Large-volume Data Compression Using Compressed Sensing for Meteorological Radar

In Japan, severe weather phenomena such as heavy rains and tornados are increasing year by year, and some of them sometimes cause meteorological disasters. In many cases, these are micro-scale phenomena in time and space, which are difficult to detect them with conventional meteorological radars due to their insufficient spatial and temporal resolutions. Therefore, we have been developing meteorological radars with high resolution such as phased array radar (PAR) and Ku-band broadband radar (BBR), and a radar network system consisting of them. By operating this system, further enhancement of the radar performance is able to be realized due to cover unobservable areas complementarily. Thus, precipitation parameters can estimate more accurately. The PAR and the BBR have 100 m and about 5 m spatial resolutions and temporal resolution of 30 sec for volume scan within a radius of 60 km and that of about 1 minute for volume scan within a radius of 20 km, respectively.　These high-resolution radars, however, definitely produce large-volume data. At present, amount of data of the PAR per volume scan (600 ranges, 300 azimuths, 110 elevations) are 520 MB and that of the BBR (8,192 ranges, 120 azimuths, 30 elevations) are 180 MB, respectively, and those data are unacceptable in a current backbone information network. The aim of this study is to provide an information with regard to detection and warning of severe weather accomplished by integration processing of large-volume data (e.g., precipitation attenuation correction, Doppler velocity anti-aliasing or unfolding, calculation of 3-demensional wind field) observed via the radar network system. In order to solve this problem and to achieve our aim, we propose the compression of the large-volume data by using Compressed sensing (CS), which can realize highly efficient data compression for sparse signals. The radar-observed 3-demensional precipitation profile has characteristics to be transformed to sparse or compressible data by using a convenient basis such as Discrete wavelet transform (DWT) and Discrete cosine transform (DCT) since those are spatially correlated. Specifically, a procedure of the applied method is described as follows. The measured data, at first, are transformed to sparse or compressible data by using DWT or DCT, then, the transformed data are compressed by N×M random matrix with a Gaussian distribution of zero mean and unit variance (N