• Benjamin, S. G., Brewster K. , Brummer R. , Jewett B. F. , Schlatter T. W. , Smith T. L. , and Stamus P. A. , 1991: An isentropic three-hourly data assimilation system using ACARS aircraft observations. Mon. Wea. Rev., 119, 888906.

    • Search Google Scholar
    • Export Citation
  • Benjamin, S. G., Jamison B. D. , Moninger W. R. , Sahm S. R. , Schwartz B. E. , and Schlatter T. W. , 2010: Relative short-range forecast impact from aircraft, profiler, radiosonde, VAD, GPS-PW, METAR and mesonet observations via the RUC hourly assimilation cycle. Mon. Wea. Rev., 138, 13191343.

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

    • Search Google Scholar
    • Export Citation
  • de Haan, S., 2011: High-resolution wind and temperature observations from aircraft tracked by Mode-S air traffic control radar. J. Geophys. Res., 116, D10111, doi:10.1029/2010JD015264.

    • Search Google Scholar
    • Export Citation
  • Graham, R. J., Anderson S. R. , and Bader M. J. , 2000: The relative utility of current observation systems to global-scale NWP forecasts. Quart. J. Roy. Meteor. Soc., 126, 24352460.

    • Search Google Scholar
    • Export Citation
  • Moninger, W. R., Mamrosh R. D. , and Pauley P. M. , 2003: Automated meteorological reports from commercial aircraft. Bull. Amer. Meteor. Soc., 84, 203216.

    • Search Google Scholar
    • Export Citation
  • Moninger, W. R., Benjamin S. G. , Jamison B. D. , Schlatter T. W. , Smith T. L. , and Szoke E. J. , 2010: Evaluation of regional aircraft observations using TAMDAR. Wea. Forecasting, 25, 627645.

    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., Bonavita M. , Eyre J. , Goldberg M. , Järvinen H. , Serio C. , Thépaut J.-N. , and Wulfmeyer V. , 2006: Post-EPS developments on atmospheric sounding and wind profiling. EUMETSAT Position Paper, 36 pp. [Available online at http://www.eumetsat.int/groups/pps/documents/document/pdf_peps_pp_atmos_sound_wind.pdf.]

  • Undén, P., and Coauthors, 2002: HIRLAM-5 scientific documentation. HIRLAM Project Tech. Rep., 144 pp. [Available online at http://hirlam.org/index.php?option=com_content&view=article&id=64:general-description-of-the-hirlam-model&catid=48:synoptic-scale-model-hirlam&Itemid=101.]

  • WMO, 2003: Aircraft Meteorological Data Relay (AMDAR) reference manual. WMO-958, 84 pp. [Available online at http://www.wmo.int.]

  • WMO, 2004: Statement of guidance regarding how well satellites and in situ sensor capabilities meet WMO user requirements in several application areas. WMO/TD-1052, 52 pp. [Available online at http://www.wmo.int.]

  • WMO, 2006: Implementation plan for the evolution of space and surface-based sub-systems of the GOS. WMO/TD-1267, 20 pp. [Available online at http://www.wmo.int.]

  • WMO, cited 2011: Fourth WMO Workshop on the Impact of Various Observing Systems on NWP. WMO Tech. Rep. [Available online at http://www.wmo.ch/pages/prog/www/OSY/Reports/NWP-4_Geneva2008_index.html.]

  • View in gallery
    Fig. 1.

    The H11 and X11 NWP model areas.

  • View in gallery
    Fig. 2.

    Coverage and number of AMDAR observations available for assimilation for the period 1 Feb–10 Mar 2008.

  • View in gallery
    Fig. 3.

    Number of Mode-S observations available for assimilation during the period from 1 Feb to 10 Mar 2008. Note that the legend differs from that in Fig. 2.

  • View in gallery
    Fig. 4.

    Mean number of AMDAR and Mode-S observations per hour. The line with symbols denotes the AMDAR observations landing at Amsterdam Schiphol Airport.

  • View in gallery
    Fig. 5.

    Coverage of surface observations assimilated in the period 1 Feb–10 Mar 2008.

  • View in gallery
    Fig. 6.

    Coverage and number of radiosonde observations during 1 Feb–10 Mar 2008 at four synoptic hours (0000, 0600, 1200, and 1800 UTC).

  • View in gallery
    Fig. 7.

