The purpose of this paper is to investigate compression of Doppler weather radar data. The National Weather Services' network of Doppler radars continually generate massive quantities of weather information which have been recently disseminated to users across networks for real-time analysis and long-term archival as part of the CRAFT project. In order to reduce the network bandwidth necessary to transport these data (and the associated communications costs) and reduce long-term storage needs, compression of the data is required.

A comprehensive characterization of frequency distribution of individual fields (reflectivity, Doppler velocity, spectrum width) and combined fields was conducted. Based on this information, several standard compression algorithms were applied (zeroth-order Huffman and arithmetic, UNIX compress) to analyze the resulting compression ratio (CR). It turns out that algorithms based on the Burrows-Wheeler Transformation (BWT) provide a very good compromise between space/time complexity and CR. The Burrows-Wheeler algorithm is a hybrid scheme combining several strategies including move-to-front heuristic, run-length coding and Huffman/Arithmetic on top of BWT. We present CR of NEXRAD radar data for various weather regimes including clear air, squall line, hurricane, and stratiform rain using the Burrows-Wheeler algorithm bzip2. We found that CR for the precipitation events analyzed range between 6:1 to as high as 11:1. CR for the clear air events analyzed range between 10:1 and 25:1. Finally we show how simple radar data preprocessing can improve the above stated CR by 10-25%.