• Bluestein, H. B., and et al. , 2014: Radar in atmospheric sciences and related research: Current systems, emerging technology, and future needs. Bull. Amer. Meteor. Soc., 95, 18501861, doi:10.1175/BAMS-D-13-00079.1.

    • Search Google Scholar
    • Export Citation
  • Chang, S.-F., , J. Sun, , Y.-C. Liou, , S.-L. Tai, , and C.-Y. Yang, 2014: The influence of erroneous background, beam-blocking and microphysical nonlinearity on the application of a four-dimensional variational Doppler radar data assimilation system for quantitative precipitation forecasts. Meteor. Appl., 21, 444458, doi:10.1002/met.1439.

    • Search Google Scholar
    • Export Citation
  • Chung, K.-S., , I. Zawadzki, , M. K. Yau, , and L. Fillion, 2009: Short-term forecasting of a midlatitude convective storm by the assimilation of single–Doppler radar observations. Mon. Wea. Rev., 137, 41154135, doi:10.1175/2009MWR2731.1.

    • Search Google Scholar
    • Export Citation
  • Crook, N. A., 1994: Numerical simulations initialized with radar-derived winds. Part I: Simulated data experiments. Mon. Wea. Rev., 122, 11891203, doi:10.1175/1520-0493(1994)122<1189:NSIWRD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Crook, N. A., , and J. D. Tuttle, 1994: Numerical simulations initialized with radar-derived winds. Part II: Forecasts of three gust-front cases. Mon. Wea. Rev., 122, 12041217, doi:10.1175/1520-0493(1994)122<1204:NSIWRD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Crook, N. A., , and J. Sun, 2002: Assimilating radar, surface, and profiler data for the Sydney 2000 forecast demonstration project. J. Atmos. Oceanic Technol., 19, 888898, doi:10.1175/1520-0426(2002)019<0888:ARSAPD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Dudhia, J., 1989: Numerical study of convection observed during the Winter Monsoon Experiment using a mesoscale two-dimensional model. J. Atmos. Sci., 46, 30773107, doi:10.1175/1520-0469(1989)046<3077:NSOCOD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Gal-Chen, T., 1978: A method for the initialization of the anelastic equations: Implications for matching models with observations. Mon. Wea. Rev., 106, 587606, doi:10.1175/1520-0493(1978)106<0587:AMFTIO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Gao, J., , and D. J. Stensrud, 2012: Assimilation of reflectivity data in a convective-scale, cycled 3DVAR framework with hydrometeor classification. J. Atmos. Sci., 69, 10541065, doi:10.1175/JAS-D-11-0162.1.

    • Search Google Scholar
    • Export Citation
  • Gao, J., , and D. J. Stensrud, 2014: Some observing system simulation experiments with a hybrid 3DEnVAR system for storm-scale radar data assimilation. Mon. Wea. Rev., 142, 33263346, doi:10.1175/MWR-D-14-00025.1.

    • Search Google Scholar
    • Export Citation
  • Hayden, C. M., , and R. J. Purser, 1995: Recursive filter objective analysis of meteorological fields: Applications to NESDIS operational processing. J. Appl. Meteor., 34, 315, doi:10.1175/1520-0450-34.1.3.

    • Search Google Scholar
    • Export Citation
  • Hong, S.-Y., , J. Dudhia, , and S.-H. Chen, 2004: A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation. Mon. Wea. Rev., 132, 103120, doi:10.1175/1520-0493(2004)132<0103:ARATIM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Houze, R. A., Jr., , M. I. Biggerstaff, , S. A. Rutledge, , and B. F. Smull, 1989: Interpretation of Doppler weather radar displays of midlatitude mesoscale convective systems. Bull. Amer. Meteor. Soc., 70, 608619, doi:10.1175/1520-0477(1989)070<0608:IODWRD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Hu, M., , M. Xue, , J. Gao, , and K. Brewster, 2006a: 3DVAR and cloud analysis with WSR-88D level-II data for the prediction of the Fort Worth, Texas, tornadic thunderstorms. Part I: Cloud analysis and its impact. Mon. Wea. Rev., 134, 675698, doi:10.1175/MWR3092.1.

    • Search Google Scholar
    • Export Citation
  • Hu, M., , M. Xue, , J. Gao, , and K. Brewster, 2006b: 3DVAR and cloud analysis with WSR-88D level-II data for the prediction of the Fort Worth, Texas, tornadic thunderstorms. Part II: Impact of radial velocity analysis via 3DVAR. Mon. Wea. Rev., 134, 699721, doi:10.1175/MWR3093.1.

    • Search Google Scholar
    • Export Citation
  • Jung, Y., , G. Zhang, , and M. Xue, 2008a: Assimilation of simulated polarimetric radar data for a convective storm using the ensemble Kalman filter. Part I: Observation operators for reflectivity and polarimetric variables. Mon. Wea. Rev., 136, 22282245, doi:10.1175/2007MWR2083.1.

    • Search Google Scholar
    • Export Citation
  • Jung, Y., , M. Xue, , G. Zhang, , and J. M. Straka, 2008b: Assimilation of simulated polarimetric radar data for a convective storm using the ensemble Kalman filter. Part II: Impact of polarimetric data on storm analysis. Mon. Wea. Rev., 136, 22462260, doi:10.1175/2007MWR2288.1.

    • Search Google Scholar
    • Export Citation
  • Kawabata, T., , T. Kuroda, , H. Seko, , and K. Saito, 2011: A cloud-resolving 4DVAR assimilation experiment for a local heavy rainfall event in the Tokyo metropolitan area. Mon. Wea. Rev., 139, 19111931, doi:10.1175/2011MWR3428.1.

    • Search Google Scholar
    • Export Citation
  • Kessler, E., 1969: On the Distribution and Continuity of Water Substance in Atmospheric Circulation. Meteor. Monogr., No. 32, Amer. Meteor. Soc., 84 pp.

    • Search Google Scholar
    • Export Citation
  • Klemp, J. B., , and R. B. Wilhelmson, 1978: The simulation of three-dimensional convective storm dynamics. J. Atmos. Sci., 35, 10701096, doi:10.1175/1520-0469(1978)035<1070:TSOTDC>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Li, Y., , X. Wang, , and M. Xue, 2012: Assimilation of radar radial velocity data with the WRF hybrid ensemble–3DVAR system for the prediction of Hurricane Ike (2008). Mon. Wea. Rev., 140, 35073524, doi:10.1175/MWR-D-12-00043.1.

    • Search Google Scholar
    • Export Citation
  • Lin, Y., , S. R. Peter, , and W. J. Kenneth, 1993: Initialization of a modeled convective storm using Doppler radar–derived fields. Mon. Wea. Rev., 121, 27572775, doi:10.1175/1520-0493(1993)121<2757:IOAMCS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Liou, Y.-C., , J.-L. Chiou, , W.-H. Chen, , and H.-Y. Yu, 2014: Improving the model convective storm quantitative precipitation nowcasting by assimilating state variables retrieved from multiple-Doppler radar observations. Mon. Wea. Rev., 142, 40174035, doi:10.1175/MWR-D-13-00315.1.

    • Search Google Scholar
    • Export Citation
  • Miller, M. J., , and R. P. Pearce, 1974: A three-dimensional primitive equation model of cumulonimbus convection. Quart. J. Roy. Meteor. Soc., 100, 133154, doi:10.1002/qj.49710042402.

    • Search Google Scholar
    • Export Citation
  • Navon, I. M., , X. Zou, , J. Derber, , and J. Sela, 1992: Variational data assimilation with an adiabatic version of the NMC spectral model. Mon. Wea. Rev., 120, 14331446, doi:10.1175/1520-0493(1992)120<1433:VDAWAA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Pan, X., , X. Tian, , X. Li, , Z. Xie, , A. Shao, , and C. Lu, 2012: Assimilating Doppler radar radial velocity and reflectivity observations in the weather research and forecasting model by a proper orthogonal-decomposition-based ensemble, three-dimensional variational assimilation method. J. Geophys. Res., 117, D17113, doi:10.1029/2012JD017684.

    • Search Google Scholar
    • Export Citation
  • Rogers, E., , T. L. Black, , D. G. Deaver, , G. J. DiMego, , Q. Zhao, , M. Baldwin, , N. W. Junker, , and Y. Lin, 1996: Changes to the operational “early” Eta analysis/forecast system at the National Centers for Environmental Prediction. Wea. Forecasting, 11, 391412, doi:10.1175/1520-0434(1996)011<0391:CTTOEA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Schaefer, J. T., 1990: The critical success index as an indicator of warning skill. Wea. Forecasting, 5, 570575, doi:10.1175/1520-0434(1990)005<0570:TCSIAA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Snyder, C., , and F. Zhang, 2003: Assimilation of simulated Doppler radar observations with an ensemble Kalman filter. Mon. Wea. Rev., 131, 16631677, doi:10.1175//2555.1.

    • Search Google Scholar
    • Export Citation
  • Sun, J., 2005a: Initialization and numerical forecasting of a supercell storm observed during STEPS. Mon. Wea. Rev., 133, 793813, doi:10.1175/MWR2887.1.

    • Search Google Scholar
    • Export Citation
  • Sun, J., 2005b: Convective-scale assimilation of radar data: Progress and challenges. Quart. J. Roy. Meteor. Soc., 131, 34393463, doi:10.1256/qj.05.149.

    • Search Google Scholar
    • Export Citation
  • Sun, J., , and N. A. Crook, 1997: Dynamical and microphysical retrieval from Doppler radar observations using a cloud model and its adjoint. Part I: Model development and simulated data experiments. J. Atmos. Sci., 54, 16421661, doi:10.1175/1520-0469(1997)054<1642:DAMRFD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Sun, J., , and N. A. Crook, 2001: Real-time low-level wind and temperature analysis using single WSR-88D data. Wea. Forecasting, 16, 117132, doi:10.1175/1520-0434(2001)016<0117:RTLLWA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Sun, J., , and Y. Zhang, 2008: Analysis and prediction of a squall line observed during IHOP using multiple WSR-88D observations. Mon. Wea. Rev., 136, 23642388, doi:10.1175/2007MWR2205.1.

    • Search Google Scholar
    • Export Citation
  • Sun, J., , and H. Wang, 2013: Radar data assimilation with WRF 4D-Var: Part II. Comparison with 3D-Var for a squall line over the U.S. Great Plains. Mon. Wea. Rev., 141, 22452264, doi:10.1175/MWR-D-12-00169.1.

    • Search Google Scholar
    • Export Citation
  • Sun, J., , M. Chen, , and Y. Wang, 2010: Frequent-updating analysis system based on radar, surface, and mesoscale model data for the Beijing 2008 Forecast Demonstration Project. Wea. Forecasting, 25, 17151735, doi:10.1175/2010WAF2222336.1.

    • Search Google Scholar
    • Export Citation
  • Tai, S.-L., , Y.-C. Liou, , J. Sun, , S.-F. Chang, , and M.-C. Kuo, 2011: Precipitation forecasting using Doppler radar data, a cloud model with adjoint, and the Weather Research and Forecasting Model: Real case studies during SoWMEX in Taiwan. Wea. Forecasting, 26, 975992, doi:10.1175/WAF-D-11-00019.1.

