• Bi, L., Jung J. A. , Morgan M. C. , and Le Marshall J. F. , 2011: Assessment of assimilating ASCAT surface wind retrievals in the NCEP Global Data Assimilation System. Mon. Wea. Rev.,139, 3405–3421.

  • Desroziers, G., Berre L. , Chapnik B. , and Poli P. , 2005: Diagnosis of observation, background and analysis-error statistics in observation space. Quart. J. Roy. Meteor. Soc., 131, 33853396.

    • Search Google Scholar
    • Export Citation
  • Figa-Saldaña, J., Wilson J. J. W. , Attema E. , Gelsthorpe R. , Drinkwater M. R. , and Stoffelen A. , 2002: The Advanced Scatterometer (ASCAT) on the Meteorological Operational (MetOp) platform: A follow on for the European wind scatterometers. Can. J. Remote Sens., 28, 404412.

    • Search Google Scholar
    • Export Citation
  • Gustafsson, H., Berre L. , Hornquist S. , Huang X. Y. , Lindskog M. , Navascues B. , Mogensen K. S. , and Thorsteinsson S. , 2001: Three-dimensional variational data assimilation for a limited area model. Part I: General formulation and background constraint. Tellus, 53A, 425446.

    • Search Google Scholar
    • Export Citation
  • Hersbach, H., and Janssen P. , 2007: Operational assimilation of surface wind data from the MetOp ASCAT scatterometer at ECMWF. ECMWF Newsletter, No. 113, ECMWF, Reading, United Kingdom, 6–8.

  • Hersbach, H., Stoffelen A. , and de Haan S. , 2007: An improved C-band scatterometer ocean geophysical model function: CMOD5. J. Geophys. Res., 112, C03006, doi:10.1029/2006JC003743.

    • Search Google Scholar
    • Export Citation
  • Hollingsworth, A., and Lönnberg P. , 1986: The statistical structure of short-range forecast errors as determined from radiosonde data. Part 1: The wind field. Tellus, 38A, 111136.

    • Search Google Scholar
    • Export Citation
  • Järvinnen, H., and Unden P. , 1997: Observation screening and first-guess quality control in the ECMWF 3D-Var data assimilation system. Research Department Tech. Memo. 236, ECMWF, Reading, United Kingdom, 34 pp.

  • Lindskog, M., and Coauthors, 2001: Three-dimensional variational data assimilation for a limited area model. Part II: Observation handling and assimilation experiments. Tellus, 53A, 447468.

    • Search Google Scholar
    • Export Citation
  • Pirkka, O., 2010: Feasibility of assimilating ASCAT surface winds into a limited area. M.S. thesis, Dept. of Physics, University of Helsinki, 81 pp. [Available online at https://helda.helsinki.fi/bitstream/handle/10138/20968/feasibil.pdf.]

  • Portabella, M., and Stoffelen A. , 2004: A probabilistic approach for SeaWinds data assimilation. Quart. J. Roy. Meteor. Soc., 130, 126.

    • Search Google Scholar
    • Export Citation
  • Singh R., Pal P. K. , Kishtawal C. M. , and Joshi P. C. , 2008: The impact of variational assimilation of SSM/I and QuikSCAT satellite observations on the numerical simulation of Indian Ocean tropical cyclones. Wea. Forecasting,23, 460–476.

  • Stoffelen, A., 1998: Toward the true near-surface wind speed: Error modeling and calibration using triple collocation. J. Geophys. Res., 103 (C4), 77557766.

    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., and Anderson D. , 1997: Scatterometer data interpretation: Measurement space and inversion. J. Atmos. Oceanic Technol., 14, 12981313.

    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., and van Beukering P. , 1997: Improved backscatter processing and impact of tandem ERS winds on HIRLAM. HIRLAM Project Rep. 31, IMET, Dublin, Ireland, 86 pp.

  • Unden, P., 2002: HIRLAM-5 scientific documentation. HIRLAM-5 Project, Norrköpping, Sweden, 144 pp. [Available online at http://hirlam.org/index.php?option=com_docman&task=doc_details&gid=308&Itemid=139.]

  • Verspeek, J. A., Stoffelen A. , Portabella M. , Bonekamp H. , Anderson C. , and Figa J. , 2010: Validation and calibration of ASCAT using CMOD5.n. IEEE Trans. Geosci. Remote Sens., 48, 386395, doi:10.1109/TGRS.2009.2027896.

    • Search Google Scholar
    • Export Citation
  • Vogelzang, J., Stoffelen A. , Verhoef A. , de Vries J. , and Bonekamp H. , 2009: Validation of two-dimensional variational ambiguity removal on SeaWinds scatterometer data. J. Atmos. Oceanic Technol., 26, 12291245.

    • Search Google Scholar
    • Export Citation
  • Vogelzang, J., Stoffelen A. , Verhoef A. , and Figa-Saldaña J. , 2011: On the quality of high-resolution scatterometer winds. J. Geophys. Res.,116, C10033, doi:10.1029/2010JC006640.

  • View in gallery
    Fig. 1.

    HIRLAM model domains for the D11 (6 h) and H11 (3 h) cycles.

  • View in gallery
    Fig. 2.

    Typical HIRLAM data coverage for the 1200 UTC analysis window as obtained from the KNMI experimental observation monitoring tool, including (left) radiosonde profiles, (center) aircraft (AMDAR) observations, and (right) surface observations from synoptic stations, ships, and drifting buoys. Black (gray) symbols denote observations at locations used (not used) in the H11 analysis. Gray-colored symbols are either outside the H11 domain as displayed in Fig. 1 or flagged as suspect at the quality control stage.

  • View in gallery
    Fig. 3.

    Locations (triangles) and instrument identification numbers of moored buoys.

  • View in gallery
    Fig. 4.

    (left) Ascending and (right) descending ASCAT tracks for three different gray-colored overpasses. The overpass time (UTC) is displayed next to the swaths. The dots denote the WVC centroids at which the wind observations are defined. Their spatial separation is 25 km. The ASCAT swath has 21 WVCs across track and is 500 km wide. Each panel shows three overpasses on 27 April 2010: (left) at around 1951 UTC (rightmost swaths, which are in time for the 2100 UTC analysis), 2133 UTC (in time for the 2100 UTC analysis), and 2313 UTC (in time for the 0000 UTC analysis but outside the HIRLAM H11 domain; see Fig. 1) and (right) at around 1003 UTC (observations arrive too late for the 0900 UTC analysis), 1143 UTC (observations arrive too late for the 1200 UTC analysis), and 1324 UTC (observations arrive too late for the 1500 UTC analysis). See the text for further details.