    Surface and radiosonde observations at De Bilt. (top) Wind speed (kt) and direction at 1200 UTC. (bottom) Pressure (black line) and temperature (minimum =dashed; maximum = dotted) at the surface observed hourly. The labels (a)–(f) refer to the weather analysis displayed in Fig. 8.

  • View in gallery
    Fig. 8.

    Synoptic weather analysis for the period 1 Feb–10 Mar 2008.

  • View in gallery
    Fig. 9.

    Statistics for the comparison of the model forecasts against Mode-S observations at the full and half hour for experiments Ref, RefM, and M01 at the (left) 875- and (right) 400-hPa levels. The bias and RMS are plotted for (top to bottom) temperature, wind speed, and wind direction. Thick lines represent the RMS; thin lines (lowest values) denote the bias between the model and observations.

  • View in gallery
    Fig. 10.

    As in Fig. 9, but for experiments Ref, A01, and MA1.

  • View in gallery
    Fig. 11.

    Statistics from the comparison of the model forecasts against AMDAR observations at the full and half hour. Thick lines represent the RMS; thin lines (lowest values) denote the bias between the model and observations.

  • View in gallery
    Fig. 12.

    Statistics from the comparison of the model forecasts against radiosonde observations from De Bilt. Thick lines represent the RMS (highest values); thin lines (lowest values) denote the bias between the model and observations.

  • View in gallery
    Fig. 13.

    “Real time” comparison of the different forecasts against Mode-S observations. For Ref the forecast length is minimal 2 h, while for MA01 it is maximal 2 h. The shaded area denotes the minimum and maximum of the bias and RMS computed from 20 subsets of the total dataset.

  • View in gallery
    Fig. 14.

    Analysis from KNMI based on HIRLAM analysis at 0000 UTC 25 Feb 2008.

  • View in gallery
    Fig. 15.

    Profiles at Schiphol Airport showing wind speed and direction observations (Mode-S, denoted by crosses, and the De Bilt radiosonde, denoted by circles) and model runs for Ref (3-h forecast, solid line) and MA01 (1-h forecast, dashed line) for two different times.

  • View in gallery
    Fig. 16.

    Surface pressure analysis and wind at 850 hPa based on HIRLAM valid at 1200 UTC 4 Mar 2008.

  • View in gallery
    Fig. 17.

    As in Fig. 15, but for 0600–2100 UTC 4 Mar 2008.

  • View in gallery
    Fig. 18.

    Analysis from KNMI based on HIRLAM analysis of 1200 UTC 10 Mar 2008.

  • View in gallery
    Fig. 19.

    As in Fig. 15, but for 1200 and 1500 UTC 10 Mar 2008. Note that the Ref forecast is now t = 4 for 1300 UTC. The MA01 forecasts are still t = 1.

  • View in gallery
    Fig. A1.

    Observation minus background statistics for Mode-S observations.

  • View in gallery
    Fig. A2.

    Observation minus analysis statistics for Mode-S observations.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 496 176 58
PDF Downloads 289 97 6

Assimilation of High-Resolution Mode-S Wind and Temperature Observations in a Regional NWP Model for Nowcasting Applications

Siebren de HaanKNMI, De Bilt, Netherlands

Search for other papers by Siebren de Haan in
Current site
Google Scholar
PubMed
Close
and
Ad StoffelenKNMI, De Bilt, Netherlands

Search for other papers by Ad Stoffelen in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

In this paper the beneficial impacts of high-resolution (in space and time) wind and temperature observations from aircraft on very short-range numerical weather forecasting are presented. The observations are retrieved using the tracking and ranging radar from the air traffic control facility at Schiphol Airport, Amsterdam, the Netherlands. This enhanced surveillance radar tracks all aircraft in sight every 4 s, generating one million wind and temperature observations per day in a radius of 270 km around the radar. Nowcasting applications will benefit from improved three-dimensional wind fields. When these observations are assimilated into a numerical model with an hourly update cycle, the short-range three-dimensional wind field forecasts match the observations better than those from an operational forecast cycle, which is updated every 3 h. The positive impact on wind in the first hours of the forecast gradually turns into a neutral impact, when compared to other wind and temperature observations. The timeliness of the forecasts combined with the high resolution of the observations are the main reasons for the observed nowcasting benefits. All in all, the assimilation of high-resolution wind (and temperature) observations is found to be beneficial for nowcasting and short-range forecasts up to 2–3 h.

