Extended Abstract (84K)5B.4
Storm Clustering for Data-driven Weather Forecasting
Xiang Li, Univ. of Alabama, Huntsville, AL; and R. Ramachandran, S. Movva, S. Graves, B. Plale, and N. Vijayakumar
The advances in real-time observation, computational technology and model improvement are driving the paradigm shift of weather forecast from static to dynamic and adaptive forecast. The Linked Environment for Atmospheric Discovery (LEAD) project funded by the National Science Foundation is one of such research efforts. The LEAD goal is to build an infrastructure that allows users to run weather models in response to weather events in timely manner and with high accuracy. One of the key factors to the success of dynamic and adaptive model forecast is the identification of regions of interest for weather phenomena, either from observations or from model forecasts. This process may consist of two components: identifying individual weather phenomenon and grouping individual phenomenon into clusters that indicate regions of interest. The latter is important since severe weather events are normally regional clusters of individual events. Dynamic model forecasts need to be launched at these regional scales to correctly encompass areas of active weather.
We propose a clustering method that automatically groups detected weather events into logical spatial clusters. Individual storm events are detected from the WSR88-D radar measurements using a previously developed storm detection algorithm. We examined two clustering algorithms for their performances for spatial clustering. Two statistical indices were also investigated for their applicability in determining the optimal number of clusters in a storm data set. The results of our investigations will be presented in this paper.
Session 5B, Linked Environments for Atmospheric Discovery (LEAD)
Tuesday, 22 January 2008, 1:30 PM-3:00 PM, 207
Previous paper Next paper
Recorded presentation