    • Search Google Scholar
    • Export Citation
  • Tong, M., , and M. Xue, 2005: Ensemble Kalman filter assimilation of Doppler radar data with a compressible nonhydrostatic model: OSS experiments. Mon. Wea. Rev., 133, 17891807, doi:10.1175/MWR2898.1.

    • Search Google Scholar
    • Export Citation
  • Tripoli, G. J., , and W. R. Cotton, 1981: The use of ice–liquid water potential temperature as a thermodynamic variable in deep atmospheric models. Mon. Wea. Rev., 109, 10941102, doi:10.1175/1520-0493(1981)109<1094:TUOLLW>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Tsai, C.-C., , S.-C. Yang, , and Y.-C. Liou, 2014: Improving quantitative precipitation nowcasting with a local ensemble transform Kalman filter radar data assimilation system: Observing system simulation experiments. Tellus, 66A, 21 804, doi:10.3402/tellusa.v66.21804.

    • Search Google Scholar
    • Export Citation
  • Wang, H., , J. Sun, , X. Zhang, , X. Huang, , and T. Auligne, 2013: Radar data assimilation with WRF 4D-Var. Part I: System development and preliminary testing. Mon. Wea. Rev., 141, 22242244, doi:10.1175/MWR-D-12-00168.1.

    • Search Google Scholar
    • Export Citation
  • Weisman, M. L., , and J. B. Klemp, 1982: The dependence of numerically simulated convective storms on vertical wind shear and buoyancy. Mon. Wea. Rev., 110, 504520, doi:10.1175/1520-0493(1982)110<0504:TDONSC>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Weygandt, S. S., , A. Shapiro, , and K. K. Droegemeier, 2002: Retrieval of model initial fields from single-Doppler observations of a supercell thunderstorm. Part II: Thermodynamic retrieval and numerical prediction. Mon. Wea. Rev., 130, 454476, doi:10.1175/1520-0493(2002)130<0454:ROMIFF>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Wu, B., , J. Verlinde, , and J. Sun, 2000: Dynamical and microphysical retrievals from Doppler radar observations of a deep convective cloud. J. Atmos. Sci., 57, 262283, doi:10.1175/1520-0469(2000)057<0262:DAMRFD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., , and J. Sun, 2007: Multiple-radar data assimilation and short-range quantitative precipitation forecasting of a squall line observed during IHOP_2002. Mon. Wea. Rev., 135, 33813404, doi:10.1175/MWR3471.1.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., , Y.-H. Kuo, , J. Sun, , W.-C. Lee, , E. Lim, , Y. Guo, , and D. M. Barker, 2005: Assimilation of Doppler radar observations with a regional 3DVAR system: Impact of Doppler velocities on forecasts of a heavy rainfall case. J. Appl. Meteor., 44, 768788, doi:10.1175/JAM2248.1.

    • Search Google Scholar
    • Export Citation
  • Yuter, S. E., , and R. A. Houze Jr., 1995: Three-dimensional kinematic and microphysical evolution of Florida cumulonimbus. Part II: Frequency distributions of vertical velocity, reflectivity, and differential reflectivity. Mon. Wea. Rev., 123, 19411963, doi:10.1175/1520-0493(1995)123<1941:TDKAME>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Zhang, F., , Y. Weng, , J. A. Sippel, , Z. Meng, , and C. H. Bishop, 2009: Cloud-resolving hurricane initialization and prediction through assimilation of Doppler radar observations with an ensemble Kalman filter. Mon. Wea. Rev., 137, 21052125, doi:10.1175/2009MWR2645.1.

    • Search Google Scholar
    • Export Citation
  • Zhao, Q., , J. Cook, , Q. Xu, , and P. R. Harasti, 2006: Using radar wind observations to improve mesoscale numerical weather prediction. Wea. Forecasting, 21, 502522, doi:10.1175/WAF936.1.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Plot of function Φ(α) defined by Eq. (11). Test 1 uses the modified ice process with a lower threshold for the mixing ratio of snow qs. Test 2 and test 3 represent the tests in which the original scheme is applied without setting a lower threshold for the accretion and deposition of cloud ice, respectively.

  • View in gallery

    Plots representing the (left) WARM_RUN and (right) ICE_RUN. (a),(b) The precipitation mixing ratio (shading, g kg−1) and flow vectors (m s−1 at the bottom level (250 m) at simulation time t = 11 400 s. Vectors are plotted every 9 km. Several fields averaged along the y axis are depicted for (c),(d) precipitation mixing ratio (g kg−1); (e),(f) temperature perturbation (colored, °C) and vertical velocity (black contours at 2.5, 1.5, 0.2, −0.2, −0.6, and −1.0 m s−1); and (g),(h) water vapor perturbation (g kg−1). (i) The difference between (h) and (g) calculated by subtracting WARM_RUN from ICE_RUN. The positions of two pseudoradars for OSSEs are marked by solid circles in (b). The solid lines in (c) and (d) represent the heights of the freezing level.

  • View in gallery

    Illustration of VDRAS assimilation and forecast cycles for the OSSEs. The down-pointing arrows labeled T1–T6 at the top denote the assimilation times for both radar1 and radar2. The numbers at the top show the starting and finishing times for each 4DVar cycle with respect to the simulation time for the nature run.

  • View in gallery

    (a) The “true” precipitation mixing ratio (colored, g kg−1) in a vertical cross section through X = 150 km from the nature run at a simulation time of t = 11 400 s. The solid black line indicates the height of the freezing level (0°C). (b) The “true” reflectivity (dBZ) derived from the precipitation mixing ratios qr and qs shown in (a).

  • View in gallery

    The precipitation mixing ratio (g kg−1) converted from radar reflectivity using Eqs. (6) and (8) along the vertical cross section X = 150 km at t = 11 400 s. The conversion is based on the microphysical process assumption used in experiments (a) WARM, (b) ICE_NR, and (c) ICE. The dashed black lines on these cross sections indicate the heights of the refereed freezing level. (d) The shading is the precipitation mixing ratio from experiment ICE (Fig. 5c) minus the “true” one from the nature run (ICE_RUN; Fig. 4a). The solid and dashed lines in (d) are the freezing levels from the nature run ICE_RUN and experiment ICE, respectively.

  • View in gallery

    The 4DVar-retrieved precipitation mixing ratio analysis field (g kg−1) over the vertical cross section through X = 150 km at t = 11 400 s from (a) WARM, (b) ICE_NR, and (c) ICE.

  • View in gallery

    Vertical cross sections through X = 150 km for the temperature perturbation (color shading, °C) and vertical velocity (contours, m s−1) at t = 11 400 s from the 4DVar-retrieved analysis field from (a) WARM, (b) ICE_NR, and (c) ICE. (d) As in (a)–(c), but from the simulation of the nature run (ICE_RUN).

  • View in gallery

    The RMSEs for (a) precipitation mixing ratio (g kg−1) and (b) temperature perturbation (°C) computed over the whole domain between the nature run and three OSSE experimental analyses at t = 11 400 s.

  • View in gallery

    The RMSEs for predicted 1-, 2-, and 3-h accumulated rainfall (mm) for WARM (red), ICE_NR (black), ICE (blue), and NODA_ICE (yellow).

  • View in gallery

    Locations of data collected from different sources for SoWMEX IOP 8. The small dots indicate selected data points from the ECMWF reanalysis, triangles stand for radars, crosses represent surface mesonet stations, and black squares denote radiosondes. This is also the experimental domain for the VDRAS.

  • View in gallery

    The composite maximum radar reflectivity (color shading at 25, 30, 35, 40, 45, 50, and 55 dBZ) at (a) 1302, (b) 1332, (c) 1402, and (d) 1432 UTC 14 Jun 2008 for IOP 8. The jagged solid black line represents the Taiwanese coast line.

  • View in gallery

    Illustration of assimilation and forecast cycles for the R_WARM and R_ICE experiments. The numbers at the top of the diagram indicate the starting and finishing times (UTC) of each 4DVar cycle. RCKT and RCCG stand for the two CWB radars.

  • View in gallery

    The CFADs (color shading at 0.5%, 5%, 10%, 15%, and 20% dBZ−1 km−1) of reflectivity at 1302 UTC for (a) observation and experiments (b) R_WARM and (c) R_ICE. The bin size is 5 dBZ. Only those grid points where Z > 5 dBZ are counted. The CFADs in (a),(b), and (c) are truncated above the height of 10.75, 6.25, and 9.25 km, respectively, because of insufficient data points.

  • View in gallery

    (a) The updrafts and downdrafts averaged over the entire domain. Solid (dashed) line stands for updrafts (downdrafts), and red and blue lines represent results from R_WARM and R_ICE, respectively. (b) As in (a), but the average is performed only over those points containing the top 1% of the updrafts and downdrafts.

  • View in gallery

    The predicted maximum radar reflectivity (color shading at 25, 30, 35, 40, 45, 50, and 55 dBZ) at (a) 1302, (b) 1332, (c) 1402, and (d) 1432 UTC from experiment R_WARM. The time corresponds to (a) the initial condition and a forecast of (b) 0.5, (c) 1.0, and (d) 1.5 h. The jagged solid black line represents the Taiwanese coast line.

  • View in gallery

    As in Fig. 15, but for experiment R_ICE.

  • View in gallery

    The 2-h accumulated rainfall amount (color shading, mm) from (a) rain gauge observations, (b) experiment R_WARM, and (c) experiment R_ICE. The thick solid lines depict the terrain heights of 500, 1500, and 2500 m.

  • View in gallery

    (a) ETSs and (b) RMSEs of the forecast 2-h rainfall accumulations over the island of Taiwan from experiments R_WARM and R_ICE at different precipitation thresholds.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 117 117 11
PDF Downloads 79 79 12

The Implementation of the Ice-Phase Microphysical Process into a Four-Dimensional Variational Doppler Radar Analysis System (VDRAS) and Its Impact on Parameter Retrieval and Quantitative Precipitation Nowcasting

View More View Less
  • 1 Department of Atmospheric Sciences, National Central University, Taoyuan City, Taiwan
  • | 2 Department of Atmospheric Sciences, National Central University, Taoyuan City, and Taiwan Typhoon and Flood Research Institute, Taipei, Taiwan
  • | 3 National Center for Atmospheric Research, Boulder, Colorado
  • | 4 Department of Atmospheric Sciences, National Central University, Taoyuan City, Taiwan
© Get Permissions
Full access

Abstract

The microphysical process of a cloud-scale model used by a four-dimensional Variational Doppler Radar Analysis System (VDRAS) is extended from its original warm rain parameterization scheme to a cold rain process containing ice and snow. The development of the adjoint equations for the additional control variables related to ice physics is accomplished by utilizing the existing four-dimensional variational (4DVar) minimization framework employed by VDRAS. Experiments are conducted to examine the accuracy of the new 4DVar system with the ice physics scheme implemented and to explore the impact of the ice-phase process on numerical simulations, parameter retrievals, and the model’s quantitative precipitation nowcasting (QPN) capability.

