• Berre, L., 2000: Estimation of synoptic and mesoscale forecast error covariances in a limited area model. Mon. Wea. Rev., 128 , 644667.

    • Search Google Scholar
    • Export Citation
  • Berre, L., S. E. Stefánescu, and M. B. Pereira, 2006: The representative of the analysis effect in three error simulation techniques. Tellus, 58A , 196209.

    • Search Google Scholar
    • Export Citation
  • Bousquet, O., T. Montmerle, and P. Tabary, 2008: Using operationally synthesized multiple-Doppler winds for high resolution horizontal wind forecast verification. Geophys. Res. Lett., 35 , L10803. doi:10.1029/2008GL033975.

    • Search Google Scholar
    • Export Citation
  • Brousseau, P., and Coauthors, 2008: A prototype convective-scale data assimilation system for operation: The Arome-RUC. HIRLAM Tech. Rep. 68, 23–30.

    • Search Google Scholar
    • Export Citation
  • Browning, K., and R. Wexler, 1968: The determination of kinematic properties of a wind field using Doppler radar. J. Appl. Meteor., 7 , 105113.

    • Search Google Scholar
    • Export Citation
  • Cardinali, C., S. Pezzulli, and E. Andersson, 2004: Influence-matrix diagnostic of a data assimilation system. Quart. J. Roy. Meteor. Soc., 130 , 28272849.

    • Search Google Scholar
    • Export Citation
  • Caumont, O., and Coauthors, 2006: A radar simulator for high-resolution nonhydrostatic models. J. Atmos. Oceanic Technol., 23 , 10491067.

    • Search Google Scholar
    • Export Citation
  • Caumont, O., and V. Ducrocq, 2008: What should be considered when simulating Doppler velocities measured by ground-based weather radars? J. Appl. Meteor. Climatol., 47 , 22562262.

    • Search Google Scholar
    • Export Citation
  • Caya, A., J. Sun, and C. Snyder, 2005: A comparison between the 4DVAR and the ensemble Kalman filter techniques for radar data assimilation. Mon. Wea. Rev., 133 , 30813094.

    • Search Google Scholar
    • Export Citation
  • Crook, A., and J. Tuttle, 1994: Numerical simulations initialized with radar-derived winds. Part II: Forecasts of three gust-front cases. Mon. Wea. Rev., 122 , 12041217.

    • Search Google Scholar
    • Export Citation
  • Doviak, R., and D. Zrnic, 1993: Doppler Radar and Weather Observation. 2nd ed. Academic Press, 562 pp.

  • Fischer, C., T. Montmerle, L. Berre, L. Auger, and S. Stefanescu, 2005: An overview of the variational assimilation in the ALADIN/France numerical weather-prediction system wave–driven circulation of the mesosphere. Quart. J. Roy. Meteor. Soc., 131 , 34773492.

    • Search Google Scholar
    • Export Citation
  • Gao, J., and M. Xue, 2008: An efficient dual-resolution approach for ensemble data assimilation and tests with simulated Doppler radar data. Mon. Wea. Rev., 136 , 945963.

    • Search Google Scholar
    • Export Citation
  • Lafore, J., and Coauthors, 1998: The Meso-NH atmospheric simulation system. Part I: Adiabatic formulation and control simulations. Ann. Geophys., 16 , 90109.

    • Search Google Scholar
    • Export Citation
  • Lin, Y-L., R. Farley, and H. Orville, 1983: Bulk parameterization of the snow field in a cloud model. J. Climate Appl. Meteor., 22 , 10651092.

    • Search Google Scholar
    • Export Citation
  • Lindskog, M., K. Salonen, H. Järvinen, and D. Michelson, 2004: Doppler radar wind data assimilation with HIRLAM 3DVAR. Mon. Wea. Rev., 132 , 10811092.

    • Search Google Scholar
    • Export Citation
  • Montmerle, T., A. Caya, and I. Zawadzki, 2001: Simulation of a mid-latitude convective storm initialized with bistatic Doppler radar data. Mon. Wea. Rev., 129 , 19491967.

    • Search Google Scholar
    • Export Citation
  • ParentDuChâtelet, J., 2003: Aramis, le réseau français de radars pour la surveillance des précipitations (The French ARAMIS radar network for the sampling of precipitations). Meteorologie, 40 , 4452.

    • Search Google Scholar
    • Export Citation
  • Probert-Jones, J., 1962: The radar equation in meteorology. Quart. J. Roy. Meteor. Soc., 88 , 485495.

  • Salonen, K., H. Järvinen, and M. Lindskog, 2003: Model for Doppler radar radial winds. Preprints, 31st Conf. on Radar Meteorology, Seattle, WA, Amer. Meteor. Soc., 2B.1. [Available online at http://ams.confex.com/ams/pdfpapers/64105.pdf].

    • Search Google Scholar
    • Export Citation
  • Salonen, K., H. Järvinen, R. Eresmaa, and S. Niemelä, 2007: Bias estimation of Doppler-radar radial-wind observations. Quart. J. Roy. Meteor. Soc., 133 , 15011507.

    • Search Google Scholar
    • Export Citation
  • Sauvageot, H., 1992: Radar Meteorology. Artech House, 336 pp.

  • Snyder, C., and F. Zhang, 2003: Assimilation of simulated Doppler radar observations with an ensemble Kalman filter. Mon. Wea. Rev., 131 , 16631677.

    • Search Google Scholar
    • Export Citation
  • Sun, J., and N. 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.

    • Search Google Scholar
    • Export Citation
  • Sun, J., and N. Crook, 1998: Dynamical and microphysical retrieval from Doppler radar observations using a cloud model and its adjoint. Part II: Retrieval experiments of an observed Florida convective storm. J. Atmos. Sci., 55 , 835852.

    • Search Google Scholar
    • Export Citation
  • Tabary, P., 2007: The new French operational radar rainfall product. Part I: Methodology. Wea. Forecasting, 22 , 393427.

  • Tabary, P., L. Perier, J. Gagneux, and J. Parent-Du-Chatelet, 2005: Test of a staggered PRT scheme for the French radar network. J. Atmos. Oceanic Technol., 22 , 352364.

    • 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.

    • Search Google Scholar
    • Export Citation
  • Wattrelot, E., O. Caumont, S. Pradier-Vabre, M. Jurasek, and G. Haase, 2008: 1d + 3DVar assimilation of radar reflectivities in the pre-operational AROME model at Météo-France. Proc. Fifth Conf. on Radar in Meteorology and Hydrology. Helsinki, Finland, Finnish Meteorological Institute, 097. [Available online at http://erad2008.fmi.fi/proceedings/extended/erad2008-0097-extended.pdf].

    • Search Google Scholar
    • Export Citation
  • Weygandt, S., A. Shapiro, and 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.

    • 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.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., Y-H. Kuo, J. Sun, W-C. Lee, E. Lim, Y-R. Guo, and D. 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.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., Y-H. Kuo, J. Sun, W-C. Lee, D. Barker, and E. Lim, 2007: An approach of radar reflectivity data assimilation and its assessment with the inland QPF of Typhoon Rusa (2002) at landfall. J. Appl. Meteor. Climatol., 46 , 1422.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., and Coauthors, 2008: Doppler radar data assimilation in KMA’s operational forecasting. Bull. Amer. Meteor. Soc., 89 , 3943.

