• Bannister, R. N., 2008: A review of forecast error covariance in atmospheric variational data assimilation. I: Characteristics and measurements of forecast error covariances. Quart. J. Roy. Meteor. Soc., 134, 19511970, doi:10.1002/qj.339.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bech, J., U. Gjertsen, and G. Haase, 2007: Modelling weather radar beam propagation and topographical blockage at northern high latitudes. Quart. J. Roy. Meteor. Soc., 133, 11911204, doi:10.1002/qj.98.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Caumont, O., V. Ducrocq, E. Wattrelot, G. Jaubert, and S. Pradier-Vabre, 2010: 1D+3DVAR assimilation of radar reflectivity data: A proof of concept. Tellus, 62, 173187, doi:10.1111/j.1600-0870.2009.00430.x.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Haase, G., and T. Landelius, 2004: Dealiasing of Doppler radar velocities using a torus mapping. J. Atmos. Oceanic Technol., 21, 15661573, doi:10.1175/1520-0426(2004)021<1566:DODRVU>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Haiden, T., A. Kann, C. Wittman, G. Pistotnik, B. Bica, and C. Gruber, 2011: The Integrated Nowcasting through Comprehensive Analysis (INCA) system and its validation over the eastern alpine region. Wea. Forecasting, 26, 166183, doi:10.1175/2010WAF2222451.1.

    • Crossref
    • 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, doi:10.1175/1520-0493(2004)132<1081:DRWDAW>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Michelson, D., and A. Henja, 2012: Opera work package 3.6: Odyssey additions, task 3: Tuning and evaluation of “andre” tool. OPERA Tech. Rep. Working Doc. WD-2012-02c, 20 pp.

  • Montmerle, T., and C. Faccani, 2009: Mesoscale assimilation of radial velocities from Doppler radars in a preoperational framework. Mon. Wea. Rev., 137, 19391953, doi:10.1175/2008MWR2725.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Salonen, K., G. Haase, R. Eresmaa, R. Hohti, and H. Järvinen, 2011: Towards the operational use of Doppler radar radial winds in HIRLAM. Atmos. Res., 100, 190200, doi:10.1016/j.atmosres.2010.06.004.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Seity, Y., P. Brousseau, S. Malardel, G. Hello, P. Bernard, F. Bouttier, C. Lac, and V. Masson, 2011: The AROME-France convective-scale operational model. Mon. Wea. Rev., 139, 976991, doi:10.1175/2010MWR3425.1.

    • Crossref
    • 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. J. Atmos. Sci., 54, 16421661, doi:10.1175/1520-0469(1997)054<1642:DAMRFD>2.0.CO;2.

    • Crossref
    • 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.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, H., J. Sun, X. Zhang, X.-Y. Huang, and T. Auligné, 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.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wattrelot, E., O. Caumont, and J.-F. Mahfouf, 2014: Operational implementation of the 1D+3D-Var assimilation method of radar reflectivity data in the AROME model. Mon. Wea. Rev., 142, 18521873, doi:10.1175/MWR-D-13-00230.1.

    • Crossref
    • 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.

    • Crossref
    • 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, doi:10.1175/JAM2439.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • View in gallery

    Map representation of the current radar installations in Europe included in the OPERA network.

  • View in gallery

    (right) Example of a radar reflectivity scan quality controlled using the BALTRAD toolbox. (left) The original reflectivity field as it is delivered to OPERA. The color scale of the radar echoes ranges from 0 to 45 dBZ.

  • View in gallery

    Data flowchart for the radar data production chain used in the HIRLAM community.

  • View in gallery

    Radar coverage as it is used in the model. The model domain is also shown.

  • View in gallery

    Example from 0300 UTC 24 Aug 2014 of the radar reflectivity observations that enter the model after the preprocessing but before the final thinning and quality control (screening).

  • View in gallery

    Example from 0300 UTC 24 Aug 2014 of the radar reflectivity observations that enter the minimization, i.e., what is left after the final thinning and quality control (screening). Note that the markers are slightly larger than in Fig. 5 for display purposes.

  • View in gallery

    Difference of the humidity field at model level 50, ~850 hPa, between an analysis including radar reflectivity observations and one without radars starting from the same first-guess field. Solid lines indicate a moistening of the model, and dashed lines show that the model becomes dryer. The example is from 0300 UTC 24 Aug 2014.

  • View in gallery

    Verification against observations for relative humidity at the surface (2-m level) for two experiments: the reference experiment (red) and the experiment including reflectivity from radar observations (green). The upper two lines show the standard deviation, and the lower two show the systematic errors.