It is shown that the ice-phase microphysical process can significantly alter the kinematic and thermodynamic structure of deep convection and provide a better description of the contents of the hydrometeors. During the 4DVar minimization, using the VDRAS-predicted freezing level after the previous assimilation cycle to replace the true but unknown 0°C line is found to be a feasible approach for separating the rain and snow and, at the same time, allowing the 4DVar minimization algorithm to converge to an optimal solution. A real case study from intensive observation period 8 of the 2008 Southwest Monsoon Experiment shows that, with the added ice-phase process, VDRAS is more capable of capturing the actual evolution of the reflectivity field than the original scheme. The model’s QPN skill is also improved significantly. Thus, the benefits of adding the ice-phase process into a 4DVar radar data assimilation system on the convective-scale weather analysis and forecast are demonstrated.

Corresponding author address: Dr. Yu-Chieng Liou, Department of Atmospheric Sciences, National Central University, No. 300 Jhongda Rd., 320 Jhongli, Taoyuan City, Taiwan. E-mail: tyliou@atm.ncu.edu.tw

Abstract

The microphysical process of a cloud-scale model used by a four-dimensional Variational Doppler Radar Analysis System (VDRAS) is extended from its original warm rain parameterization scheme to a cold rain process containing ice and snow. The development of the adjoint equations for the additional control variables related to ice physics is accomplished by utilizing the existing four-dimensional variational (4DVar) minimization framework employed by VDRAS. Experiments are conducted to examine the accuracy of the new 4DVar system with the ice physics scheme implemented and to explore the impact of the ice-phase process on numerical simulations, parameter retrievals, and the model’s quantitative precipitation nowcasting (QPN) capability.

It is shown that the ice-phase microphysical process can significantly alter the kinematic and thermodynamic structure of deep convection and provide a better description of the contents of the hydrometeors. During the 4DVar minimization, using the VDRAS-predicted freezing level after the previous assimilation cycle to replace the true but unknown 0°C line is found to be a feasible approach for separating the rain and snow and, at the same time, allowing the 4DVar minimization algorithm to converge to an optimal solution. A real case study from intensive observation period 8 of the 2008 Southwest Monsoon Experiment shows that, with the added ice-phase process, VDRAS is more capable of capturing the actual evolution of the reflectivity field than the original scheme. The model’s QPN skill is also improved significantly. Thus, the benefits of adding the ice-phase process into a 4DVar radar data assimilation system on the convective-scale weather analysis and forecast are demonstrated.

Corresponding author address: Dr. Yu-Chieng Liou, Department of Atmospheric Sciences, National Central University, No. 300 Jhongda Rd., 320 Jhongli, Taoyuan City, Taiwan. E-mail: tyliou@atm.ncu.edu.tw

1. Introduction

A Doppler radar is capable of collecting observational data, including radial velocity and reflectivity with high temporal and spatial resolutions, and is thus a powerful instrument for monitoring and studying severe weather systems (Houze et al. 1989; Bluestein et al. 2014). A variety of methods have been developed to utilize radar data to provide proper initial conditions needed for initializing numerical weather prediction models, with the purpose of reducing the model spinup time and improving the forecast accuracy, especially at convective scales. Gal-Chen (1978) first tried to infer the unobserved thermodynamic parameters (i.e., pressure and temperature) from the radar-derived three-dimensional winds. Since then, many studies have been carried out utilizing this concept to initialize a numerical model based upon retrieved state variables from radar measurements (e.g., Lin et al. 1993; Crook 1994; Crook and Tuttle 1994; Weygandt et al. 2002; Zhao et al. 2006; Liou et al. 2014).

The three-dimensional variational data assimilation (3DVar) technique, because of its efficiency and stability, has been widely adopted to assimilate Doppler radar data for severe weather studies (e.g., Xiao et al. 2005; Hu et al. 2006a,b; Xiao and Sun 2007; Chung et al. 2009; Gao and Stensrud 2012). The ensemble Kalman filter (EnKF) technique has also been used to assimilate Doppler and/or dual-polarimetric radar observations in numerous studies (e.g., Snyder and Zhang 2003; Tong and Xue 2005; Jung et al. 2008a,b; Zhang et al. 2009; Tsai et al. 2014). A distinguishing feature of the EnKF is its capability to provide flow-dependent background error covariance using ensemble forecast statistics. Recently, active efforts have been made to develop hybrid techniques by combining the EnKF with the variational approaches for the assimilation of radar data (e.g., Li et al. 2012; Pan et al. 2012; Gao and Stensrud 2014).

Compared to 3DVar, the four-dimensional variational (4DVar) adjoint data assimilation technique is a more sophisticated algorithm, which considers the model trajectory over one or multiple assimilation windows. Examples of 4DVar-based radar assimilation systems include the Variational Doppler Radar Analysis System (VDRAS) by Sun and Crook (1997, hereafter SC97), the Japan Meteorological Agency’s 4DVar (Kawabata et al. 2011), and the Weather Research and Forecasting (WRF) Model 4DVar (Sun and Wang 2013; Wang et al. 2013). VDRAS is distinguished from the others by its ability to perform rapid update cycles with the frequency of less than 15 min (Sun and Crook 2001; Sun and Zhang 2008; Sun et al. 2010; Chang et al. 2014), which is very crucial for nowcasting applications.

VDRAS was developed at the National Center for Atmospheric Research (NCAR). It uses a cloud-resolving model as a forward forecast model. Using this prognostic model as a constraint and its adjoint model during the 4DVar minimization procedure, VDRAS assimilates radar observations with short assimilation windows and therefore is particularly suitable for studying convective-scale weather phenomena. The 4DVar radar data assimilation enables the retrieval of the three-dimensional winds, unobserved thermodynamic fields, microphysical parameters, and other model variables.

VDRAS has been successfully applied to demonstrate its potential for the short-term forecasting of severe storms in a number of field projects, such as the Sydney 2000 Forecast Demonstration Project (Crook and Sun 2002), Severe Thunderstorm Electrification and Precipitation Study (STEPS; Sun 2005a), International H2O Project (IHOP_2002; Sun and Zhang 2008), Beijing 2008 Forecast Demonstration Project (Sun et al. 2010), and 2008 Southwest Monsoon Experiment (SoWMEX; Tai et al. 2011; Chang et al. 2014). However, the microphysical scheme used by VDRAS in the aforementioned studies considered only the warm rain processes. To remove this limitation, Wu et al. (2000) added one additional prognostic equation for graupel–hail in their study of a summertime hail-producing convective storm. They demonstrated that, using an idealized observing system simulation experiment (OSSE), the distribution of graupel/hail can be reasonably retrieved. However, when applied to the forecast of the real case, their system failed to capture the evolution of the actual storms, indicating that the model was not able to describe the true evolution of the cloud physics in the atmosphere.

The main purpose of this research is to develop the 4DVar adjoint for an ice microphysical scheme and explore whether there are positive impacts of the ice physics on the convective-scale analyses and forecasts. Since the maintenance and computational costs of a full 4DVar system are high, in this study the ice-phase microphysical process implemented into VDRAS, different from that of Wu et al. (2000), is a simple ice scheme. The development of the additional adjoint equations for ice physics fully takes advantage of the original 4DVar minimization framework already adopted by VDRAS. Both OSSEs and a real case study are conducted to examine the accuracy of the newly implemented cold rain process and to investigate the impact of adding the ice-phase microphysics on the retrieval of the state variables and short-term quantitative precipitation nowcasting (QPN) capability of VDRAS.

It is demonstrated in this research that the ice-phase microphysical process can provide a better description of the contents of the hydrometeors and enable VDRAS to capture the actual evolution of the reflectivity field better than the original scheme. The model’s QPN skill can be improved significantly. A feasible approach is also proposed to determine the true but unknown 0°C line for separating the rain and snow while, at the same time, allowing the 4DVar minimization algorithm to converge to an optimal solution.

This paper is organized as follows. Section 2 describes the VDRAS system and the method of implementing the ice-phase microphysical process. Section 3 introduces the verification of the adjoint model, followed by the definitions of the verification indices in section 4. The differences between the forward simulations with and without the ice-phase microphysics are demonstrated in section 5, while the OSSE results from the 4DVar experiments are examined in section 6. A real case study using the data collected during the intensive observation period (IOP) 8 of the 2008 SoWMEX is presented in section 7. The summary and conclusions are offered in section 8. The appendixes include the microphysical scheme and the derivatives needed for coding the adjoint equations of the ice process.

2. Description of VDRAS

VDRAS is composed of four major components. They are data acquisition, data preprocessing, 4DVar assimilation–minimization, and model outputs (Sun and Zhang 2008; Sun et al. 2010). The 4DVar radar assimilation system in VDRAS includes a forward cloud-resolving numerical model, the adjoint of the numerical model, a cost function, and a minimization algorithm. The numerical model is formulated within a Cartesian coordinate system with a flat surface. There are six prognostic variables in the original VDRAS forward model: wind velocity components (u, υ, w), rainwater mixing ratio (), total water mixing ratio (), and liquid water potential temperature [, defined in Eq. (A7)], as suggested in Tripoli and Cotton (1981). The total water mixing ratio is the sum of rainwater, cloud water, and water vapor [Eq. (A2)]. The temperature and the cloud water mixing ratio are diagnosed from the prognostic variables by assuming that all vapor greater than the saturation value is converted to cloud water. The perturbation pressure is diagnosed by solving the Poisson equation:
e1
where is the velocity vector, is the basic-state density of air, and are perturbation of temperature and water vapor mixing ratio, and denote the cloud water and rainwater mixing ratio, respectively.

The 4DVar technique is known to be very computationally expensive. Thus, saving the computing time is often one of the top considerations in determining which physical scheme to use. This is particularly important for VDRAS since one of its applications is to provide rapidly updated atmospheric fields for diagnoses and forecasts. Under this condition, the reason to choose the simple ice scheme in this research is to keep the number of control variables as small as possible so that a fast convergence in the 4DVar minimization procedure can still be achieved without requesting extra computational resources.

The modification of the VDRAS microphysical process is an extension of that reported by SC97. The liquid-phase microphysics is kept the same as the original scheme documented in SC97. The additional ice-phase microphysics is based on Dudhia (1989, hereafter D89) and Hong et al. (2004). This ice-phase microphysics scheme assumes no supercooled cloud water or rain below 0°C and no superwarmed snow or ice crystals above 0°C. Therefore, the snow mixing ratio is stored in the same array that is used for storing the rainwater mixing ratio . Similarly, the mixing ratio of cloud ice shares the same array with the cloud water mixing ratio . The hydrometeor type is unambiguously determined at each grid box by the 0°C freezing temperature. In addition, the counterpart of the liquid water potential temperature becomes the ice potential temperature [as defined in Eq. (A8)]. The detailed description of the new microphysics parameterization scheme is described in appendix A. It should be emphasized that this research represents a new trial, as the microphysics scheme proposed by D89 and Hong et al. (2004) had not been tested in a 4DVar system before.