  • View in gallery
    Fig. 5.

    ASCAT observation usage in HIRLAM. (bottom) Total number of available assimilations (gray) and number used for assimilation (black) of ASCAT observations per day in the H11 domain depicted in Fig. 1 for the period 24 Apr–7 Jul 2010. (top) The differences in the gray and black curves seen in the bottom panel corresponding to the number of unused ASCAT observations. The missing part of the curves near 26 Jun is due to an interruption in operations because of a system upgrade.

  • View in gallery
    Fig. 6.

    ASCAT observations used in the experimental HIRLAM suite (X11) for the 2100 UTC analysis on 27 Apr 2010. Black areas denote locations where observations were used in the analysis, and dark gray areas denote locations not used in the analysis because of wind retrievals that did not pass the quality control [e.g., because of proximity to land, unusual sea state or other surface wind flow regimes not well modeled during the scatterometer wind retrieval; Stoffelen et al. (1997)].

  • View in gallery
    Fig. 7.

    Bias and standard deviation of (of ) for 10-m wind speed from ASCAT and forecasts from the reference H11 run (gray) and experimental (including ASCAT) X11 run (black) for the experimental period 24 Apr–7 Jul 2010. The black dashed line shows the percentage change in standard deviation, with negative values denoting a standard deviation reduction, i.e., improved skill by assimilating ASCAT.

  • View in gallery
    Fig. 8.

    As in Fig. 7, but now verified against the (left) zonal (u) and (right) meridional (υ) wind components of moored buoys located as displayed in Fig. 3.

  • View in gallery
    Fig. 9.

    Background departure standard deviation of the HIRLAM model wind from buoy (left) zonal and (right) meridional wind components for all buoys separately. The locations of the buoys are in Fig. 3. The statistics are based on 14 535 observations over the complete 10-week experimental period. Black and gray bars correspond to the operational H11 and experimental X11 cycles, respectively.

  • View in gallery
    Fig. 10.

    As in Fig. 7, but now for pressure observations from (left) moving ships and (right) moored buoys and ships. Black dotted lines denote the differences between the absolute bias of the reference (H11) run and the absolute bias of the experimental ASCAT (X11) run. Positive differences imply positive impact.

  • View in gallery
    Fig. 11.

    HIRLAM reference model (H11) surface pressure (hPa) error standard deviation against drifting buoys. Some of the buoy locations are displayed in the right panel of Fig. 12.

  • View in gallery
    Fig. 12.

    (left) As in Fig. 10, but now for (right) the four drifting buoys and their displacement over the 10-week experimental period.

  • View in gallery
    Fig. 13.

    Analyses (top row) and successive forecasts (rows 2–6) of 10-m wind speed (grayscale) and mean sea level pressure (black contours) for the (left) reference (H11) run and (middle) the experimental (X11) ASCAT run. (right) The differences (X11 − H11) in mean sea level pressure (contours with values) and in wind speed. The amplitude of the latter is in the grayscale at the bottom with plus signs (+) denoting areas of increased wind speed in the X11 run. The analysis time is 2100 UTC 6 Jun 2010. Forecast are shown every 3 h up to FC + 15 h.

  • View in gallery
    Fig. 14.

    (top left) The 10-m wind speed background— the circle in southwest Ireland marks the location of the observatory on Valentia, (top right) analysis, and (middle) analysis increment, i.e., analysis minus background for the reference (H11) run with plus signs denoting areas with a positive increment. (bottom) As in (top), but for the experimental X11 ASCAT run. (bottom right) The ASCAT coverage and measured wind speeds.

  • View in gallery
    Fig. 15.

    Observation minus forecast for radiosondes launched at Valentia, marked by the circle in Fig. 14. Gray (black) curves correspond to the reference (experimental ASCAT) run. Forecasts are initiated from the 2100 UTC 6 Jun 2010 analysis. (left) The 3-h forecast minus the 0000 UTC 7 Jun radiosonde launch. (middle) The 9-h forecast minus the 0600 UTC 7 Jun radiosonde launch. (right) The 15-h forecast minus the 1200 UTC 7 Jun radiosonde launch.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 184 80 4
PDF Downloads 101 34 0

Impact of ASCAT Scatterometer Wind Observations on the High-Resolution Limited-Area Model (HIRLAM) within an Operational Context

Siebren de HaanKNMI, De Bilt, Netherlands

Search for other papers by Siebren de Haan in
Current site
Google Scholar
PubMed
Close
,
Gert-Jan MarseilleKNMI, De Bilt, Netherlands

Search for other papers by Gert-Jan Marseille in
Current site
Google Scholar
PubMed
Close
,
Paul de ValkKNMI, De Bilt, Netherlands

Search for other papers by Paul de Valk in
Current site
Google Scholar
PubMed
Close
, and
John de VriesKNMI, De Bilt, Netherlands

Search for other papers by John de Vries in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

Denial experiments, also denoted observing system experiments (OSEs), are used to determine the impact of an observing system on the forecast quality of a numerical weather prediction (NWP) model. When the impact is neutral or positive, new observations from this observing system may be admitted to an operational forecasting system based on that NWP model. A drawback of the method applied in most denial experiments is that it neglects the operational time constraint on the delivery of observations. In a 10-week twin experiment with the operational High-Resolution Limited-Area Model (HIRLAM) at KNMI, the impact of additional ocean surface wind observations from the Advanced Scatterometer (ASCAT) on the forecast quality of the model has been verified under operational conditions. In the experiment, the operational model was used as reference, parallel to an augmented system in which the ASCAT winds were assimilated actively. Objective verification of the forecast with independent wind observations from moored buoys and ASCAT winds revealed a slight improvement in forecast skill as measured by a decrease in observation-minus-forecast standard deviation in the wind components for the short range (up to 24 h). A subjective analysis in a case study showed a realistic deepening of a low pressure system over the North Atlantic near the coast of Ireland through the assimilation of scatterometer data that were verified with radiosonde observations over Ireland. Based on these results, the decision was made to include ASCAT in operations at the next upgrade of the forecasting system.