  • View in gallery

    As in Fig. 8, but against radiosondes at the 500-hPa level.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 179 87 7
PDF Downloads 111 36 4

Assimilation of Multinational Radar Reflectivity Data in a Mesoscale Model: A Proof of Concept

View More View Less
  • 1 Swedish Meteorological and Hydrological Institute, Norrköping, Sweden
  • | 2 Danish Meteorological Institute, Copenhagen, Denmark
Full access

Abstract

Radar reflectivity observations have proven to be beneficial for improving the skill of numerical weather prediction (NWP) models. A few countries around the world use radar reflectivity observations in their operational runs. The majority of experiments and usage are still only including the local radar observations from the country in which the forecasts are made. The model domains, on the other hand, cover areas far greater than this, and therefore observations from surrounding countries need to be included. As of today there is no central collection and redistribution of volume data in Europe. In recent years, there has been an initiative to collect and harmonize European radar observations, but the redistribution of data for this purpose has only been of centrally constructed composites. This study describes the efforts to collect volume reflectivity data from several data providers and make them available for use in an NWP model. A preprocessing of the reflectivity data has been set up to handle the different incoming data and to make a first data reduction for the NWP models to be able to include the new observations. Assimilation experiments have been performed that prove it is possible to assimilate operational radar reflectivity data from several countries, with a neutral to positive impact.

Denotes content that is immediately available upon publication as open access.

© 2017 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Martin Ridal, martin.ridal@smhi.se

Abstract

Radar reflectivity observations have proven to be beneficial for improving the skill of numerical weather prediction (NWP) models. A few countries around the world use radar reflectivity observations in their operational runs. The majority of experiments and usage are still only including the local radar observations from the country in which the forecasts are made. The model domains, on the other hand, cover areas far greater than this, and therefore observations from surrounding countries need to be included. As of today there is no central collection and redistribution of volume data in Europe. In recent years, there has been an initiative to collect and harmonize European radar observations, but the redistribution of data for this purpose has only been of centrally constructed composites. This study describes the efforts to collect volume reflectivity data from several data providers and make them available for use in an NWP model. A preprocessing of the reflectivity data has been set up to handle the different incoming data and to make a first data reduction for the NWP models to be able to include the new observations. Assimilation experiments have been performed that prove it is possible to assimilate operational radar reflectivity data from several countries, with a neutral to positive impact.

Denotes content that is immediately available upon publication as open access.

© 2017 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Martin Ridal, martin.ridal@smhi.se

1. Introduction

Improving the skill to forecast significant weather situations, such as heavy precipitation events and flash floods, has come into greater focus in recent years because of their impact on modern-day society through the threat to the infrastructure and human safety. High-resolution numerical weather prediction (NWP) models are increasingly important in forecasting such significant weather. The possibility to capture and correctly describe these situations in the initial conditions of the models depends greatly on frequent and detailed observations, as they are usually localized and fast moving. Examples of such observations are radar data, GNSS Global Navigation Satellite System (GNSS), Secondary Surveillance Radar in Selective Mode for Interrogation of Aircraft (MODE-S), and satellite observations.

Radar data provide a unique dataset that is high resolution both temporally as well as spatially and hence can provide information on a subkilometer scale with an interval of typically 10 min. Furthermore, these dense sets of observations provide both wind and humidity information that traditional observation networks cannot and that is of pivotal importance for a mesoscale weather model that tries to correctly capture the situations mentioned above.

Radar observations have proven to be beneficial to numerical weather prediction in several studies using the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (NCAR) Mesoscale Model (MM5; Xiao et al. 2007), the Weather Research and Forecasting (WRF) modeling system (e.g., Sun and Crook 1997; Xiao and Sun 2007; Wang et al. 2013), the High Resolution Limited Area Model (HIRLAM) (Lindskog et al. 2004; Salonen et al. 2011), and the Application of Research to Operations at Mesoscale (AROME) model (Wattrelot et al. 2014; Montmerle and Faccani 2009).

From radar observations there are two parameters that can be used for data assimilation in NWP models, radial velocity, and reflectivity. The reflectivity is the radar echo that is returned from any object in the line of sight of the radar beam. These objects are in most cases linked to precipitation, but radar echoes can also be reflected by birds or insects. Since the radar echoes can be returned by objects other than hydrometeors, the quality control of the data is an important issue. Before the data are introduced into the model system, it is necessary to remove any nonprecipitating echoes such as echoes due to ground or sea clutter, birds and insects, or other anomalous echoes. It is also important to identify areas around each radar where the beam is blocked by known obstacles, like a mountain or a building. If the blockage is total, any echo from the area beyond the obstacle is obviously a false echo. If the radar beam is only partially blocked, the signal can be used depending on how extensive the blocking is (Bech et al. 2007).

The radial velocity is obtained from the motion of the echoes by measuring the Doppler shift in the sent and received signals. This yields a measure of how fast the reflecting objects are moving toward or away from the radar. This information can then be translated into wind observations. Since the radial velocity observations are determined from the phase difference between successive transmitted pulses, there is a maximum velocity that can be determined unambiguously. This maximum velocity is called the Nyquist velocity or unambiguous velocity. If the wind speed is too high relative to the Nyquist velocity of the radar, aliasing effects will occur. The effect will be an erroneous wind speed, sometimes in the wrong direction. It is therefore important to either disregard observations from radars with a too-low Nyquist velocity from the data assimilation or apply a dealiasing algorithm (e.g., Haase and Landelius 2004) to the wind data.