The application of the 4DVar technique in VDRAS allows the use of prognostic equations as constraints to minimize the cost function. VDRAS assimilates the Doppler radar data and then finds an optimal initial condition through the minimization algorithm. The cost function J for measuring the misfit between the model state variables and the radar observations can be written as
e2
where is the model state variables at the beginning of the current assimilation window, is the background field forecasted from the previous cycle, and denotes the background error covariance matrix. According to Sun and Crook (2001) and Sun and Zhang (2008), the parameters in the analysis field are correlated in 4DVar through the dynamic model; therefore, the background error covariance can be approximated by a relatively simple correlation model. A recursive filter of Hayden and Purser (1995) is applied in VDRAS to estimate the background error covariance. The second term on the right-hand side of Eq. (2) represents the discrepancy between the model-produced radial velocity and rain–snow mixing ratio from their radar-observed counterparts and , respectively. Note that the subscripts r and s stand for rain and snow, while the superscript o denotes observation. It is assumed that there is no spatial correlation between observations. The summation is over space υ and time t. The coefficients and are selected so that they are proportional to the inverse of the observational error variances for radial velocity and rain–snow, respectively. According to Sun (2005b), the observation errors are estimated by computing the local standard deviations of both the radial velocity and reflectivity using the adjacent two ranges and two azimuths.
The simulated radial wind is calculated from the model’s Cartesian velocity components using the following relation:
e3
where represents the distance between a grid point and the radar location , and (m s−1) is the terminal velocity (subscript x indicates rain r or snow s), estimated through the following relations as proposed by SC97 and D89:
e4
e5
where is the basic state pressure, is the pressure at the ground, and represents the density of air. The mixing ratio of rain is estimated from the radar reflectivity Zr (dBZ) by the formula adopted in SC97:
e6
The mixing ratio of snow is estimated by employing the following formula used in Tong and Xue (2005):
e7
where Zs (mm6 m−3) is the radar reflectivity, represents the density of snow, is the density of ice, and and are the dielectric factors for ice and water, respectively. From D89, the intercept parameter for snow is specified as . Equation (7) can be further transferred to the following form (Zs in dBZ):
e8
for snow. By comparing Eqs. (6) and (8), it is realized that, for a given radar reflectivity, the total amount of snow converted from reflectivity is greater than that of rain. On the other hand, given the same mixing ratio value, rain (snow) produces stronger (weaker) radar reflectivity.

In Eq. (2), a spatial and temporal smoothness penalty term is applied by which the minimized results are forced to smoothly fit the observations (SC97). The final term denotes a mesoscale background field, which is used as a first guess for the minimization process to begin with. This term ensures that the 4DVar analysis does not drift too far from a larger-scale background field and is needed particularly in the radar data-void regions. The mesoscale background field in VDRAS can be obtained through an objective analysis made by combining the data from in situ observations, such as radiosondes, wind profilers, surface stations, and radar velocity–azimuth display (VAD) analyses, with that from mesoscale model forecasts and reanalyses. The weighting coefficient of this term is given to make its magnitude similar to the observational term. Sun and Zhang (2008), Sun et al. (2010), and Tai et al. (2011) discussed this procedure in detail.

The open lateral boundary condition is applied in the VDRAS numerical model. The along-beam and cross-beam components of the horizontal inflow wind along the boundaries are determined by a combination of the radar radial velocity observation and a background wind field, respectively. The outflow wind is extrapolated using the values at the closest two inner grid points. The top and bottom boundary conditions are set to zero for vertical velocity, and all other variables are defined such that their normal derivatives vanish. The detailed model setup of the boundary conditions can also be found in SC97 and Sun and Crook (2001).

3. Verification of the adjoint model for the ice-phase microphysics process

The tangent linear model is obtained by linearizing the forward model, and the adjoint model is the transpose of the tangent linear model. In appendix A, the derivatives of the ice-phase microphysics process needed for coding the adjoint equations are introduced. The discussion in SC97 for the warm rain process indicates that, when an extremely small amount of rainwater occurs, it would lead to abnormal gradients when computing the derivatives. Similarly, it is also found in this study that a small mixing ratio of snow would cause huge gradients for accretion and deposition of cloud ice. This can be illustrated by examining Eqs. (B2) and (B19), in which the derivatives of accretion and deposition of cloud ice with respect to snow mixing ratio qs are found to be inversely proportional to the snow mixing ratio as expressed in the following:
e9
e10

It is clear that a large gradient can be caused by a very small snow mixing ratio, and this would make the minimization difficult. Thus, after a series of tests, we found that the same lower limit for rainwater suggested in SC97, a minimum value of 0.001 g kg−1, was also appropriate for snow mixing ratio. Therefore, in Eqs. (A13) and (A15), this threshold would be imposed on the snow content (embedded in parameter λ) when computing the accretion and deposition, respectively. Furthermore, when qs is given a threshold value, the gradients expressed by Eqs. (B2) and (B19) would be specified to be zero.

To verify the correctness of the adjoint model, a test is done following Navon et al. (1992) to define a function :
e11
where J stands for the cost function from the forward model integration, G represents the gradient of the cost function with respect to the model initial state computed from the adjoint backward integration, represents the initial conditions of the prognostic variables, h is a normalized random vector, and is a scalar. Three tests are conducted to examine the performance of the adjoint model. Test 1 is an experiment using the modified ice-physics process with a threshold value of snow imposed for accretion and deposition. In test 2 and test 3, the original scheme without imposing any lower limit is employed for accretion and deposition, respectively. It can be found from Fig. 1 that, with the modified ice-phase microphysics process, the values of Φ from test 1 vary from 0.998 to 1.001 over a wide range of α, indicating that the adjoint model is consistent with the forward model. In contrast, the values of Φ from both test 2 and test 3 are unreasonably approaching zero, implying the inconsistency between a forward and an adjoint model being caused by the original accretion and deposition processes in the numerical model.
Fig. 1.
Fig. 1.

Plot of function Φ(α) defined by Eq. (11). Test 1 uses the modified ice process with a lower threshold for the mixing ratio of snow qs. Test 2 and test 3 represent the tests in which the original scheme is applied without setting a lower threshold for the accretion and deposition of cloud ice, respectively.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

Another important issue when implementing the ice-phase microphysical process is to determine the freezing level so that the liquid-phase and ice-phase processes can be distinguished during the model forward and backward integrations. Furthermore, the freezing level needs to remain unchanged during the iterative procedure for minimization, or it is difficult for the cost function to converge to a minimum. The experiments in section 5 will be designed to discuss this issue.

4. Verification indices

One of the purposes of this study is to assess the rainfall forecast capability of VDRAS after adding ice physics to the 4DVar minimization procedure; thus, the accuracy of the predicted accumulated rainfall is used to evaluate the model performance. In each time step (Δt), the accumulated precipitation (PR; mm) for each grid at the lowest model level can be written as
e12
where (kg m−3) stands for the density of liquid water. Based on Eq. (12), one can compute the required accumulated rainfall over any given period of time for verification.
For a quantitative comparison of the accuracy of the predicted precipitation obtained in different experiments, the equitable threat score (ETS) proposed by Schaefer (1990) and Rogers et al. (1996) is chosen to verify the forecast performance. This index can be defined as follows:
e13
where H stands for the number of correctly predicted points, F is the number of model forecast points, O denotes the number of observed points, and R represents the number of random hits by chance and is written as
e14
where N is the total number of points in the verification domain. The number of points is counted in H, F, and O only when the precipitation is above certain prescribed thresholds. When ETS reaches one, the forecast is considered perfect.
The root-mean-square error (RMSE) is also applied to conduct a quantitative comparison between the retrieval–forecast from VDRAS and the observations, expressed as follows:
e15
where subscripts R and O, respectively, represent the retrieved/forecasted and observed values for a certain parameter X, and N is the total number of grid points used for the computation.

5. Forward simulations with and without ice-phase microphysics

This section discusses the differences in the model simulations with and without the ice physics. The cloud-resolving model built in VDRAS is used to conduct two simulations, named WARM_RUN and ICE_RUN, to demonstrate the influence of the ice-phase physics on simulating a squall line. The microphysical scheme in WARM_RUN utilizes the original warm rain process, as mentioned in SC97. On the other hand, ICE_RUN contains both liquid and the new ice-phase schemes, as introduced in section 3. Except for the microphysical schemes, the experimental designs of these two simulations are identical.

The model domain is 330 × 330 × 15 km3. The horizontal and vertical grid spaces are 3 km and 500 m, respectively. The time step for the model integration is 5 s. The sounding used for the initial temperature and moisture profiles are from Weisman and Klemp (1982). This initial sounding provides a favorable environment for the development of convection. The initial wind shear profile is unidirectional, with a magnitude of 11.5 m s−1 from surface to Z = 3 km. The initial u component of wind at surface is −11.5 m s−1, while the υ component of wind on all levels is 0 m s−1. The initial precipitation (i.e., rain and snow) mixing ratio and vertical velocity are set to 0. A thermal perturbation field is implemented to the initial field. It is along a north–south-oriented line with a 2.0-K maximum potential temperature excess (Klemp and Wilhelmson 1978), superimposed by small (0.1 K) random perturbations.

Figures 2a and 2b present the distribution at the lowest model level (Z = 250 m) of the rainwater mixing ratio from the experiments WARM_RUN and ICE_RUN at t = 11 400 s. The rainwater from ICE_RUN exhibits a more compact line pattern, while in WARM_RUN the initial line-shaped convection evolves into isolated convective cells. Figures 2c–h depict the averaged structures of the precipitation mixing ratio, temperature perturbation (from the initial state), vertical velocity, and water vapor perturbations over a vertical cross section. The average is performed along a north–south direction, or the y axis of the domain. In WARM_RUN, the major precipitation area (defined as where the values of qr or qs exceed 2.5 g kg−1) extends vertically from 3250 to 13 750 m, as shown in Fig. 2c. In contrast, in ICE_RUN, the major precipitation area covers a shallower depth from the freezing level at Z ~ 3650–9250 m, as illustrated in Fig. 2d. The cold pools near the surface from both simulations are similar, as shown in Figs. 2e and 2f. However, the temperature perturbation at the upper levels in ICE_RUN is relatively warmer than that in WARM_RUN. The area with temperature perturbation greater than 1.5°C is also wider in ICE_RUN than that in WARM_RUN. These differences are caused by the release of more latent heat during the formation of snow through deposition compared to that of rainwater through condensation. The excessive amount of latent heat is also expected to trigger stronger vertical velocities, as is confirmed by the larger area and stronger updraft in ICE_RUN (Fig. 2f) in comparison with those in WARM_RUN (Fig. 2e). The simulated water vapor fields are also illustrated in Figs. 2g and 2h. It is found that stronger and wider updrafts produced in experiment ICE_RUN (Fig. 2f) are capable of bringing more vapor to the middle level (Fig. 2h) than experiment WARM_RUN (Fig. 2g). On the other hand, stronger downdrafts also make the near-surface atmosphere drier if compared with the result from experiment WARM_RUN (Fig. 2g). This feature is even more evident when displaying the difference between experiments ICE_RUN and WARM_RUN by subtracting WARM_RUN from ICE_RUN, as shown in Fig. 2i.

