-
The national meteorological weather services of Sweden, Norway, Finland, and Estonia have joined forces in a collaboration, named MetCoOp, around a common operational km-scale forecasting system (Müller et al., 2017). It is a configuration of the shared Aire Limitée Adaptation dynamique Developpement InterNational (ALADIN)-High Resolution Limited Area Model (HIRLAM) NWP system. This system can be run with different configurations and in MetCoOp the HIRLAM-ALADIN Regional Meso-scale Operational NWP In the Europe Application of Research to Operations at Mesoscale (HARMONIE-AROME) is used (Bengtsson et al., 2017) and is run as an ensemble forecasting system. In our study we use the cy43 version of the MetCoOp forecasting system. The northern European model domain is illustrated in Fig. 1. It has 900
$ \times $ 960 horizontal grid points with a grid distance of 2.5 km, and 65 vertical model levels, extending from roughly 12 m above ground up to approximately 33 km (10 hPa).The three main components of this forecasting system are surface DA, upper-air DA, and the forecast model. The system is run with a 3-hourly assimilation cycle and launching forecasts at 0000, 0300, 0600, 0900, 1200, 1500, 1800, and 2100 UTC. Only the forecasts launched at the synoptic hours 0000, 0600, 1200, and 1800 UTC are being used by the duty forecasters. Therefore, these have a relatively strict observation cut-off time of 1 h and 15 min compared to the 3 h and 20 min for the asynoptic cycles 0300, 0900, 1500, and 2100 UTC. For the asynoptic cycles, the only time-constraint is to produce a 3 h forecasts serving as background state for the DA at the following synoptic cycle. Thus, this 3 h forecast can be produced just prior to the synoptic cycle DA. Due to the operational cut-off constraints mentioned above, in practice only observations within a time-range from −1 h and 30 min to +1 h and 15 min are used for the cycles at 0000, 0600, 1200, and 18 UTC. For the asynoptic cycles observations within the entire time-range −1 h and 30 min to + 1 h and 29 min are used.
A detailed description of the forecast model setup is given in Seity et al. (2011) and Bengtsson et al. (2017). It is a non-hydrostatic model formulation with a spectral representation of the model state (Bubnová et al., 1995; Bénard et al., 2010). Sub-grid scale parameterization of clouds, including shallow convection is handled by the EDMF (Eddy Diffusitivity Mass Flux) originating from de Rooy and Siebesma (2008) and Neggers et al. (2009). Deep convection is resolved by the model. Turbulence and vertical diffusion is represented using the so-called HARATU scheme, which is based on a Turbulent Kinetic Energy scheme by Lenderink and Holtslag (2004). The radiative transfer is modelled as described by Fouquart and Bonnel (1980) and Mlawer et al. (1997) for short- and long-wave radiation processes, respectively. Surface processes are modeled using the SURFEX (Surface Externalisée) scheme (Masson et al., 2013). Global forecasts provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) are used as lateral boundary conditions. These forecasts are launched every 6 h with a 1 h output frequency. Global model information is also used to replace larger-scale information in the background state with lateral boundary information (Müller et al., 2017). The idea is to make use of high-quality large-scale information from the ECMWF global fields, in the MetCoOp analysis.
In the surface DA synoptic observations of two-meter temperature, two-meter relative humidity and snow cover are used to estimate the initial state of the soil temperature, soil moisture, and snow field. The DA is comprised of a horizontal optimal interpolation (Taillefer, 2002), which for soil moisture and temperature is followed by a vertical optimal interpolation procedure (Giard and Bazile, 2000). The upper-air DA is based on a 3-dimensional variational (3D-Var) approach (Fischer et al., 2005). Many types of observations are assimilated including conventional in-situ measurements (pilot–balloon wind, radiosonde, aircraft, buoy, ship, and synop), Global Navigation Satellite System (GNSS) Zenith Total Delay (ZTD) data, weather radar reflectivity information, as well as infrared (IR) and PMW radiances from satellite-based instruments. The IR radiances are sensed by the Infrared Atmospheric Sounding Interferometer (IASI) placed on board the Metop satellites. PMW radiances are traditionally provided by the Advanced TIROS Operational Vertical Sounder (ATOVS) instrument family, including AMSU-A and MHS. Recently the DA system has been prepared to also utilize data from the MWHS-2 instrument on board the FY-3C and FY-3D polar orbiting satellites. Background error covariances are based on a climatological assumption and their representation is based on a multivariate formulation under the assumptions of horizontal homogeneity and isotrophy. They are calculated from an ensemble of forecast differences (Berre, 2000; Brousseau et al., 2012). These are produced by Ensemble Data Assimilation (EDA) experiments carried out with the HARMONIE-AROME system. The HARMONIE-AROME EDA uses perturbed observations and ECMWF global EDA (Bonavita et al., 2012) forecasts as lateral boundary conditions. Scaling is applied to the derived statistics in order to be in agreement with the amplitude of HARMONIE-AROME + 3 h forecast errors (Brousseau et al., 2012).