The mesoscale model domains are increasing in size to cover not only the country in which the National Meteorological Service (NMS) is located but surrounding countries as well. This implies there is a need to include observations from surrounding countries. When it comes to conventional observations, this problem is solved through the World Meteorological Organization (WMO). Data are shared and put on the Global Telecommunication System (GTS). For radar data, on the other hand, the situation is much more problematic. Radar data have, so far, not been harmonized within Europe. Each country, or even data provider within each country, could have its own output format in the data delivery.

The European Meteorological Network (EUMETNET) Operational Programme for the Exchange of Weather Radar Information (OPERA) is an attempt to solve some of these problems and to collect radar data centrally. The purpose of OPERA is to deliver reflectivity composites, which means that the data are not redistributed in its raw format, that is, the volume data. There is, furthermore, no demand for members to send in the radial velocity information. However, as OPERA is aware of the needs for NWP usage, they have started to work for a change regarding these matters. Members are encouraged to send their radial velocity data, and the redistribution of volume data is in the plans.

Within the HIRLAM community, work is ongoing to assimilate radar data and to overcome the outstanding differences in the radar data volume files. The harmonization by OPERA opens the possibility for a more straightforward inclusion of radar observations from all countries covered by the model domain of the local mesoscale model. There will be no need for format conversions and interpolations. There are still, however, differences in the radar data sent to OPERA from different data suppliers even when following the OPERA standards. Such differences include number of radars, scan strategies, the size of the volume, and so forth. A preprocessing step that handles these differences is therefore necessary to be able to read the OPERA data into the model database. Such a preprocessing, which also includes a thinning of the data and a sanity check, has been developed by the Swedish Meteorological and Hydrological Institute (SMHI) and the Danish Meteorological Institute (DMI). A common quality control of the radar data is also of great importance to make sure all the data included are of the same quality.

Radar reflectivity data from OPERA are already used in operational forecasts by the Swedish and Norwegian cooperation (MetCoOp) for making numerical weather forecasts on a common domain. The data from the Swedish radars are the same as the data sent to OPERA. In Denmark, OPERA data from 10 countries are used in preoperational runs. Both meteorological services are using the data stream of OPERA dataset up by HIRLAM.

This paper describes the ongoing work within the HIRLAM community to utilize the OPERA data in order to be able to include radar reflectivity observations from multiple countries in the data assimilation, with a focus on the preprocessing and the technical solutions. How multinational OPERA radar reflectivity observations are used within the HIRLAM Aire Limitée Adaptation Dynamique Développement International (ALADIN) Regional Mesoscale Operational NWP in Europe (HARMONIE) is described. A full evaluation of the impact of the radar reflectivity data will be given in a follow-up paper. Section 2 outlines the structure, preprocessing, and utilization of the radar reflectivity observations. The data assimilation of radar reflectivity in the HARMONIE context is outlined in section 3 and the functionality is demonstrated in section 4 as a case study. Conclusions are drawn in section 5.

2. Radar reflectivity data

When using radar data from surrounding countries the need for homogeneous datasets is much desired, and for radar data this has been a big problem. In recent years, however, the EUMETNET program OPERA has put a lot of effort into collecting radar data in a predefined format following the OPERA data information model (ODIM). Most European countries deliver radar reflectivities to OPERA using the ODIM today. Figure 1 shows all the radars currently included in the OPERA network. Today, very few radial wind observations are available through OPERA since there is no demand to send these data. There is, however, a strong wish to make the wind data available, so this will hopefully change in the near future.

Fig. 1.
Fig. 1.

Map representation of the current radar installations in Europe included in the OPERA network.

Citation: Journal of Applied Meteorology and Climatology 56, 6; 10.1175/JAMC-D-16-0247.1

At the moment, the only product from OPERA that is redistributed to the members is a two-dimensional composite of reflectivity. The volume data that are used for most NWP applications are not currently redistributed; however, there are plans to make this available in the future. In the meantime, the data exchange of volume data has been solved through a multilateral agreement that started with the HIRLAM countries but has now been extended (see section 2c).

a. Preprocessing

The datasets from different producers can be very different in terms of scan strategies and number of data. Some radars even have different scan strategies for different altitudes. Many radars are configured to sample the wind and the reflectivity differently to optimize the quality of both. Receiving the wind and reflectivity in separate files is not a problem for the assimilation system. However, if the two quantities are sent in the same file the scan strategy can differ between observations times. If it differs, it is important, both in the reading of the data but also in the preprocessing, to use the most suitable observation times for the respective quantity. A few examples of scan strategies that need to be taken into account are listed here:

  • Swedish radar data, both radial wind and reflectivity, change the scan strategy from the fourth elevation. The number of observed range bins remains the same, but the bin size is 2 km for the lower scans and 1 km for the higher scans. This means that the measured distance is only half for higher scans. This needs to be taken into account.

