There is a paper by
Lavers, D., L. Luo, and E. F. Wood (2009), A multiple model assessment of seasonal climate forecast skill for applications, Geophys. Res. Lett., 36, L23711, doi:10.1029/2009GL041365. [thanks to Jos de Laat for alerting us to it!].
The paper includes a very important conclusion on what is achievable in terms on seasonal (and thus) longer term forecast skill.
The abstract reads
“Skilful seasonal climate forecasts have potential to affect decision making in agriculture, health and water management. Organizations such as the National Oceanic and Atmospheric Administration (NOAA) are currently planning to move towards a climate services paradigm, which will rest heavily on skilful forecasts at seasonal (1 to 9 months) timescales from coupled atmosphere-land-ocean models. We present a careful analysis of the predictive skill of temperature and precipitation from eight seasonal climate forecast models with the joint distribution of observations and forecasts. Using the correlation coefficient, a shift in the conditional distribution of the observations given a forecast can be detected, which determines the usefulness of the forecast for applications. Results suggest there is a deficiency of skill in the forecasts beyond month-1, with precipitation having a more pronounced drop in skill than temperature. At long lead times only the equatorial Pacific Ocean exhibits significant skill. This could have an influence on the planned use of seasonal forecasts in climate services and these results may also be seen as a benchmark of current climate prediction capability using (dynamic) couple models.”
The discussion part of the paper has the very important finding (which comments on climate predictions)
“Given the actual skill demonstrated by operational seasonal climate forecasting models, it appears that only through significant model improvements can useful long-lead forecasts be provided that would be useful for decision makers – a quest that may prove to be elusive.”