    • Search Google Scholar
    • Export Citation
  • Zhang, F., C. Snyder, and J. Sun, 2004: Impacts of initial estimate and observation availability on convective-scale data assimilation with an ensemble Kalman filter. Mon. Wea. Rev., 132 , 12381253.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Map of the French radar network ARAMIS over the AROME domain. Dashed lines denote radars that are not yet Doppler converted, numbers indicate the number of PPIs performed every 15 min. TRAP, ARCI, BOLL, TREI, ABBE, and AVES correspond to the radars of Trappes, Arcis, Bollène, Treillière, Abbeville, and Avesnes, France, respectively. The ring radius for each radar is ∼100 km.

  • View in gallery

    Example of radial velocities (m s−1) from the first elevation performed by the BOLL radar (Fig. 1): (top) raw data and after the application of (middle) median and (bottom) cleaner filters. Positive velocities point toward the radar.

  • View in gallery

    Radar geometry using the earth’s effective radius: ϕ is the azimuth angle from north N, θ is the beam elevation, α is the angle between the earth’s radius a and the target point, h is the antenna height, and d is the radar-target distance. The radial velocity Vr is the projection of the horizontal velocity Vh along the ray path.

  • View in gallery

    Innovation bias estimate for 4 months of wind speed (m s−1) and direction (°) for (left) TRAP and (right) TREI radars (see Fig. 1) deduced from the computations of VAD profiles of observed and simulated radial velocities. Squares are the number of observations used for the computation (top axis).

  • View in gallery

    Example of the daily monitoring of radial velocities for two different radars: (top) TRAP and (bottom) ARCI (see Fig. 1), in November 2007. The black and light gray histograms denote the number of observations before and after screening (values on the right axis), the gray histogram shows the number of elevations considered (values on the left axis), and the plain black line is the radial velocities standard deviation computed within segments of 10° of azimuth and then averaged over all elevations (values on the left axis).

  • View in gallery

    Time series of analysis bias (solid lines) and standard deviations (dashed lines) for CNTRL (gray) and RADAR (black) against ground based measurements from 15 Nov to 10 Dec 2007 for the (top) wind speed, (middle) wind direction, and (bottom) relative humidity. The black histograms show the number of observations used for the statistics (right y axis).

  • View in gallery

    Time series of POD values for 3-h forecast of accumulated rain for CNTRL (gray line) and RADAR (black line) for 3 thresholds (from top to bottom: 0.5, 2, and 5 mm h−1). PODs are computed against rain gauge measurements over the AROME domain from 0000 UTC 30 Nov to 0000 UTC 5 Dec 2007. Histogram shows the number of rain gauge measurements taken into account in the computations (right y axis).

  • View in gallery

    Accumulated rain (top) POD and (bottom) FAR computed at different thresholds (x axis) for 3-h forecasts performed by RADAR (black) and CNTRL (gray) averaged from 0000 UTC 30 Nov to 0000 UTC 5 Dec 2007. A good forecast is characterized by a POD and FAR close to 1 and 0, respectively.

  • View in gallery

    Zoom of horizontal cross sections at 950 hPa of the divergence field for (a) CNTRL and (b) RADAR [contours are in 10−5 s−1, negative values (convergence) are plotted in dashed contours], and (c) rain rates (mm h−1; light blue and yellow showing rates around 10 and 50 mm h−1, respectively) at 1800 UTC 8 Nov 2007. In (b), dots correspond to active profile of radial velocities that are taken into account in the analysis, red squares show the location of the Doppler radars, and the black box corresponds to the area where the wind displayed in Fig. 10 is averaged.

  • View in gallery

    Hodograph representing the horizontal wind (m s−1) at different pressure levels (numbers, hPa) averaged within the black box displayed in Figs. 9 for CNTRL and RADAR.

  • View in gallery

    Simulated rain rates (mm h−1) after (left) 1 and (right) 2 h of forecast for (top) CNTRL and (middle) RADAR experiments using their respective 1800 UTC analyses on 8 Nov 2007. (bottom) The corresponding observed rain rates (mm h−1) using the same color scale.

  • View in gallery

    QPF scores against rain gauge data for the 6-h accumulated rainfall forecast from 1800 UTC 8 Nov 2007 analyses by RADAR (solid) and CNTRL (dashed). The x and y axes show POD and FAR scores for different thresholds (numbers), respectively. Values between parentheses denote the number of samples considered in the computations. The closer to 1 and to 0 for the POD and FAR, respectively, the better the forecast.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 74 65 6
PDF Downloads 57 50 1

Mesoscale Assimilation of Radial Velocities from Doppler Radars in a Preoperational Framework

View More View Less
  • 1 Centre National de Recherches Météorologiques, Toulouse, France
© Get Permissions
Full access

Abstract

This paper presents the results of a preoperational assimilation of radial velocities from Doppler radars of the French Application Radar la Météorologie InfraSynoptique (ARAMIS) network in the nonhydrostatic model, the Application of Research to Operations at Mesoscale (AROME). For this purpose, an observation operator, which allows the simulation of radial winds from the model variables, is included in the three-dimensional variational data assimilation (3DVAR) system. Several data preprocessing procedures are applied to avoid as much as possible erroneous measurements (e.g., due to dealiasing failures) from entering the minimization process. Quality checks and other screening procedures are discussed. Daily monitoring diagnostics are developed to check the status and the quality of the observations against their simulated counterparts. Innovation biases in amplitude and in direction are studied by comparing observed and simulated velocity–azimuth display (VAD) profiles. Experiments over 1 month are performed. Positive impacts on the analyses and on precipitation forecasts are found. Scores against conventional data show mostly neutral results because of the much-localized impact of radial velocities in space and in time. Significant improvements of low-level divergence analysis and on the resulting forecast are found when specific sampling conditions are met: the closeness of convective systems to radars and the orientation of the low-level horizontal wind gradient with respect to the radar beam. Focus on a frontal rainband case study is performed to illustrate this point.

Corresponding author address: Thibaut Montmerle, Météo-France/CNRM/GMAP, 42 av. G. Coriolis, Toulouse 31057, France. Email: thibaut.montmerle@meteo.fr

Abstract

This paper presents the results of a preoperational assimilation of radial velocities from Doppler radars of the French Application Radar la Météorologie InfraSynoptique (ARAMIS) network in the nonhydrostatic model, the Application of Research to Operations at Mesoscale (AROME). For this purpose, an observation operator, which allows the simulation of radial winds from the model variables, is included in the three-dimensional variational data assimilation (3DVAR) system. Several data preprocessing procedures are applied to avoid as much as possible erroneous measurements (e.g., due to dealiasing failures) from entering the minimization process. Quality checks and other screening procedures are discussed. Daily monitoring diagnostics are developed to check the status and the quality of the observations against their simulated counterparts. Innovation biases in amplitude and in direction are studied by comparing observed and simulated velocity–azimuth display (VAD) profiles. Experiments over 1 month are performed. Positive impacts on the analyses and on precipitation forecasts are found. Scores against conventional data show mostly neutral results because of the much-localized impact of radial velocities in space and in time. Significant improvements of low-level divergence analysis and on the resulting forecast are found when specific sampling conditions are met: the closeness of convective systems to radars and the orientation of the low-level horizontal wind gradient with respect to the radar beam. Focus on a frontal rainband case study is performed to illustrate this point.

