CGAM Tropical Group

The presentations in Session 5 highlighted the huge gap in the expectations of users of seasonal forecasts and the current level of skill. This is particularly true for precipitation, which, as in numerical weather prediction, proves to be one of the most challenging parameters to simulate and forecast. Users require quantitative information with more detail than the standard tercile forecast, and many applications need information on the distribution of rainfall through the season, in other words, the statistics of the weather.

But QPF on seasonal timescales is also important for other, more fundamental reasons. The mean distribution of tropical heating, associated primarily with precipitation, drives the planetary scale circulation patterns, and it is changes in the distribution of rainfall, particularly in the tropics, that drives the major teleconnections which seasonal forecasting seeks to exploit. Because of the non-linearities between the heating distribution and the circulation response, we know that the linear approach to removing model biases, used in many seasonal forecasting systems, is fundamentally unsound.

So improving model performance has to be a priority, if not the priority in the coming years. Systematic errors in the simulation of key aspects of the weather, such as the diurnal cycle, mesoscale convective systems, and the Madden Julian Oscillation, have been topics of discussion this week in the context of short-range forecasting and extreme events. They are highly relevant also to seasonal prediction because there is good evidence that interactions between the whole range of space and time scales are important for the mean climate and its variability. There is little doubt that increasing the resolution of the seasonal forecast models could lead to significant improvements in the representation of the weather and in handling these scale interactions. There is clearly much to be learned from the experience of those working in numerical weather prediction and with detailed process models, such as CRMs.

In many ways, seasonal forecasting is still in its infancy, at the stage that NWP was 20 plus years ago, but I think our expectations are much higher because of the success of NWP. But we need to remember that seasonal forecasting is inherently difficult because we have to deal with the coupled ocean-atmosphere system, which has many more degrees of freedom.

There is no doubt that operational weather forecasting has had a huge impact on the development of better atmospheric models and more accurate representations of key physical processes. As operational seasonal prediction develops, it is clear that it will, in turn, provide an excellent test-bed for coupled models. By continually confronting the models with observations and by continually being challenged by new events (such as the Eastern European floods), seasonal forecasting raises important questions about what drives climate variability and what factors influence its predictability.

Even at this early stage in the development of operational seasonal forecasting, there are causes for optimism. We already know that there is useful skill when the boundary forcing is strong (El Nino). We also know that estimates of potential predictability are probably conservative because they are based on imperfect models. It is likely that by reducing model systematic error, we will find increased levels of predictability even for regions of, currently, rather marginal skill, such as the Atlantic/European sector. As the models become more skillful and the information levels in the seasonal forecasts improve, particularly with the move to higher resolution and a better representation of the statistics of the weather, then more sophisticated approaches to QPF on seasonal timescales will undoubtedly emerge.