Presentation on theme: "Munehiko Yamaguchi Typhoon Research Department, Meteorological Research Institute of the Japan Meteorological Agency 9:00 – 12:00 2011.12.15 (Thr) Topic."— Presentation transcript:
Munehiko Yamaguchi Typhoon Research Department, Meteorological Research Institute of the Japan Meteorological Agency 9:00 – 12: (Thr) Topic No. 2 Initial Condition Sensitivity of Typhoon Track Prediction in the Western North Pacific Tropical Cyclone Ensemble Forecast Nanjing, China
Performance of typhoon track predictions by NWP model (Yamaguchi et al. 2009, MWR) The accuracy of typhoon track predictions has improved steadily over the last few decades Time series of 3-yr running mean of position errors by JMA’s Global Spectral Model from 1997 to 2007 The position error of 5-day forecasts in 2007 is smaller than that of 3-day forecasts in 1997.
Verification for individual cases Position error (km) Sample Number Distance between Beijing and Shanghai Position errors of 3-day predictions by JMA/GSM. Verification period: 3 years from 2008 to 2010.
Some issues to be addressed The accuracy of typhoon track forecasts has steadily improved. Chan (2010, GPTC) Since Chan et al. (2002) paper, research on the physics of general TC motion has been almost non-existent, which suggests that most scientists are quite content with the current theories of TC motion. Present Issues In reality, however, significant errors still exist and there are prediction cases where the position error exceeds 1000 km at 3 days. There are few studies focusing on the cause of prediction errors. (e.g. Carr and Elsberry 2000a, 2000b ）. Approach Obs. Data assimilation NWP User Forecaster Flow of typhoon forecasting Various sources of prediction errors Any approach to separate them to some extent?
Design of numerical experiments JMA’s global spectral model (JMA/GSM) is run from the ECMWF’s initial conditions, which are available through the YOTC (Year of Tropical Convection) dataset, to distinguish TC track prediction errors attributed to initial conditions from those attributed to the NWP model. Initial conditions are thought to be essential for accurate predictions in cases where the prediction is significantly improved by replacing the initial condition. Meanwhile, hints for modifications of the NWP model will be given by cases where the replacement of the initial condition does not help improve the prediction while the prediction by the other NWP system is accurate. Black: Observed track Blue: JMA’s model + JMA’s initial condition Green: ECMWF’s model + ECMWF’s initial condition Red: JMA’s model + ECMWF initial condition
Results Experiment period: to (4 months) Verified TCs: 16 TCs in the west Pac. over the 4 months Forecast time (hours) Position error (km) Number of samples JM-JI: JMA’s model + JMA’s initial condition EM-EI: EC’s model + EC’s initial condition JM-EI: JMA’s model + EC’s initial condition
The difference between JM-JI and EM-EI is similar to that seen in the verification for TCs over 3 years (see figure on the right), that is, EM-EI is better than JM-JI by a lead time of one day. Comparison with verification results over 3 years
Results Replacing the original initial condition of JMA/GSM with the ECMWF analysis reduces the TC track prediction errors by 5 %, 11 %, 9 %, 11 % and 15 % at 1 to 5 days, respectively, and explains 20 %, 29 %, 29 %, 38 % and 68 % of the difference in the errors between JMA and ECMWF at 1 to 5 days, respectively. Forecast time (hours) Position error (km) Number of samples
Individual cases There are prediction cases where the replacement of the initial condition significantly improves the track prediction. Typhoon Dujuan initiated at 12 UTC 5 th Sep Typhoon Lupit initiated at 12 UTC 21 st Oct Black: Observed track Blue: JMA’s model + JMA’s initial condition (JM-JI) Green: ECMWF’s model + ECMWF’s initial condition (EM-EI) Red: JMA’s model + ECMWF initial condition (JM-EI) Orange: JMA’s model + low wavenumber component ( ≤ T42) of ECMWF initial condition + high wavenumber component ( ≥ T42) of JMA initial condition (JM-EI2) Error reduction at 3 day: 595 km to 122 km Error reduction at 3 day: 720 km to 280 km
Ensemble prediction for those cases Typhoon Dujuan initiated at 12 UTC 5 th Sep Typhoon Lupit initiated at 12 UTC 21 st Oct Ensemble track prediction by the JMA Typhoon EPS (TEPS) that deals with initial condition uncertainties based on singular vectors. TEPS captures the scenario of the observed track. It implies that TEPS is successful in expressing the uncertainties of TC track predictions when they are sensitive to initial conditions.
Another individual cases Typhoon Morakot initiated at 12 UTC 4 th Aug Typhoon Parma initiated at 12 UTC 30 th Sep There are prediction cases where the replacement of the initial condition does not help improve the prediction while the ECMWF’s prediction is accurate. Black: Observed track Blue: JMA’s model + JMA’s initial condition (JM-JI) Green: ECMWF’s model + ECMWF’s initial condition (EM-EI) Red: JMA’s model + ECMWF initial condition (JM-EI)
Ensemble prediction for those cases TEPS cannot capture the observed track, either, implying need for modifications of JMA/GSM and/or dealing with model uncertainties in TEPS. Typhoon Morakot initiated at 12 UTC 4 th Aug Typhoon Parma initiated at 12 UTC 30 th Sep. 2009
Northward bias –Typhoon Conson (2010) - Northward bias is not a problem only in JMA but also in other major NWP centers. It is noteworthy that such northward bias tends to appear in the east of Philippines. JMA CMC ECMWF UKMO
Northward bias –Typhoon Nanmadol (2011) - JMAECMWF It would be of great importance to identify the cause of the bias and modify the NWP systems including EPSs for better deterministic and probabilistic forecasts. CMCNCEP
Summary 1.The representation of the steering flow formed by the synoptic environment around the TCs is important for accurate TC track predictions as demonstrated by various previous studies (e.g. Chan and Gray 1982). 2.Ensemble prediction, which deals with initial condition uncertainties, is successful in expressing the uncertainties of TC track predictions when they are sensitive to initial conditions. 3.There are systematic errors in NWP models. The northward bias that tends to appear to the east of the Philippines would be common systematic errors among many NWP models.