  • Data from the Netherlands have different sizes of the number of observations in each scan. This means that we need to be very careful when trying to get the data into the model database. Another problem is that the largest number of range bins is not in the lowest elevation but a few elevations above.

  • The lowest elevation in the Norwegian reflectivity data contains twice as many azimuth angles as the rest of the elevations.

The big challenge in the preprocessing step is to harmonize the datasets with as little impact as possible on the data. One such recurring example is filling the upper, or lower, elevations with “no data,” as in the example with data from the Netherlands. More difficult is to make sure that the number of azimuth angles is the same for all elevations from one radar. This is necessary since the assimilation software expects the data in a certain way.

Another problem is that the radar reflectivity data are very high resolution—often higher than the model resolution. To utilize these data optimally and not run in to computer memory problems, the amount of observations need to be reduced. There are several ways to do this. In this study we do a simple spatial thinning in the preprocessing step, that is, before the data even enter the model, to reduce the horizontal spacing. The data are thinned to a spatial distance of 6 km in range and 3° in azimuth. How many data are removed depends on the original resolution, which can be very different. By specifying a distance and an azimuth angle for the thinned data, the problem with varying input data is solved, as for the case with Norwegian data. Additionally, this will make sure that the vertical columns of the observed volumes are nearly vertical. If instead we choose, for example, every fifth observation in range, the vertical column would tilt toward the radar as elevation angle increases. This is of importance as radar reflectivity observations are treated columnwise as described in section 3b.

When using these data in NWP models, it has been discovered that even though the data producers are obliged to follow the ODIM format, there are a few inconsistencies in the files. The main reason why these features have not been identified earlier and corrected is that the two-dimensional composites created and delivered by OPERA do not require the same metadata as NWP usage does. For the composites, these inconsistencies are of no significance, but for NWP usage, however, they can be of great importance. For example, there can be missing parameters such as antenna gain, pulse width, or the Nyquist velocity. It can also be unit errors or simply erroneous data. In the preprocessing, a sanity check of the input files is performed where these inconsistencies are identified and corrected. In the cases where known values can be obtained from the data provider (most cases) the missing parameters are filled in. If not, a default value is set while other parameters will be recalculated or rescaled if possible. In some cases, the metadata are found at different levels in the Hierarchical Data Format (HDF5) files, which complicates the reading of the data. To keep the code in the NWP model as clean as possible, the metadata are rearranged (if needed) to be the same for all data producers.

To highlight problems like these and communicate them to OPERA an OPERA user group has been established. Within this group the needs for NWP usage can be discussed. The group includes users of European radar data for several purposes, for example, nowcasting, early warning systems, and bird migration studies—all with different needs and demands on the data.

b. Quality control

There are a few different widespread quality-control packages available, for example, part of the Integrated Nowcasting through Comprehensive Analysis (INCA) system (e.g., Haiden et al. 2011) and the Advanced Weather Radar Network for the Baltic Sea Region (BALTRAD) toolbox (Michelson and Henja 2012), as well as a few systems used by only one or two countries.

For this study, the radar reflectivity is quality controlled using the BALTRAD toolbox. The main reason for this choice is that BALTRAD is the official software provider for OPERA. The BALTRAD toolbox contains various quality indicators that can be applied to the reflectivity data. The toolbox is described in detail and evaluated in Michelson and Henja (2012).

The quality-control measures used in this study include detection of land and sea clutter, nonprecipitation echoes (birds, insects, etc.) and wireless communication disturbances. The result is a quality index field for each observation point stating the probability of a disturbance in the observation. The quality index ranges from 0 to 1 where 1 means that it is very likely that the observation is precipitation or clear sky. The user can then decide the threshold of the probability that can be accepted before disregarding the data. In addition, a beam blockage map for each radar is made and any echoes from the blocked areas are removed. If the blockage is only partial, the radar beam is weakened. By assuming a uniform beamfilling, this loss can be compensated. Beams with a blocking up to 70% can be filled to obtain a good observation of reflectivity (Bech et al. 2007). For more complete blockings, the observations are indicated as “no data.”

An example of a quality-controlled reflectivity field from a radar in Sweden is shown in Fig. 2, right panel. It can be compared with the left panel, which shows the uncorrected field. Numerous small isolated echoes that are likely not to be precipitation are removed.

Fig. 2.
Fig. 2.

(right) Example of a radar reflectivity scan quality controlled using the BALTRAD toolbox. (left) The original reflectivity field as it is delivered to OPERA. The color scale of the radar echoes ranges from 0 to 45 dBZ.

