Development of Nowcasting Method Based on Spatial Scale Analysis of Precipitation Distribution Observed by X-band Polarimetric Radar

In Japan, to reduce disaster damages caused by localized heavy rainfall in urban area, Ministry of Land, Infrastructure, Transport and Tourism (MLIT) provides X-band polarimetric (multi paramater) RAdar Information Network (XRAIN) data experimentally from 2010. Specifications of XRAIN data are as follows. Spatial resolution is 250m, time resolution is 1minute, and operational update interval is 1minute. As for MLIT operational C-band radars, spatial resolution is 1km, and time resolution is 5minutes, and update interval is 5minutes. Thus, XRAIN data has an advantage over conventional C-band radars. Furthermore, XRAIN data has high quality, because multi-parameter radar data, such as KDP and Zdr, is used in estimating rainfall intensity. We have developed the new nowcasting method utilizing XRAIN data to improve accuracy of localized heavy rainfall prediction. Developed nowcasting method is as follows. First, precipitation distribution is separated into small- and large-scale components, corresponding to meso-ã scale convective precipitation and stratiform precipitation through wavelet transformation and inverse reconstruction using Mexican hat wavelet. Small-scale precipitation distribution (SPD) contains only components of less than a threshold length. Threshold length is set to 16km, considering typical spatial scale of meso-ã convective cells. Large-scale precipitation distribution (LPD) is calculated by subtracting SPD from total precipitation distribution. A key of developed method is to separate radar observed precipitation distribution to SPD and LPD. This separation is accomplished only after by using XRAIN data of high resolution in space and time. Next, SPD and LPD are predicted separately. SPD is divided into precipitation cells, and movement vectors of divided cells are identified by pattern matching method using three SPDs (at present, 2minutes before and 4minutes before). Each cell in SPD is moved by the identified vector. In the prediction, growth and decay of each cell is considered using statistical relationship between total precipitation in a cell and lapse time from the cell occurrence. LPD is moved by the vector field. Movement vector fields are identified to minimize RMS of time lagged precipitation filed among the three LPDS (at present, 6minutes before and 12minutes before). In the prediction of LPD, growth and decay of precipitation distribution is not considered. Finally, separately predicted precipitation distributions are composed to gain total precipitation distribution. Prediction lead time of SPD and LPD is 60minutes, and prediction update interval is 1minute. We applied developed nowcasting method to rainfall events occurred in Kinki region in Japan. Compared with the nowcasting method without considering horizontal scale separation, developed method shows better skill score.