Events / Media
Data Assimilation Seminar Series
Dr. Guo-Yuan Lien (May. 18, 2017, 15:30-)
| Affiliation | RIKEN AICS |
|---|---|
| Title | 30-second-cycle convection-resolving data assimilation of dense phased array weather radar data |
| Abstract |
Assimilation of meteorological radar data has been proven useful for short-range numerical weather prediction (NWP) of convective storms. However, to fully utilize the high spatial and temporal resolution radar data in a convection-permitting model is not an easy task. In particular, advanced radar technique, the phased array weather radar (PAWR), can scan 100-m resolution 3-dimensional volumes at a 30-second frequency, resulting in an observation dataset much bigger than that a typical convective scale data assimilation system can use. Therefore, there is an imperative need to explore the use of these high-resolution observations in data assimilation and in the short-range NWP. We develop the SCALE-LETKF system, utilizing the Scalable Computing for Advanced Library and Environment-Regional Model (SCALE-RM) and the Local Ensemble Transform Kalman Filter (LETKF), aiming to conduct convection-resolving ensemble data assimilation of the high-frequency and high-resolution PAWR data. The targeted model resolution ranges from 1 kilometer to 100 meters and the targeted update frequency is every 30 seconds. We carefully design the system to improve the I/O performance and reduce the memory usage, which are critical for our application. In addition, we study on an "implicit localization" method by imposing observation number limit in the LETKF. This method was firstly proposed by Hamrud et al. (2015), and a similar method was used in Schraff et al. (2016). We apply this method with the very dense PAWR data and found that with this method, assimilating only a fraction of the original dense data can lead to even better analyses and forecasts than assimilating all data. The optimal observation number assimilated at a local model grid is roughly proportional to the ensemble size. These results emphasize the importance of the limitation of the observation numbers when using an ensemble data assimilation method. Meanwhile, the method can also significantly reduce the computational time. Therefore, we suggest that the method is a simple and very effective way to perform the LETKF assimilation of dense observation data. |
| Link to PDF file |