Corresponding author address: Siebren de Haan, KNMI, Wilhelminalaan 10, De Bilt 3732 GK, Netherlands. E-mail: siebren.de.haan@knmi.nl

Abstract

In this paper the beneficial impacts of high-resolution (in space and time) wind and temperature observations from aircraft on very short-range numerical weather forecasting are presented. The observations are retrieved using the tracking and ranging radar from the air traffic control facility at Schiphol Airport, Amsterdam, the Netherlands. This enhanced surveillance radar tracks all aircraft in sight every 4 s, generating one million wind and temperature observations per day in a radius of 270 km around the radar. Nowcasting applications will benefit from improved three-dimensional wind fields. When these observations are assimilated into a numerical model with an hourly update cycle, the short-range three-dimensional wind field forecasts match the observations better than those from an operational forecast cycle, which is updated every 3 h. The positive impact on wind in the first hours of the forecast gradually turns into a neutral impact, when compared to other wind and temperature observations. The timeliness of the forecasts combined with the high resolution of the observations are the main reasons for the observed nowcasting benefits. All in all, the assimilation of high-resolution wind (and temperature) observations is found to be beneficial for nowcasting and short-range forecasts up to 2–3 h.

Corresponding author address: Siebren de Haan, KNMI, Wilhelminalaan 10, De Bilt 3732 GK, Netherlands. E-mail: siebren.de.haan@knmi.nl

1. Introduction

Advances in mesoscale numerical weather prediction are very relevant for nowcasting and short-range forecasting of, among others, extreme weather events. Of particular concern is the initialization of such models on small scales through the assimilation of a set of high-resolution observations that is fit for this purpose. Not only is higher spatial density required for analyzing small scales, but because these scales develop relatively fast, timely observations at high temporal density are also required. Furthermore, to specify the atmospheric dynamics, wind observations are most important for mesoscale numerical weather prediction (NWP). Most notably, high-resolution upper-air profile wind observations are lacking (Stoffelen et al. 2006; WMO 2004, 2006). An improvement in the initial wind field is expected to result in a better forecast of both wind and temperature. Upper-air temperature on the contrary is a large-scale parameter and is thus generally of less importance for mesoscale numerical weather prediction.

Radiosonde and aircraft are the main sources of information about the upper-air wind (WMO 2011), temperature, and humidity. The aircraft data used operationally at the Royal Netherlands Meteorological Institute (KNMI) are collected through the Aircraft Meteorological Data Relay (AMDAR) system. In this paper we will exploit another source of upper-air atmospheric wind and temperature information from aircraft obtained by the tracking and ranging surveillance radar of air traffic control operations at Schiphol Airport in Amsterdam, the Netherlands. Using the Rapid Update Cycle (RUC) model on a 3-h cycle from the National Centers for Environmental Prediction (NCEP), Benjamin et al. (1991) showed that the assimilation of aircraft data led to significant improvements in 3- and 6-h forecasts. Wind forecast errors in that range were reduced by approximately 10%.

AMDAR is the most important data source for winds over the Atlantic Ocean, and the second most important data source after radiosondes over North America in the Met Office model (Graham et al. 2000). Moninger et al. (2003) showed also that AMDAR data improve both the short- and the long-term weather forecasts. In Moninger et al. (2010) multiyear evaluation of Tropospheric AMDAR (TAMDAR) data showed a positive impact on 3-h RUC forecasts of temperature, relative humidity, and wind. Recent assimilation experiments (Benjamin et al. 2010) were performed with eight different observation data types. Assimilation experiments with the hourly RUC showed that aircraft observations had the largest overall impact on the forecast quality.

Only recently has a method for retrieving wind and temperature information from all aircraft in the vicinity of a tracking and ranging (TAR-1) radar been developed (de Haan 2011). Using TAR-1 radar data from air traffic control at Schiphol airport, de Haan (2011) showed that the wind information from this source has a quality comparable to AMDAR wind observations. Temperature information can also be inferred from TAR observations; however, the quality of these observations is lower than from AMDAR. TAR observations are gathered using the selective mode of the radar and are, therefore, called Mode-S observations. All aircraft within the range of the radar are polled, and the flight information obtained can be used to derive wind and temperature observations.

