Introduction
The Hype
The Impacts
Interactive Timeline
La Niņa
Prediction Methods
> benefits of predictions
> TOPEX/Poseidon
> TOPEX/Poseidon data
> uses for predictions
> types of
models
> obstacles
The Preparation
About the Site
Archives
Search
Learn Anything?
Forum
______________________
Obstacles in Forecasting
Problems to Overcome
At the present time, routine predictions of ENSO are being made with a variety of
statistical and dynamical models. Each forecast system has its unique problems and
limitations, but all are limited by observational data in one or more ways. For some
fields, such as SST and surface winds, gridded data are available for a few decades, but
the quality of the data deteriorates rapidly prior to 1970. For other fields, such as
thermocline depth or ocean currents, the situation is worse yet. The lack of comprehensive
and reliable data for the climate system over a period of several decades is a major
limitation for all forecast systems. Not only does it limit the number of retrospective
forecasts with which the systems can be developed and validated, but it also limits the
ability to initialize accurately the forecasts that can be attempted.
Another problem concerns the injection of multiple sources of data into models – the
so-called assimilation problem. The basic problem is that a typical collection of data is
incompatible with any forecast model, due to data errors and model shortcomings. To date
no assimilation scheme specific to the coupled atmosphere-ocean system has been utilized
in the ENSO predictions. Given the coupled nature of ENSO, such a scheme would be the
optimal one, without it, predictions are unnecessarily limited. For all the forecast
models that have been analyzed, forecast skill is found to vary in time. One of the most
striking variations concerns the seasonal cycle. It is found that predicting through the
period March-May of each year is considerably more difficult than other seasons. For some
models, the so-called "spring barrier" is severe enough to completely invalidate
many forecasts; for other, it simply limits some forecasts relative to others. In
addition, forecast skill typically varies from year to year. One possibility is that the
predictability of the real climate system is variable (in which case no prediction system
can overcome it). Another is that systematic errors or omissions in the models, or the
lack of adequate data are limiting the models’ potential skill. It is likely that
both contribute to some degree.
Most believe that the best prospects for climate prediction lie ultimately with the most
complete dynamical models – coupled with general circulation models. An inherent
problem with the development of complex models is precisely their complexity. In practice,
it is very difficult to "tune" a complex model, even if one is aware of
systematic errors. An added problem is the computational expense. Complex models require
an enormous amount of computing resources, even with present-day technology. The inability
to produce large ensembles of predictions over an acceptable period of time seriously
slows progress. Typically, analyses are done on marginal or totally inadequate samples,
and the conclusions are in danger of error. The net result is that real progress is very
slow indeed.
A final and very important problem involves the extension from ENSO proper to global
climate. One methodology is to invoke statistical relationships between ENSO
manifestations (such as east Pacific SST) and climate variables in other regions of the
world. Such analyses have been done , and are steadily being refined. They suggest
important connections between ENSO and temperature and precipitation extremes in many
regions. The statistical approach is, of course, limited by the observational data base,
as well as the inability to produce new patterns – ones not appearing in the data
base from which the statistical analyses is based. Clearly, nature is not similarly
constrained. The other approach is to attempt to predict the regional climate signals
directly with a global atmospheric model. The problems here consist of those discussed
above, as well as the following: even given perfect knowledge of ENSO, the ability to
predict the seasonal climate in many parts of the world (particularly outside the tropics)
is extremely limited. The chaotic nature of the atmosphere in the extra-tropics can result
in large seasonal anomalies even the absence of ENSO. From the climate prediction point to
view, this internal "noise" adds considerable uncertainty to any prediction, and
requires large ensembles of forecasts to achieve any statistical significance.
Previous Section: Types of Models