-
The NWP system uses atmospheric temperature and moisture information extracted from the PMW radiances sensed by satellite instruments summarized in Table 1. The AMSU-A instrument is primarily used to retrieve information on the vertical distribution of atmospheric temperature. The MHS instrument, on the other hand, provides information on the vertical structure of water vapour. Similarly to MHS, the MWHS-2 instrument has capability to retrieve information on atmospheric moisture. In addition, it has some capacity to retrieve information on temperature. To produce the model counterparts of the observed PMW radiances an observation operator,
$ H $ , is applied. This operator is based on the radiative transfer model RTTOV (Radiative Transfer for TOVS) version 11.2.0 (Saunders et al., 2018) as developed under the EUMETSAT Satellite Application Facility to support NWP. At present, only clear-sky radiances with small contributions from surface can be efficiently handled by the observation operator. Furthermore, the observation operator requires that the measurement is sensitive mainly to atmospheric conditions below the model top. It implies that a clear-sky radiance with a major part (roughly 90%) of the integral of the Jacobians below the model top can be assimilated, if influence by the surface is small enough. The resulting channel usage from the different instruments is presented in Table 2. The AMSU-A channels used are the temperature sensitive ones located around 55 GHz while the MHS channels utilized are centered around$ 183 $ GHz. The used channels from MWHS-2 are mainly the moisture sensitive ones around the$ 183 $ GHz water vapor absorption line (channels 11 to 15). MWHS-2 channels 5 and 6, located around$ 118 $ GHz and sensitive also to temperature, are used with very low weight within the assimilation system. Some low-peaking channels (MHS channel 5 and MWHS-2 channel 15) affected by the surface are used over sea only. The reason for this is that the surface emissivity and skin temperature contributions are better represented in the observation operator for radiances over sea than for radiances over land. Due to the problem of contamination by the surface, many of the PMW instrument channels (AMSU-A channel 6, MHS channels 3, 4, and MWHS-2 channels 5, 6, 11, 12, 13, and 14) are used over low level terrain only. The discrimination of low and high level terrain is done using the model orography over the field of view of the observation.Instrument Satellites AMSU-A Metop-A, Metop-B, Metop-C, NOAA-18, NOAA-19 MHS Metop-A, Metop-B, Metop-C, NOAA-19 MWHS-2 FY-3C, FY-3D Table 1. PMW radiance observation usage.
Channel number Channel Frequency (GHz) AMSU-A MWHS-2 MHS AMSU-A MWHS-2 MHS 6 − − 54.400 (H) − − 7 − − 54.940 (V) − − 8 − − 55.500 (H) − − 9 − − 57.290344 (H) − − − 5 − − 118.75 ± 0.8 (H) − − 6 − − 118.75 ± 1.1 (H) − − 11 3 − 183 ± 1.0 (H) 183 ± 1.0 (H) − 12 − − 183 ± 1.8 (H) − − 13 4 − 183 ± 3.0 (H) 183 ± 3.0 (H) − 14 − − 183 ± 4.5 (H) − − 15 5 − 183 ± 7.0 (H) 190.31 (V) Table 2. PMW radiance channels used in the DA (window channels used in the data filtering are not included).