Corresponding author address: Thibaut Montmerle, Météo-France/CNRM/GMAP, 42 av. G. Coriolis, Toulouse 31057, France. Email: thibaut.montmerle@meteo.fr

1. Introduction

To improve local meteorological forecasts of potentially dangerous events, Météo-France is currently developing a system based on the Cloud Resolving Model (CRM) in the nonhydrostatic model, Application of Research to Operations at Mesoscale (AROME) with a high spatial resolution over France. The impact of radial velocities from French Doppler radars will be addressed in this context.

Since volumetric-scanning Doppler radars are, at present, the only observing system capable of sampling the air circulation within precipitating systems (and even sometimes in clear-air conditions within the boundary layer, depending on atmospheric conditions) with a high horizontal and temporal resolution, research efforts have been carried out in recent years to assimilate those measurements in CRMs. Most of these studies have been performed in research mode on specific convective case studies. The seminal papers of Sun and Crook (1997, 1998) presents the use of a four-dimensional variational data assimilation (4DVAR) to provide initial fields balanced with a CRM by minimizing the difference between real and simulated radar observations (radial velocities and reflectivities) of a deep convective storm. Although encouraging results were obtained, a number of difficulties, mainly linked to the moist adjoint retrieval model, have been pointed out: computational expenses, linearizations of the microphysical scheme to construct accurate tangent linear and adjoint models, and specification of background error statistics. To circumvent these points, several methods have been developed to retrieve, at first, dynamical and thermodynamical fields from a time series of multiple (Lin et al. 1983), bistatic (Montmerle et al. 2001), or single (Crook and Tuttle 1994; Weygandt et al. 2002) Doppler radar data. The retrieved fields are then used as initial conditions for CRM. Such methods have shown promising results for short-term (e.g., less than 1 h) precipitation forecast, but they aim mainly to retrieve variables that are not directly described by radar in quasi–real time (e.g., the thermodynamical and the microphysical variables). To address the problem of the background error statistics specification pointed out in the 4DVAR techniques, the (ensemble Kalman filter) EnKF-based methods have also been applied to Doppler radar data describing convective storms (Snyder and Zhang 2003; Zhang et al. 2004; Tong and Xue 2005; Gao and Xue 2008). Indeed, these methods use flow-dependent model error statistics, by iteratively integrating in time an ensemble of model states initially generated using a Monte Carlo method. EnKF and 4DVAR approaches have been recently compared for a simulated case by Caya et al. (2005) and similar performances have been found. Implementing such methods in a cloud-resolving model within an operational framework are, however, difficult, mainly because of their large computational cost and of the requested very short assimilation window (between 5 and 10 min in the cited papers). Forecasts performed by CRMs are indeed time expensive and, for instance, AROME, in its 3-h Rapid Update Cycle (RUC) configuration, considers many other data types, such as ground-based data, radiosoundings, and satellite observations, which requires a cutoff time of 1.5 h (Brousseau et al. 2008).

In this context, the three-dimensional variational data assimilation (3DVAR) approach seems very attractive, especially for Doppler radar data. As a matter of fact, as in the 4DVAR, the 3DVAR integrates model information through continuous cycling and has a much lower computational cost. Such attempts have been made by Lindskog et al. (2004) who tested the impact of the assimilation of radial wind “superobservations” and velocity–azimuth display (VAD) profiles, deduced from the Swedish radar network observations, using the 3DVAR of the hydrostatic High-Resolution Limited-Area Model (HIRLAM). A small positive impact on verification scores has been found on temperature and on wind fields during a 10-day assimilation cycle experiment. Xiao et al. (2005) have shown some improvement in rainfall forecasts for two particular convective cases observed over Korea, using the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (NCAR) Mesoscale Model (MM5) and its 3DVAR on single-Doppler data sampled by one radar performing volumetric scans. Then this approach has been improved in order to include reflectivity in the assimilation process. The new configuration considers the rainwater mixing ratio as a control variable and warm rain process in the 3DVAR (Xiao et al. 2007). These new features have been imported in the 3DVAR of the Weather Research and Forecasting (WRF) modeling system (Xiao and Sun 2007). Two cycled assimilations of multiple Doppler radar data have been successfully applied in that latter study to simulate a squall line observed in the southern Great Plains. This technique has then been implemented operationally in Korea (Xiao et al. 2008) on observations from four Doppler radars. Statistically, a significant positive impact of the assimilation of such data on the short-range quantitative precipitation forecast (QPF) has been found. These promising results and the fact that metropolitan France is nowadays well covered by the 24 radars that form the Application Radar la Météorologie InfraSynoptique (ARAMIS) network, 17 of which with Doppler capacity, has encouraged us to take advantage of this huge amount of information for our numerical weather prediction (NWP) system at the convective scale, based on the AROME model. Therefore, the main objective of this study is to evaluate the impact of radial velocities sampled by these 17 volumetric-scanning radars, on analyses, on forecast statistics, and on QPF scores using AROME 3DVAR in a preoperational framework. In parallel, some work is going on to assimilate reflectivity in AROME (Wattrelot et al. 2008), but the results are not addressed herein.

This manuscript is organized as follows. The French radar network ARAMIS and the different preprocessings applied to the raw Doppler wind data are presented in section 2. Details about the model and the 3DVAR system, as well as discussions on data monitoring, are given in section 3. Section 4 presents results of the experimental preoperational model over 25 days of continuous cycled assimilations. Emphasis is made on one typical case study in order to point out the potential impact of radial velocities on analyses and on the resulting forecast. The last section provides a summary and draws some perspectives.

2. The radar data

a. The French radar network

The ARAMIS network (ParentDuChâtelet 2003) is composed of 24 C- and S-band radars that cover metropolitan France and Corsica (Fig. 1), using wavelengths of 5.6 and 2.7 GHz, respectively. In total, 17 of them are Doppler, and the remaining will be Doppler radars by the end of 2008. Each radar makes a complete volume scan every 15 min, performing different plan position indicators (PPIs) for a minimum of 2 elevation angles to a maximum of 11, priority being given to low angles to sample the boundary layer. The 15-min duration of the volume scan has been chosen to limit mechanical maintenance costs and to ensure data quality. Because of the signal processing, degradations of the Doppler and of the dual-polarization products can indeed occur with higher rotation speeds. The spatial resolution of radar measurements is of 1 km, within an area of 250 km of radius. As the beam broadening makes their reliability decreasing with distance, only measurements performed 150 km from the antenna are considered in this study.

Raw data are available online in binary universal form of representation of meteorological data (BUFR) format, with a delay of 1 or 2 h. Each file contains values for one elevation of reflectivity, radial wind (for data coming from a Doppler radar), and data status. Twelve categories of data status are possible, including clutter, clear-sky echoes, sea clutter, or hydrometeor type. The latter status is provided in expectation of the gradual setup of the dual-polarization capability for the whole network in the near future.

b. Doppler wind preprocessing

Before their assimilation into 3DVAR, three main steps of preprocessing are undertaken: 1) velocity dealiasing, 2) removal of noise and unrealistic echoes, and 3) data setup.

