LETKF assimilation of high-density radar observations
-- with Takemasa Miyoshi, Juan Ruiz, Masaru Kunii, and the JST/CREST Big Data Assimilation team
Assimilation of meteorological radar data has been proven useful for analyses and short-range forecasts of convective storms. However, the resolution of radar observations can be higher than that one can easily use in data assimilation. In particular, an advanced radar such as the phased array weather radar (PAWR) can scan a 3-dimensional volume at resolution as high as 100 meters in 30 seconds. We explore the assimilation of such high-density PAWR data at their original spatial and temporal resolution using the local ensemble transform Kalman filter (LETKF) (Miyoshi et al. 2016a, b). In the innermost nested model domain, both the reflectivity and radial velocity data are assimilated at model resolution up to 100 m with a 30-s rapid update cycle. We find that 1) setting all reflectivity data below 10 dBZ to a constant 5 dBZ, 2) limiting the observation numbers per grid in the LETKF (initially proposed by Hamrud et al. 2015) contribute to the improvement of the PAWR assimilation results (Lien et al. 2016).
LETKF assimilation of satellite precipitation
-- with Eugenia Kalnay, Takemasa Miyoshi, Shunji Kotsuki, and Daisuke Hotta
Many in-situ and satellite based precipitation observations have been made available; however, assimilating these precipitation data into numerical weather prediction (NWP) models is difficult because of the nonlinearity and non-Gaussianity of the precipitation variable, and large model and observation errors. We propose to use the LETKF to assimilate precipitation observations. The LETKF features the flow-dependent background error covariance, which is able to relate the precipitation variable to other “master” dynamical variables based on the original nonlinear moist physical parameterization in the model. In addition, we also propose two changes in the precipitation assimilation process: a) transform precipitation into a variable with a Gaussian distribution used in the assimilation, and b) only assimilate precipitation at the locations where at least some ensemble members have positive precipitation. With these techniques, we show promising results of precipitation assimilation in both observing system simulation experiments (OSSE) using a simplified model (Lien et al. 2013) and realistic configuration using real data and real global NWP models (Lien et al. 2016a, b: GFS model/TMPA data; Kotsuki et al. 2017: NICAM model/GSMaP data). We also show that the ensemble forecast sensitivity to observation (EFSO) method can be used to accelerate the assimilation development for new observing systems such as the satellite precipitation data (Lien et al. 2018).
Tropical cyclone simulation with controlled tracks and axisymmetric structure based on the EnKF
-- with Chun-Chieh Wu, Yi-Hsuan Huang, and Tzu-Hsiung Yen
The track (i.e., central position) and the axisymmetric wind profile are very useful information to describe a mature tropical cyclone (TC), and they are commonly used in delivering the TC observations. However, they are usually not directly used in numerical weather prediction models. Based on the ensemble Kalman filter, we assimilate these special TC observation parameters into the Weather Research and Forecasting (WRF) model in the same way as other conventional observations (Wu et al. 2010). This method leads to a broad range of applications “softly” controlling the observational (Wu et al. 2012) or hypothetical (Yen et al. 2011) tracks and structure of a TC in numerical simulations while maintaining acceptable balance of model variables.
Development of the SCALE-LETKF data assimilation system
-- with Takemasa Miyoshi, the SCALE model development team, and the JST/CREST Big Data Assimilation team
To facilitate the study of data assimilation with high-resolution regional NWP models, we develop the SCALE-LETKF data assimilation package that couples the LETKF data assimilation module with the Scalable Computing for Advanced Library and Environment (SCALE)-Regional Model (RM). The SCALE-RM is a regional NWP model designed for high-resolution simulation. We carefully design the system to make it able to efficiently conduct rapid-update-cycle data assimilation with big model and observational data (Lien et al. 2016; Liao et al. 2017). The system has been tested on Japan’s flagship supercomputer, the K computer. It is also highly configurable so that it can be conveniently used in a wide range of research. The system has been used for a stable 1-year near-real-time NWP with a broader domain around the Japan region, producing reasonable precipitation forecast for a heavy rainfall case associated with a typhoon (Lien et al. 2017). It is also used for the studies of high-resolution radar assimilation.
The SCALE code and documentation are available at its website.
The SCALE-LETKF code is available at its GitHub page.
Development of the GFS-LETKF data assimilation system
-- with Eugenia Kalnay and Takemasa Miyoshi
We build the GFS-LETKF system that couples the LETKF data assimilation module to the National Centers for Environmental Prediction (NCEP) Global Forecasting System (GFS) model. The system is tested on a Linux cluster at a T62L64 resolution, assimilating real observations from NCEP PREPBUFR data (Lien et al. 2013). It is also used for the precipitation assimilation studies (Lien et al. 2016b).
The GFS-LETKF code is available at the LETKF GitHub page.