Numerical simulation with radar data assimilation over the Korean Peninsula Seoul National University Ji-Hyun Ha, Gyu-Ho Lim and Dong-Kyou Lee
Introduction The forecast skill associated with warm season rainfall is relatively low, both by absolute standards and relative to predictions of winter season weather systems with strong baroclinicity (Olson et al., 1995; Fritsch et al. 1998). However, with the improving performance of numerical prediction models and increasing computational resources, there is a renwed interest in the predictability of the daily weather, especially at the mesosale (Ehrendorfer 1997; Errico et al. 2002). Several studies have suggested that data assimilation is needed to improve the heavy rainfall prediction and more experimental studies on the assimilation should be conducted (Wee, 1999; Lee and Lee, 2003;Liu et al., 2005; Yu, 2007). Radar data assimilation is a key scientific issue in numerical weather prediction of convective systems for short-range forecasting (Wilson et al., 1998). In recent years considerable progress has been made in the assimilation of radar observations into convective-scale numerical models for heavy rainfall prediction. The objective of this study is to investigate short-range forecasting of the WRF model through the 3DVAR data assimilation of radar data and impact of radar data assimilation for improving the accuracy of heavy rainfall forecast.
Description for radar observations - Radars, which provide observations of radial velocity and reflectivity of hydrometeors, are about 120 km apart on averaged with the observable range for each exceeding 100 km cover the entire the southern Korean Peninsula - A few km spatial resolution and 6-10 min time interval Process of radar data for analysis and data assimilation (a) KAF (5) KMA (12) USAF (2) (b) Preprocessing : noise filtering and dealiasing radial velocity UF (Universal Format) output Interpolation into XYZ by SPRINT -Extract the radial velocity and reflectivity for data assimilation -Synthesis of reflectivity and wind retrieval for analysis (from Park and Lee, 2009) High resolution radial velocity and reflectivity
Heavy rainfall case On 11-12 July 2006, a heavy rainfall event associated with MCSs occurred over the Korean Peninsula. One of the reasons for studying this case is that operational forecasts failed to predict the amount of precipitation. MTSAT Enhanced IR satellite image 2100 UTC 11 2200 UTC 11 2300 UTC 11-35 -45-55 An isolated storm moved eastward while developing quickly from 2200-2300 UTC. The size of the most intensive convective system at 2300 UTC was approximately 2000 km², which corresponded to the meso-ß scale.
12 h accumulated rainfall amount Synoptic environment (2006071118) 850 hpa 1000 hpa divergence Hourly Precipitation (mm) 90 80 70 60 50 40 30 20 10 0 Goyang 1121 1123 1201 1203 1205 1207 1209 11-12 July 2006 (UTC) * Geopotential height (solid line) Equivalent potential temperature (shaded) Wind speed greater than 12.5 m/s (dahsed line) - 12 h accumulated precipitation at Goyang : 335.0 mm - 1 hour maximum rainfall amount : 77.5 mm/h at 2300 UTC 11 July 2006
- Evolution of convective cell Reflectivity Divergence (shaded) and vertical velocity (line) Reflectivity (shaded) and convergence (line) 2150 UTC 2150 UTC 2150 UTC 2200 UTC 2200 UTC M 2200 UTC M2 M1 M2 2210 UTC M3 M2 M1 M1 2210 UTC 2210 UTC M3 M2 M1 2220 UTC A 2230 UTC M1 M2+M3 = 2220 UTC 2220 UTC B M1 M1 2230 UTC 2230 UTC M1 2240 UTC 2240 UTC 2240 UTC 2250 UTC M1 2250 UTC 2250 UTC 2300 UTC A B The propagation of the convective system shows the development of back-building MCS, such as stagnation of the entire convective system oriented in the east-west direction.
Numerical simulations and results Model domain and configuration description Domain 1 (D01) Domain 2 (D02) Horizontal resolution 18 km 6 km Horizontal grid number 170 150 211 211 Vertical layers / Model top Explicit moisture 31 sigma layers / 50 hpa WSM6 Cumulus parameterization scheme Boundary layer Long-wave radiation Short-wave radiation Kain-Fritsch scheme YSU scheme RRTM radiation Dudhia scheme NO Surface physics Thermal diffusion scheme Model initial and boundary data: FNL 1 ⅹ1 data
Radar observations 1km level 1.5km level 3 km level Radar name Wavelength (cm) USAF (2) RKJK, RKSG S band (10 cm) KAF (5) RKWJ, RSCN, RTAG, RWNJ, RYCN C band (5 cm) KMA (12) RGDK, RKWK, RJNI, RKSN, RGSN, RSSP S band (10 cm) Information at the low level is limited due to complex RBRI, RIIA, RPSN, RMYN, RDNH, RCJU topography. C band Thus, (5 cm) we assimilate the surface data for information at the low level.
