Detection of Hazardous Weather Phenomena Using Data Assimilation Techniques.

Currently, the most widely used paradigm for the automated detection of tornadoes and other hazardous weather events involves identifying patterns in “raw” Doppler radar reflectivity and velocity data. The patterns include gate-to-gate shear, strong convergence oriented in lines, descending areas of high reflectivity, etc. The Warning Decision Support System-Integrated Information (WDSS-II), developed by the National Severe Storms Laboratory, uses this general methodology, incorporating near storm environment analyses for hail detection, satellite data for clutter suppression, and a variety of other advanced tools. The major limitation of tools such as WDSS-II, however, is that new detection algorithms must be created, or existing ones adapted, each time a new observation system is deployed (e.g., TDWR). Further, they operate principally on data directly measured by the sensor (e.g., radial velocity and reflectivity) and thus do not make use of other important fields that are potentially available to them (e.g., pressure and temperature). Finally, such systems are limited in their ability to synthesize data from other observing platforms in a dynamically consistent manner.

An alternative approach that has the potential to overcome these limitations involves using advanced data assimilation and retrieval techniques to generate dynamically consistent, 3D gridded analyses of all key observed and unobserved meteorological quantities to which data mining tools can be applied. The potential advantages include the ability to interrogate quantities not available from raw data and the use of geometrically simple 3D grids. The most important advantage, however, is that the mining algorithms do not depend upon the data sources and needn't be changed when new sources are added (e.g., new types of radars). Rather, the incorporation of other data simply improves the quality of the analysis. Potential limitations of this approach include reliability of the retrieved fields and degraded spatial fidelity owing to the resolution constraint of the assimilation process.

To examine these tradeoffs, we compare the conventional detection algorithms of WDSS-II to features in assimilated analyses produced using the ensemble Kalman filter for an observed tornadic storm that occurred on 29 May 2004 and that was observed by NEXRAD radar. Our study includes examination of sensitivities to a variety of variable factors including grid spacing, data frequency, ensemble size, and quantities assimilated.