Citation: Journal of Applied Meteorology and Climatology 56, 6; 10.1175/JAMC-D-16-0247.1

For the experiments presented in this paper, a quality index threshold of 0.6 has been chosen. This threshold was chosen a bit conservatively to make sure that only good data enter the assimilation system. With a threshold of 0.6, about 30% of the observations are sorted out as “bad” observations. Decreasing this threshold to 0.50, that is, a lower-quality demand on the data, decreases the “bad data” to around 20% of the observations. For a quality threshold of 0.7 the “bad data” rate is 36%–40%. These numbers are for radars with significant amounts of precipitation in the field of view, as in Fig. 2. For radars and times with very little precipitation the numbers can be very different.

Work is ongoing to optimize the use of the radar reflectivity observations, including the settings of the quality controls. In this study we use the same quality indices and thresholds for all radars included in the experiments. Depending on the location of the radar and the radar hardware, this may not be the optimal settings for all radars. Another way to optimize the use of the data is through a better data reduction, as creating super observations or a more creative thinning. This will be further investigated in later studies.

c. Operational data flow

OPERA does not currently redistribute the raw volume data of reflectivity that is sent by the member countries. The plan is to do so in the future; however, the need to get access to radar reflectivity data from surrounding countries is a current problem. Therefore, the HIRLAM countries together with a few other countries have developed an intermediate solution. An operational data flow has been set up through BALTRAD for those countries that have agreed to share their data with other countries for NWP usage. The data that are sent to OPERA bounce immediately to the BALTRAD server in Sweden. At this server, the data are quality controlled using the BALTRAD toolbox, as described in section 2b. The data are then made available at an FTP area for NMSs or research institutes to pick up. The data are available within 15 min after observation time. The flow of reflectivity data is visualized in Fig. 3. There is no archiving at this server, but data from 2010 and forward can be made available through BALTRAD upon request.

Fig. 3.
Fig. 3.

Data flowchart for the radar data production chain used in the HIRLAM community.

Citation: Journal of Applied Meteorology and Climatology 56, 6; 10.1175/JAMC-D-16-0247.1

d. Radial winds

Since the start of OPERA, the goal has been to deliver reflectivity composites. The only demand on delivered data has been on reflectivity. A few data providers send radial velocity data together with the reflectivity, but there are still many that do not. However, interest for using wind data is increasing, so there is a strong wish from OPERA that all its members should send wind data. In addition, a few of the centers that deliver wind data provide observations that are not collocated with the reflectivity observations. This leads to a problem with the quality control since most quality filters are applied to the reflectivity data. There is thus no quality information for the radial velocity observations in these cases.

For radial velocities, there is also the problem with aliasing effects if the wind speed increases above the unambiguous, or Nyquist, velocity υN. This velocity depends on the pulse repetition time of the radar measurement. Velocity aliasing usually can be detected from abrupt velocity changes of about 2υN relative to neighboring observation points. The observations can be dealiased using a dealiasing algorithm. In the BALTRAD toolbox, such an algorithm is implemented (Haase and Landelius 2004), although it needs to be used with care. If the Nyquist velocity is too low, there may be problems with dealiasing. Work is ongoing to investigate if wind data with a low υN should be disregarded or if they can be dealiased and where this limit should be in that case.

Even if the quality issue should be solved, the wind information from OPERA suffers from the same difficulties as reflectivity. This means, for example, different data providers deliver data with different volume size. A few of the missing parameters, as described above, are also needed for the wind assimilation. Therefore, the same preprocessing steps can be used in the same way for radial winds as for reflectivity.

3. Radar reflectivity data assimilation

Very few institutes use radar reflectivity data operationally in their daily NWP runs, but a lot of research is going on in this area. Radar reflectivity observations have been used in both three- and four-dimensional variational data assimilation (3D-Var and 4D-Var, respectively). Direct assimilation of radar reflectivity using 4D-Var (Sun and Crook 1997) and 3D-Var (Xiao et al. 2007) has been developed as well as indirect assimilation via pseudo-observations with 4D-Var (Wang et al. 2013; Sun and Wang 2013) and 3D-Var (Wang et al. 2013; Wattrelot et al. 2014).

Common for all these experiments and operational usage is that they are using only data from their own national networks or from specific measurement campaigns where the data are provided on a special agreement and in a specific format. In this study, radar reflectivity data delivered on an operational basis from several countries are used.

a. NWP model

HARMONIE is a model system that includes a nonhydrostatic mesoscale model (Seity et al. 2011). It can be run with different model configurations using, for example, different physics packages or surface schemes. These types of models on finer scales are assumed to give a better description of the meteorological processes because of the possibility to resolve the more elaborated prediction of vertical motion and convection.

The HARMONIE experiments presented in this study are run with AROME physics and are based on code version cy38h1.1. The model domain has 648 grid points in both east–west and north–south directions. The distance between grid points is 2.5 km, and there are 65 levels in the vertical with the model top at 10 hPa. The geographical coverage of the domain is shown in Fig. 4.

Fig. 4.
Fig. 4.

Radar coverage as it is used in the model. The model domain is also shown.