In this paper we show the impact of high-resolution wind (and temperature) observations from aircraft by assimilation into the High-Resolution Limited-Area Model (HIRLAM) NWP model (Undén et al. 2002). The impact is assessed by performing NWP experiments without and with the Mode-S data in HIRLAM. An hourly assimilation cycle will be applied to exploit the high resolution (in space and time) of these new observations. For reference, a 3-h cycle of HIRLAM will be used. This reference is kept close to the operational configuration. Although all experiments are performed offline, the settings (external data sources, observations, boundaries, etc.) are kept as close as possible to operational practice to have an assessment of the impact of these observations that is representative for operational practice at KNMI.

The next section gives a description of the NWP model used in this study followed by a description of the observations used in the assimilation experiments. A description of the general weather situation during the experiment from 1 February to 10 March 2008 is presented subsequently. This is followed by the results of the experiments and a discussion of the impacts on selected cases. The final sections are devoted to the conclusions and recommendations.

2. Numerical weather prediction

At KNMI, HIRLAM (Undén et al. 2002) is run operationally for the short-term weather forecast. The version used in this study is HIRLAM 7.0. This version was operational from November 2006 up to mid-January 2010.

Regional NWP models require lateral boundaries conditions providing information on the transient synoptic-scale flow. In the configuration used in this study, a double nesting of NWP models is used, close to the operational set up at KNMI. Two types of model cycles are used; the first model has a 3-h cycle of assimilation and forecast and uses European Centre for Medium-Range Weather Forecasts (ECMWF) boundaries on a domain covering large parts of Europe and the northern Atlantic Ocean (H11). The second model setup has an hourly cycle on a smaller domain (X11) and uses the boundaries from the 3-h-cycle forecasts. Figure 1 shows the H11 and X11 NWP regions. Both model configurations have a horizontal resolution of 11 km and 60 vertical levels, ranging from the surface to the top of the model atmosphere at 0.1 hPa.

Fig. 1.
Fig. 1.

The H11 and X11 NWP model areas.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

For both types of models different types of experiments are conducted through assimilating different sets of observations. Assimilation is performed with the aim of finding the best possible analyzed state of the atmosphere given the observations, an initial guess (also called background or first guess), and a priori defined constraints. The background state is a forecast from the previous analysis cycle.

The assimilation technique used here is three-dimensional variational data assimilation (3DVAR), which is widely used in NWP. The analysis is determined by minimizing a cost function including the background error covariances () and observations error covariances () together with observations (y) and the background state (xb). The cost function is defined as
e1
where H, the observation operator, maps the model fields onto the observations. Generally, H is nonlinear and a linearized approximation is used to find the minimum for J(x). The background error covariance matrix is balanced as described in Berre (2000). The matrix is adjusted to incorporate observations from Mode-S, according to the values found in de Haan (2011). Table 1 shows the values used in this study.
Table 1.

Observation errors.

Table 1.

Within the HIRLAM assimilation scheme, observations are preprocessed to remove redundancies. The aircraft observation dataset is reduced to obtain at least a separation between observations of 0.5 grid point distance (i.e., approximately 6 km) and a height difference of at least 50 hPa. AMDAR and Mode-S observations are thinned simultaneously (however with slightly different error characteristics).

In Table 2 the five different experiments are described. Two experiments have a 3-h cycle, while the other three have an hourly cycle. The boundaries used in the 3-hourly cycles (domain H11) are obtained from the operational ECWMF run, while the boundaries of the hourly runs are forecast fields of (at least) a 1-h forecast from the reference run (Ref-run). Because the domain on which the hourly experiments are performed is small (the X11 region; see Fig. 1), the lateral boundary conditions are updated hourly to reduce detrimental boundary effects. The main difference between the experiments is, apart from the domain and cycling period, the type and amount of observations used.

Table 2.

Experiments descriptions and observations (rapid AMDAR observations are collected within 10 min of analysis time).

Table 2.

The goal of this research is to assess the impact of Mode-S observations on very short-range numerical weather forecasts, with a focus on future (operational) nowcasting applications. It is thus essential to compare the operational forecast quality from the 3-h cycle with more advanced NWP configurations, both with higher density observations (Mode-S) and with faster cycling. For this reason the following experiments were set up: a reference run (3-h cycle, abbreviated as Ref) that mimics the operational HIRLAM runs, the reference run with additional Mode-S observations assimilated (RefM), an hourly reference run (with rapid available AMDAR but without Mode-S, called A01), an hourly run with only Mode-S (M01), and a hourly run with all available data (Mode-S and AMDAR, called MA01).