Corresponding author address: Gert-Jan Marseille, KNMI, Wilhelminalaan 10, 3732 GK De Bilt, Netherlands. E-mail: gert-jan.marseille@knmi.nl

Abstract

Denial experiments, also denoted observing system experiments (OSEs), are used to determine the impact of an observing system on the forecast quality of a numerical weather prediction (NWP) model. When the impact is neutral or positive, new observations from this observing system may be admitted to an operational forecasting system based on that NWP model. A drawback of the method applied in most denial experiments is that it neglects the operational time constraint on the delivery of observations. In a 10-week twin experiment with the operational High-Resolution Limited-Area Model (HIRLAM) at KNMI, the impact of additional ocean surface wind observations from the Advanced Scatterometer (ASCAT) on the forecast quality of the model has been verified under operational conditions. In the experiment, the operational model was used as reference, parallel to an augmented system in which the ASCAT winds were assimilated actively. Objective verification of the forecast with independent wind observations from moored buoys and ASCAT winds revealed a slight improvement in forecast skill as measured by a decrease in observation-minus-forecast standard deviation in the wind components for the short range (up to 24 h). A subjective analysis in a case study showed a realistic deepening of a low pressure system over the North Atlantic near the coast of Ireland through the assimilation of scatterometer data that were verified with radiosonde observations over Ireland. Based on these results, the decision was made to include ASCAT in operations at the next upgrade of the forecasting system.

Corresponding author address: Gert-Jan Marseille, KNMI, Wilhelminalaan 10, 3732 GK De Bilt, Netherlands. E-mail: gert-jan.marseille@knmi.nl

1. Introduction

In this paper we assess the impact of ocean surface wind observations from the Advanced Scatterometer (ASCAT) in the operational Royal Netherlands Meteorological Office (KNMI) High-Resolution Limited-Area Model (HIRLAM) setting (Unden 2002). HIRLAM is a consortium of European meteorological institutes for cooperative research with the aim of developing and maintaining a short-range weather forecasting system for operational use by the participating meteorological institutes. However, the exact implementation of the forecasting system differs between institutes. In the remainder, when we refer to the HIRLAM model, we refer to the implementation at KNMI that is discussed in section 2.

Many weather centers have assimilated scatterometer wind data operationally in their global models since the early 1990s and have demonstrated improved forecast skill (Bi et al. 2011; Hersbach and Janssen 2007). Some studies were carried out for regional models as well, like the European Remote Sensing Satellites-1 and -2 (ERS-1/2) in HIRLAM (Stoffelen and van Beukering 1997; Pirkka 2010) and the Quick Scatterometer (QuikSCAT) in the fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5; Singh et al. 2008), and also showed positive impacts. An increasing number of institutes use scatterometer wind data operationally in their regional models, yet reporting on impact assessment for NWP in an operational context is limited.

The impact of observing systems in a NWP model is generally determined by denial experiments; that is, by comparing analyses and forecasts of a control experiment using all observations in the analysis with a similar experiment denying the observation type under investigation. Most of these experiments are performed offline, not considering the time latency of observations, that is, the time delay between the observation time and the actual availability of the observation. However, in daily operations, latency in data availability has a direct effect on the performance of regional models with short assimilation windows of typically 1–6 h. Data latency is therefore taken into account in the experiments discussed in the following sections. For synoptic surface observations the latency is on the order of minutes, while for radiosondes it can be up to 1 h from the moment the balloon is launched. The availability of satellite data for assimilation is related to the downlink moment of the data to the ground station. ASCAT is situated on the polar-orbiting meteorological operational satellite (MetOp-A) giving global coverage of ocean surface winds (Figa-Saldaña et al. 2002). Only during a satellite overpass can data be downlinked to a ground station. At the time of our experiments (2010), the only ground station available for MetOp-A was at Spitsbergen (Svalbard, Norway). The satellite orbit period of about 100 min implied that only half of the observations were available within the required 1 h, 15 min required for use in HIRLAM (see section 2). The so-called level-0 data need processing, which adds to the latency. As a consequence, the limited number of available satellite observations due to data latency reduces the amount of information from the observing system for the operational HIRLAM model and regional models in general. This was recognized and direct data broadcasts over several areas have been implemented since spring 2011 with the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) Advanced Retransmission Service (EARS; http://www.eumetsat.int/Home/Main/Satellites/GroundNetwork/EARSSystem/EARS-ASCAT/index.htm?l=en) to enhance the Atlantic Ocean coverage through additional data dumps from ascending–descending passes over several stations covering the HIRLAM area, including Maspalomas in the Canary Islands; Lannion, France; and Athens, Greece. Also, since spring 2011, the data are dumped twice per orbit, at an additional ground station in Antarctica (McMurdo).

As a result of KNMI’s ongoing effort to incorporate and optimize the use of observations in NWP, we demonstrate the impact of assimilating ASCAT winds in the operational KNMI HIRLAM environment of 2010, taking into account the limited availability of observations due to latency for the 2010 situation. The remainder of this paper is organized as follows. First, an overview of the HIRLAM operational environment at KNMI is given in section 2, including the nesting of the various model domains, the time scheduling implemented to execute the model over the different domains, the use of observations, and the assimilation method. This is followed by a description of the ASCAT ocean surface wind product in section 3. Section 4 discusses the experimental setup and the impact of assimilating ASCAT data as demonstrated through the verification of wind forecasts against independent wind data from buoys and ASCAT. Section 5 presents a case study of the actual deepening of a low pressure system in the model simulation through the addition of ASCAT data. The last section summarizes the conclusions and provides an outlook on future work.

2. KNMI HIRLAM operational environment

The HIRLAM forecasting system employs a limited-area primitive equations forecast model for the time evolution of the atmospheric state and a three-dimensional variational data assimilation system (3DVAR; Gustafsson et al. 2001; Lindskog et al. 2001) in cycling mode to blend prior information on the state of the atmosphere from the forecast model with available observations to provide the initial state for a subsequent short-range forecast. The operational HIRLAM system at KNMI in 2010 was version 7.2.1 (www.hirlam.org).

The HIRLAM system is operated in a nested configuration. Model simulations are performed on two domains, the outline of which is shown in Fig. 1. The simulation runs have different cycling schedules: 4 times per day on the larger domain, in the remainder called the D11 cycle (or D11 run or suite), and 8 times per day on the smaller domain, in the remainder called the H11 cycle (or H11 run or suite). The model equations are identical for both cycles. For both D11 and H11, analyses and forecasts are conducted; for D11 every 6 h with a forecast length of 48 h, and for H11 every 3 h with a forecast length of 24 h. Both cycles have a horizontal grid box size of 11 km. D11 has 60 vertical levels and H11 has 40 vertical levels, both ranging from the surface to the top of the model atmosphere at 0.1 hPa.

Fig. 1.
Fig. 1.

HIRLAM model domains for the D11 (6 h) and H11 (3 h) cycles.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

Assimilation of observations into NWP models aims at finding the best possible estimate of the atmospheric state, also denoted as analysis, given a short-term forecast, also denoted as the first guess or background, from a previous analysis, new available observations, and their respective error characteristics. The analysis serves as forecast initial state from which the forecast is obtained through the integration of the model equations in time and space. Observations used in HIRLAM are discussed below.