The first point is linked to the fact that the maximum speed recorded without ambiguity (e.g., the Nyquist velocity) is inversely dependent on the radar’s pulse repetition time (PRT). As a consequence, an increase of the Nyquist velocity reduces the distance between targets (Doppler dilemma) and consequently the range of measurements. To solve this problem, the staggered PRT approach (Tabary et al. 2005), which retrieves the radial velocity as the result of a dealiasing process based on different Nyquist velocities (corresponding to different PRTs), is used. For the ARAMIS network, three different PRTs are considered and the extended Nyquist velocity measured without uncertainty is ±60 m s−1. In the case of highly turbulent flows associated with large Doppler power spectrum widths, an important amount of dealiasing algorithm failures can, however, be noticed. Different filtering procedures are thus mandatory to accomplish step 2, as unrealistic data may potentially have negative effects on wind analysis.

First of all, echoes classified as ground or sea clutters are excluded. Clear-sky measurements, due to several nonmeteorological targets such as insects or dust, are also not considered at present.

Two different filters are then applied to identify and to correct, as much as possible, the failures of the dealiasing algorithm. The first one is a median filter applied to boxes of 5 × 5 pixels, which removes almost all the spurious echoes. Figure 2 (middle) shows an example of the effect of this filtering procedure on raw data (Fig. 2, top). This picture also shows that unrealistic echoes may remain after this filtering (red spots), mainly in highly turbulent areas where the Doppler power spectrum width is broadening. To completely remove these unwanted echoes, a second filter, called a cleaner is applied. This nonlinear filter shares the same principle of ordered values than the median filter: each observation i is the center of a 5 × 5 gridpoint box. Differences between i and the remaining N points are computed and ordered. The mean of the two (for an even number of data into the box) or three (for an odd number of data) values at the middle of the ordered set D is calculated, with O being the mean of the corresponding observations. If the difference between i and O is larger than the fixed threshold of 10 m s−1, i is substituted by O. If not, the quality of the other N observations is evaluated comparing D to their difference to i. If this value is larger than 10 m s−1, the observation is removed. An example of a final result after this filtering is displayed in Fig. 2 (bottom).

Using a smaller threshold in the comparison with D, this cleaner filter has furthermore proven to be robust and efficient in testing the validity of the elevation before applying the median filter: if more than 85% of data fails the two successive tests, the entire elevation is rejected. Thus, an elevation can be discarded if no observation can be extracted from the raw data because either they are too noisy, or if the file contains just noise [typical from datasets affected by automatic frequency control problem (Sauvageot 1992)].

The remaining pixels are stored as vertical profiles of radial wind, the vertical profiles being produced by stacking measurements at the same location, but at different elevations. Since these profiles have a spatial resolution (1 km) higher than AROME’s grid (2.5 km), and in order to save computer memory, their density is reduced and only one profile out of five is kept.

3. Doppler winds in AROME 3DVAR

After a brief description of the AROME model, this section aims to present how radial winds are managed in its 3DVAR assimilation system. The observation operator that simulates radial velocities from forecasted winds is first described. Discussions on measurement errors and on bias follow. Screening procedures and monitoring of the data quality are finally presented.

a. The AROME model

The AROME model is a nonhydrostatic model running at a horizontal resolution of 2.5 km over France (Fig. 1). AROME’s main aim is to improve the local meteorological forecasts of potentially dangerous convective events (e.g., storms, unexpected floods, wind bursts) and of lower-tropospheric phenomena (e.g., wind, temperature, turbulence, visibility).

AROME has inherited (i) the physical parameterizations from the nonhydrostatic Méso-NH model (Lafore et al. 1998), which considers, in particular, a comprehensive microphysical scheme based on five hydrometeors and a 3D turbulence scheme; and (ii) the dynamical core and the 3DVAR data assimilation system derived from the regional Aire Limitée Adaption Dynamique Développement International (ALADIN) system that has run operationally at Météo-France since 2005 with a 10-km horizontal resolution (Fischer et al. 2005). ALADIN also provides the lateral boundary conditions to AROME every 3 h.

AROME 3DVAR uses a specific error covariance matrix (i.e., the so-called 𝗕 matrix), which shares the same multivariate formulation as ALADIN-France (Berre 2000), using errors of vorticity, divergence, temperature, surface pressure, and humidity with scale-dependant statistical regressions to compute cross covariances. These statistics have been calculated using an ensemble-based method (Berre et al. 2006). AROME will hopefully run operationally by the end of 2008 using cycled assimilation/forecast steps, based on a specific 3-h RUC configuration [see Brousseau et al. (2008) for more details].

b. Radial wind observation operator

The 3DVAR assimilation system is updated in order to take into account radial velocities from the ARAMIS network. An observation operator, based on Salonen et al. (2003) and Caumont et al. (2006), is specifically developed in order to simulate radial wind measurements from AROME’s control variables. Its geometry is displayed in Fig. 3.

At first, a bilinear interpolation is used to compute the horizontal model wind Vh at the observation location:
i1520-0493-137-6-1939-e1
where the azimuth ϕ is the horizontal angle between north and the observation. Considering the surface curvature of the earth, the correct estimation of the radial component of Vh is
i1520-0493-137-6-1939-e2
where θ is the scanning elevation angle and α is the angular correction that takes into account the bending of the radar beam, due to the change of the atmospheric refractivity index. The expression of α is
i1520-0493-137-6-1939-e3
where d is the distance of each observation from the radar, h is the antenna height, and ae = 4a/3 is the effective radius of the earth. This assumption implies that the refractivity index varies linearly with the height (Doviak and Zrnic 1993), for example, the radar beam propagates in a straight line through the atmosphere (Fig. 3). As shown in Caumont et al. (2006), this hypothesis is valid in most cases when vertical gradients of humidity are not too strong. Furthermore, this representation of the bending beam is computationally much cheaper than a straightforward analytical computation of each radar gate height.

Equation (2) neglects the contribution of the vertical component of the wind, which also depends on the fall speed of hydrometeors. Caumont and Ducrocq (2008) showed that for relatively low elevation angles used by the radars of the ARAMIS network (14° maximum), the fall velocity can be neglected. Therefore, since the vertical component of the wind is usually smaller than the horizontal one, only the horizontal terms are considered in Eq. (2).

The broadening of the radar beam is taken into account in the operator, but only the main lobe is considered. It is represented by a Gaussian function (Probert-Jones 1962), which peaks at the observation height. The velocities are vertically averaged within the beam borders, using as upper and lower limits the 200-hPa level and the radar horizon, respectively. Finally, to perform the minimization, the tangent linear and the adjoint of the tangent linear of this radial wind observation operator are coded.

c. Radial wind innovation biases

As Salonen et al. (2007) have exposed, bias correction of radial wind innovations (e.g., observation–guess value) is not straightforward: by averaging in azimuth, systematic biases in amplitude and in direction can indeed be missed. Therefore, these biases have to be treated separately. In this paper, a method based on the rotation of the observed and simulated wind toward an arbitrary reference direction is used, in order to get nominal wind directions for both datasets. In a different way, we have chosen to make use of the VAD technique (Browning and Wexler 1968) on observed and simulated datasets, considering only data within a radius of 36 km around radars, in agreement with Tabary et al. (2005). Observed and modeled profiles of wind speed (FF) and direction (DD) can indeed be retrieved for each radar. Four months of computations, from September to December 2007, have been undertaken. VAD profiles are computed at each assimilation time (e.g., every 3 h), with a vertical resolution of 200 m. To reject empirically nonlinear measurements due to turbulence, convection, or the remaining unfolded data, radial velocities characterized with a 3 m s−1 standard deviation within the considered slice of altitude are not considered in the calculations.

