Tropical cyclone forecasts for the Western North Pacific with high-resolution atmosphere and coupled atmosphere-ocean models

This work quantified the skills of high-resolution regional nonhydrostatic models in forecasting tropical cyclones (TCs) in the Western North Pacific. The selected cases were almost all TCs during 2012–2014 with an initial time of 1200 UTC. The Japan Meteorological Agency (JMA)-nonhydrostatic model with a horizontal grid spacing of 5 km (NHM5km_atm) and its atmosphere-ocean coupled version (NHM5km_cpl) were used to conduct three-day forecasts. The JMA-global spectral model (GSM) outputs interpolated to a horizontal grid spacing of 0.5 degree were used for initial and lateral boundary conditions of the NHM5km_atm and NHM5km_cpl. The skills and GSM forecast skill were validated with respect to the Regional Specialized Meteorological Center Tokyo best track dataset. Results showed that use of the NHM5km_atm and NHM5km_cpl generally improved track forecasts at forecast times of 24–60 h. Track forecasts improved by as much as 20% for TCs with strong vertical shears of horizontal winds. However, a two-tailed test for the mean value revealed that the improvements were not statistically significant above the 90% confidence level. Use of the NHM5km_atm and NHM5km_cpl significantly improved TC intensity forecasts of 2–3 days by more than 20% with respect to the GSM, but strong TC intensities were not well predicted by short-term forecasts because of initialization deficiencies. Although the NHM5km_cpl tended to seriously underestimate TC intensities, it tended to produce the greatest increase in the correlation coefficient between observed and predicted intensity changes. This study also showed that the method used to determine the TC center position affects the track forecast error by up to a few percent and that the maximum wind speed forecast error depends on the best track dataset selected as a reference.