For the H11 cycle, a 3-h forecast from the previous H11 run is used as the background in the analysis. The D11 cycle uses a 6-h forecast from the previous D11 run as the background. Because HIRLAM is a limited-area model, the forecast at the domain boundaries is constraint by the model in which it is embedded (i.e., D11 for the H11 run). The D11 run is nested in the global forecast fields from the European Centre for Medium-Range Weather Forecasts (ECMWF). H11 and D11 have different observation cutoff times (i.e., the time span between the actual start of the model run, which defines the end of the time window of used observations, and the analysis time). H11 (D11) uses an observation window of 2 h, 30 m (5 h) with a cutoff time of 1 h (2 h). Table 1 summarizes the main characteristics of the H11 and D11 runs. As an example, the 1200 UTC analysis for the H11 run uses the 3-h forecast from the 0900 UTC analysis as background. All observations in the time window from 1030 until 1300 UTC are used in the analysis. The analysis calculation starts at the window end at 1300 UTC. Forecasts (FCs) are produced from the analyses with a maximum range of 24 h.

Table 1.

Main characteristics of the HIRLAM D11 and H11 cycles. Analyses for D11 are obtained for 0000, 0600, 1200, and 1800 UTC (tan) (i.e., a 6-h cycle). H11 analyses are also obtained at 0300, 0900, 1500, and 2100 UTC (i.e., a 3-h cycle). The observation time window around the analysis time is obtained from the cycle interval tc and the cutoff time tcut through [tantc/2, tan + tcut]. The start of the analysis is at tan+ tcut.

Table 1.

Observations used for the operational run at KNMI are generally obtained through the Global Telecommunication System (GTS), which is dedicated to the exchange of meteorological observations. Other observation types are also available, such as GPS atmospheric delay, wind and pressure observations from drifting and moored buoys, and wind profiler observations, but these are currently not used in the H11 and D11 suites at KNMI. At present the observation set used in the assimilation is composed of Aircraft Meteorological Data Relay (AMDAR) data (wind and temperature), radiosonde data (wind, temperature, and humidity), and synoptic station observations over land and sea (pressure). Figure 2 shows the typical observation coverage for the 1200 UTC analysis time observation window. AMDAR observations are collected from civil aviation aircraft yielding wind and temperature observations of high spatial density during the ascent and descent flight stages near airports. Over the Atlantic Ocean, at cruise level, observations are obtained only at 7-min intervals corresponding to about 105-km spatial separation. Radiosonde launch sites and surface synoptic observation (synop) stations are generally located over land, with some sparse stations over the ocean. From Fig. 2 it is clear that the oceanic region of the model domain is poorly sampled, in particular for wind and temperature. Satellite data fill this gap.

Fig. 2.
Fig. 2.

Typical HIRLAM data coverage for the 1200 UTC analysis window as obtained from the KNMI experimental observation monitoring tool, including (left) radiosonde profiles, (center) aircraft (AMDAR) observations, and (right) surface observations from synoptic stations, ships, and drifting buoys. Black (gray) symbols denote observations at locations used (not used) in the H11 analysis. Gray-colored symbols are either outside the H11 domain as displayed in Fig. 1 or flagged as suspect at the quality control stage.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

Observation handling at HIRLAM involves observation screening, reformatting of observation data structures, storage, and the generating and plotting of observation statistics (e.g., Lindskog et al. 2001). In the screening, logical checks are performed such as a location check for the observations in the model domain. Consistency checks (e.g., no ship observations over land), blacklisting (whitelisting) to forcefully exclude (include) observations from particular stations and/or in particular areas, quality control in the form of a background- or first-guess check to reject observations not consistent with prior information, etc. Additional observation screening is present in the 3DVAR analysis. Variational quality control (Järvinnen and Unden 1997) is built in to reduce or eliminate the influence of observations that are inconsistent with the current solution for the analyzed model state in the early stages of the minimization. The 3DVAR assimilation scheme assumes that all observations are valid at analysis time. Observations from synop stations are available at 1-h intervals, but only those observations closest to the analysis time are used in the analysis. Spatial observation thinning is not applied in the operational 3DVAR environment at KNMI.

Pressure and wind observations from drifting and moored buoys are not used in the HIRLAM analyses. Yet, as will be shown later, it was found that wind observations from a selection of moored buoys are of good quality and can be used for the verification of model forecast fields. In general, buoy wind observations are time averaged over 10-min intervals to reduce observation noise. Moored buoys have also been used for the validation of scatterometer winds (Stoffelen 1998; Hersbach et al. 2007; Vogelzang et al. 2011). Figure 3 shows the locations of moored buoys whose observations are used for verification in section 4. Table 2 summarizes the usage of the baseline observations in the operational HIRLAM environment.

Fig. 3.
Fig. 3.

Locations (triangles) and instrument identification numbers of moored buoys.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

Table 2.

Summary of baseline observations and usage in the HIRLAM analysis, with Y and N denoting yes and no, respectively.

Table 2.

3. ASCAT scatterometer ocean surface wind observations

The ASCAT scatterometer on board the low-earth polar-orbiting satellite MetOp-A is designed to measure the electromagnetic backscatter by the wind-roughened ocean surface (Figa-Saldaña et al. 2002). ASCAT has six antennas: three measuring to the left and three measuring to the right of the satellite track. The system covers two 500-km swaths that are separated from the satellite ground track by about 360 km (see Fig. 4).

Fig. 4.
Fig. 4.

(left) Ascending and (right) descending ASCAT tracks for three different gray-colored overpasses. The overpass time (UTC) is displayed next to the swaths. The dots denote the WVC centroids at which the wind observations are defined. Their spatial separation is 25 km. The ASCAT swath has 21 WVCs across track and is 500 km wide. Each panel shows three overpasses on 27 April 2010: (left) at around 1951 UTC (rightmost swaths, which are in time for the 2100 UTC analysis), 2133 UTC (in time for the 2100 UTC analysis), and 2313 UTC (in time for the 0000 UTC analysis but outside the HIRLAM H11 domain; see Fig. 1) and (right) at around 1003 UTC (observations arrive too late for the 0900 UTC analysis), 1143 UTC (observations arrive too late for the 1200 UTC analysis), and 1324 UTC (observations arrive too late for the 1500 UTC analysis). See the text for further details.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

ASCAT wind information is organized into wind vector cells (WVCs) projected on the instrument swath. The number of WVCs determines the sampling resolution for the surface wind field and the wind information is considered to be independent from one WVC to the next. Each WVC contains two to four ambiguous wind vector solutions that result from the inversion of CMOD-5, the geophysical model function for C-band radars also known as the “wind cone,” for a given set of backscatter values and a given scanning geometry (Stoffelen et al. 1997). Each wind ambiguity is characterized by a solution probability that is determined based on the distance-to-cone residual in the inversion.