Beside the obvious fact that the number of data taken into account during the retrieval depends on the number of sampled precipitating events, this number also depends on the number of elevations: the more PPIs are performed, the more the VAD at each altitude slice is statistically reliable.

Results for Trappes (TRAP) and Treillière (TREI), France, radars (see Fig. 1 for radar locations) are displayed in Fig. 4. Two opposite situations are illustrated: TRAP has 11 elevations of scanning, while there are only 2 for TREI. This difference in scanning strategies implies that TRAP considers approximately 5 times more observations than TREI, at each level, with a higher concentration at lower altitudes, for geometry reasons. As a consequence, the estimations of the wind profiles are statistically much less reliable for TREI than for TRAP. Nevertheless, small innovation biases compared to the maximum allowed (60 m s−1) recorded velocities, both on wind speed and on direction, are shown for both radars. Comparable results have been found for all radars (not shown). Overall, these results suggest that AROME’s 3-h forecasts give realistic linear wind estimation near radars and that radars do not present systematic errors. For these reasons, bias correction is not applied to radar wind data in this study.

d. Radial wind errors

The error for Doppler wind mainly relies on the quality of the phase measurement and on contaminations by moving targets that are not related to meteorological phenomena. Three main sources of errors can be highlighted:

  • The contamination by ground clutters. These clutters are characterized upstream using a dynamic identification scheme based on the analysis of the pulse-to-pulse fluctuation of the radar reflectivity (Tabary 2007). If this identification scheme succeeds, the data status is updated and the corresponding pixel is rejected.

  • The Doppler power spectral width related to each measurement, which depends on the atmospheric conditions (a steady atmosphere has a spectral width equal to zero). As discussed in section 3b, the spectral width broadening, due to strongly turbulent air conditions, often leads to failures in the dealiasing algorithm and consequently the appearance of noisy pixels. These errors are difficult to quantify and are indirectly managed through the different filters during the data preprocessing (as discussed in section 3b).

  • The distance of the target from the radar, which increases the probability that the radar beam is not homogeneously filled by meteorological targets because of its broadening. These errors are indirectly taken into account by the observation error that depends linearly on the distance d to the radar:
    i1520-0493-137-6-1939-e4

Using this formula, the observation error ranges from 1 to 3.4 m s−1 at 150 km from the radar. No cross correlations of observation errors between adjacent pixels are considered.

e. Data screening

Every observation of each profile is put through a data screening procedure. This process removes observations with innovation vectors larger than 20 m s−1. This quite large value is chosen as a compromise between (i) numerical stability in the minimization process and (ii) the possibility of repositioning misplaced convergence structures that are present in the background, and/or of analyzing unpredicted wind circulation sampled by radars, which is expected from radial wind assimilation.

The final step before minimization consists in performing an observation thinning to avoid as much as possible error correlations between adjacent pixels, in order to ensure the optimality of the system (cross correlation of observation errors being neglected, as seen in the previous section). As discussed in Cardinali et al. (2004), the direct consequence of this thinning is an increase of the individual impact of one single observation of radial wind in the final analysis. Data thinning is done by retaining the best profile within 15 × 15 km2 boxes, the criterion of selection being based on the distance from the radar [e.g., on the error variance σo(d)] and on the number of valid observations within each profile. Within each box, the selected profiles will thus have the best compromise between observation error and number of data per profile. The value of 15 km has been chosen according to the background error correlation lengths in the low troposphere, which correspond approximately to this value (see Brousseau et al. 2008).

f. Monitoring diagnostics

In an operational configuration, the quality of the data entering the system and the results of the minimization step has to be carefully checked by monitoring observations and their modeled counterparts. Such monitoring is also of great interest to weather radar users to evaluate the status of the entire Doppler radar network against model outputs in quasi–real time (i.e., 8 times a day).

Badly unfolded data, nonmeteorological echoes, or data sent by deficient radars have to be carefully identified and removed. As a matter of fact, even if the preprocessing and the quality control are quite efficient in removing unwanted observations, they do not make certain that the final dataset is free from spurious data. For this purpose, monitoring diagnostics, as those displayed in Fig. 5 for TRAP (top) and ARCI (bottom) radars over two weeks in November and December 2007, are developed. The displayed averaged standard deviation index (black line) is particularly robust and useful, since it gives a direct overview of the measurement quality. This index is given by
i1520-0493-137-6-1939-e5
where nelev is the number of PPIs and σnaz is the radial velocity variance within a slice of naz azimuth. A value of 10° for naz is chosen. Small σVr values (typically lower than 5 m s−1) and coherent ratio of data entering and exiting the screening procedure (as displayed by the histograms) indicate a correct behavior of the radar. High values of σVr, which correspond to uncommon meteorological situation, as those displayed for TRAP on at 1200 UTC 2 December, can, however, occur. In this particular case, it reflects the strong horizontal and vertical wind shears generated by a midlatitude storm. On other cases, however, high values of σVr indicates Doppler radar deficiencies. For instance, measurement problems are obviously displayed in the monitoring of Arcis (ARCI), France, which shows high standard deviation values for several successive assimilation times (Fig. 5, bottom), whereas adjacent radars display low values (not shown). Indeed, hardware failures related to the ARCI Doppler system were encountered at that time. As a consequence, this particular radar has been temporarily blacklisted.

Time series of observation − guess and observation − analysis biases and standard deviations performed for each elevations have been also elaborated to check the performances of the minimization process for each analysis time (not shown). Close to zero biases and the reduction of rms errors after the minimization have been displayed for each radar, which indicates correct behaviors of the assimilation process.

4. Assimilation experiments

The setup of the two experiments performed for this study are detailed in the first section. Forecast scores against conventional data, computed over a 3-week experiment, are then discussed. Focus is finally made for one particular case of frontal rainband in order to point out (i) how radial velocities impact the analyses and the resulting forecasts, and (ii) what are the limitations of using this kind of observation in a NWP system.

a. Experimental setup

An experimental preoperational configuration of AROME with assimilation of radial velocities is tested for 25 days, from 15 November to 10 December 2007. Two experiments are prepared for this purpose:

  • CNTRL performs two forecasts of 24 and 12 h at 0000 and 1200 UTC, respectively, starting from analyses obtained using AROME 3DVAR in a 3-h RUC mode (e.g., a 3-h forecast is used as background field for the next assimilation step and so on). The same observation types as the French operational ALADIN 3DVAR are considered [conventional observations, temperature, and humidity at 2 m, wind at 10 m, radiances from Advanced Television and Infrared Observation Satellite (TIROS) Operational Vertical Sounder (ATOVS), Spinning Enhanced Visible and Infrared Imager (SEVIRI), and Special Sensor Microwave Imager (SSM/I) instruments, winds from atmospheric moving vectors (AMV), and scatterometers]. The ALADIN model provides lateral boundary conditions every 3 h.