Fig. 2.
Fig. 2.

Plots representing the (left) WARM_RUN and (right) ICE_RUN. (a),(b) The precipitation mixing ratio (shading, g kg−1) and flow vectors (m s−1 at the bottom level (250 m) at simulation time t = 11 400 s. Vectors are plotted every 9 km. Several fields averaged along the y axis are depicted for (c),(d) precipitation mixing ratio (g kg−1); (e),(f) temperature perturbation (colored, °C) and vertical velocity (black contours at 2.5, 1.5, 0.2, −0.2, −0.6, and −1.0 m s−1); and (g),(h) water vapor perturbation (g kg−1). (i) The difference between (h) and (g) calculated by subtracting WARM_RUN from ICE_RUN. The positions of two pseudoradars for OSSEs are marked by solid circles in (b). The solid lines in (c) and (d) represent the heights of the freezing level.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

The results shown in this section clearly demonstrate that the implementation of the ice-phase microphysics process can significantly alter the kinematic and thermodynamic structure of a simulated deep convection.

6. Impact of the ice microphysics scheme on 4DVar analysis and forecast—OSSE experiments

a. Experimental design

This section uses a series of OSSE tests to investigate the performance of the VDRAS 4DVar radar data assimilation algorithm with the newly implemented ice physics. The simulation of the ICE_RUN introduced in the previous section, also referred to as the “nature run,” is selected to imitate a true atmosphere. Thus, the simulated results from ICE_RUN will be considered as the observations in this study.

Three OSSE experiments (i.e., WARM, ICE_NR, ICE) are first conducted. They all are initialized by the sounding adopted by Weisman and Klemp (1982), with details described in section 5. The standard assimilation and forecast cycles are illustrated in Fig. 3. The assimilation window begins at t = 9900 s, followed by two 10-min 4DVar cycles. Each 4DVar cycle assimilates both radial wind and radar reflectivity (computed from the nature run) at three time levels, with an interval of 5-min. A 5-min forecast is inserted between 4DVar cycles and provides the background and first guess for the second cycle. A 3-h forecast follows the assimilation period.

Fig. 3.
Fig. 3.

Illustration of VDRAS assimilation and forecast cycles for the OSSEs. The down-pointing arrows labeled T1–T6 at the top denote the assimilation times for both radar1 and radar2. The numbers at the top show the starting and finishing times for each 4DVar cycle with respect to the simulation time for the nature run.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

Two pseudoradars (radar1 and radar2) are placed to the east and west of the squall line, respectively, with a distance of 170 km (see Fig. 2b). The maximum range of detection is set to 250 km so that the squall line can be completely observed under the coverage of two radars. The observational data are prepared using the model outputs from the nature run through Eqs. (3), (4), and (5) for the radial velocity and Eqs. (6) or (8) for the reflectivity, respectively. The reflectivity is available at each grid point within the radar coverage. The radial velocity is considered observable only at the places where the reflectivity exceeds 0.0 dBZ. The data from radar1 and radar2 are assimilated synchronously.

Figure 4a shows a vertical cross section of the model-generated “true’ precipitation mixing ratio along X = 150 km from the nature run (i.e., ICE_RUN) at t = 11 400 s. The thin solid line in Fig. 4a indicates the freezing level (0°C). The “true” radar reflectivity observations derived from precipitation (rain or snow) mixing ratio of the nature run (Fig. 4a) using Eqs. (6) and (8) are displayed in Fig. 4b. It can be seen that rainwater, even with lesser amounts, can produce stronger radar reflectivity below freezing level than the snow does above the freezing level. This can also be realized by comparing Eqs. (6) and (8). According to Eq. (2), VDRAS assimilates the rain and snow mixing ratio calculated from observed radar reflectivity over the entire depth. Unfortunately, the position of the true freezing level that separates the rain from snow is not always known in a real case scenario. Furthermore, the height of the freezing level is expected to vary both in time and space. These factors need to be taken into account when designing the new VDRAS ice physics 4DVar assimilation algorithm and will be discussed in the following experiments.

Fig. 4.
Fig. 4.

(a) The “true” precipitation mixing ratio (colored, g kg−1) in a vertical cross section through X = 150 km from the nature run at a simulation time of t = 11 400 s. The solid black line indicates the height of the freezing level (0°C). (b) The “true” reflectivity (dBZ) derived from the precipitation mixing ratios qr and qs shown in (a).

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

Table 1 gives a list of the three OSSE experiments and a brief introduction for each. Experiment WARM assimilates radar data but employs only the SC97 liquid-water-phase scheme. In other words, all radar reflectivity will be converted to liquid water, no matter where it is. This is for the intercomparison of the results with and without using the ice physics. The experiments ICE_NR and ICE are conducted to explore the impact of using ice physics scheme. The radar reflectivity at each grid point, when its location is lower or higher than the freezing level, will be converted to rainwater or snow, respectively. In addition, experiments ICE_NR and ICE are also used to investigate how the specification of the reference freezing level could influence the analyses and QPN. In experiment ICE_NR, two 4DVar cycles are used, with the height of the reference freezing level directly taken from the nature run in both cycles. Therefore, experiment ICE_NR represents a purely idealized case for the purpose of examining the accuracy of the newly implemented ice-phase 4DVar assimilation algorithm. In experiment ICE, the reference freezing levels in two 4DVar cycles are estimated by different methods, respectively. In the first cycle, the height of the reference freezing level is obtained from the background radiosonde observation; thus, it is static and homogeneous horizontally. In contrast, after the first 4DVar cycle, the model-forecasted three-dimensional temperature fields at each time step extending to the time covering the second cycle from t = 10 800 to 11 400 s are saved to determine the reference freezing level. This allows the height of the 0°C line to vary spatially and temporally. Nevertheless, it should be emphasized that, in order for the minimization algorithm to converge to an optimal solution, the aforementioned freezing-level height needs to remain unchanged during the following iterative forward and backward model integration in the 4DVar minimization procedure.

Table 1.

List of experimental designs.

Table 1.

b. Converted mixing ratio using radar reflectivity based on different microphysical process assumptions

Figure 5 shows the precipitation mixing ratios [ and in Eq. (2)] converted from the “true” radar reflectivity (see Fig. 4b) over the vertical cross section along X = 150 km at t = 11 400 s. The conversion is based on the different assumptions regarding the microphysical process employed by the three OSSE experiments. Those are viewed as the observed hydrometeor fields ready to be assimilated into the model.

Fig. 5.
Fig. 5.

The precipitation mixing ratio (g kg−1) converted from radar reflectivity using Eqs. (6) and (8) along the vertical cross section X = 150 km at t = 11 400 s. The conversion is based on the microphysical process assumption used in experiments (a) WARM, (b) ICE_NR, and (c) ICE. The dashed black lines on these cross sections indicate the heights of the refereed freezing level. (d) The shading is the precipitation mixing ratio from experiment ICE (Fig. 5c) minus the “true” one from the nature run (ICE_RUN; Fig. 4a). The solid and dashed lines in (d) are the freezing levels from the nature run ICE_RUN and experiment ICE, respectively.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

Experiment WARM includes only the liquid-phase process; therefore, the radar reflectivity at any height, even at upper levels, is still considered to be produced by rainwater. As a result, Fig. 5a reveals that the observed rainwater mixing ratio in WARM, after being converted from radar reflectivity using Eq. (6), is significantly underestimated above Z = 3750 m.

The reference freezing levels for experiments ICE_NR and ICE are displayed in Figs. 5b and 5c, respectively. In ICE_NR, since the reference freezing level is directly from the nature run, the precipitation field displayed in Fig. 5b is the same as the one shown in the nature run (Fig. 4a). For experiment ICE, Fig. 5c indicates that the “observed” precipitation mixing ratio over the entire domain generally agrees well with that in the nature run (Fig. 4a). However, noticeable differences with significantly large mixing ratio values are found along the reference freezing level. Further analyses are conducted to explain this phenomenon.

In experiment ICE, the VDRAS-forecast reference freezing levels at each time step, obtained after the first 4DVar cycle and within the second 4DVar assimilation cycle, is used to separate the ice phase from liquid phase. The heights of these simulated freezing levels, although they vary with time and space, do not completely match the true 0°C line. Consequently, if the freezing level forecasted by VDRAS is lower (higher) than the true one, the actual region occupied by liquid (ice) would be mistakenly classified as ice (liquid). This would lead to a recovery of an excessive (insufficient) amount of precipitation mixing ratios when they are converted from radar reflectivity using Eq. (8) [Eq. (6)]. This phenomenon is clearly illustrated in Fig. 5d, which shows the “true” and the VDRAS-simulated reference freezing levels and the difference of the precipitation mixing ratio between the two experiments (i.e., experiment ICE minus nature run from ICE_RUN). It can be seen that when the VDRAS-produced freezing level (denoted by dashed line) is lower (higher) than the “true” freezing level (denoted by solid line), the difference of the mixing ratio is positive (negative).

c. Analyses and forecasts

The 4DVar-retrieved analysis fields of the precipitation mixing ratio (qr and qs), temperature perturbations, and vertical velocities from the three OSSEs at t = 11 400 s are displayed in Figs. 6 and 7, respectively. Figure 7d also depicts the “true” temperature perturbations and vertical velocities from the nature run (i.e., ICE_RUN) for comparison. Figures 6a and 7a show that experiment WARM retrieves a lesser precipitation mixing ratio, smaller temperature perturbations, and weaker vertical velocities in the mid- and upper levels than those from the other two experiments ICE and ICE_NR because of the usage of warm rain process only and lesser latent heat release. It is shown that the retrieved precipitation mixing ratio (Fig. 6b) and temperature perturbations and vertical velocities (Fig. 7b) from experiment ICE_NR are in very good agreement with those from the nature run, as depicted in Figs. 4a and 7d, respectively. This indicates that the newly implemented ice-phase microphysics scheme in the 4DVar assimilation/minimization procedure is functioning properly. Furthermore, it is encouraging to see that using the VDRAS-forecasted freezing level obtained after the first 4DVar cycle, as described earlier in this section, the retrieved fields of precipitation mixing ratio (Fig. 6c), temperature perturbations, and vertical velocity (Fig. 7c) by experiment ICE are also very similar to their counterparts from ICE_NR (Figs. 6b and 7b). Because of the misjudged freezing-level heights, the isolated regions with strong magnitudes found in the mixing ratio field shown in Fig. 5c no longer exist in Fig. 6c. By comparing the same mixing ratio field at the beginning of the minimization procedure (not shown), the disappearance of the excessive mixing ratio can be attributed to the diffusion effect during the model forward–backward integrations.

Fig. 6.
Fig. 6.

The 4DVar-retrieved precipitation mixing ratio analysis field (g kg−1) over the vertical cross section through X = 150 km at t = 11 400 s from (a) WARM, (b) ICE_NR, and (c) ICE.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

Fig. 7.
Fig. 7.