The wind ambiguities, solution probabilities, and prior information from the ECMWF model 10-m background winds are used in a 2D variational ambiguity removal procedure (Vogelzang et al. 2009) to produce an analyzed surface wind field. This wind field is then used to select the wind vector ambiguity in each WVC that is closest to the analysis, based on vector difference, as the solution for the observed surface wind. A wind vector solution flag is set to the index of the selected wind ambiguity in each WVC.

Finally, the backscatter measurements, wind ambiguities, scanning geometry, and wind vector solution flag, among others, are made available as an ASCAT wind product (see http://www.knmi.nl/scatterometer) in Binary Universal Form for the Representation of Meteorological Data (BUFR) and Network Common Data Form (NetCDF) formats. Current ASCAT products include wind speed and direction information at either 25- or 12.5-km spacing. The experiments described in this paper use the 25-km product that has an accuracy of 1.3 m s−1 in wind speed and 16° in wind direction when compared to collocated ECMWF model winds (Verspeek et al. 2010; Vogelzang et al. 2011).

The orbital period of MetOp-A is about 100 min. At the time of the experiments full orbit data were downlinked only once per orbit to the Svalbard ground station. In addition to the downlink time, some further latency is introduced due to processing of the backscatter signals to ocean surface winds and the dissemination of observations through the GTS. As a consequence, the total time latency for descending orbits is more than 120 min, so these data arrive too late for assimilation in HIRLAM. On the other hand, observations from ascending orbits inside the model domain may be assimilated if they fit within the observation window.

Given the above constraints, the HIRLAM observation cutoff time of 1 h (see Table 1), and the H11 model domain (see Fig. 1), effectively about half of the total number of ASCAT observations in the model domain can be used in the analysis (see Fig. 5). The left panel of Fig. 4 shows a typical example, with three overpasses from the ascending orbit node. Data from the first overpass (two rightmost swaths) around 1951 UTC and the second overpass at around 2133 UTC are in time for the 2100 UTC analysis with the observation window [1930, 2200]. The leftmost overpass at 2313 UTC is outside the HIRLAM H11 domain, so the data cannot be used for the 0000 UTC analysis. Data from the descending orbit nodes in the right panel of Fig. 4 are downlinked to Svalbard about 1.5 h after observation time and therefore arrive too late for the analysis.

Fig. 5.
Fig. 5.

ASCAT observation usage in HIRLAM. (bottom) Total number of available assimilations (gray) and number used for assimilation (black) of ASCAT observations per day in the H11 domain depicted in Fig. 1 for the period 24 Apr–7 Jul 2010. (top) The differences in the gray and black curves seen in the bottom panel corresponding to the number of unused ASCAT observations. The missing part of the curves near 26 Jun is due to an interruption in operations because of a system upgrade.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

The HIRLAM implementation of ASCAT winds is de-signed to assimilate the components of the wind vector ambiguities. Before ASCAT observations are admitted to the analysis, they first have to undergo a screening procedure. The screening of ASCAT observations consists of a location check for each WVC against the HIRLAM domain and a check on the observation time for each across-track row of WVCs against the observation time window of a given assimilation cycle. In addition, a threshold check is performed for the presence of sea ice and land. Finally, the WVC quality flag from the ASCAT wind product is used to ensure the use of winds based on high-quality and complete backscatter measurements and a successful inversion. Because the ASCAT wind information consists of wind ambiguities, no first-guess check is carried out and, in the analysis, variational quality control is not active for ASCAT data.

In 3DVAR, the cost function
e1
is minimized. The component terms in (1) are quadratic forms expressing the “distance” between the analysis state and the prior or background state and the observations, respectively. The cost function Jo comprises the contributions of individual observation types:
e2
For ASCAT, the cost function is defined as
e3
where
e4
is the cost of the ith ambiguity; Nj is the number of ambiguities in observation j; (u, υ) and (ui, υj) are the analysis and ASCAT wind vector ambiguity components, respectively; σo,ASCAT is the expected standard deviation of the error in the ASCAT wind components with a value of 1.8 m s−1 (Vogelzang et al. 2009); Pi is the a priori solution probability (Portabella and Stoffelen 2004); and p is an empirical weight factor for the ambiguities, which currently has the value of four. This weight factor emphasizes the discrimination between the ambiguities and makes the expression for the cost function behave more as an “if” statement.

4. Experimental setup and results

In this section we demonstrate the added value of ASCAT winds for the operational KNMI HIRLAM cycle. Heretofore, we conducted an experimental model run (in the remainder denoted X11) parallel in time to the operational H11 run for the 10-week period of 24 April–7 July 2010, with the only difference being the additional use of ASCAT winds in the experimental run. The model domains for H11 and X11 are identical and the boundaries for both H11 and X11 are obtained from D11. The observation cutoff time of 1 h (see Table 1) and the model domain used (see Fig. 1) imply that only about half of the ASCAT data in the model domain can be used as discussed in the previous section. As an example, Fig. 6 shows the ASCAT observations used in the 2100 UTC experimental (X11) HIRLAM analysis of 27 April 2010. The assimilated swaths correspond to the first two (in time) overpasses in the left panel of Fig. 4.

Fig. 6.
Fig. 6.

ASCAT observations used in the experimental HIRLAM suite (X11) for the 2100 UTC analysis on 27 Apr 2010. Black areas denote locations where observations were used in the analysis, and dark gray areas denote locations not used in the analysis because of wind retrievals that did not pass the quality control [e.g., because of proximity to land, unusual sea state or other surface wind flow regimes not well modeled during the scatterometer wind retrieval; Stoffelen et al. (1997)].

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

The impact of ASCAT is assessed by comparing the model forecasts from both experiments with independent observations. Within an operational context it is also important to monitor the production chain and the timeliness of the model output for customer delivery, since adding observations to the assimilation will increase the computation time. The parallel setup of the ASCAT experiment guarantees a realistic assessment of the impact obtained from ASCAT observations in an operational environment.

The assimilation experiment was performed in a semi-operational environment. Both runs were started almost simultaneously, using exactly the same boundary conditions and baseline set of observations. The parallel run was performed from 24 April to 7 July 2010. Short interruptions did not cause a break in the assimilation chain of the X11 run through restarts in offline mode to catch up with the operational model. The restarts were initiated with the latest X11 analysis and identical boundaries and conventional observations as in the operational H11 run plus ASCAT.