Data assimilation using WRF 3DVAR Experiment design 11 JUL 12 UTC 15 18 21 12 JUL 00 03 06 09 12 D01 Forecast D02 GTS data assimilation Forecast AWS and/or RADAR data assimilation Experiment name CNTL RADAR+AWS RADAR AWS RV RF Radar3km Radar1.5km Reference Without data assimilation Radial velocity + Reflectivity + Surface data Radial velocity + Reflectivity Surface data Radial velocity Reflectivity 3km horizontal interval 1.5km horizontal interval
Incremental Analysis Update (IAU) Linear balance in the variational system are often insufficient to prevent the development of spurious energy on the fastest time scale of numerical forecast (Polavarapu et al. 2004). A separate filtering procedure is required to remove spurious high-frequency gravity wave noise, which can have a detrimental effect on the first few hours of the forecast, and on the data assimilation cycle as a whole. Thus, we apply the incremental analysis update (IAU) method for data assimilation of the WRF model. By gradually incorporating analysis increments, the IAU method removes high frequencies (Lee et al., 2006). Increments generated by WRF 3DVAR are transformed into tendencies of the model variables (u, v, t and q). 1 p NoiseLevel N t dx() t dt Model variables = F( X) + W( X X ) Original model forcing a b IAU forcing Analysis increment N sp without IAU with IAU assimilation cycle - The inclusion of data cause a fluctuating curve without IAU method. However, the noise is removed by IAU method.
12 h accumulated rainfall OBS CNTL RADAR AWS RADAR+AWS Experiments with data assimilation produce a better precipitation forecast than the experiment without data assimilation. RADAR+AWS has captured well the concentration of the heavy rainfall. Radar data contributes to the pattern of the precipitation, while, surface data improves the intensity of precipitation.
Impact of radar data assimilation for heavy rainfall forecast Time series of the precipitation at the grid point of maximum accumulated 12-h rainfall OBS CNTL RADAR AWS RADAR+AWS Even though there exists phase error, the simulated rainfall in RADAR begins and ends in the early hours of the forecast, but in AWS it begins in the late hours of the forecast and continues up until the final hours. Radar data assimilation contributes to storm development in the early hours of the forecast.
2300 UTC 11 July (reflectivity (shaded) and wind speed (lines)) RADAR+AWS RADAR AWS Rainwater mixing ratio difference (2300UTC) (a) RADAR+AWS - AWS (b) RADAR - AWS 0100 UTC 12 July RADAR+AWS RADAR AWS The rainwater mixing ratio shows a positive difference over the west coast of the central Korean Peninsula, which is consistent with the area of strong reflectivity. Strong reflectivity occurs along the northern edge of the LLJ in RADAR+AWS, RADAR AWS Interactions existinbetween the MCS ratio and LLJ. and These positive differences rainwater mixing seem to cause highly concentrated convection over the west coast of the central Korean Peninsula, and Strong reflectivity in the east-west direction over 40 dbz near the west coast of the central contribute to the development of the convection. Korean Peninsula is simulated by RADAR+AWS and RADAR, but spread out by AWS Radar data, rather than the surface data, contributes to the development of the convective cells in the model.
Impact of horizontal resolution of radar data 12-h accumulated rainfall OBS 1 RADAR Threat scores RADAR3km RADAR1.5km 0.8 0.6 0.4 0.2 0 10 30 50 70 The experiments with high-density RADAR RADAR3km radar data improve RADAR1.5km the 12-h accumulated rainfall amount and distribution compared with the experiment using the low-density (5km in this study). The experiment with 1.5km horizontal interval shows better agreement with the observations in rainfall amount even though the rainfall distribution of RADAR1.5km is slightly shifted northward compared with the observed rainfall.
Summary and conclusion An active MCS produces heavy rainfall over the Korean Peninsula on 11-12 July 2006. In order to predict the heavy rainfall, WRF 3DVAR data assimilation and WRF model are adopted to generate optimal initial and subsequent numerical simulations. In data assimilation, the WRF 3DVAR cycling model with incremental analysis increment is used to remove high-frequency gravity wave. The assimilation of radar data shows better agreement with the observations than without data assimilation in terms of rainfall distribution and amount. The simulation using radar data contributes to the development of convective storms in the early hours of the forecast. In the sensitivity test, radial velocity from the radar data shows larger impact in simulating the heavy rainfall than reflectivity. The experiments with highdensity radar data improve the accumulated rainfall amount and distribution compared with the experiment with low-density radar data.