Citation: Journal of Applied Meteorology and Climatology 56, 6; 10.1175/JAMC-D-16-0247.1

The assimilation scheme in HARMONIE is a 3D-Var assimilation scheme that creates an analysis by minimizing a cost function:
e1
where x is the model state vector to be determined by the minimization; xb is a model background state, for example, a short-range forecast; and y is the observation vector. The term H is the observation operator that transforms the model state into the observed quantities. The matrices and are the covariance matrices containing the errors of the background field and observations, respectively. A general difficulty in the formulation of 3D-Var algorithms is the large dimension of the covariance matrix and required inversion of this matrix. In many of the 3D-Var schemes, this has been solved by introducing simplifying assumptions about the covariance of forecast errors (e.g., horizontal homogeneity) and by transformation of the model state vector to an assimilation control vector, of which the error covariance matrix can be assumed to be diagonal.

Apart from the conventional observations, such as surface data from synoptic stations (SYNOP), radiosondes, drifting buoys, and aircraft observations, satellite data from AMSU-A and AMSU-B/Microwave Humidity Sounder (MHS) are used in the assimilation for the experiments presented here.

b. Radar observation treatment

The reflectivity itself is used to get a measure of the precipitation intensity and to analyze where the precipitation is located. An observation of reflectivity will moisten the model in areas where the model indicates no precipitation or adjust the intensity in precipitation areas. The reflectivity is not directly assimilated into the model since there is a complicated, nonlinear relation between the model variables and reflectivity. This includes parameterizations of microphysical processes and non-Gaussian error distributions. Instead a vertical moisture profile is retrieved through a one-dimensional (1D) Bayesian retrieval based on a comparison between observed and simulated reflectivities. This humidity profile is then used in the 3D-Var assimilation scheme, where the profiles are used to modify the humidity first-guess fields. However, in this process the humidity information also indirectly influences other model variables, mainly through the intrinsic properties of the matrix (Bannister 2008); see Eq. (1). The method is described in detail in Caumont et al. (2010) and Wattrelot et al. (2014).

An important part of the radar assimilation is to assimilate, not only precipitation seen by the radar but also areas where the model indicates precipitation but the radar does not. The difficulty is to determine if the radar actually measures clear sky or if it is unable to detect precipitation because of the sensitivity of the radar. The sensitivity and which reflectivity values to trust can be estimated for each radar individually depending on the properties of the radar. Not all data providers supply all the necessary metadata for this calculation, and it is complicated to transfer this information into the observation operators. To get around these problems and to be on the safe side, we simply do not use observed reflectivity values below 0 dBZ. This includes a risk of losing a few good observations of drizzle, but the advantages of not including noise are bigger.

The assimilation of winds is more straightforward. There is an observation operator to calculate the corresponding radial velocity for each radar position using the model wind fields (Montmerle and Faccani 2009).

4. A simple proof of functionality

This paper focuses on showing that it is technically possible to include radar reflectivity data from several countries. The results shown here are only briefly evaluated to show that reasonable fields are obtained and with no negative impact on the forecast scores. A full evaluation of results will be given in coming papers.

The examples shown here are run over an area centered over Denmark including parts of many surrounding countries. It is ideal for testing the multinational approach. It is of great importance for a country like Denmark to include observations from surrounding countries in order to get an impact since the country itself is rather small.

In this study, radar reflectivity from nine countries—Belgium, Germany, Denmark, Estonia, Finland, France, The Netherlands, Norway, and Sweden—are used. No radial velocity data are used, primarily because of the lack of wind information sent to OPERA as mentioned in section 2d. The reflectivity data from all these countries are collected from OPERA and quality controlled as described above. The model area and the coverage of the included radars are shown in Fig. 4. Observations close to the boundaries are not used, which is clearly seen in Fig. 4. The reason is that the effective influence radius imposed by the structure functions will cause a wrapping effect if observations are used too close to the domain edges. An observation at the western boundary will thus cause an increment at the eastern boundary. A solution to this problem exists and will be introduced.

Figure 5 shows the radar reflectivity that enters the model, that is, everything that is not excluded by the preprocessing. After this step, the data go through the so-called screening step in which a further thinning is made to avoid the effects of spatially correlated errors. For the experiments presented here, the final horizontal resolution of the radar observations is 15 km since the models effective resolution is closer to 8 km than 2.5 km. In the screening step, there is also a redundancy check and a comparison with the model first guess. The latter is a quality control item to make sure that the observations do not differ too much from the model. The size of the difference allowed varies with observation type.

Fig. 5.
Fig. 5.

Example from 0300 UTC 24 Aug 2014 of the radar reflectivity observations that enter the model after the preprocessing but before the final thinning and quality control (screening).

Citation: Journal of Applied Meteorology and Climatology 56, 6; 10.1175/JAMC-D-16-0247.1

Figure 6 shows the same dataset as in Fig. 5 but after the screening. Note that the markers are slightly thicker for display purposes. These are the observations that are actually used in the minimization; that is, they affect the final analysis together with observations from other sources that passed the screening.

