The clearest evidence for predictability in the seasonal climate is from the proven impact of sea-surface temperature anomalies on the seasonal large-scale atmospheric circulations, as has been demonstrated in the first two lectures. Most operational seasonal forecast methods attempt to utilize this discovery. Five classes of methods are described below, whose forecasts have been and will continue to be available:

(i) Coupled Ocean-Atmosphere-Land General Circulation Models (CGCMs)

(ii) Atmospheric GCMs (as discussed in Part 2) forced with forecast SST fields

(iii) High resolution regional climate models (RCMs) forced with the large scale forecast fields from the GCMs/CGCMs

(iv) Statistical transformations of the output and/or statistical combinations of the output from the above model types

(v) Statistical models designed to capture the physics and dynamics of the climate system evolution in a set of statistical relationships based on observed historical data. The fundamental equations of physics are not used.

The first lecture noted that there does exist other possible sources of seasonal predictability beyond the coupling of the ocean and the atmosphere. We here pause to review the ones that are currently considered most likely to add to the predictability currently being achieved with SST-based methods. The reader should not be led into a sense of over optimism that these sources will dramatically increase the accuracy of seasonal predictions, but they may in some regions have a noticeable contribution to make.

a) Land surface state. Research has demonstrated that the state of the land surface at the start of a season can impact the evolution of the climate of the coming season. In particular, the soil moisture content has been shown to be important in semi-arid regions. One of the difficulties in quantifying how important the effect is, relates to the difficulty in making reliable observations of soil moisture distributions. Another candidate is snow cover. For example, the winter snow pack of the Himalayas has been proposed to impact the evolution of the subsequent Indian summer monsoon.

b) We mentioned in Part 1 that there is an effective lid on approximately the lowest 10km of the atmosphere, where most of the processes that impact our weather develop. The atmosphere below the "lid" is called the troposphere and the next zone above the lid is called the stratosphere. It is known that there are periods of strong anomalous winds in the stratosphere, especially varying in a quasi-regular two year cycle, know as the stratospheric quasibiennial oscillation (QBO). It is possible that these stratospheric wind anomalies can impact developments in the lowest 10km of the atmosphere (the troposphere). If they do, then this can further contribute to our ability to anticipate seasonal climate variations, because the QBO is sufficiently regular to be projected quite accurately on seasonal to interannual timescales. At this stage, evidence for the impact of the QBO on the seasonal climate in the troposphere is mainly empirical. One of the difficulties in establishing the relationship empirically is that there are possible sources of climate variation within the coupled ocean-atmosphere system on similar timescales, and these variation could statistically appear to be associated with the QBO, even without a real physical mechanism linking them. Ideally, we would like to see evidence using GCMs, modelling the impact of the stratospheric wind variations on tropospheric seasonal climate, analogous to modeling the impact of the SST variations on the seasonal climate (Part 2). Research is underway, but it is a difficult task to model the effect.

c) Within the troposphere itself, it has been proposed that the large scale structures of the initial atmospheric state can have some bearing on the future seasonal evolution of the atmosphere. It is not proposed that the precise location of a storm today can bring predictability to the coming seasonal climate anomaly (the deterministic influence of this a weather event on the atmosphere's evolution is indeed typically considered to be limited toin the next 10-14 days). Rather, it is considered that large-scale atmospheric structures such as the strength and broad location of jet streams and monsoon flows may have a modest influence on the evolution of the coming climate. Evidence using GCMs has so far not been found definitively.

d) Atmospheric Composition. Part 1 noted that changes in atmospheric composition can act as a forcing on climate and bring predictability to the system, if the forcing can be anticipated. The most widely discussed is the changes in greenhouse gas composition that have occurred and continue to occur due to human activity, bringing a degree of predictability to the climate system over long timescales (e.g. the average globale temperature expected for 2050-2100).For seasonal prediction, in some regions of the world, anticipated atmospheric pollution during the coming season may come to be an important additional factor in the climate to expect in the coming season. In this context, recent work on the Asian Brown Cloud may be an example.

The above factors (a-d) are being actively researched and may in the future contribute additional predictability to the widely used operational systems employed today. It should be noted that some forecast systems, especially statistical methods, already attempt to include these factors, such as the QBO. However, it is generally considered that the most reliable skill of current operational seasonal forecast systems is almost wholly rooted in the coupling of the slowly varying ocean with the atmosphere.