Further Thoughts on ENSO
Arun Kumar
NOAA/NCEP/Climate Modeling Branch
Camp Springs, MD. USA
wd01ak@sun10.wwb.noaa.gov
Why Categorize?
It seems that the naming (or categorizing) of tropical Pacific SST events is necessary only because of the limited historical data. In principle, with respect to year-to-year variations in the tropical Pacific SSTs, the impact(s) of anomalous SSTs form a continuous spectrum and can be predicted for all anomalous SST states. For example, let us assume that the tropical SST anomalies have a single preferred spatial structure and that the interannual variations in the ENSO events are due only to the changes in the amplitude of this pattern. Let us further assume that the atmospheric response is linear and proportional to the amplitude of the anomalous SSTs.
Within this paradigm, based on our favorite choice of threshold, if we categorize the tropical Pacific SST events into El Niño and La Niña, it does not imply that the impacts of an SST anomaly below that threshold (i.e., when the SSTs are categorized as normal) are absent. It is only that the impacts are smaller and the forecast may not be of any societal benefit, on the average. The concept of “benefit,” however, depends on how long one wants to (or can) use the predictions to benefit their cause. For example, benefits from the use of predicted information for specific users may accrue over small numbers of events for large amplitude tropical SST events. For the weaker SST events, however, positive gains require the application of information over a larger number of events.
Within this scenario one can question what we are gaining by categorizing the events. If the information is targeted to a user who has a basic understanding of the scope of seasonal prediction, the thresholds may not be of much use. For now, it seems that the biggest advantage of the categorization is so the we can use an appropriate composite for the predictions.
Loaded Dice and Seasonal Predictions
Probabilistic forecasts seem to be a hard concept to convey to individual users (e.g. individual farmers or fishermen) who are directly effected by the use of information (as opposed to the use of predictions by large corporations). Human memory (or the life span) is not long enough to see the average benefits from the use, if seasonal predictions occur over a long period. Also, possible non-linearity in the economic benefits at the individual level may result in bankruptcy in case of failure (if the probabilistic prediction does not happen to be correct) but may result in equal riches if the prediction happens to be right (an example, once again, would be individual farmers).
The analogy of a seasonal forecast as “loaded dice” is often given, but we should not forget that one can toss the dice a hundred times within the span of an hour, but one can only play maybe 50 “games” based on seasonal prediction in a human life time (needless to say the average length of the game over which one will win depends on how loaded the dice are, or in terms of the SSTs, how strong the SST events were).
It is just a thought, but maybe most of the beneficiaries of the seasonal predictions will be the users in the arena of “large economics,” e.g., large corporations.
Do ENSO Events Increase Weather-Related Disasters?
It will be interesting to study, for both La Niña and El Niño events whether the impact of ENSO is to increase the frequency of weather-related disasters, or whether the tropical SST events merely increase our ability to predict the geographical location where these disasters are more likely to occur. In other words, do ENSO events act like a lens which tend to focus on the “random distribution of disasters” to particular regions?
Why Do We Have So Many Seasonal Forecasts?
During the La Niña summit, the current proliferation of the SST and seasonal forecasts was cause for some concern and different mechanisms were proposed to deal with it. It is interesting to contrast this situation in seasonal forecasting with the situation in the short or medium range atmospheric forecast community where a prediction effort is dominated by a few large organizations (mostly weather centers). Is there a fundamental difference between the two either in terms of the methodology or in terms of the scientific nature of the problem?
One fundamental difference which comes to mind is the use of atmospheric initial conditions for the short and medium range forecasts. To gather and analyze this information over a large geographic region requires a large infrastructure, and hence the paucity of the availability of a multitude of short range atmospheric forecasts. To date, detailed initial conditions for the ocean 3-dimensional structures are only used in a few of the coupled models compared to a plethora of statistical (or semi-statistical) seasonal prediction models. Depending on one’s point of view, one can draw his/her own projections as to what the future in seasonal predictions might be, i.e., will it become like medium range forecast procedures where a detailed 3-dimensional specification of the ocean initial state will be crucial, or will a rough knowledge of large scale anomalies suffice and statistical models will always be a viable alternative method?
Attribution of Weather (or Climate) Events
We seem to endlessly argue about the attribution of extreme weather events to something. Given that seasonal forecasts are probabilistic (even more so on sub-seasonal time scales), a basic question is whether it makes any sense to seek causes for individual events. Is it more meaningful to attribute climate anomalies to boundary forcing in a statistical sense? One can go back to the analogy of the loaded dice again, and where the probability is higher towards one particular outcome. For a particular draw, how can we attribute the outcome to the fact that the dice was loaded?
The bottom line for any attribution should be that it will be helpful in prediction at some future stage.