To indicate whether radiance data from particular instruments and channels are affected by clouds, radiances in associated window channels within the instrument, capable of identifying clouds, are compared with the corresponding model state equivalents. If the window channel departures are larger than a particular threshold value, the corresponding non-window channel radiances are considered to be affected by clouds, and are therefore rejected from use in the DA. The window channel for AMSU-A channels 6 and 7 is AMSU-A channel 4 (
$ 52.8 $ GHz). The window channel used for MHS channels 3, 4, and 5 is MHS channel 2 at$ 157 $ GHz. The window channel used for MWHS-2 channels 5 and 6 is MWHS-2 channel 1 ($ 89 $ GHz) and for MWHS-2 channels 11, 12, 13, 14, and 15 the window channel is MWHS-2 channel 10 at$ 150 $ GHz. Note that for the rather high-peaking AMSU-A channels 8 and 9 no cloud detection is applied. For AMSU-A channel 6, a complementary check, aiming at identifying cloud and rain is applied, using liquid water path and scattering index. In case of too high liquid water path or enhanced scattering from precipitation-sized particles, radiance observations from AMSU-A channel 6 are considered contaminated by large hydrometeors and are therefore rejected.Systematic errors that might be present in the clear-sky radiances that have passed the cloud detection are handled by applying an adaptive Variational Bias Correction (VARBC) as proposed by Dee (2005) and further adapted by Auligné et al. (2007). A linear model of the following form is applied to describe the bias
$ {\boldsymbol{b}} $ in the PMW observations:Here
$ {\boldsymbol{p}}_i $ represents the predictors,$ {\boldsymbol{x}} $ represents the model state,$\, \beta_i $ represents the bias parameters, and$N_{\rm p}$ describes the number of predictors. Predictors are adopted to the channels and typically include atmospheric layer thicknesses and satellite instrument viewing angle. The bias parameters and the model initial state are simultaneously derived within the variational DA framework.A quality control procedure is applied to remove radiance observations that are considered to be of poor quality. Based on the operational monitoring experiences, we do not use AMSU-A and MHS radiances from field of views close to scan-line edges. Channels from instruments on specific satellite platforms might be temporarily or permanently blacklisted in the DA system. This blacklisting may occur due to known problems reported by the satellite agencies (NOAA-18 MHS channels 3-5, NOAA-19 MHS channel 3), by other collaborating NWP partners assimilating the same satellite radiances (NOAA-19 AMSU-A channels 7–8 and Metop-B channel 7), or due to quality limitations found during observation monitoring (i.e., where observed radiances are compared with corresponding model state equivalents over a longer period). It should be mentioned that the Metop-B AMSU-A channel 7 blacklisting is based on experiences with noisy radiances starting already in 2017. However, this noise is not presently observed, and the plan is to re-activate the assimilation of this channel. Further, as part of the quality control, a filtering is applied to get rid of radiance observations affected by gross errors. In this gross error check, a radiance observation,
$ y_i $ , is rejected if it satisfies the following inequality:where
$ \lambda = 1 + \sigma_{{\rm{o}},i}^2 / \sigma_{{\rm{b}},i}^2 $ ,${{\rm{FgLim}}}$ is the rejection limit and$ [H({{\boldsymbol{x}}_{{\rm{b}}}})]_i $ denotes the projection of the model state on$ y_i $ observation, where the potential observation bias has been accounted for.$ \sigma_{{\rm{o}},i} $ and$ \sigma_{{\rm{b}},i} $ are the standard deviation of the observation error and background error equivalent, respectively.The bias corrected radiances that have passed the data selection and quality control are assumed to have a Gaussian error distribution. The associated observation error standard deviations are derived from long-term observation monitoring. The observation errors are comprised of instrument errors, representativity errors, persistence errors, and errors in the observation operator, and they are slightly inflated to account for the lack of representation of observation error correlations (Bormann and Bauer, 2010). For the AMSU-A channels, the estimated error standard deviations are approximately
$ 0.2 $ K. For the MHS channels, the values are$ 1.8 $ K, for MWHS-2 channels 11–15 the values are$ 1.8-2.0 $ K, and for MWHS-2 channels 5 and 6 the estimated error standard deviations are$ 0.9 $ K. To further alleviate the effects on the initial state of spatially correlated observation errors, a thinning of radiances is applied. For radiances from the AMSU-A and MHS instruments, a thinning distance of$ 80 $ km is used. This choice is based on previous experiences by Randriamampianina (2006). Due to lack of previous experience with the MWHS-2 instrument, as a first step we chose to apply the larger thinning distance of$ 160 $ km. This could probably be reduced at a later stage, based on a posteriori diagnosis of observation error correlations following Bormann and Bauer (2010).The rejections in the data selection and quality control are dominated by the cloud detection and the thinning procedure. Roughly 3% of the observations that have passed the data selection (except thinning, since it is applied after the quality control) are identified as gross errors and rejected by the gross error check described by Eq. 2.
Instrument | Satellites |
AMSU-A | Metop-A, Metop-B, Metop-C, NOAA-18, NOAA-19 |
MHS | Metop-A, Metop-B, Metop-C, NOAA-19 |
MWHS-2 | FY-3C, FY-3D |