The operational ASCAT 25-km product was used in the experiments. The specified observation error for both wind components, which determines the weight of the observations in the analysis, is 1.8 m s−1, similar to that used at ECMWF. It is well known that global models lack variance on scales below 200 km (Vogelzang et al. 2011). As a consequence, the representativeness error of closely spaced observations (separated by less than 200 km) is correlated. However, nowadays assimilation systems assume uncorrelated observations. To account for this inconsistency, NWP centers apply data thinning and/or inflate the observation error variances to reduce the weight of the observations in the analysis. ECMWF thins ASCAT data to 100-km observation separation. No data thinning or error inflation was applied in the HIRLAM X11 experiment; thus giving substantially more weight to the ASCAT observations in the analysis than is done by ECMWF. It should be noted that the forecast range of interest differs for both models, with ECMWF focusing on the midterm (5–10 days) and HIRLAM X11–H11 focusing on the short range (0–24 h). This might and most probably will lead to a different deployment of the observations in global and mesoscale models, as is further discussed in section 5.

Results of the assimilation of scatterometer observations is discussed in section 4a by comparing forecasts of wind and pressure from the H11 and X11 experiments with available independent observations from buoys, ships, and ASCAT. We focus on wind and pressure forecasts over the ocean, where maximum impact is expected. The last part of this section describes a case study.

a. Objective verification

Forecasts have been compared to independent scatterometer winds for the entire experimental period. Figure 7 shows a more than 3% reduction in the 10-m wind speed error standard deviation for the X11 experiment, including ASCAT winds, relative to the operational H11 experiment at FC + 6 h that gradually decreases to close to neutral at FC + 24 h. The boundary conditions are equal in both the H11 and X11 experiments, so the forecasts from both experiments will be nearly identical near the boundaries. Therefore, only scatterometer observations sufficiently remote from the domain boundaries have been used for a valid comparison. The most significant impact of the additional observations is expected in the first couple of hours of the forecast, because the analysis increments induced by the additional observations are advected out of the verification area for longer forecast ranges. Figure 7 shows a positive impact from the ASCAT wind assimilation through a reduction in both the 10-m wind speed bias and standard deviation over the full forecast range.

Fig. 7.
Fig. 7.

Bias and standard deviation of (of ) for 10-m wind speed from ASCAT and forecasts from the reference H11 run (gray) and experimental (including ASCAT) X11 run (black) for the experimental period 24 Apr–7 Jul 2010. The black dashed line shows the percentage change in standard deviation, with negative values denoting a standard deviation reduction, i.e., improved skill by assimilating ASCAT.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

Figure 8 shows the verification of both model suites on 10-m winds for the moored buoys displayed in Fig. 3. The wind observations from buoys are currently not assimilated. Yet, the buoy winds are of good quality as can be observed in Fig. 8 from the close to zero bias and standard deviations below 1.8 m s−1 for both runs. A positive impact is observed when assimilating ASCAT wind observations. The zonal (east–west) component shows a zero bias for both runs, while the standard deviation is reduced by slightly more than 2% at FC + 3 h and about 1% at FC + 24 h. For the meridional (north–south) component of the wind, the improvement in standard deviation reduction is even larger, from about 2% at FC + 3 h, 3% at FC + 12 h, and slightly more than 1% at FC + 24 h. The meridional wind component shows a small positive bias.

Fig. 8.
Fig. 8.

As in Fig. 7, but now verified against the (left) zonal (u) and (right) meridional (υ) wind components of moored buoys located as displayed in Fig. 3.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

Another objective way of verifying the observation impact is by evaluating the background departure variances; also denoted observation minus background or (ob) variances. Background and analysis departures are calculated and archived by default for all observing systems used in HIRLAM. Figure 9 shows that the model winds are consistently closer to buoy-measured winds when ASCAT observations are assimilated. This is true for all buoy locations and for both wind components. The (ob) standard deviation significantly decreases when using ASCAT observations from 1.69 to 1.63 m s−1 for the zonal wind component and from 1.69 to 1.62 m s−1 for the meridional wind component. Statistical significance was based on the F distribution (http://en.wikipedia.org/wiki/F-distribution) with parameters d1 = d2 = 14.535 and testing the null hypothesis of equal (ob) variances when using ASCAT data. The F values for the corresponding 90% and 99% confidence levels then equal 1.027 and 1.044, respectively. The quotients of the (ob) variances for the zonal and meridional components are 1.075 and 1.088, respectively (i.e., larger than the value for the 99% confidence level), meaning that the null hypothesis should be rejected with more than 99% confidence. The ASCAT impact on the reduction of the (ob) variance is significant in the sense of the described test above.

Fig. 9.
Fig. 9.

Background departure standard deviation of the HIRLAM model wind from buoy (left) zonal and (right) meridional wind components for all buoys separately. The locations of the buoys are in Fig. 3. The statistics are based on 14 535 observations over the complete 10-week experimental period. Black and gray bars correspond to the operational H11 and experimental X11 cycles, respectively.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

The background departure variance equals by definition the sum of the background error variance and the observation error variance. The latter is composed of the measurement error (instrument noise) variance and the representativeness error variance. For ASCAT observations the standard deviation of (ob) equals 1.40 and 1.41 m s−1 for the zonal and meridional wind components, respectively (i.e., smaller than found for buoy winds, which is as expected because buoys yield point observations while ASCAT observations are spatial averages by construction). The representativeness error, which is a substantial component of the background departure, is thus larger for buoys than for ASCAT.

Figure 10 shows the statistics of observations minus forecasts for surface pressure from moving ships and the moored buoys (and ships) displayed in Fig. 3. A positive impact is found for positive differences (black lines). Compared to moving platforms, the bias is slightly increased, while the standard deviation is smaller when using ASCAT. For moored platforms, a neutral impact is observed: slightly positive in the first 10 h of the forecast and slightly negative for larger forecast ranges.

Fig. 10.
Fig. 10.

As in Fig. 7, but now for pressure observations from (left) moving ships and (right) moored buoys and ships. Black dotted lines denote the differences between the absolute bias of the reference (H11) run and the absolute bias of the experimental ASCAT (X11) run. Positive differences imply positive impact.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

The quality of the surface pressure observations from drifting buoys is nonconstant for different buoys. Figure 11 shows the standard deviation of the difference of the observed surface pressure and the corresponding HIRLAM reference forecast for a number of buoys. Clearly, the quality differs per buoy. To reduce the impact of quality differences on the results, only four buoys, which match up best with the HIRLAM model, are used for verification (buoys 62513, 62504, 62517, and 44622). Their locations during the experimental period are displayed in the right panel of Fig. 12.

Fig. 11.
Fig. 11.

