ESRL/PSD Seminar Series

An empirical modeling approach that can be used to make tropical forecasts and evaluate the processes contributing to and detracting from their predictability on subseasonal time scales is discussed. The model used, a “coupled” linear inverse model (CLIM) derived from observed simultaneous and time-lag correlation statistics of 5-day running mean anomalous SST, 20ºC thermocline depth, OLR, and 200 and 850 mb winds, which has been run at PSD in near-realtime for the last few years and is currently being transitioned to CPC for operational use. Its forecasts have skill that is comparable to CFS2. In general, geographical and temporal variations of forecast skill are quite similar between the LIM and CGCMs, making the much simpler CLIM a potentially useful forecast benchmark and also an attractive tool for assessing and diagnosing overall predictability the CFS2 and potentially other CGCMs. It is shown that certain initial conditions, derived from a singular vector analysis of the CLIM system propagator, result in both maximum anomaly amplification and greater realized forecast skill in both the CLIM and CFS2. Additionally, the eigenvectors of the system's dynamical evolution operator separate into two distinct, but nonorthogonal, subspaces: one governing the nearly uncoupled daily-to-subseasonal dynamics, and the other governing the strongly air-sea coupled longer-term dynamics. These subspaces arise naturally from the CLIM analysis; no bandpass frequency filtering is applied. A further analysis of which processes drive predictability on different time scales shows that maximum forecast skill occurs when initially destructive interference of anomalies from these two subspaces evolves to yield constructive interference.