Comparing forecasts from Ref and RefM will expose the additional value of Mode-S in the operational setting. The hourly runs are made to demonstrate the additional value of hourly cycling in the presence of either high-density Mode-S observations or more spatially distributed but relatively sparse AMDAR. The combined Mode-S and AMDAR run further tests the complementarity of high-density, but rather local, Mode-S observations and sparse, but spatially well distributed, AMDAR. Comparing experiments in real time, taking into account the production time of the forecasts will disclose the best model setup for numerical nowcasting.

3. Observations and weather situation

In the current operational version of HIRLAM at KNMI, only surface pressure observations, radiosonde (wind, temperature, and humidity) observations, and AMDAR (wind and temperature) observations are used. Other synoptic-scale information (e.g., satellite) is provided through the lateral boundary conditions.

a. Aircraft observations

Aircraft are equipped with sensors that measure the speed of the aircraft, its position and ambient temperature, and pressure. Wind information can be derived from these measurements. At present, a selection of these observations are transmitted to a ground station using the AMDAR system. An atmospheric profile can be generated when measurements are taken during takeoff and landing. See WMO (2003) and de Haan (2011) for more details.

Recently, a new type of aircraft-related meteorological information has become available, which originates from the observations inferred from a tracking and ranging radar used for air traffic control. These data are called Mode-S (because it is using the surveillance mode of the radar). Wind information from AMDAR and Mode-S instruments is derived from the position of the aircraft reported by heading, ground track, and true airspeed. Heading is the direction in which the nose of the aircraft points, true airspeed is the speed of the aircraft with respect to the air, and the ground track delineates the motion of the aircraft relative to the ground. The wind vector is the difference between the motion of the aircraft relative to the ground and its motion relative to the air (defined by the airspeed and heading).

AMDAR temperature is obtained from direct readings of the sensors on board of the aircraft, while Mode-S temperature is inferred from the reported Mach number, true airspeed (Vt), and flight level (which is directly related to pressure). The relation between speed of sound, temperature, and the ideal gas law is used to estimate the air temperature T as follows:
e2
where T is in units of K. The Mach number, M, has no dimension. The vertical coordinate of the aircraft observations is generally expressed in flight levels, which is a height related to a standard atmosphere at the observed pressure (expressed in 100 ft). For example, FL100 is at pressure 696 hPa, FL200 at 465 hPa, FL300 at 300 hPa, and FL400 at 187 hPa (approximately).

In de Haan (2011) it was shown that, when heading correction and calibration on these observations are applied, good quality wind observations can be obtained. After applying the corrections and calibration, the wind observations from Mode-S are of nearly the same quality as the wind observations from AMDAR (typically RMS difference of 2–3.5 m s−1 with background winds, depending on height). The temperature observations are of worse quality as compared to AMDAR. Below, we discuss the observation density of AMDAR and Mode-S during the experiment period. All Mode-S observations used in this study are calibrated and corrected using the methods described in de Haan (2011).

1) AMDAR observations

AMDAR uses raw data available from the aircraft monitoring systems. Raw data rates typically vary from 1 to 16 samples per second. According to WMO (2003) in order to reduce the noise, the AMDAR reports are averages of observations over a certain time depending on the phase of flight. This averaging is done internally by an aircraft computer before transmitting the data: averaging times of 10 s for ascent and descent and 30 s above flight level FL200 (20 000 ft or 465 hPa) are typically used.

The coverage of AMDAR observations shows the main flight routes (see Fig. 2). The observations are concentrated around the main airports, such as those in London, United Kingdom (Heathrow); Paris, France; and Frankfurt, Germany.

Fig. 2.
Fig. 2.

Coverage and number of AMDAR observations available for assimilation for the period 1 Feb–10 Mar 2008.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

2) Mode-S observations

Mode-S data used in this paper are collected using the TAR-1 radar at Amsterdam Schiphol (EHAM) airport. In response to a request by the radar, each individual aircraft sends information on the current speed and heading and Mach number, from which temperature and wind can be inferred. The radar performs a full scan every 4 s and covers an area of 270 km around the radar. The vertical coverage is location dependent since it is limited by the curvature of the earth. The recorded messages contain information generated by the flight computer including the transponder identification number, flight level, Mach number, roll, true airspeed, and heading. The message is complemented with information on the position and ground track from the tracking radar. All aircraft are queried, resulting in about 1.5 · 106 raw Mode-S observations per day around Schiphol Airport. These raw observations are thinned in a fashion similar to that for the AMDAR observations and about 5% of the observations are used.

