This study investigates the potential predictability of tropical cyclone (TC) activity over the Philippine Area of Responsibility (PAR) on weekly timescales using the NCEP 16-day Global Ensemble Forecast System (NCEP-GEFS). An algorithm that uses information on selected dynamic and thermodynamic criteria was utilized to detect and track TC-like vortices (TCLV) from the 6-hourly NCEP-GEFS model runs covering a two-year period from 1 January to 31 December in 2015 and 2017. A 2 × 2 contingency table was used to summarize the event forecast relative to the observed TC occurrence and subsequent tracks over the PAR on weekly timescales. A forecast hit was declared if there is at least 50% overlap between the polygons created from the drawn circles with 500 km radius centered at the identified forecast TCLV center and the observed TC track. The hindcast period covering the evaluation of NCEP-GEFS indicate a hit-rate of 0.49 and 0.19 for the 1- and 2-week TC forecasts, respectively in the PAR. It is also revealed that the stronger the TC and the farther it developed to the eastern boundary of the PAR, which typically occur during El Niño (as in the 2015 case), the higher chance it could be forecasted one week ahead of time. Furthermore, better TC predictability in the PAR is achieved when the Madden-Julian Oscillation’s active convection phase is located over Africa and the western Indian Ocean (Phase 1) for the Week-1 forecast period and over the Maritime Continent (Phase 5) for the Week-2 forecast period.
Terrestrial, Atmospheric and Oceanic Sciences 32(5), p.741-754