  • RADAR shares the same configuration as CNTRL, but includes radial velocities from 17 volumetric-scanning radars as additional data, following the procedures described in the previous sections.

b. Forecast scores

The evaluation of the impact of radial winds on forecasts against conventional data can be somewhat difficult, radial wind observations only being available if precipitating clouds occur near radars. As a consequence, the signal brought by radial velocities is smoothed when averaged over the whole domain and over a long period of time, which partly explains the relatively neutral scores against conventional data. Such scores against ground-based measurements indeed show no significant improvements for temperature and pressure at the analysis time, and at all forecast ranges for all prognostic variables (not shown). However, except for the standard deviations of the wind direction that shows more variability, Fig. 6 shows positive scores for RADAR in comparison with CNTRL at the analysis time; smaller biases and a slight reduction of standard deviations during most of the entire evaluation period for both wind and humidity are displayed.

More signals are found in forecast scores against rain gauge measurements, where positive impact is found for many cases. For instance, 6-day QPF scores computed for the 3-h RUC forecasts for different precipitation thresholds are plotted in Fig. 7. A couple of active lows, accompanied by important precipitations and strong winds, were crossing mainland France at the time. When including radial winds, the detection skills, measured by the probability of detection (POD), are improved for several analysis times and for different thresholds. QPF scores averaged over the whole period (Fig. 8) confirm this improvement, except for the 15 mm h−1 thresholds for which only few samples were taken into account in the computation. In most of cases, the RADAR experiment shows an increase of false-alarm rate (FAR) for large thresholds. This characteristic is noticed for other time periods, but the low number of encountered samples makes it difficult to draw any conclusion.

As will be described in detail in the next section, where focus is made on a particular case study, precipitation forecast improvements generally occur when the onset of a potentially raining structure is well detected through the sampling of pronounced low-level convergence structures, and/or when the vertical shear of horizontal wind is well depicted. These conditions are the key factors to correctly reproduce suitable initial conditions, which allow a realistic development of convection at mesoscale. On the contrary, degradations are often observed in areas not covered by radars, where the background could not be corrected in a realistic way. Thus, the Doppler conversion of the few non-Doppler remaining radars of ARAMIS (cf. Fig. 1), and the evolution of scanning strategies toward a more volumetric scanning of the troposphere in a near future should hopefully improve results.

c. Case study: Frontal rainband

At around 1800 UTC 8 November 2007, a frontal rainband, resulting from an active cold front located over the south of the United Kingdom gathering humid boundary layers over the Channel, approached the northern French coast. Figure 9c shows the corresponding composite pattern of rain rates. Since no precipitating events occurred over the simulation domain for several assimilation times before the 1800 UTC analysis, background fields were almost identical for both experiments (not shown). Thus, the added value of radial velocities on wind analysis is obvious for this case: RADAR shows a narrow line of convergence (Fig. 9b) that coincides well with the maximum of observed reflectivity, while convergence is much less organized for CNTRL (Fig. 9a). Bousquet et al. (2008) have compared such kind of low-level convergence structures produced by AROME 3DVAR with results given by an independent multiple Doppler retrieval. Similar features and values are found, which somehow validate AROME’s analyses. In their case study, the assimilation of radial velocities has, in particular, permitted to shift a misplaced squall line to a more realistic location.

Higher elevations from Abbeville (ABBE) and Avesnes (AVES), France, radars (corresponding to the two red squares at the top of Fig. 9b, see Fig. 1 for radar’s location) also impact the wind analysis up to the midtroposphere (Fig. 10). Mean winds computed within the displayed boxes show indeed an increase in the vertical shear of the horizontal wind from 1000 to 600 hPa for RADAR, which favor the development of a more organized convection. As a consequence, more realistic features are found for RADAR for the cumulated rainfalls (Figs. 11). The assimilation of radial wind produces more intense and better structured precipitations during the first hours of the simulation, through the triggering of convective cells produced by the initially more realistic low-level convergence structures. For this particular case, QPF scores confirm the positive impact of the assimilation of radial velocities for thresholds below 10 mm h−1 during the first 6 h of simulation (Fig. 12), especially for the 5 mm h−1 threshold for which the skill of detection almost doubles with less false alarm.

Since the 3DVAR technique has limited ability in retrieving the tangential wind, Doppler wind measurements efficiently capture the low-level convergence structures localized near radars when the maximum gradient of horizontal wind is oriented roughly parallel to the ray path. The resulting wind analysis becomes more realistic and has a direct positive effect on forecasts. On the contrary, when the gradient of horizontal wind is oriented more perpendicular to the ray path (typically when a convergence line passes along right over the radar) and, obviously, when convection occurs too far from radars to allow the sampling of boundary layer dynamics, the impact of Doppler winds on the analyses and on the resulting forecasts is less obvious. Therefore, the timing between optimal observation and assimilation time plays a key role in the resulting forecast.

5. Conclusions

In recent years, Météo-France has built a comprehensive network of Doppler radars over France. Such Doppler radars give, when performing volumetric scans and combined to an a priori estimation of the wind field such as a model output, access to detailed 3D air circulation within precipitating systems. This ability is of great interest for weather forecasting at the mesoscale, since no other observation type, currently used in operational NWP models, has this capability. As a consequence, efforts are undertaken to include radial velocities in the 3DVAR assimilation system of the nonhydrostatic AROME model, which will be operational over France by the end of 2008, with a 2.5-km horizontal resolution.

Several procedures are applied to raw data in order to avoid as much as possible erroneous measurements entering the minimization. Dealiasing failures and/or noisy data are managed through successive filtering steps. Monitoring diagnostics are developed to routinely detect potential anomalies in the measurements performed by each radar.

In parallel, an observation operator, which allows us to simulate radial winds measurements from the model horizontal wind, is developed. Then, the quality of observed winds is evaluated against its simulated counterpart. Innovation biases in wind amplitude and in direction are inferred from the computations of observed and simulated VAD profiles deduced from 4 months of analyses. Statistically stable results are obtained for radars performing at least five PPIs per scanning cycle. Nonsignificant biases are found for all radars, showing that (i) AROME’s 3-h forecasts give realistic linear wind estimation in the vicinity of radars; (ii) radars are well calibrated; and (iii) the dealiasing algorithm, based on a triple-PRT scheme, coupled with the filtering procedures, gives satisfactory results compared to model outputs. Thinning within 15 × 15 km2 boxes is applied to avoid correlations of observation errors between adjacent pixels. Finally, the remaining vertical profiles of radial winds are weighted in the minimization by an observation error variance proportional to their distance from the radar, in order to take into account the decrease of the data reliability due to beam broadening.

Two experiments, based on the preoperational version of AROME, are performed during a 3-week period, in order to quantify the impact of the assimilation of Doppler winds on short- to medium-range weather forecasts at the mesoscale. Radial velocities are very useful to analyze structures of low-level convergence associated with convective systems, leading to more structured and more realistic precipitation forecasts. Optimal results are obtained when the maximum gradient of horizontal wind within the convective system is oriented toward the radar. In other cases, the dynamics of the sampled radial velocities is weak and the resulting wind analysis could not be improved. The timing of the radars sampling with respect to the assimilation times is thus a key factor for a potentially positive impact on forecasts. Furthermore, for radial winds having a very local effect on wind analyses and measurements being limited to precipitating areas, impacts on forecast scores against conventional data on large samples in space and in time are mostly neutral. However, improvements against surface measurements for the analysis of wind and humidity are noticeable. Scores on precipitations, such as classical QPF scores, have more signals and show clear improvements of the probability of detection for many cases.

