May 22 - July 30
Determining Which Polarimetric Variables are Important for Weather/Non-Weather Discrimination Using Statistical Methods
Samantha Berkseth, Valliappa Lakshmanan, Chris Karstens, and Kiel Ortega
What is already known:
- Non-weather targets detected by weather radar can have a negative effect on the quality of the data being analyzed.
- Quality control algorithms can help to remove these contaminants and make the data easier to interpret.
- The dual polarization upgrade to the WSR-88D network allows for additional radar variables to be analyzed and implemented in these quality control algorithms.
What this study adds:
- The importance of different dual polarization variables are assessed using statistical methods.
- Some statistical methods are more telling than others in revealing how unique a variable is in aiding with quality control.
Abstract:
Weather radar is a useful tool for the meteorologist in examining the atmosphere and determining what types of weather are occurring, how large an area a weather event might cover, and how severe that event might be. It is also widely used for automated applications. However, weather radar can pick up on objects other than just weather, causing the data to become cluttered and harder for forecasters to decipher. Quality control algorithms can help to identify which echoes returning to the radar are meteorological and which are not, and they can then remove such contaminants to create a clearer image for the meteorologist. With the recent widespread upgrade to dual polarization technology for the WSR-88D (Weather Surveillance Radar 1988 Doppler) radars, polarimetric variables can be used in these quality control algorithms, allowing for more aspects of the data to be analyzed and more of the contamination to be removed. This study analyzes those polarimetric variables in order to determine which are the most important for weather/non-weather discrimination. Such research serves to help rank variable importance and prevent the quality control algorithm from being overfit, thus aiding in developing the most efficient algorithm for operational use.