HIRLAM reference model (H11) surface pressure (hPa) error standard deviation against drifting buoys. Some of the buoy locations are displayed in the right panel of Fig. 12.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

Fig. 12.
Fig. 12.

(left) As in Fig. 10, but now for (right) the four drifting buoys and their displacement over the 10-week experimental period.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

The pressure observation minus forecast statistics for the selected four drifting buoys are shown in Fig. 12 (left panel). ASCAT data have a small negative impact on the bias and a small positive impact on the standard deviation with forecast time.

b. Case study: 6 June 2010

On 6 June 2010 at around 2100 UTC, a low pressure system was situated west of Ireland that moved eastward while slowly filling. On 7 June, the low pressure system reached Ireland around 1200 UTC. Figure 13 shows a similar positioning of the low pressure system for both operational H11 and experimental X11, including ASCAT, runs but with small differences in wind speed in the region of high wind speeds south and east of the low pressure system. The lowest values for the mean sea level pressure are 1000 and 997 hPa for the reference and experimental runs, respectively.

Fig. 13.
Fig. 13.

Analyses (top row) and successive forecasts (rows 2–6) of 10-m wind speed (grayscale) and mean sea level pressure (black contours) for the (left) reference (H11) run and (middle) the experimental (X11) ASCAT run. (right) The differences (X11 − H11) in mean sea level pressure (contours with values) and in wind speed. The amplitude of the latter is in the grayscale at the bottom with plus signs (+) denoting areas of increased wind speed in the X11 run. The analysis time is 2100 UTC 6 Jun 2010. Forecast are shown every 3 h up to FC + 15 h.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

The positioning of the low pressure system at analysis time is close to the HIRLAM domain boundary. The effect of assimilating the ASCAT winds is shown by the difference plots of wind speed and mean sea level pressure, depicted in the last column of Fig. 13. The low pressure system is deepened and maintained throughout the complete forecast range. Also, the wind speed increase close to the low pressure system is maintained. The used ASCAT observations are located east of the low pressure system (see Fig. 14). From the analysis increment it is clear that the maximum impact of the ASCAT observations is obtained near the ASCAT track but the analysis also spatially spreads the observation information to regions outside the swath. In addition, regions with plus signs in the increment indicate an increase in the simulated model wind through the addition of ASCAT observations.

Fig. 14.
Fig. 14.

(top left) The 10-m wind speed background— the circle in southwest Ireland marks the location of the observatory on Valentia, (top right) analysis, and (middle) analysis increment, i.e., analysis minus background for the reference (H11) run with plus signs denoting areas with a positive increment. (bottom) As in (top), but for the experimental X11 ASCAT run. (bottom right) The ASCAT coverage and measured wind speeds.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

Model wind forecasts from the operational and experimental run were verified against radiosonde data at Valentia (off the southwest coast of Ireland); the 3-, 6-, and 9-h forecasts initiated at 2100 UTC 6 June 2010 and verifying at 0000, 0600, and 1200 UTC 7 June 2010, respectively, were compared with successive radiosonde launches at these verification times. Figure 15 shows, on average a better match of winds from the experimental (ASCAT) run with the radiosonde winds, in particular for the meridional wind component.

Fig. 15.
Fig. 15.

Observation minus forecast for radiosondes launched at Valentia, marked by the circle in Fig. 14. Gray (black) curves correspond to the reference (experimental ASCAT) run. Forecasts are initiated from the 2100 UTC 6 Jun 2010 analysis. (left) The 3-h forecast minus the 0000 UTC 7 Jun radiosonde launch. (middle) The 9-h forecast minus the 0600 UTC 7 Jun radiosonde launch. (right) The 15-h forecast minus the 1200 UTC 7 Jun radiosonde launch.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00056.1

The assimilation of ASCAT wind observations resulted in a slight deepening of the low pressure system, as well as modestly stronger southerly winds at the Irish coast. Unfortunately, the deepening of the low pressure system could not be confirmed by pressure observations because of missing moored and drifting buoys close to the location of the low pressure system.

5. Summary, conclusions, and outlook

ASCAT ocean surface wind observations have been proven to be beneficial for the forecast quality of the regional HIRLAM model operational at KNMI. This was demonstrated by comparing the operational cycle with a parallel run over a 10-week period with additional ASCAT wind observations consisting of the ASCAT 25-km dataset provided by KNMI within the context of the Ocean and Sea Ice Satellite Application Facility project (OSI SAF; http://www.osi-saf.org/). All other conditions in the operational and parallel runs were kept constant, including the conditions at the domain boundaries. Generally, observing system impact experiments do not take into account the operational time constraint on the delivery of observations, which may not be a limitation for most global models, but is an important aspect for regional models with short cutoff times. This paper studied a clean impact experiment taking into account the operational conditions at KNMI, including the assimilation window cutoff time and observation delivery through the GTS. Observations arriving too late (later than the window cutoff time) were thus ignored in the analysis. It was shown that at the time that the experiments were conducted in 2010 about half of the ASCAT observations in the model domain could not be used due to these time constraints. Another aspect to consider within an operational context is the monitoring of the production chain and timeliness of the model output for customer delivery, since adding new observations for assimilation will increase the computation time.

Despite these limitations, wind forecasts verified better against independent (i.e., not assimilated) moored buoys and ASCAT observations with the use of available additional ASCAT data in the analysis. The most substantial improvement was found in the first 12 h with a 2%–3% reduction in the wind component observation-minus-forecast standard deviation, which decreased to 1% after 24 h. The impact on the 24-h forecast of surface pressure was neutral to slightly positive when compared to ship observations and drifting buoys. A case study showed a realistic deepening of a low pressure system in the North Atlantic near the coast of Ireland through the assimilation of scatterometer data that were verified with radiosonde observations over Ireland.

The ASCAT experiment was performed without additional steps to optimize the information content of the ASCAT observations for the HIRLAM model. The observations were used at the highest available density with 25-km spacing between observations. No data thinning or error inflation was applied, which is common practice at most NWP centers. It was found that the ASCAT background departure standard deviation is smaller than the specified 1.8 m s−1 observation error standard deviation, meaning that the actual observation error is substantially smaller than specified. The conservative use of observations in the analysis is attributed to the assumption of uncorrelated observation errors. However, nowadays models do not resolve scales below 200 km (Vogelzang et al. 2011), meaning that the representativeness error of closely spaced (<200-km separation) observations is correlated. For this reason error inflation is applied to observations with correlated errors to reduce their weight in the analysis.

