Real-time ensemble prediction and data assimilation of tropical cyclones with COAMPS-TC

During the 2011 Atlantic hurricane season the recently developed Coupled Ocean/Atmosphere Mesoscale Prediction System--Tropical Cyclones (COMAPS-TC) was run in real-time with an 80-member ensemble data assimilation system and a 10-member forecast system. The data assimilation and forecasting system was part of a demonstration ensemble system for the Hurricane Forecasting Improvement Project (HFIP). From 1 August to 30 September ensemble data assimilation was performed four times daily with the Data Assimilation Research Testbed (DART) for every tropical cyclone in the Atlantic, Eastern Pacific, and Western Pacific ocean basins. Two times a day, a 10-member subset of the 80-member data assimilation ensemble was forecast to 120-h. Track and intensity forecast were digested by the Automated Tropical Cyclone Forecasting (ATCF) system and were made available to forecasters at the Joint Typhoon Warning Center as well as the Developmental Testbed Center at NCAR. Furthermore, probabilistic graphics from the 10-member ensemble of the storm track, intensity, and structure, as well as plots mean fields were made available to the general public via the World Wide Web. This forecasting and data assimilation system was part of the HFIP stream 2 program and designed to demonstrate the feasibility of real-time ensemble regional prediction systems. In this talk the design and implementation of the real-time ensemble COAMPS-TC system will be discussed. The system used three nested domains with horizontal resolution of 45-15- and 5-km with the 15- and 5-km nests designed to automatically track and follow the storm. Results from the two-month experiment will be presented. With the exception of the data assimilation system, the configuration of the ensemble COAMPS-TC was identical to that of the real-time deterministic COAMPS-TC system. A homogenous comparison of the two systems will be presented. The ensemble-mean track mean absolute error (MAE) was competitive to the MAE deterministic track forecast to 60-hour lead time and as much as 100 km lower beyond 60-h. Despite improvement to the track forecasts, the ensemble forecasts contained a significant weak bias for storm intensity. The source of this bias will be discussed.