Fig. 6.
Fig. 6.

Example from 0300 UTC 24 Aug 2014 of the radar reflectivity observations that enter the minimization, i.e., what is left after the final thinning and quality control (screening). Note that the markers are slightly larger than in Fig. 5 for display purposes.

Citation: Journal of Applied Meteorology and Climatology 56, 6; 10.1175/JAMC-D-16-0247.1

To investigate the impact of the radar reflectivity observations, experiments are run over a longer period of time. The examples shown below are from a two-week period in the end of August 2014 when a strong precipitation event hit Denmark and southern Sweden. This event, which caused major floods, was rather well predicted in the longer forecast but not so well predicted in the shorter forecasts.

First, to see if the reflectivity data are used properly by the model, we look at the impact after only one assimilation cycle. The radar experiment is therefore compared with the same analysis made without the radar reflectivity, that is, the reference experiment. These two will have the same first guess, and the difference will be caused only by the reflectivity observations. The example from 0300 UTC 24 August 2014 is shown in Fig. 7. It shows the difference in humidity at model level 50, which is rather close to the ground, corresponding to ~850 hPa. The location of the impact is well collocated with the radars that measure reflectivity for this time: along the coast of Norway, in northern Denmark, and in southern Sweden. The radar at the island of Bornholm, Denmark, is seen clearly, as well as one radar with observations in Germany. This follows well with the reflectivity data that enter the assimilation, as shown in the previous figures. There are areas with both positive (solid lines, light gray) and negative (dashed lines, dark gray) increments seen in Fig. 7. Positive means there is none or too little precipitation in the model, while the areas with negative increments mean there is weaker precipitation in the observations or observations of clear sky at the same location that the model generates precipitation. This example indicates that the radar observations are able to both moisten and to dry the model, that is, relocate the precipitating areas.

Fig. 7.
Fig. 7.

Difference of the humidity field at model level 50, ~850 hPa, between an analysis including radar reflectivity observations and one without radars starting from the same first-guess field. Solid lines indicate a moistening of the model, and dashed lines show that the model becomes dryer. The example is from 0300 UTC 24 Aug 2014.

Citation: Journal of Applied Meteorology and Climatology 56, 6; 10.1175/JAMC-D-16-0247.1

To determine if this impact from the radar observations is better or worse, the model was run for a one-week period, starting at the date shown above. One week may seem a bit too short, but during this period an extreme precipitation event took place almost at the center of the domain. The experiments here are also mainly run to show that the radar reflectivity data do not cause any problems. A full verification and evaluation will be made in coming papers. Two experiments were run, one reference and one including radar reflectivity observations. The results were then verified against observations. Two examples are shown here for relative humidity for which the impact is expected to be the largest since no wind information from the radars is assimilated. Figure 8 show verification of relative humidity at the surface (Rh2m), while Fig. 9 shows relative humidity verified against radiosondes at the 500-hPa level. For Rh2m there is a neutral to slightly negative impact in standard deviation (the upper two lines) when radar observations are included (shown as green), while the model gets drier and thereby reduces the systematic errors (bias, shown as the lower two lines). At higher altitudes, on the other hand, there is a clear improvement both in the standard deviation as well as in the bias as seen in Fig. 9. A larger impact at higher levels is to be expected since there are very few other observations in the free atmosphere as compared with the surface.

Fig. 8.
Fig. 8.

Verification against observations for relative humidity at the surface (2-m level) for two experiments: the reference experiment (red) and the experiment including reflectivity from radar observations (green). The upper two lines show the standard deviation, and the lower two show the systematic errors.

Citation: Journal of Applied Meteorology and Climatology 56, 6; 10.1175/JAMC-D-16-0247.1

Fig. 9.
Fig. 9.

As in Fig. 8, but against radiosondes at the 500-hPa level.

Citation: Journal of Applied Meteorology and Climatology 56, 6; 10.1175/JAMC-D-16-0247.1

5. Conclusions

In this study it is shown that it is possible (and beneficial) to use radar reflectivity data from different data producers in the assimilation of an NWP model. There are, however, a few obstacles that need to be overcome; for example, a standardized input format is of great importance. Additionally, a common quality control for the data is desired so that all ingoing radar observations are of the same quality.

Such a standard format for radar observations has been developed by the EUMETNET program OPERA. It turns out, however, that the data files that are sent to OPERA are not completely following these standards. Small but important differences still exist. One reason why this has not been discovered earlier is that no one, not even OPERA products, is using the radar reflectivities in such an advanced way. For this reason the radar preprocessing and quality control is a very important part of the observation usage. In a preprocessing step, such differences and any missing parameters are adjusted and added with real values if they exist and default values if not. The quality control is made for all data using the BALTRAD toolbox. This ensures that all the reflectivity data entering the NWP model assimilation system are harmonized and of the same quality.