Vertical cross sections through X = 150 km for the temperature perturbation (color shading, °C) and vertical velocity (contours, m s−1) at t = 11 400 s from the 4DVar-retrieved analysis field from (a) WARM, (b) ICE_NR, and (c) ICE. (d) As in (a)–(c), but from the simulation of the nature run (ICE_RUN).

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

The retrieved precipitation and temperature fields at t = 11 400 s after the 4DVar minimization procedure from the three OSSEs are compared against their true counterparts, with the RMSEs displayed in Fig. 8. As expected, the results from WARM have the highest errors, while those from ICE_NR show the lowest errors. The retrieved fields from experiment ICE have similar accuracy with those from ICE_NR, implying the use of the VDRAS-forecast freezing level during the 4DVar minimization procedure to imitate the true but unknown 0°C line is a feasible approach.

Fig. 8.
Fig. 8.

The RMSEs for (a) precipitation mixing ratio (g kg−1) and (b) temperature perturbation (°C) computed over the whole domain between the nature run and three OSSE experimental analyses at t = 11 400 s.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

After the radar data assimilation is completed at t = 11 400 s, the VDRAS continues its model time integration and successfully makes a 3-h forecast of the rainfall. Figure 9 reveals a comparison of the model forecast skill by showing the RMSEs of the predicted 1-, 2-, and 3-h accumulated rainfall. It can be seen that the performance of ICE is generally comparable to the idealized experiment ICE_NR. However, with only warm rain microphysics process included in the VDRAS cloud-resolving model while the “true” atmosphere contains the ice-phase microphysics process, the forecast error of experiment WARM turns out to be significantly higher than the other two experiments over the entire forecast time.

Fig. 9.
Fig. 9.

The RMSEs for predicted 1-, 2-, and 3-h accumulated rainfall (mm) for WARM (red), ICE_NR (black), ICE (blue), and NODA_ICE (yellow).

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

To clarify the respective role played by the radar data assimilation and the use of the ice physics scheme in improving the rainfall forecast, one more OSSE experiment is designed. In experiment NODA_ICE, the proposed ice physics scheme is adopted. However, the model is initialized at t = 11 400 s using the same sounding data described in section 5 and by the radar reflectivity produced by the nature run ICE_RUN at the same time. In other words, NODA_ICE represents an experiment equipped with the newly implemented cold rain process but without assimilating any radar data into the model. The RMSEs shown in Fig. 9 indicate that the error of the rainfall forecast from NODA_ICE is the largest among all experiments in the first hour and becomes comparable to those from experiment WARM at the second and third hour. Note that WARM is a radar data assimilation experiment that considers only the warm rain process. The results from Fig. 9 demonstrate that, in addition to the use of the ice physics scheme, radar data assimilation still plays an important role in improving the model’s rainfall forecast skill. The OSSE experiments conducted in this section show that a synergy of both radar data assimilation and an ice microphysical scheme yields the best rainfall forecast results (experiment ICE).

Overall, this section introduces the implementation of the ice-phase process in the VDRAS radar data assimilation procedure. From the OSSE experiments presented in this study, one finds positive impacts of the ice-phase process employed during the radar data assimilation on the recovery of the kinematic, thermodynamic, and microphysical structures inside a convective system, as well as on the improvement of the model’s rainfall forecast.

7. A real case study of 2008 SoWMEX IOP 8

A real case from SoWMEX IOP 8, a field experiment conducted in 2008 in southern Taiwan, is selected to further explore the influence of the ice-phase process under a realistic scenario. This particular IOP was for a prefrontal squall line, whose influence on the precipitation of Taiwan lasted for about two days. Thus, the entire observation period started at 0000 UTC 14 June and ended at 0000 UTC 16 June.

Figure 10 illustrates the locations of two Taiwan Central Weather Bureau (CWB) operational S-band Doppler radars (RCCG and RCKT in Cigu and Kenting, respectively), nine surface stations, and two radiosonde stations. Reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF) are also utilized to fill in the data-void region over the ocean. Note that the data from the two radiosonde stations were not included in the ECWMF reanalysis data. The data from different sources shown in Fig. 10 for this real case study were independent of each other. The in situ observations and the ECMWF reanalysis data are combined to establish a mesoscale background field. Before radar data are ingested into VDRAS, the radial velocity folding and ground clutter contamination are carefully corrected. The entire domain size is 528 × 432 × 15 km3, with the horizontal and vertical grid space set to 2.0 and 0.5 km, respectively.

Fig. 10.
Fig. 10.

Locations of data collected from different sources for SoWMEX IOP 8. The small dots indicate selected data points from the ECMWF reanalysis, triangles stand for radars, crosses represent surface mesonet stations, and black squares denote radiosondes. This is also the experimental domain for the VDRAS.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

Figure 11 shows the composite maximum radar reflectivity on 14 June from 1302 to 1432 UTC, observed by RCCG and RCKT (see Fig. 10 for their geographic locations). The quasi-linear shape convective rainbands were elongated in a northeast–southwest direction, moving toward the east and southeast, and produced a significant amount of rainfall (>100 mm day−1) over the southwestern part of Taiwan.

Fig. 11.
Fig. 11.

The composite maximum radar reflectivity (color shading at 25, 30, 35, 40, 45, 50, and 55 dBZ) at (a) 1302, (b) 1332, (c) 1402, and (d) 1432 UTC 14 Jun 2008 for IOP 8. The jagged solid black line represents the Taiwanese coast line.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

Experiments R_WARM and R_ICE (see Table 1) are equivalent to their OSSE counterparts WARM and ICE, respectively. The two experiments have the same assimilation procedure as shown in Fig. 12. Two 16-min 4DVar cycles are included with a short 6-min forecast period inserted in between. Radar data from three volume scans including radial winds and reflectivity are assimilated in each 4DVar cycle. The assimilation for all experiments ends at 1302 UTC, followed by a 2-h model forecast.

Fig. 12.
Fig. 12.

Illustration of assimilation and forecast cycles for the R_WARM and R_ICE experiments. The numbers at the top of the diagram indicate the starting and finishing times (UTC) of each 4DVar cycle. RCKT and RCCG stand for the two CWB radars.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

Figure 13 shows the contoured frequency by altitude diagram (CFAD; Yuter and Houze 1995) of reflectivity at 1302 UTC after the radar data have been assimilated through two 4DVar cycles. As described in Yuter and Houze (1995), if the available data are insufficient to be representative of the structure, the plots should be removed. This condition often occurs at higher altitudes to which it is harder for the convection or the radar scans to reach. We compute the number of available data points at each horizontal layer in the analysis domain and obtain the maximum value. In this study, the criterion to truncate the plot is that when the number of data points at a given layer is less than 15% of the aforementioned maximum number.

Fig. 13.
Fig. 13.

The CFADs (color shading at 0.5%, 5%, 10%, 15%, and 20% dBZ−1 km−1) of reflectivity at 1302 UTC for (a) observation and experiments (b) R_WARM and (c) R_ICE. The bin size is 5 dBZ. Only those grid points where Z > 5 dBZ are counted. The CFADs in (a),(b), and (c) are truncated above the height of 10.75, 6.25, and 9.25 km, respectively, because of insufficient data points.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

The observed reflectivity (Fig. 13a) with a range from 15 to 25 dBZ occurs most frequently at the heights from 4 to 10 km. In contrast, the analysis field shown in Fig. 13b reveals that the reflectivity in experiment R_WARM extends to only below the height of 5.0 km, suggesting that the model-produced convection reaches a much lower altitude than the true one does. However, when the ice process is considered in R_ICE, the statistical distribution of radar reflectivity can be dramatically improved. The frequency of occurrence, as well as the vertical extension of major reflectivity (i.e., 15–25 dBZ) produced by R_ICE, are more consistent with the observations than R_WARM.

The averages of positive and negative vertical velocities at 1302 UTC over the entire domain are shown in Fig. 14a, while the averages over these points with the top 1% of the updrafts and downdrafts are displayed in Fig. 14b. It is found that R_ICE can retrieve stronger down- and updrafts than R_WARM does. According to the conclusion from the numerical experiments presented in section 5, this is due to the release of more latent heat when employing the ice physics scheme in R_ICE.

Fig. 14.
Fig. 14.

(a) The updrafts and downdrafts averaged over the entire domain. Solid (dashed) line stands for updrafts (downdrafts), and red and blue lines represent results from R_WARM and R_ICE, respectively. (b) As in (a), but the average is performed only over those points containing the top 1% of the updrafts and downdrafts.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

The evolutions of predicted maximum radar reflectivity from R_WARM and R_ICE are shown in Figs. 15 and 16, respectively. Compared to the observations (Fig. 11), Fig. 15 illustrates that, in R_WARM, radar reflectivity with smaller strength (<35 dBZ) dissipates very quickly even within the first 30 min. In contrast, experiment R_ICE (Fig. 16) is able to maintain the weak reflectivity for a much longer period of time. It is believed that, since the ice process can generate stronger updrafts, it helps to keep the hydrometeors in the air for a longer period of time and prevent the reflectivity from scattering too quickly.

Fig. 15.
Fig. 15.

The predicted maximum radar reflectivity (color shading at 25, 30, 35, 40, 45, 50, and 55 dBZ) at (a) 1302, (b) 1332, (c) 1402, and (d) 1432 UTC from experiment R_WARM. The time corresponds to (a) the initial condition and a forecast of (b) 0.5, (c) 1.0, and (d) 1.5 h. The jagged solid black line represents the Taiwanese coast line.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

Fig. 16.
Fig. 16.

As in Fig. 15, but for experiment R_ICE.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

Figure 17 exhibits the spatial distributions of a 2-h accumulated precipitation from the rain gauge observations and the predictions from the numerical experiments R_WARM and R_ICE. The observed major precipitation [>12 mm (2 h)−1] regions can be identified along the western slopes of the mountains, the southwestern plain, and the southern tip of the island. Figures 17b and 17c show that, in both experiments, the principal rainfall area near the mountains and southern tip are well forecasted. However, the rainband that stretches to the southwestern plain area is underestimated in R_WARM. In contrast, the forecast of this rainband turns out to be better in R_ICE (Fig. 17c), with more rainfall over the southwestern plain near the coastal area than that of R_WARM. Furthermore, the area of weak rainfall [<6 mm (2 h)−1] predicted in R_ICE is larger than that of R_WARM and is more consistent with the observations. It should be pointed out that the current VDRAS does not have the capability to resolve the terrain. This may contribute to the northward shift of the forecasted rainfall maximum when compared with the surface observations.

Fig. 17.
Fig. 17.

The 2-h accumulated rainfall amount (color shading, mm) from (a) rain gauge observations, (b) experiment R_WARM, and (c) experiment R_ICE. The thick solid lines depict the terrain heights of 500, 1500, and 2500 m.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

The comparison of ETSs and RMSEs under different rainfall thresholds [2, 6, 10, 14, and 18 mm (2 h)−1] are shown in Fig. 18. Among all thresholds, R_ICE indeed has significantly higher ETSs and lower RMSEs than R_WARM, indicating that, in the case shown in this study, experiment R_ICE has superior QPN capability after the implementation of the additional ice-phase microphysical process into VDRAS.

