The combined state-parameter estimation problem and its relevance to tropical cyclone prediction

When a NWP forecast fails, the failure may be imputed to two sources: that due to errors in the initial and boundary conditions and that due to intrinsic deficiencies within the model itself. The latter error may be further subdivided into sources due to discretization, unresolved physics, etc.

Traditional data assimilation strategies, be they variational or ensemble in character, have restricted their solutions to the state vector itself. Here we present the case that a number of “tunable” model parameters (as well as a number not normally considered tunable) may be estimated using an ensemble Kalman filter (EnKF) with the ultimate result that both the state estimates (i.e. analyses) and forecasts obtained from said state estimates are improved relative to the more simplistic state-only method.

To do so, we consider a selection of Atlantic basin hurricanes from the 2005 season. EnKF-based analyses are generated using both the state-only and the combined state-parameter method, assimilating data available from the current constellation of low-earth orbiting (LEO) and geostationary (GEO) satellites. The combined state-parameter analyses in turn serve as truth for an observing system simulation experiment (OSSE) designed to illustrate the impact of future platforms such as Nexrad-in-space (NIS) and GEOSTAR.