Development of Assimilation Methods for Polarimetric Radar Data Takuya Kawabata1, Hiroshi Yamauchi2, Nobuhiro Nagumo1, Ahoro Adachi1 1: Meteorological Research Institute / JMA 2: Japan Meteorological Agency
Conventional data assimilation system Target : meso-α, β scale Horizontal resolution : 10 km ~ Observation : conventional data, satellite data Large scale convergence by meso-α, β front Updraft by large scale orography
Storm scale data assimilation system Target : meso-γ scale (e.g. cumulonimbi) Horizontal resolution : ~2 km Observation : Remote sensing Small horizontal scale Large temporal variance Large difficulty on forecast Assimilation system, observation High resolution High frequency
Observing cumulonimbus Himawari Cloud Wind GNSS Water vapor Rain Wind Wind Water vapor Wind Lidar H. R. Byers and R. R. Braham. Jr. (1949) Modified by Shoji and Kawabata Radar
NHM-4DVAR (Kawabata et al. 2007; 2011; 2013; 2014a, b) Model Forward model : JMANHM with full physics (JMA operational meso-scale nonhydrostatic model) Adjoint model :Dynamical core Cloud microphysical process (warm rain; w/o parameterization) Observation Doppler radial wind, radar reflectivity, GPS precipitable water vapor, GPS slant delay, zenith delay, Doppler wind lidar, RASS, wind profiler, surface wind, surface temperature, surface pressure Horizontal resolution 2.0 km polarimetric parameters (0.5 km)
Radial Vector (1 min) and GNSS-PWV (5 min) Radar observation Forecast result Reflectivity elevation = 0.7degree reflectivity : dBZ elevation = 0.7degree reflectivity : dBZ Initiation and evolution of the MCS were predicted. Surrounding air flow of the MCS was observed by clear air echoes.
Radar Reflectivity, Radial Vector (1 min), and GNSS-PWV (5 min) Reflectivity 4DVAR with assimilation of radar reflectivity 1st guess 2130 JST Obs 4DVAR without radar reflectivity Radar reflectivity was assimilated added to radial wind and GPS PWV. Forecast of the rainband resulted only for 1 hour. The reason may be why the surrounding air flow was not observed.
GNSS Slant Total Delay (5 min) 1-hr rainfall STD PWV Obs Bck ZTD
Doppler Lidar (1 min) OBS Assimilation First guess 1-hr rainfall A A A
Dual Polarimetric Radar Conventional radar: only horizontally polarized Dual pol. : both horizontally and vertically polarized Reflectivity and phase vary as to oblateness of raindrops! Reflectivity: Zh, Zv, ZDR =(Zh-Zv) Phase: φDP (φh- φv), KDP =(spatial derivation of φDP) Dual Pol. provides information on not only reflectivity but also shapes of rain drops. It is possible to estimate more accurate rainfall intensity with phase observations. Horizontal polarization Vertical polarization Different sectional areas with respect to shapes of rain drops.
Forward Operators for Polarimetric Radars (1) Model Two-moment microphysical scheme (Hashimoto 2008) N0 Intercept parameter Slope parameter µ Spectral shape parameter Predict (3) Space interpolator Prognostic equations for number N and mass concentration q of hydrometeors (2) Two types of variable converters Model variables -> pol. observation - Fitting of scattering amplitude (Zhang et al. 2001) Pol. observation -> model variable Qr (Kdp) (Bringi and Chandrasekar 2001) Beam broadening with Gaussian weights in both horizontal and vertical directions Consider the bending effect Focus on warm rain in this study
Variable converters (1) (model to observation) Oblateness (Brandes et al. 2005) From spherical rain in the model to spheroid 𝑟=0.9951+2.51× 10 −2 𝐷−3.644× 10 −2 𝐷 2 +5.303× 10 −3 𝐷 3 −2.492× 10 −4 𝐷 4 FIT Used by Jung et al. (2009) Drop size D and number of rainwater are provided by the model. Fitting (indirect scattering calculation) (Zhang et al. 2001) 𝑆 ℎ,𝑣 𝐷 = 𝛼 ℎ,𝑣 𝐷 𝛽 ℎ,𝑣 𝑍 𝐻,𝑉 = 4 𝜆 4 𝜋 4 𝐾 𝑤 2 𝛼 ℎ,𝑣 2 𝑁 0 Λ − 2 𝛽 ℎ,𝑣 +1 Γ 2 𝛽 ℎ,𝑣 +1 𝐾 𝐷𝑃 = 180𝜆 𝜋 𝑁 0 𝛼 𝑘 Λ − 𝛽 𝑘 +1 Γ 𝛽 𝑘 +1 𝐴 𝐻 = 𝛼 𝐻 𝐾 𝐷𝑃 𝛽 𝐻 𝐴 𝐷𝑃 = 𝛼 𝑑 𝐾 𝐷𝑃 𝛽 𝑑 α, β are fitting coefficients. Fit from scattering amplitude to a function with number and mixing ratio of rain water
Variable converters (2) (observation to model) Physically based methods Derived theoretically or fitting with scattering calculation Suggested by radar meteorologists (Bringi and Chandrasekar 2001) Used by Li and Mecikalski (2012) Qr= 𝑐 1 𝑍 ℎ 𝑎 1 10 0.1 𝑏 1 𝑍 𝐷𝑅 KD Model Qr= c 2 𝐾 𝐷𝑃 𝑓 𝑏 2 Used by Yokota et al. (2014) and in this study Observation Used by Li and Mecikalski (2013) Qr= c 3 𝐾 𝐷𝑃 𝑎 3 10 0.1 𝑏 3 𝑍 𝐷𝑅 ax, bx, cx are constants Zh, ZDR and KDP are actually observed by polarimetric radars and converted to model variable (Qr).
Horizontal distributions of Pol. parameters (model to obs) FIT OBS Zh ZDR KDP
Comparison with observations (model to obs) Zh (dBZ) ZDR (dB) KDP (° km-1) AVG STD OBS 18.99 11.10 0.93 1.05 1.31 1.54 FIT 19.23 10.37 0.55 0.41 0.25 0.18 FIT - OBS -0.10 14.68 -0.30 0.99 -0.91 1.47 Zh in FIT shows good agreement with the observations. ZDR in fit differs a bit from the observations. KDPs in FIT and the observations are different, due to neither model or observational errors.
Attenuation effect (obs to model) Super cell case Qr(Zh, Zdr) KD: Qr(KDP) Qr(Kdp, Zdr) TRUE Since Zh and ZDR are affected by Ah and ADP, respectively, only KD is not affected by the attenuation. A virtual radar is located at beyond the left-bottom corner. KD is the best
Fraction Skill Score FIT and KD show similar scores in weak intensity region both in wide and narrow verification grids. However, FIT shows better skill in strong intensity region than KD.
Data assimilation with NHM-4DVAR analysis increment (shade) and obs(contour) FIT Zh KDP ZDR KD
Summary A fitting function is employed for direct assimilation of pol. radar data (FIT). A retrieval method only with KDP is useful for DA (KD). FIT seems to be better than KD in intense rain region. Both of FIT and KD are implemented into NHM- 4DVAR and being tested.