Fig. 18.
Fig. 18.

(a) ETSs and (b) RMSEs of the forecast 2-h rainfall accumulations over the island of Taiwan from experiments R_WARM and R_ICE at different precipitation thresholds.

Citation: Journal of the Atmospheric Sciences 73, 3; 10.1175/JAS-D-15-0184.1

8. Summary and future work

In this study, a simple ice physics scheme and its adjoint containing cloud ice and snow are successfully developed. The implementation of this ice physics scheme into VDRAS is accomplished by taking advantage of the existing 4DVar minimization framework of VDRAS so that the increase of the computational cost can remain insignificant. A series of OSSEs and a real case experiment are conducted to verify the forward–adjoint model integration and evaluate the influence of the ice-phase microphysics on the parameter retrievals, radar data assimilation, and model QPN skill. Based on the experiments conducted in this research, the major conclusions are summarized as follows:

  1. The forward simulation indicates that using an ice-phase microphysical scheme can produce more latent heat release and thus trigger stronger updrafts in upper levels than the original warm rain process does. The kinematic and thermodynamic structures of the deep convection obtained from the simulations with and without the ice-phase microphysical process are significantly different.
  2. With ice-phase physics implemented in the model, the retrieved precipitation (rain and snow) mixing ratio, temperature perturbation fields, and vertical velocity through 4DVar radar data assimilation are also stronger than those from the experiment without ice physics.
  3. To separate the rain from snow, it is feasible to use the VDRAS-forecasted freezing level obtained after the previous 4DVar assimilation cycle to replace the true but unknown 0°C line and, at the same time, allow the 4DVar technique to converge to an optimal solution during the second 4DVar cycle.
  4. The conclusions from OSSE tests are further verified by using real observational data from the IOP 8 of 2008 SoWMEX. In this particular real case study, adding ice phase improves the model’s forecast of the vertical and horizontal distribution of the radar reflectivity and helps the model to capture the actual evolution of the convective system. The new VDRAS with ice microphysical scheme also demonstrates a higher QPN capability.

In the future, the new VDRAS developed in this research will be applied to more real case studies so that its performance can be evaluated comprehensively under different scenarios. Furthermore, the current version of VDRAS is also subject to the limitation of using a Cartesian grid system and is unable to deal with complex terrain. Thus, the implementation of terrain-resolving capability to VDRAS will be the focus of future work.

Acknowledgments

This research is supported by the Ministry of Science and Technology of Taiwan under MOST103-2119-M-008-008,MOST103-2625-M-008-008-MY2, MOST104-2119-M-008-007, and the Central Weather Bureau (CWB) of Taiwan under MOTC-CWB-103-M-06. The authors are very grateful to the CWB for providing RCCG and RCKT radar data. The data from radars, surface stations, and soundings are obtained from the data bank archived by the Taiwan Typhoon and Flood Research Institute ( http://www.ttfri.narl.org.tw/service03.html).

APPENDIX A

Microphysics Scheme

The original liquid-phase and the implemented ice-phase scheme are described in this section. The hydrometeor fields include the mixing ratios of water vapor , rainwater , snow , cloud water , and cloud ice . The definition of mixing ratio r is the ratio of the mass of a hydrometeor variable to the mass of dry air and can be written as
ea1
In addition to rainwater (or snow), the total mixing ratio is another prognostic hydrometeor variable and can be expressed as
ea2
and
ea3
The parameters and are the saturated vapor mixing ratio with respect to water and ice, respectively, and are given by
ea4
ea5
The liquid–ice water potential temperatures (i.e., and ) are used as the prognostic temperature variables in VDRAS. The thermodynamics equation is written with an additional diffusion term:
ea6
where stands for the diffusivity of , is the initial unperturbed state of the air density, T is the temperature, and denotes the liquid–ice water potential temperature expressed by
ea7
ea8
In Eqs. (A7) and (A8), and are the latent heat of evaporation and sublimation, respectively.
As an extension of SC97, the equation governing precipitation (i.e., rainwater and snow) mixing ratio can be expressed by
ea9
For the prediction of the rainwater mixing ratio , terms stand for accretion of cloud water by rainwater, autoconversion of cloud water to rainwater, and evaporation of raindrops in a nonsaturated atmosphere, respectively. Accretion is expressed by
ea10
where is set to 0.002 s−1 (Miller and Pearce 1974). Following Kessler (1969), autoconversion and evaporation of rain can be written as
ea11
ea12
where and . The critical mixing ratio of cloud is 1.5 g kg−1.
On the other hand, for predicting snow mixing ratio , terms represent the accretion of cloud ice by snow, autoconversion of cloud ice to snow, and deposition of vapor to snow, respectively. The processes for ice phase are explained in the following. According to D89, the accretion of cloud ice by snow is expressed by
eaa13a
eab13b
in which and are constants used in the snowfall-speed formula, , where is the diameter of snow. The Γ is the gamma function for snow. The accretion efficiency is from Hong et al. (2004). The autoconversion based on Hong et al. (2004) is written as
eaa14a
eab14b
where is the critical mixing ratio of ice crystal, and represents the model time step. The formula for deposition from D89 is given by
eaa15a
and
eab15b
where is the thermal conductivity of air and is the diffusivity of water vapor in air. The Sc = 0.6 is the Schmidt number, is the dynamic viscosity of air, and Si equals .

APPENDIX B

Tangent Linear Model of Ice Physics

The tangent linear model is obtained by linearizing the forward model. This section shows the derivatives of the ice process needed for coding the adjoint equations.

Accretion of cloud ice by snow [Eq. (A13)] can be rewritten as follows:
eba1a
ebb1b
The derivatives of Eq. (B1) with respect to qs, qt, and are given as
eb2
eb3
eb4
Within Eqs. (B2), (B3), and (B4), the derivatives of cloud ice mixing ratio qi and temperature T with respect to qt, qs, and would yield the following relationships shown from Eqs. (B5)(B8). Similar to cloud water, the cloud ice mixing ratio is also diagnosed from total mixing ratio qt for ice phase by assuming that, when the temperature is below 0°C, all vapor greater than the saturation value is converted to cloud ice [Eq. (A3)]. The cloud ice mixing ratio can be defined by
eb5
The derivatives of cloud ice mixing ratio can be expressed as follows:
eb6
eb7
eb8
The derivative of saturation ice mixing ratio with respect to temperature T in Eqs. (B6) and (B8) is expressed as
eb9
The temperature for ice process is defined by
eb10
We multiply the temperature on both sides of Eq. (B10) and use Eq. (B5) for (qi + qs), then rewrite Eq. (B10) as
eb11
The derivatives of Eq. (B11) with respect qs, qt, and are computed and rearranged to yield
eb12
eb13
eb14
The derivations of the autoconversion process expressed in Eq. (A14) with respect to qs, qt, and are expressed as
eb15
eb16
eb17
The formula for deposition shown by Eq. (A15) is rearranged as
eba18a
ebb18b
ebc18c
ebd18d
Then the derivatives of deposition with respect to qs, qt, and can be written as
eb19
eb20
eb21

REFERENCES

  • Bluestein, H. B., and et al. , 2014: Radar in atmospheric sciences and related research: Current systems, emerging technology, and future needs. Bull. Amer. Meteor. Soc., 95, 18501861, doi:10.1175/BAMS-D-13-00079.1.

    • Search Google Scholar
    • Export Citation
  • Chang, S.-F., , J. Sun, , Y.-C. Liou, , S.-L. Tai, , and C.-Y. Yang, 2014: The influence of erroneous background, beam-blocking and microphysical nonlinearity on the application of a four-dimensional variational Doppler radar data assimilation system for quantitative precipitation forecasts. Meteor. Appl., 21, 444458, doi:10.1002/met.1439.

    • Search Google Scholar
    • Export Citation
  • Chung, K.-S., , I. Zawadzki, , M. K. Yau, , and L. Fillion, 2009: Short-term forecasting of a midlatitude convective storm by the assimilation of single–Doppler radar observations. Mon. Wea. Rev., 137, 41154135, doi:10.1175/2009MWR2731.1.

    • Search Google Scholar
    • Export Citation
  • Crook, N. A., 1994: Numerical simulations initialized with radar-derived winds. Part I: Simulated data experiments. Mon. Wea. Rev., 122, 11891203, doi:10.1175/1520-0493(1994)122<1189:NSIWRD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Crook, N. A., , and J. D. Tuttle, 1994: Numerical simulations initialized with radar-derived winds. Part II: Forecasts of three gust-front cases. Mon. Wea. Rev., 122, 12041217, doi:10.1175/1520-0493(1994)122<1204:NSIWRD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Crook, N. A., , and J. Sun, 2002: Assimilating radar, surface, and profiler data for the Sydney 2000 forecast demonstration project. J. Atmos. Oceanic Technol., 19, 888898, doi:10.1175/1520-0426(2002)019<0888:ARSAPD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Dudhia, J., 1989: Numerical study of convection observed during the Winter Monsoon Experiment using a mesoscale two-dimensional model. J. Atmos. Sci., 46, 30773107, doi:10.1175/1520-0469(1989)046<3077:NSOCOD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Gal-Chen, T., 1978: A method for the initialization of the anelastic equations: Implications for matching models with observations. Mon. Wea. Rev., 106, 587606, doi:10.1175/1520-0493(1978)106<0587:AMFTIO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Gao, J., , and D. J. Stensrud, 2012: Assimilation of reflectivity data in a convective-scale, cycled 3DVAR framework with hydrometeor classification. J. Atmos. Sci., 69, 10541065, doi:10.1175/JAS-D-11-0162.1.

    • Search Google Scholar
    • Export Citation
  • Gao, J., , and D. J. Stensrud, 2014: Some observing system simulation experiments with a hybrid 3DEnVAR system for storm-scale radar data assimilation. Mon. Wea. Rev., 142, 33263346, doi:10.1175/MWR-D-14-00025.1.

    • Search Google Scholar
    • Export Citation
  • Hayden, C. M., , and R. J. Purser, 1995: Recursive filter objective analysis of meteorological fields: Applications to NESDIS operational processing. J. Appl. Meteor., 34, 315, doi:10.1175/1520-0450-34.1.3.

    • Search Google Scholar
    • Export Citation
  • Hong, S.-Y., , J. Dudhia, , and S.-H. Chen, 2004: A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation. Mon. Wea. Rev., 132, 103120, doi:10.1175/1520-0493(2004)132<0103:ARATIM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Houze, R. A., Jr., , M. I. Biggerstaff, , S. A. Rutledge, , and B. F. Smull, 1989: Interpretation of Doppler weather radar displays of midlatitude mesoscale convective systems. Bull. Amer. Meteor. Soc., 70, 608619, doi:10.1175/1520-0477(1989)070<0608:IODWRD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Hu, M., , M. Xue, , J. Gao, , and K. Brewster, 2006a: 3DVAR and cloud analysis with WSR-88D level-II data for the prediction of the Fort Worth, Texas, tornadic thunderstorms. Part I: Cloud analysis and its impact. Mon. Wea. Rev., 134, 675698, doi:10.1175/MWR3092.1.