The coverage of Mode-S, shown in Fig. 3, immediately exposes the limited range of the observations due to the observation method. However, the number of observations is very large (note the differences in scale between the AMDAR and Mode-S coverage figures) thus potentially enabling observations of small- and fast-scale weather phenomena. The concentration of the observations around Schiphol is clearly visible and Brussels, Belgium, can also be detected from the coverage.

Fig. 3.
Fig. 3.

Number of Mode-S observations available for assimilation during the period from 1 Feb to 10 Mar 2008. Note that the legend differs from that in Fig. 2.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

Aircraft observations (both AMDAR and Mode-S) are limited to the operation hours of airports. This implies that fewer observations are gathered just after midnight and in the early morning hours. In Fig. 4 the mean numbers of AMDAR and Mode-S observations per hour are shown. Also shown in Fig. 4 are the numbers of AMDAR observations at Schiphol airport during the period under consideration. Clearly, the lowest numbers of observations are between 0000 and 0300 UTC. Note that in this period no Mode-S observations were collected between 0000 and 0200 UTC because the system was not fully operational at that time.

Fig. 4.
Fig. 4.

Mean number of AMDAR and Mode-S observations per hour. The line with symbols denotes the AMDAR observations landing at Amsterdam Schiphol Airport.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

b. Surface and radiosonde observations

Every hour, pressure observations from the available surface synoptic observation (SYNOP) stations can be used for assimilation. In Fig. 5, the locations of the pressure observations used in this study are shown. The coverage is mainly restricted to land, with some observations from platforms in the North Sea. Note that the SYNOP observations from Denmark, the Czech Republic, and Sweden are not included in the operational dataset for part of the period.

Fig. 5.
Fig. 5.

Coverage of surface observations assimilated in the period 1 Feb–10 Mar 2008.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

Besides SYNOP observations, radiosonde observations are used. Radiosondes measure a profile of temperature, wind, and humidity. A radiosonde system consists of a ground segment and a balloon to which a small lightweight container is attached. A temperature sensor and a humidity sensor are attached to the container. The radiosonde is launched approximately 40–50 min before the main synoptic hours (which are 0000, 0600, 1200, and 1800 UTC) to assure that it reaches the tropopause (i.e., 500 hPa) around the synoptic main hour. Because of the long duration of the flight (approximately 2 h), this observation method implies that the actual observation time at a certain height does not correspond to the profile time stamp and may differ by about an hour. Because of the wind, the radiosonde drifts away, and the observed profile does not correspond exactly to a profile above the launch site. This drift is ignored in the current HIRLAM 3DVAR system.

In Fig. 6 the coverage and number of radiosonde observations for the period 1 February–10 March 2008 are shown for four launch times. The main synoptic hours (0000 and 1200 UTC) show the largest number of launches; however, there are some locations where a radiosonde is launched four times per day.

Fig. 6.
Fig. 6.

Coverage and number of radiosonde observations during 1 Feb–10 Mar 2008 at four synoptic hours (0000, 0600, 1200, and 1800 UTC).

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

While excellent in the vertical, the distribution of the radiosonde network in Europe is too coarse in time and horizontal coverage for mesoscale modeling. Moreover, the observation frequency of a large number of radiosonde launches in Europe was reduced from four times per day to two in order to reduce the costs of the total meteorological observing network. Despite the coarse temporal and horizontal resolutions, radiosonde observations are a valuable source of information on temperature, humidity, and wind profiles in the atmosphere.

c. Synoptic weather situation

The experiments presented in this article span the period from 1 February to 10 March 2008. During this period the atmospheric flow displayed consistent variabilities. Figure 7 shows the observed surface pressure and temperature, as well as the upper-air winds.

Fig. 7.
Fig. 7.

