Improvements in the probabilistic prediction of tropical cyclone rapid intensification resulting from inclusion of satellite passive microwave observations

The physical processes responsible for rapid intensification (RI, defined as an increase in the analyzed maximum sustained surface winds by at least 25 kt in 24 h) in tropical cyclones (TCs) are primarily unresolved. The prediction of these events remains one of the most challenging aspects of TC forecasting. Probabilistic RI schemes incorporating environmental data and parameters derived from geostationary satellite infrared imagery have substantially improved objective RI prediction. Infrared data describe structural aspects of the clouds in TCs. Hence, predictors developed from such data improve statistical forecast models. Nonetheless, it is often the case that infrared data cannot detect important structural details of inner-core precipitating structures underneath overlying cirrus outflow clouds.

Unlike infrared data, passive microwave (MW) sensors aboard low-earth orbiting satellites can observe the precipitation-related structure underneath the TC's oft-present cirrus canopy. While the temporal resolution of MW data is considerably lower than the regular observations made from geostationary satellites, ongoing investigations are indicating promise for this data source to provide unique information on TC structure changes that should lead to further improvements in the statistical prediction of RI by including MW-based predictors. In this presentation, we will describe the impacts of including microwave data data in a variety of statistical forecast models.

In this study, storm-centered structural predictors were developed from 19, 37, and 85 GHz brightness temperatures from WindSat, the Advanced Microwave Scanning Radiometer – Earth Observing System (AMSR-E), the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI), the DMSP Special Sensor Microwave/Imagers (SSM/I and SSMI/S), and the Advanced Microwave Sounding Unit B (AMSU-B) data for 1995-2009. Each of these instruments measure similar aspects of precipitation structures throughout the depth of the troposphere, although at varying spatial resolutions. The three different frequencies can each depict and provide unique information on those structures. Therefore, a wide-variety of structure-based predictors is developed from this large dataset of microwave data. These predictors generally describe things like the degree of symmetry, vigor, and distribution of precipitation. One set of predictors is developed through use of an objective ring-search algorithm that attempts to find an annular structure coinciding with the likely location of an incipient or fully developed eyewall. Another set of predictors is obtained from fixed-pattern type structures. These predictors are added to and evaluated in three probabilistic forecast models, which include logistic regression and Bayesian models (Rozoff and Kossin 2011), and the Statistical Hurricane Intensity Prediction Scheme (SHIPS) Rapid Intensification Index (RII) (Kaplan et al. 2010) currently used at the National Hurricane Center.

Analyses of microwave observations indicate that TCs about to undergo RI tend to have more symmetric, vigorous, and tightly clustered convection. Moreover, in a favorable environment, RI is more likely when an eyewall begins to appear in microwave imagery. Independent testing of the logistic regression, Bayesian, and SHIPS-RI models in both the Atlantic and East Pacific Ocean basins shows that adding selected microwave-based predictors improves both the Brier skill score and reliability of forecasts significantly. While there is a degradation of skill in forecasts made at synoptic forecast times (i.e., 0000, 0600, 1200, and 1800 UTC) compared to forecasts made at the instance of a satellite overpass, the microwave predictors still improve the skill of RI prediction at synoptic forecast times in each RI model. Also, predictors from higher resolution sensors (e.g., AMSR-E and TMI) are significantly more beneficial to forecasts than predictors from lower resolution sensors (e.g., SSM/I).