    • Search Google Scholar
    • Export Citation
  • Hu, M., , M. Xue, , J. Gao, , and K. Brewster, 2006b: 3DVAR and cloud analysis with WSR-88D level-II data for the prediction of the Fort Worth, Texas, tornadic thunderstorms. Part II: Impact of radial velocity analysis via 3DVAR. Mon. Wea. Rev., 134, 699721, doi:10.1175/MWR3093.1.

    • Search Google Scholar
    • Export Citation
  • Jung, Y., , G. Zhang, , and M. Xue, 2008a: Assimilation of simulated polarimetric radar data for a convective storm using the ensemble Kalman filter. Part I: Observation operators for reflectivity and polarimetric variables. Mon. Wea. Rev., 136, 22282245, doi:10.1175/2007MWR2083.1.

    • Search Google Scholar
    • Export Citation
  • Jung, Y., , M. Xue, , G. Zhang, , and J. M. Straka, 2008b: Assimilation of simulated polarimetric radar data for a convective storm using the ensemble Kalman filter. Part II: Impact of polarimetric data on storm analysis. Mon. Wea. Rev., 136, 22462260, doi:10.1175/2007MWR2288.1.

    • Search Google Scholar
    • Export Citation
  • Kawabata, T., , T. Kuroda, , H. Seko, , and K. Saito, 2011: A cloud-resolving 4DVAR assimilation experiment for a local heavy rainfall event in the Tokyo metropolitan area. Mon. Wea. Rev., 139, 19111931, doi:10.1175/2011MWR3428.1.

    • Search Google Scholar
    • Export Citation
  • Kessler, E., 1969: On the Distribution and Continuity of Water Substance in Atmospheric Circulation. Meteor. Monogr., No. 32, Amer. Meteor. Soc., 84 pp.

    • Search Google Scholar
    • Export Citation
  • Klemp, J. B., , and R. B. Wilhelmson, 1978: The simulation of three-dimensional convective storm dynamics. J. Atmos. Sci., 35, 10701096, doi:10.1175/1520-0469(1978)035<1070:TSOTDC>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Li, Y., , X. Wang, , and M. Xue, 2012: Assimilation of radar radial velocity data with the WRF hybrid ensemble–3DVAR system for the prediction of Hurricane Ike (2008). Mon. Wea. Rev., 140, 35073524, doi:10.1175/MWR-D-12-00043.1.

    • Search Google Scholar
    • Export Citation
  • Lin, Y., , S. R. Peter, , and W. J. Kenneth, 1993: Initialization of a modeled convective storm using Doppler radar–derived fields. Mon. Wea. Rev., 121, 27572775, doi:10.1175/1520-0493(1993)121<2757:IOAMCS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Liou, Y.-C., , J.-L. Chiou, , W.-H. Chen, , and H.-Y. Yu, 2014: Improving the model convective storm quantitative precipitation nowcasting by assimilating state variables retrieved from multiple-Doppler radar observations. Mon. Wea. Rev., 142, 40174035, doi:10.1175/MWR-D-13-00315.1.

    • Search Google Scholar
    • Export Citation
  • Miller, M. J., , and R. P. Pearce, 1974: A three-dimensional primitive equation model of cumulonimbus convection. Quart. J. Roy. Meteor. Soc., 100, 133154, doi:10.1002/qj.49710042402.

    • Search Google Scholar
    • Export Citation
  • Navon, I. M., , X. Zou, , J. Derber, , and J. Sela, 1992: Variational data assimilation with an adiabatic version of the NMC spectral model. Mon. Wea. Rev., 120, 14331446, doi:10.1175/1520-0493(1992)120<1433:VDAWAA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Pan, X., , X. Tian, , X. Li, , Z. Xie, , A. Shao, , and C. Lu, 2012: Assimilating Doppler radar radial velocity and reflectivity observations in the weather research and forecasting model by a proper orthogonal-decomposition-based ensemble, three-dimensional variational assimilation method. J. Geophys. Res., 117, D17113, doi:10.1029/2012JD017684.

    • Search Google Scholar
    • Export Citation
  • Rogers, E., , T. L. Black, , D. G. Deaver, , G. J. DiMego, , Q. Zhao, , M. Baldwin, , N. W. Junker, , and Y. Lin, 1996: Changes to the operational “early” Eta analysis/forecast system at the National Centers for Environmental Prediction. Wea. Forecasting, 11, 391412, doi:10.1175/1520-0434(1996)011<0391:CTTOEA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Schaefer, J. T., 1990: The critical success index as an indicator of warning skill. Wea. Forecasting, 5, 570575, doi:10.1175/1520-0434(1990)005<0570:TCSIAA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Snyder, C., , and F. Zhang, 2003: Assimilation of simulated Doppler radar observations with an ensemble Kalman filter. Mon. Wea. Rev., 131, 16631677, doi:10.1175//2555.1.

    • Search Google Scholar
    • Export Citation
  • Sun, J., 2005a: Initialization and numerical forecasting of a supercell storm observed during STEPS. Mon. Wea. Rev., 133, 793813, doi:10.1175/MWR2887.1.

    • Search Google Scholar
    • Export Citation
  • Sun, J., 2005b: Convective-scale assimilation of radar data: Progress and challenges. Quart. J. Roy. Meteor. Soc., 131, 34393463, doi:10.1256/qj.05.149.

    • Search Google Scholar
    • Export Citation
  • Sun, J., , and N. A. Crook, 1997: Dynamical and microphysical retrieval from Doppler radar observations using a cloud model and its adjoint. Part I: Model development and simulated data experiments. J. Atmos. Sci., 54, 16421661, doi:10.1175/1520-0469(1997)054<1642:DAMRFD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Sun, J., , and N. A. Crook, 2001: Real-time low-level wind and temperature analysis using single WSR-88D data. Wea. Forecasting, 16, 117132, doi:10.1175/1520-0434(2001)016<0117:RTLLWA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Sun, J., , and Y. Zhang, 2008: Analysis and prediction of a squall line observed during IHOP using multiple WSR-88D observations. Mon. Wea. Rev., 136, 23642388, doi:10.1175/2007MWR2205.1.

    • Search Google Scholar
    • Export Citation
  • Sun, J., , and H. Wang, 2013: Radar data assimilation with WRF 4D-Var: Part II. Comparison with 3D-Var for a squall line over the U.S. Great Plains. Mon. Wea. Rev., 141, 22452264, doi:10.1175/MWR-D-12-00169.1.

    • Search Google Scholar
    • Export Citation
  • Sun, J., , M. Chen, , and Y. Wang, 2010: Frequent-updating analysis system based on radar, surface, and mesoscale model data for the Beijing 2008 Forecast Demonstration Project. Wea. Forecasting, 25, 17151735, doi:10.1175/2010WAF2222336.1.

    • Search Google Scholar
    • Export Citation
  • Tai, S.-L., , Y.-C. Liou, , J. Sun, , S.-F. Chang, , and M.-C. Kuo, 2011: Precipitation forecasting using Doppler radar data, a cloud model with adjoint, and the Weather Research and Forecasting Model: Real case studies during SoWMEX in Taiwan. Wea. Forecasting, 26, 975992, doi:10.1175/WAF-D-11-00019.1.

    • Search Google Scholar
    • Export Citation
  • Tong, M., , and M. Xue, 2005: Ensemble Kalman filter assimilation of Doppler radar data with a compressible nonhydrostatic model: OSS experiments. Mon. Wea. Rev., 133, 17891807, doi:10.1175/MWR2898.1.

    • Search Google Scholar
    • Export Citation
  • Tripoli, G. J., , and W. R. Cotton, 1981: The use of ice–liquid water potential temperature as a thermodynamic variable in deep atmospheric models. Mon. Wea. Rev., 109, 10941102, doi:10.1175/1520-0493(1981)109<1094:TUOLLW>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Tsai, C.-C., , S.-C. Yang, , and Y.-C. Liou, 2014: Improving quantitative precipitation nowcasting with a local ensemble transform Kalman filter radar data assimilation system: Observing system simulation experiments. Tellus, 66A, 21 804, doi:10.3402/tellusa.v66.21804.

    • Search Google Scholar
    • Export Citation
  • Wang, H., , J. Sun, , X. Zhang, , X. Huang, , and T. Auligne, 2013: Radar data assimilation with WRF 4D-Var. Part I: System development and preliminary testing. Mon. Wea. Rev., 141, 22242244, doi:10.1175/MWR-D-12-00168.1.

    • Search Google Scholar
    • Export Citation
  • Weisman, M. L., , and J. B. Klemp, 1982: The dependence of numerically simulated convective storms on vertical wind shear and buoyancy. Mon. Wea. Rev., 110, 504520, doi:10.1175/1520-0493(1982)110<0504:TDONSC>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Weygandt, S. S., , A. Shapiro, , and K. K. Droegemeier, 2002: Retrieval of model initial fields from single-Doppler observations of a supercell thunderstorm. Part II: Thermodynamic retrieval and numerical prediction. Mon. Wea. Rev., 130, 454476, doi:10.1175/1520-0493(2002)130<0454:ROMIFF>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Wu, B., , J. Verlinde, , and J. Sun, 2000: Dynamical and microphysical retrievals from Doppler radar observations of a deep convective cloud. J. Atmos. Sci., 57, 262283, doi:10.1175/1520-0469(2000)057<0262:DAMRFD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., , and J. Sun, 2007: Multiple-radar data assimilation and short-range quantitative precipitation forecasting of a squall line observed during IHOP_2002. Mon. Wea. Rev., 135, 33813404, doi:10.1175/MWR3471.1.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., , Y.-H. Kuo, , J. Sun, , W.-C. Lee, , E. Lim, , Y. Guo, , and D. M. Barker, 2005: Assimilation of Doppler radar observations with a regional 3DVAR system: Impact of Doppler velocities on forecasts of a heavy rainfall case. J. Appl. Meteor., 44, 768788, doi:10.1175/JAM2248.1.

    • Search Google Scholar
    • Export Citation
  • Yuter, S. E., , and R. A. Houze Jr., 1995: Three-dimensional kinematic and microphysical evolution of Florida cumulonimbus. Part II: Frequency distributions of vertical velocity, reflectivity, and differential reflectivity. Mon. Wea. Rev., 123, 19411963, doi:10.1175/1520-0493(1995)123<1941:TDKAME>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Zhang, F., , Y. Weng, , J. A. Sippel, , Z. Meng, , and C. H. Bishop, 2009: Cloud-resolving hurricane initialization and prediction through assimilation of Doppler radar observations with an ensemble Kalman filter. Mon. Wea. Rev., 137, 21052125, doi:10.1175/2009MWR2645.1.

    • Search Google Scholar
    • Export Citation
  • Zhao, Q., , J. Cook, , Q. Xu, , and P. R. Harasti, 2006: Using radar wind observations to improve mesoscale numerical weather prediction. Wea. Forecasting, 21, 502522, doi:10.1175/WAF936.1.

    • Search Google Scholar
    • Export Citation
Save