Sensitivity of Global Weather Forecasts during the 2015–16 El Niño Event to Additional Observations, Data Assimilation Methods, and Model Stochastic Parameterizations

Four sets of 7-day 80-member ensemble forecasts, started every 12 hours in the mid-January to mid-March 2016 El Niño Rapid Response (ENRR) Field Campaign period, were made using NCEP’s Global Forecast System (GFS). The sets differed in their inclusion or exclusion of the ENRR radiosonde and dropsonde observations over the east Pacific in the initial conditions, using a hybrid EnKF/4DVar versus a pure EnKF data assimilation method, and including or excluding stochastic parameterizations in the forecast model. The control forecast set was defined as the set that used the ENRR observations and the hybrid data assimilation method and included stochastic parameterizations in the forecast model. The control forecast root-mean-square (RMS) errors (defined with respect to the hybrid analyses that included the ENRR observations) of upper tropospheric geopotential height, upper tropospheric vorticity, mid-tropospheric vertical velocity, and column-integrated precipitable water in this control set were then compared to the errors in the other sets. In general the forecast errors were found to be only slightly sensitive to the ENRR observations, and more sensitive to the data assimilation method used. The 7-day forecast errors of upper tropospheric vorticity, mid-tropospheric vertical velocity, and column-integrated precipitable water are most sensitive to the inclusion or exclusion of stochastic parameterizations in the forecast model.

Including the ENRR observations generally strengthened El Niño-related features in the extratropical upper tropospheric circulation. However, the effect was weak and did not reduce the RMS forecast errors everywhere. Interestingly, the ENRR observations also affected the evolution of extratropical synoptic waves. This was most clearly manifested in the time-evolving ensemble-mean differences between the control forecasts and the forecasts without the ENRR observations as an eastward propagating synoptic wavetrain. Such a wavetrain was also evident in the difference fields of the control and other forecast sets, but with much weaker amplitude. The wavetrain strongly resembled the dominant perturbation eigenmode of the extratropical circulation. The simplest interpretation of these results is that the perturbations in the initial conditions provided by the regionally localized ENRR observations systematically excited this eigenmode with a preferred phase, whereas the perturbations in the other forecast sets were not systematic in this way.