Biases in global reanalysis datasets undermine the forecasting skill of tropical intraseasonal variability

Sensitivities of tropical intraseasonal forecasting skill to different initial conditions and air-sea coupling are examined with the University of Hawaii Hybrid Coupled Model (UH_HCM). The target period is 2004-2008 boreal summer. We found that significant biases exist in various reanalysis datasets (such as, NCEP R1, R2; ERA-Interm). For example, the amplitudes of the convective activities associated with tropical intraseasonal variability (TISV) in NCEP R1 are smaller than the observed counterpart by a factor of two to three. Motivated by this fact, we carried out a suite of forecasting experiments to explore the impacts of initial conditions on intraseasonal forecast skills.

Our results reveal that with the original NCEP R1 as initial conditions the intraseasonal forecast skills of 850-hPa zonal winds and rainfall are only about a week over the global tropics (30oS-30oN) and the Southeast Asia (10oN-30oN, 60oE-120oE). The predictability increases steadily with increased amplitudes of TISV in the initial conditions. When the TISV signals in initial conditions are recovered to a level similar to that in the observations, intraseasonal forecast skills reach 25 days for 850-hPa zonal winds and 15 days for rainfall over both the global tropics and the Southeast Asia. It is also found that the improved forecasting skills with enhanced TISV in the initial conditions are largely realized by interactive air-sea coupling.