The results shown here indicate that radar reflectivity data from nine countries can be used with neutral or positive impact on the forecast verification scores. For future experiments, efforts will be made to improve the input radar reflectivity. In particular, the data reduction needs to be made more satisfactory. In this study, a simple thinning is performed without taking the quality or data information into account. There are plans to create super observations, that is, making an average over an area, to use all that information in the observation dataset. However, this is not trivial and needs to be made carefully.

In addition to improving the quality and optimal use of the radar observations, there is other research ongoing concerning the data assimilation in general, for example, the work with the extension zone to avoid the wraparound effects mentioned in section 4. Another topic that would clearly benefit the assimilation of high-resolution data, such as radar, is the background error statistics and structure functions to make these both smaller scale but also more flow dependent.

The wind information was not included in the experiments shown here. There are currently ongoing investigations on how to make the best use of the wind information. So far, only a few countries send the radial wind information to OPERA so the usage is rather limited. The radial winds also have other problems regarding the quality information, for example, aliasing if the wind is too strong in relation to the observation sampling.

Radar reflectivity observations are used operationally in MetCoOp, which is the operational cooperation between Sweden and Norway. Experiments prior to the operationalization were run for the MetCoOp domain with similar results as presented here. Preoperational experiments are run at the Danish Meteorological Institute, including radar reflectivity data from 10 countries, with encouraging results. It will be transferred into the operational runs in the near future.

Acknowledgments

The authors acknowledge Eric Wattrelot and Thibaut Montmerle at MeteoFrance for developing and helping to understand the assimilation software and Daniel Michelson and Gunther Haase for setting up the radar data exchange service. Thanks are also given to Nils Gustafsson for valuable comments to the manuscript. The work was conducted within the international HIRLAM-B cooperation.

REFERENCES

  • Bannister, R. N., 2008: A review of forecast error covariance in atmospheric variational data assimilation. I: Characteristics and measurements of forecast error covariances. Quart. J. Roy. Meteor. Soc., 134, 19511970, doi:10.1002/qj.339.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bech, J., U. Gjertsen, and G. Haase, 2007: Modelling weather radar beam propagation and topographical blockage at northern high latitudes. Quart. J. Roy. Meteor. Soc., 133, 11911204, doi:10.1002/qj.98.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Caumont, O., V. Ducrocq, E. Wattrelot, G. Jaubert, and S. Pradier-Vabre, 2010: 1D+3DVAR assimilation of radar reflectivity data: A proof of concept. Tellus, 62, 173187, doi:10.1111/j.1600-0870.2009.00430.x.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Haase, G., and T. Landelius, 2004: Dealiasing of Doppler radar velocities using a torus mapping. J. Atmos. Oceanic Technol., 21, 15661573, doi:10.1175/1520-0426(2004)021<1566:DODRVU>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Haiden, T., A. Kann, C. Wittman, G. Pistotnik, B. Bica, and C. Gruber, 2011: The Integrated Nowcasting through Comprehensive Analysis (INCA) system and its validation over the eastern alpine region. Wea. Forecasting, 26, 166183, doi:10.1175/2010WAF2222451.1.

    • Crossref
    • 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, doi:10.1175/1520-0493(2004)132<1081:DRWDAW>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Michelson, D., and A. Henja, 2012: Opera work package 3.6: Odyssey additions, task 3: Tuning and evaluation of “andre” tool. OPERA Tech. Rep. Working Doc. WD-2012-02c, 20 pp.

  • Montmerle, T., and C. Faccani, 2009: Mesoscale assimilation of radial velocities from Doppler radars in a preoperational framework. Mon. Wea. Rev., 137, 19391953, doi:10.1175/2008MWR2725.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Salonen, K., G. Haase, R. Eresmaa, R. Hohti, and H. Järvinen, 2011: Towards the operational use of Doppler radar radial winds in HIRLAM. Atmos. Res., 100, 190200, doi:10.1016/j.atmosres.2010.06.004.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Seity, Y., P. Brousseau, S. Malardel, G. Hello, P. Bernard, F. Bouttier, C. Lac, and V. Masson, 2011: The AROME-France convective-scale operational model. Mon. Wea. Rev., 139, 976991, doi:10.1175/2010MWR3425.1.

    • Crossref
    • 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. J. Atmos. Sci., 54, 16421661, doi:10.1175/1520-0469(1997)054<1642:DAMRFD>2.0.CO;2.

    • Crossref
    • 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.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wang, H., J. Sun, X. Zhang, X.-Y. Huang, and T. Auligné, 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.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wattrelot, E., O. Caumont, and J.-F. Mahfouf, 2014: Operational implementation of the 1D+3D-Var assimilation method of radar reflectivity data in the AROME model. Mon. Wea. Rev., 142, 18521873, doi:10.1175/MWR-D-13-00230.1.

    • Crossref
    • 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.

    • Crossref
    • 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, doi:10.1175/JAM2439.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
Save