Over the western North Pacific Ocean during June-October, a tropical cyclone forms approximately every 6-7 days. However, there are time periods during which tropical cyclones occur more (less) frequently than the climatological average. Periods of increased or decreased tropical cyclone formations occur over time scales that extend beyond synoptic-scale variability to intraseasonal time scales, Therefore, there is hope that extended-range forecasts of periods of above (below) normal tropical cyclone activity could have predictive skill. In this study, the physical basis and forecast procedure for predictions of intraseasonal variability in western North Pacific tropical cyclone activity are defined.

A wavelet analysis of the variability in 850 mb circulation over the western North Pacific during June-October 1979-1998 identifies significant power in frequency intervals between 30-90 days, 10-25 days, and 2-8 days. The influence of the variability in low-level circulation in each frequency range on tropical cyclone activity is examined. Representative patterns of circulation and outgoing longwave radiation (OLR) data in each frequency range are defined by Singular Value Decomposition. Intraseasonal patterns in the 30-90 day range identify variability associated with a modulation of the Asian and western North Pacific monsoons via well-defined influences of the Madden-Julian Oscillation. The primary patterns at 10-25 days represent a combination of northwestward-moving circulations over the Philippine and East China Seas and northeastward-moving circulations that extend into the subtropical central North Pacific. Synoptic-scale patterns are dominated by combinations of equatorial mixed-Rossby gravity waves over the central Pacific and northwestward-moving waves over the western North Pacific. Individually, significant circulation patterns in the three frequency bands do not explain a significant amount of the variability in tropical cyclone activity.

Based on circulation indices from data filtered for each frequency range, a cross-wavelet analysis defines periods of significant covariability between the frequency ranges. New circulation and OLR modes are defined based on separation of periods of significant interactions between frequency bands from periods when there are no interactions. The addition of these conditioned modes to the individual modes defined above increases the amount of explained variability in tropical cyclone activity.

A framework for a canonical correlation prediction of tropical cyclone activity is established based on initially identifying the contribution from the slowly-varying 30-90 day mode. The contribution from each higher-frequency mode is then conditioned on the state of the lower-frequency modes.