Surface and radiosonde observations at De Bilt. (top) Wind speed (kt) and direction at 1200 UTC. (bottom) Pressure (black line) and temperature (minimum =dashed; maximum = dotted) at the surface observed hourly. The labels (a)–(f) refer to the weather analysis displayed in Fig. 8.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

During the first days of February, a number of fronts passed the Netherlands, as can be seen in Fig. 8a, in which the weather analysis is displayed for 1200 UTC 4 February. During these days the surface pressure gradually increases, while daily temperatures were between 1° and 10°C. This period of unsteady weather was followed by a high pressure ridge (see Fig. 8), which resulted in a blocking high pressure system over the British Islands (see Fig. 8c). The surface temperature showed strong daily fluctuation.

Fig. 8.
Fig. 8.

Synoptic weather analysis for the period 1 Feb–10 Mar 2008.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

Low pressure systems were able to pass the area around the Netherlands after the decay of the high pressure blockade (see Fig. 8d). The low pressure systems were followed by a high pressure region (see Fig. 8e), which was followed again by unsteady weather with some passing fronts.

The upper-air wind clearly showed changes with the passing of the different weather systems. Periods of high winds are followed by low wind periods. Although the experiment period was a winter period, different wind regimes were present.

Additional information on the weather for some days can be found in section 5, in which a number of selected cases are described.

4. Assimilation impact results

This section presents the assimilation impact results by comparing the wind and temperature forecast with AMDAR, Mode-S, and radiosonde observations. The forecasts are bilinear interpolated in space and linear interpolated in time; the minimum time interval between two forecasts is 1 h.

a. Comparison against Mode-S

Mode-S observations ±10 min around the full and half hour are used for validation. For the half-hour comparisons, a linear interpolation in time of the forecast fields is applied. In Fig. 9 the results are shown for the forecasts from zero (i.e., the analysis) to 4 h ahead. The statistics only show data from runs started at 0000, 0300, 0600, 0900, 1200, 1500, 1800, and 2100 UTC to have a one-to-one comparison between the forecast fields from the hourly cycles and the 3-hourly cycles.

Fig. 9.
Fig. 9.

Statistics for the comparison of the model forecasts against Mode-S observations at the full and half hour for experiments Ref, RefM, and M01 at the (left) 875- and (right) 400-hPa levels. The bias and RMS are plotted for (top to bottom) temperature, wind speed, and wind direction. Thick lines represent the RMS; thin lines (lowest values) denote the bias between the model and observations.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

In Fig. 9 the statistics of the model forecasts for the experiments Ref, RefM, and M01 are shown at two levels (400 and 875 hPa) to assess the impact of the Mode-S observations on the reference run and on the hourly run with only Mode-S assimilated. The first observation from Fig. 9 is that bias during the forecasts for the levels shown is reduced when Mode-S observations are assimilated (i.e., experiment M01). Obviously at analysis time, the two experiments using Mode-S observations show the lowest bias and RMS when compared to Mode-S. The bias shows no jump during the first hour, indicating consistency between the model and observations. For the two experiments using Mode-S data, it turns out that after a few hours the RMS converges to the RMS of the Ref-run, except for wind speed for the M01 run where the RMS becomes slightly larger. However, in the first hours of the forecast at the 875-hPa level, a positive impact on wind speed and direction is observed.

The slight degradation of the wind speed RMS with forecast time probably has its origin in poor analysis outside the Mode-S coverage. This is confirmed in the Fig. 10 where the impact of assimilation of additional AMDAR observations is shown. Figure 10 shows the bias and RMS for the experiments Ref (for reference), A01 (hourly assimilation of SYNOP and AMDAR), and MA01 (hourly assimilation of SYNOP, AMDAR, and Mode-S) at the 875- and 400-hPa levels. Clearly, A01 shows the worst statistics, while MA01 has better RMS scores than Ref, even at forecast lengths +3 and +4, especially for wind direction.

Fig. 10.
Fig. 10.

As in Fig. 9, but for experiments Ref, A01, and MA1.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