These promising results have encouraged the incorporation of radial velocities in the preoperational suite of the AROME system since April 2008. Impacts on analyses and forecasts are henceforth evaluated on a daily basis. Monitoring diagnostics are still under progress in order to routinely summarize the information brought by the huge amount of Doppler radar data entering the assimilation process. Having the opportunity to evaluate observed radial winds against their model counterparts every 3 h is also of great interest for radarists: corrupted elevations, hardware problems, and other deficiencies can indeed be quickly depicted. The promising results of this study are an incentive to carry on the Dopplerization of the whole ARAMIS network and the progressive addition of new elevations, which will hopefully happen before the end of 2008. Evaluation of this new information in our NWP system will be undertaken as soon as they become operational.

Acknowledgments

The work of Claudia Faccani has been founded by the European FLYSAFE Project. The authors are also thankful to Jean Maziejewski for his careful reading of this manuscript.

REFERENCES

  • Berre, L., 2000: Estimation of synoptic and mesoscale forecast error covariances in a limited area model. Mon. Wea. Rev., 128 , 644667.

    • Search Google Scholar
    • Export Citation
  • Berre, L., S. E. Stefánescu, and M. B. Pereira, 2006: The representative of the analysis effect in three error simulation techniques. Tellus, 58A , 196209.

    • Search Google Scholar
    • Export Citation
  • Bousquet, O., T. Montmerle, and P. Tabary, 2008: Using operationally synthesized multiple-Doppler winds for high resolution horizontal wind forecast verification. Geophys. Res. Lett., 35 , L10803. doi:10.1029/2008GL033975.

    • Search Google Scholar
    • Export Citation
  • Brousseau, P., and Coauthors, 2008: A prototype convective-scale data assimilation system for operation: The Arome-RUC. HIRLAM Tech. Rep. 68, 23–30.

    • Search Google Scholar
    • Export Citation
  • Browning, K., and R. Wexler, 1968: The determination of kinematic properties of a wind field using Doppler radar. J. Appl. Meteor., 7 , 105113.

    • Search Google Scholar
    • Export Citation
  • Cardinali, C., S. Pezzulli, and E. Andersson, 2004: Influence-matrix diagnostic of a data assimilation system. Quart. J. Roy. Meteor. Soc., 130 , 28272849.

    • Search Google Scholar
    • Export Citation
  • Caumont, O., and Coauthors, 2006: A radar simulator for high-resolution nonhydrostatic models. J. Atmos. Oceanic Technol., 23 , 10491067.

    • Search Google Scholar
    • Export Citation
  • Caumont, O., and V. Ducrocq, 2008: What should be considered when simulating Doppler velocities measured by ground-based weather radars? J. Appl. Meteor. Climatol., 47 , 22562262.

    • Search Google Scholar
    • Export Citation
  • Caya, A., J. Sun, and C. Snyder, 2005: A comparison between the 4DVAR and the ensemble Kalman filter techniques for radar data assimilation. Mon. Wea. Rev., 133 , 30813094.

    • Search Google Scholar
    • Export Citation
  • Crook, A., and J. Tuttle, 1994: Numerical simulations initialized with radar-derived winds. Part II: Forecasts of three gust-front cases. Mon. Wea. Rev., 122 , 12041217.

    • Search Google Scholar
    • Export Citation
  • Doviak, R., and D. Zrnic, 1993: Doppler Radar and Weather Observation. 2nd ed. Academic Press, 562 pp.

  • Fischer, C., T. Montmerle, L. Berre, L. Auger, and S. Stefanescu, 2005: An overview of the variational assimilation in the ALADIN/France numerical weather-prediction system wave–driven circulation of the mesosphere. Quart. J. Roy. Meteor. Soc., 131 , 34773492.

    • Search Google Scholar
    • Export Citation
  • Gao, J., and M. Xue, 2008: An efficient dual-resolution approach for ensemble data assimilation and tests with simulated Doppler radar data. Mon. Wea. Rev., 136 , 945963.

    • Search Google Scholar
    • Export Citation
  • Lafore, J., and Coauthors, 1998: The Meso-NH atmospheric simulation system. Part I: Adiabatic formulation and control simulations. Ann. Geophys., 16 , 90109.

    • Search Google Scholar
    • Export Citation
  • Lin, Y-L., R. Farley, and H. Orville, 1983: Bulk parameterization of the snow field in a cloud model. J. Climate Appl. Meteor., 22 , 10651092.

    • Search Google Scholar
    • Export Citation
  • Lindskog, M., K. Salonen, H. Järvinen, and D. Michelson, 2004: Doppler radar wind data assimilation with HIRLAM 3DVAR. Mon. Wea. Rev., 132 , 10811092.

    • Search Google Scholar
    • Export Citation
  • Montmerle, T., A. Caya, and I. Zawadzki, 2001: Simulation of a mid-latitude convective storm initialized with bistatic Doppler radar data. Mon. Wea. Rev., 129 , 19491967.

    • Search Google Scholar
    • Export Citation
  • ParentDuChâtelet, J., 2003: Aramis, le réseau français de radars pour la surveillance des précipitations (The French ARAMIS radar network for the sampling of precipitations). Meteorologie, 40 , 4452.

    • Search Google Scholar
    • Export Citation
  • Probert-Jones, J., 1962: The radar equation in meteorology. Quart. J. Roy. Meteor. Soc., 88 , 485495.

  • Salonen, K., H. Järvinen, and M. Lindskog, 2003: Model for Doppler radar radial winds. Preprints, 31st Conf. on Radar Meteorology, Seattle, WA, Amer. Meteor. Soc., 2B.1. [Available online at http://ams.confex.com/ams/pdfpapers/64105.pdf].

    • Search Google Scholar
    • Export Citation
  • Salonen, K., H. Järvinen, R. Eresmaa, and S. Niemelä, 2007: Bias estimation of Doppler-radar radial-wind observations. Quart. J. Roy. Meteor. Soc., 133 , 15011507.

    • Search Google Scholar
    • Export Citation
  • Sauvageot, H., 1992: Radar Meteorology. Artech House, 336 pp.

  • Snyder, C., and F. Zhang, 2003: Assimilation of simulated Doppler radar observations with an ensemble Kalman filter. Mon. Wea. Rev., 131 , 16631677.

    • Search Google Scholar
    • Export Citation
  • Sun, J., and N. 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.

    • Search Google Scholar
    • Export Citation
  • Sun, J., and N. Crook, 1998: Dynamical and microphysical retrieval from Doppler radar observations using a cloud model and its adjoint. Part II: Retrieval experiments of an observed Florida convective storm. J. Atmos. Sci., 55 , 835852.

    • Search Google Scholar
    • Export Citation
  • Tabary, P., 2007: The new French operational radar rainfall product. Part I: Methodology. Wea. Forecasting, 22 , 393427.

