Medium range ensemble sensitivity analyses of two extreme Pacific extratropical cyclones. Part II: Validating ensemble sensitivity using perturbed initial conditions experiments

Extratropical cyclones often cause high-impact weather such as strong winds, heavy rainfall and severe snowfall. However, medium range (3-7days) forecasts of extratropical cyclones, especially exact locations of their centers and strength, are often quite inaccurate. These forecast errors may be due to model errors or uncertainties in initial conditions, thus it is of value to investigate how these forecasts depend on the initial conditions. With the routine use of ensemble forecast systems, ensemble sensitivity techniques have been developed to examine how forecasts depend on the initial conditions. However, previous studies have mostly evaluated the sensitivity of short-term (1-3 days) ensemble forecasts of cyclones. In this study, we employ the ensemble sensitivity approach to analyze medium range forecasts.

A global ensemble Kalman filter data assimilation and forecasting system has been used to generate ensemble forecasts for selected cases. Medium range forecasts for two extreme Pacific cyclones are diagnosed to objectively determine the relationship between the forecast uncertainties and the initial upstream flow fields, In part I of this study, sensitivities to different forecast metrics related to cyclone intensity and locations have been tested and compared. In Part II, those sensitivity results are validated using model experiments.

A perturbation is generated by regressing the forecast metric (e.g. cyclone central pressure) with the ensemble initial condition. This same perturbation is then added to the initial condition of each ensemble member and the perturbed ensemble is integrated and the results compared to the control ensemble. The change in forecast metric that is realized in the perturbed ensemble is then compared to the original forecast metric (cyclone central pressure in this example) to validate whether initial condition perturbations derived using the linear ensemble sensitivity approach can actually produce the desired change in the forecast metric under the highly non-linear evolution of the forecast flow field.

Consistent with previous studies, our results suggest that for 60-hour forecasts, the perturbed ensembles largely realized the expected changes in forecast metric when small initial condition perturbations are applied. However, as the lead time is increased, the amplitude of the realized change decreases, and at 7.5-day lead time, no significant changes in the forecast metric are realized in the perturbed ensemble for initial perturbations generated using cyclone pressure, latitude, or longitude as the forecast metric. However, our results suggest that there are other forecast metrics that may work better and whose predictions (partly) survive for at least 7.5 days, and these results will be discussed in the presentation.