The Hollingsworth and Lönnberg method (Hollingsworth and Lönnberg 1986) and/or the Desroziers diagnostic method (Desroziers et al. 2005) may be used to separate the observation error and background error from the background departures. In addition, the respective error correlation length scales may be obtained. A better characterization of observation (and model) errors will lead to a more optimal use of observations in NWP models. This will be an active area of research for the coming years, thereby keeping in mind the different focal points of global models (medium-range forecasts) and mesoscale models (nowcasting, short-range forecasts).

Data latency is an important aspect for the operational use of observations, in particular for regional models, which have a more frequent cycling than global models. At KNMI, experiments are ongoing with a rapid update cycle of 1 h. Latency is not taken into account in standard observing system experiments (OSEs) for observation system impact assessment. Standard OSEs therefore tend to overestimate the observation impact, in particular when applied to regional models. ASCAT-B observations from the European MetOp-B satellite, which was launched in September 2012, and expected future access to observations from Indian and Chinese scatterometers will substantially increase the coverage of ocean surface winds, but their optimal use for regional models will benefit from additional ground stations by delivering the data in time. For ASCAT, and other satellite instruments, the situation has improved substantially thanks to the EARS system. The addition of ground stations since 2011 in the HIRLAM area for the fast data delivery of satellite data has reduced the data latency substantially. The current thinking is that ASCAT’s impact on the operational HIRLAM model would be larger than has been demonstrated in this paper.

Future research activities at KNMI are related to the transition from the hydrostatical HIRLAM model to the nonhydrostatic HARMONIE model. Planned assimilation efforts will focus on the HARMONIE model, including the use of ASCAT, GPS water vapor, high-resolution aircraft observations from radar tracking (Mode-S) and radar winds, and reflectivity.

Acknowledgments

We thank our colleague Anton Verhoef for the composition of Fig. 4.

REFERENCES

  • Bi, L., Jung J. A. , Morgan M. C. , and Le Marshall J. F. , 2011: Assessment of assimilating ASCAT surface wind retrievals in the NCEP Global Data Assimilation System. Mon. Wea. Rev.,139, 3405–3421.

  • Desroziers, G., Berre L. , Chapnik B. , and Poli P. , 2005: Diagnosis of observation, background and analysis-error statistics in observation space. Quart. J. Roy. Meteor. Soc., 131, 33853396.

    • Search Google Scholar
    • Export Citation
  • Figa-Saldaña, J., Wilson J. J. W. , Attema E. , Gelsthorpe R. , Drinkwater M. R. , and Stoffelen A. , 2002: The Advanced Scatterometer (ASCAT) on the Meteorological Operational (MetOp) platform: A follow on for the European wind scatterometers. Can. J. Remote Sens., 28, 404412.

    • Search Google Scholar
    • Export Citation
  • Gustafsson, H., Berre L. , Hornquist S. , Huang X. Y. , Lindskog M. , Navascues B. , Mogensen K. S. , and Thorsteinsson S. , 2001: Three-dimensional variational data assimilation for a limited area model. Part I: General formulation and background constraint. Tellus, 53A, 425446.

    • Search Google Scholar
    • Export Citation
  • Hersbach, H., and Janssen P. , 2007: Operational assimilation of surface wind data from the MetOp ASCAT scatterometer at ECMWF. ECMWF Newsletter, No. 113, ECMWF, Reading, United Kingdom, 6–8.

  • Hersbach, H., Stoffelen A. , and de Haan S. , 2007: An improved C-band scatterometer ocean geophysical model function: CMOD5. J. Geophys. Res., 112, C03006, doi:10.1029/2006JC003743.

    • Search Google Scholar
    • Export Citation
  • Hollingsworth, A., and Lönnberg P. , 1986: The statistical structure of short-range forecast errors as determined from radiosonde data. Part 1: The wind field. Tellus, 38A, 111136.

    • Search Google Scholar
    • Export Citation
  • Järvinnen, H., and Unden P. , 1997: Observation screening and first-guess quality control in the ECMWF 3D-Var data assimilation system. Research Department Tech. Memo. 236, ECMWF, Reading, United Kingdom, 34 pp.

  • Lindskog, M., and Coauthors, 2001: Three-dimensional variational data assimilation for a limited area model. Part II: Observation handling and assimilation experiments. Tellus, 53A, 447468.

    • Search Google Scholar
    • Export Citation
  • Pirkka, O., 2010: Feasibility of assimilating ASCAT surface winds into a limited area. M.S. thesis, Dept. of Physics, University of Helsinki, 81 pp. [Available online at https://helda.helsinki.fi/bitstream/handle/10138/20968/feasibil.pdf.]

  • Portabella, M., and Stoffelen A. , 2004: A probabilistic approach for SeaWinds data assimilation. Quart. J. Roy. Meteor. Soc., 130, 126.

    • Search Google Scholar
    • Export Citation
  • Singh R., Pal P. K. , Kishtawal C. M. , and Joshi P. C. , 2008: The impact of variational assimilation of SSM/I and QuikSCAT satellite observations on the numerical simulation of Indian Ocean tropical cyclones. Wea. Forecasting,23, 460–476.

  • Stoffelen, A., 1998: Toward the true near-surface wind speed: Error modeling and calibration using triple collocation. J. Geophys. Res., 103 (C4), 77557766.

    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., and Anderson D. , 1997: Scatterometer data interpretation: Measurement space and inversion. J. Atmos. Oceanic Technol., 14, 12981313.

    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., and van Beukering P. , 1997: Improved backscatter processing and impact of tandem ERS winds on HIRLAM. HIRLAM Project Rep. 31, IMET, Dublin, Ireland, 86 pp.

  • Unden, P., 2002: HIRLAM-5 scientific documentation. HIRLAM-5 Project, Norrköpping, Sweden, 144 pp. [Available online at http://hirlam.org/index.php?option=com_docman&task=doc_details&gid=308&Itemid=139.]

  • Verspeek, J. A., Stoffelen A. , Portabella M. , Bonekamp H. , Anderson C. , and Figa J. , 2010: Validation and calibration of ASCAT using CMOD5.n. IEEE Trans. Geosci. Remote Sens., 48, 386395, doi:10.1109/TGRS.2009.2027896.

    • Search Google Scholar
    • Export Citation
  • Vogelzang, J., Stoffelen A. , Verhoef A. , de Vries J. , and Bonekamp H. , 2009: Validation of two-dimensional variational ambiguity removal on SeaWinds scatterometer data. J. Atmos. Oceanic Technol., 26, 12291245.

    • Search Google Scholar
    • Export Citation
  • Vogelzang, J., Stoffelen A. , Verhoef A. , and Figa-Saldaña J. , 2011: On the quality of high-resolution scatterometer winds. J. Geophys. Res.,116, C10033, doi:10.1029/2010JC006640.

Save