  • Tabary, P., L. Perier, J. Gagneux, and J. Parent-Du-Chatelet, 2005: Test of a staggered PRT scheme for the French radar network. J. Atmos. Oceanic Technol., 22 , 352364.

    • 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.

    • Search Google Scholar
    • Export Citation
  • Wattrelot, E., O. Caumont, S. Pradier-Vabre, M. Jurasek, and G. Haase, 2008: 1d + 3DVar assimilation of radar reflectivities in the pre-operational AROME model at Météo-France. Proc. Fifth Conf. on Radar in Meteorology and Hydrology. Helsinki, Finland, Finnish Meteorological Institute, 097. [Available online at http://erad2008.fmi.fi/proceedings/extended/erad2008-0097-extended.pdf].

    • Search Google Scholar
    • Export Citation
  • Weygandt, S., A. Shapiro, and 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.

    • 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.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., Y-H. Kuo, J. Sun, W-C. Lee, E. Lim, Y-R. Guo, and D. 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.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., Y-H. Kuo, J. Sun, W-C. Lee, D. Barker, and E. Lim, 2007: An approach of radar reflectivity data assimilation and its assessment with the inland QPF of Typhoon Rusa (2002) at landfall. J. Appl. Meteor. Climatol., 46 , 1422.

    • Search Google Scholar
    • Export Citation
  • Xiao, Q., and Coauthors, 2008: Doppler radar data assimilation in KMA’s operational forecasting. Bull. Amer. Meteor. Soc., 89 , 3943.

    • Search Google Scholar
    • Export Citation
  • Zhang, F., C. Snyder, and J. Sun, 2004: Impacts of initial estimate and observation availability on convective-scale data assimilation with an ensemble Kalman filter. Mon. Wea. Rev., 132 , 12381253.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Map of the French radar network ARAMIS over the AROME domain. Dashed lines denote radars that are not yet Doppler converted, numbers indicate the number of PPIs performed every 15 min. TRAP, ARCI, BOLL, TREI, ABBE, and AVES correspond to the radars of Trappes, Arcis, Bollène, Treillière, Abbeville, and Avesnes, France, respectively. The ring radius for each radar is ∼100 km.

Citation: Monthly Weather Review 137, 6; 10.1175/2008MWR2725.1

Fig. 2.
Fig. 2.

Example of radial velocities (m s−1) from the first elevation performed by the BOLL radar (Fig. 1): (top) raw data and after the application of (middle) median and (bottom) cleaner filters. Positive velocities point toward the radar.

Citation: Monthly Weather Review 137, 6; 10.1175/2008MWR2725.1

Fig. 3.
Fig. 3.

Radar geometry using the earth’s effective radius: ϕ is the azimuth angle from north N, θ is the beam elevation, α is the angle between the earth’s radius a and the target point, h is the antenna height, and d is the radar-target distance. The radial velocity Vr is the projection of the horizontal velocity Vh along the ray path.

Citation: Monthly Weather Review 137, 6; 10.1175/2008MWR2725.1

Fig. 4.
Fig. 4.

Innovation bias estimate for 4 months of wind speed (m s−1) and direction (°) for (left) TRAP and (right) TREI radars (see Fig. 1) deduced from the computations of VAD profiles of observed and simulated radial velocities. Squares are the number of observations used for the computation (top axis).

Citation: Monthly Weather Review 137, 6; 10.1175/2008MWR2725.1

Fig. 5.
Fig. 5.

Example of the daily monitoring of radial velocities for two different radars: (top) TRAP and (bottom) ARCI (see Fig. 1), in November 2007. The black and light gray histograms denote the number of observations before and after screening (values on the right axis), the gray histogram shows the number of elevations considered (values on the left axis), and the plain black line is the radial velocities standard deviation computed within segments of 10° of azimuth and then averaged over all elevations (values on the left axis).

Citation: Monthly Weather Review 137, 6; 10.1175/2008MWR2725.1

Fig. 6.
Fig. 6.

Time series of analysis bias (solid lines) and standard deviations (dashed lines) for CNTRL (gray) and RADAR (black) against ground based measurements from 15 Nov to 10 Dec 2007 for the (top) wind speed, (middle) wind direction, and (bottom) relative humidity. The black histograms show the number of observations used for the statistics (right y axis).

Citation: Monthly Weather Review 137, 6; 10.1175/2008MWR2725.1

Fig. 7.
Fig. 7.

Time series of POD values for 3-h forecast of accumulated rain for CNTRL (gray line) and RADAR (black line) for 3 thresholds (from top to bottom: 0.5, 2, and 5 mm h−1). PODs are computed against rain gauge measurements over the AROME domain from 0000 UTC 30 Nov to 0000 UTC 5 Dec 2007. Histogram shows the number of rain gauge measurements taken into account in the computations (right y axis).

Citation: Monthly Weather Review 137, 6; 10.1175/2008MWR2725.1

Fig. 8.
Fig. 8.

Accumulated rain (top) POD and (bottom) FAR computed at different thresholds (x axis) for 3-h forecasts performed by RADAR (black) and CNTRL (gray) averaged from 0000 UTC 30 Nov to 0000 UTC 5 Dec 2007. A good forecast is characterized by a POD and FAR close to 1 and 0, respectively.

Citation: Monthly Weather Review 137, 6; 10.1175/2008MWR2725.1

Fig. 9.
Fig. 9.

Zoom of horizontal cross sections at 950 hPa of the divergence field for (a) CNTRL and (b) RADAR [contours are in 10−5 s−1, negative values (convergence) are plotted in dashed contours], and (c) rain rates (mm h−1; light blue and yellow showing rates around 10 and 50 mm h−1, respectively) at 1800 UTC 8 Nov 2007. In (b), dots correspond to active profile of radial velocities that are taken into account in the analysis, red squares show the location of the Doppler radars, and the black box corresponds to the area where the wind displayed in Fig. 10 is averaged.

Citation: Monthly Weather Review 137, 6; 10.1175/2008MWR2725.1

Fig. 10.
Fig. 10.

Hodograph representing the horizontal wind (m s−1) at different pressure levels (numbers, hPa) averaged within the black box displayed in Figs. 9 for CNTRL and RADAR.

Citation: Monthly Weather Review 137, 6; 10.1175/2008MWR2725.1

Fig. 11.
Fig. 11.

Simulated rain rates (mm h−1) after (left) 1 and (right) 2 h of forecast for (top) CNTRL and (middle) RADAR experiments using their respective 1800 UTC analyses on 8 Nov 2007. (bottom) The corresponding observed rain rates (mm h−1) using the same color scale.

Citation: Monthly Weather Review 137, 6; 10.1175/2008MWR2725.1

Fig. 12.
Fig. 12.

QPF scores against rain gauge data for the 6-h accumulated rainfall forecast from 1800 UTC 8 Nov 2007 analyses by RADAR (solid) and CNTRL (dashed). The x and y axes show POD and FAR scores for different thresholds (numbers), respectively. Values between parentheses denote the number of samples considered in the computations. The closer to 1 and to 0 for the POD and FAR, respectively, the better the forecast.

Citation: Monthly Weather Review 137, 6; 10.1175/2008MWR2725.1

Save