" are well set up to deal with uncertainty in a forecast. Even people not formally trained in statistics, and without cost-benefit models for their industry, will often have an intuitive appreciation of probabilistic forecasts - I believe that gambling is one of the most universal components of human societies.

I do not know what the other forecast systems - beyond Frontier and POAMA - are predicting for this coming season. Using the spread between the forecasts may be a decent estimate of uncertainty, right now that gives a 50/50 forecast - but that's not a good estimate, since with two elements it's either 100% or 50/50. What do the other centers say?

Another way would be to have probabilities of "false positives" and "false negatives" for both forecast systems, and use those to assess the combined forecast. Does anyone have those probabilities for the POAMA and Frontier systems?

It is my opinion that these forecasts can be usefully communicated to society, but it is essential that the uncertainty inherent in them is communicated as well. The probabilistic forecasts will allow for the forecasts to have some potential utility, as well as help minimize some of the potential for the negative unintended consequences.

Given how unusual it is for IODs to follow eachother, and how unusual it is for IODs to occur with La Niña events (one of which is limping along), I'm feeling cautious about the forecasts. But anyone who listened to my forecast last year (I said no IOD) knows better than to pay attention to what I say on this.

Regardless, it's very exciting to see the forecast systems working and the in situ data being used (nice plots, Yukio). Such a change in a few years.



Dear Toshio,

I remember reading the Neumann and Charney paper, some years ago, probably about 10 years after it was first published. I still marvel at the progress numerical weather prediction has made, and I think seasonal climate prediction is on a similar long term track. The question is how do we interact with the public now, in this early stage? We need to take into account both the best interest of the public and the advancement of our science. Along with R&D on the coupled systems there has to be research on how to best make societal use of the Indian Ocean predictions.

If you make an announcement now, based on SINTEX-F, it will be a result/forecast that is quite simple and clear. It will be interesting and probably it will be picked up by the media. Also, it probably will be sensationalized. We have to ask ourselves, what can people in the affected areas do with such information? Maybe they can do something but it will have to be tentative. If strong action is taken, there is a risk that the forecast is wrong and that also causes damage.

If a group of operational and/or experimental forecasters put together a statement based on multi-model information, it will represent the state of our science better. It will have to be scientific in nature and may not be so interesting or easy to read for the public. It may or may not be picked up by the media.

As I see it we have a dilemma, as you point out. It is even a double edged dilemma--properly informing the public and properly representing our science.

I still would like to see the small group of seasonal forecasters--perhaps only the ones running state of the art experimental systems--put together a consensus statement.

Best regards, Gary


Dear Gary:

Thank you for your thoughtful comments and for raising a very important issue. I understand your anxiety; it is quite reasonable particularly from the viewpoint of weather bureaux that are responsible for outcomes of their <forecast>.

Once we release the information of another pIOD in 2007, suicide rate of Australian farmers, for example, might increase at a significant level. If our <prediction> turns out to be wrong, we might be accused of this. If the Earth climate follows our prediction, on the other hand, no release of such future information will lead to loss of many lives, particularly in developing countries, as Swadhin mentioned. No one except for our frontier scientist conscience will blame us in the latter case.

Just like the global warming issue, we need to discuss whether a <precautional approach> of frontier scientists (rather than operational agencies' official ones) in the seasonal <prediction> is beneficial to the society or not. We need to evaluate this scientifically. Personally I think it is useful as a whole even at the present level. The difference between the evolving seasonal <prediction> in the tropics and the weather <forecast> is becoming only quantitative because of the steady progress (even if it is very slow) of our <prediction> skill. (Please remember the weather <forecast> in the 1960s right after the innovation of Neumann and Charney.) The situation seems to be even much better compared to the global warming <projection> that has qualitative difference from the seasonal <prediction> as well as the weather <forecast>.

Best wishes, Toshio


Gary, I was thinking about the official seasonal forecast issued from BoM (which is currently based entirely on empirical relationships). They don't even factor in the IOD. They assess the likely state of El Niño, and the likely state of the subtropical Indian ocean (the so called Nicholls dipole), which was shown to have a relationship with Australian winter rainfall in an older Nicholls paper (which I think is now regarded as a bit of a red herring as far as predictive capability is concerned). The importance and predictability of the IOD has not percolated up (or down) to those who issue the forecast. I guess that's the fault of the researchers (like me) in BMRC.

For those interested, I include a link to the official BoM seasonal forecast issued last July 2006, when we knew we going to have a pIOD. As you can see from reading the justification of the BoM forecast,

the warm loading onto the subtropical IO was an important factor. This link shows what this SST pattern is in the IO , which you can see has nothing to do with an IOD. So, in the official BoM forecasts, there is no mention of the IOD. Of course as it turned out last year, the official forecast for the SE of Australia wasn't too good: we had a pIOD and record drought.

Ideally, in the near future the Bureau will switch their official seasonal forecast to the direct output from the dynamical ensemble system (single model or multi-model), for which the hindcast skill is well documented. In this case, we won't selectively call out a dipole event or El Niño event, rather, we will directly provide the regional climate predictions (e.g. pdf of eastern Australian temperature and rainfall) that result from prediction of the entire coupled system.