Overall, in the lowest level both hourly cycles using Mode-S observations show a positive impact (bias and RMS) when compared to Mode-S. The hourly run with only AMDAR observations shows a clear negative impact due to the absence of observations close to the Netherlands. At higher levels a positive impact of assimilating Mode-S is observed for RMS; however, this impact disappears after a certain forecast length, where the duration of the positive impact differs for different levels and parameters. Experiment M01 has a negative impact after 2 h (for wind speed RMS, see Fig. 9). The fact that Mode-S winds have a beneficial impact initially, when verified in the Mode-S cone of measurements, indicates that the weather situation is well analyzed and propagated initially. However, its deterioration beyond a few hours in the forecast, suggests that the initial conditions outside the Mode-S cone of measurements have deteriorated with the introduction of Mode-S winds and which clearly affects the forecast range beyond 2–3 h. This would mean that the Mode-S observation weights are reasonable, but that Mode-S information is spread erroneously due to the simplified background error covariance structures. This probably results in overfitting of the Mode-S observations away from the Mode-S cone of measurements. Adding information from AMDAR (mainly from surrounding airports, experiment MA01) also improves the impact for longer forecasts. Adding Mode-S information to the reference (RefM) shows a positive impact in the first hours of the forecast; however, for nowcasting applications the impact is small due to the latency of the forecasts from the reference run, which is between 2 and 5 h.

b. Comparison with AMDAR observations

Figure 11 shows the statistics for experiments Ref, M01, and MA01 when compared to AMDAR observations. Note that AMDAR data are present in different parts of the X11 domain than Mode-S and essentially have effects in domain areas away from the Mode-S data (partly near the X11 boundaries). The results of the other experiments are not shown, because the statistics were similar to the comparison with Mode-S presented earlier. Note that not all AMDAR observations used in the comparison are actually assimilated in the hourly runs due to the time of presence in the observation database with respect to the observation cutoff time of a model run (see Table 2).

Fig. 11.
Fig. 11.

Statistics from the comparison of the model forecasts against AMDAR observations at the full and half hour. Thick lines represent the RMS; thin lines (lowest values) denote the bias between the model and observations.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

By comparing Fig. 11 and Figs. 9 and 10, it is clear that the RMS for temperature is lower for all levels and models. This is in line with the results found in de Haan (2011): Mode-S temperature observations are of lower quality than are the AMDAR temperature observations (see also Table 1). For temperature, the benefit of assimilating additional AMDAR data is most visible at the lowest levels in the first 2 h of the forecast.

For wind speed we see a similar effect. The Ref-run performs better in the lowest levels up to 2 h into the forecast; after 2 h, no significant difference can be observed. The wind direction has worse statistics for the hourly runs in the lowest level. Figure 11 shows that the Ref-run has the best statistics in the first hours of the forecast. Apart from the lowest level, the run MA01 has the best statistics of the hourly runs with equal quality to those of the 3-h runs after 2 h into the forecast time. So, at the AMDAR locations, generally away from the Mode-S measurements, the assimilation of only Mode-S data has a detrimental effect. This is consistent with the above-mentioned overfitting of Mode-S winds away from the rather confined Mode-S cone of measurements. Assimilation of AMDAR in addition to Mode-S suppresses this effect and provides improved verification against AMDAR. The spatially well distributed AMDAR observations aid in controlling the detrimental effects of the assimilation of the confined Mode-S observations (see also the appendix).

Note that the wind statistics at the lowest level are puzzling. Because of the limited number of observations used in the comparison (around 1000 for AMDAR at 875 hPa over a 6-week period) it might still be that some undiscovered outliers influence the statistics. It is important to note that the number of AMDAR observations involving landings at Amsterdam Schiphol airport is very low (see Fig. 4, line with symbols). The decrease in wind direction RMS at 875 hPa is remarkable for a forecast range of 4 h for all model runs, while a local maximum in RMS occurs at around 2–3 h in the forecast. The relative scores at each forecast time are similar to those in the other panels though.

c. Comparison with radiosonde observations

Radiosonde observations obtained at De Bilt are used to evaluate the forecasts of the different experiments. Note that the De Bilt radiosonde profiles are almost completely within the Mode-S measurement cone and thus, given the aircraft verifications above, we would expect an independent and favorable verification of the MA01 experiments. The full (10 s) resolution radiosonde observations are used for the comparison. Note that only a subset of these observations are assimilated (i.e., 50 so-called significant levels). So, at t = 0 h some dependency of the analysis on the verification data may exist, most strongly in Ref since few other observation sources are used.

Because the Ref and RefM runs have a 3-h cycle and the radiosondes are launched only two times a day, it is only possible to compare radiosonde observations at the analysis time and at a 3-h forecast. For the other experiments, a comparison can be made with other forecasts ranges (starting from analysis times 0800, 0900, 1000, 1100, 1200, and 2000, 2100, 2200, 2300, and 0000 UTC).

In Fig. 12 the statistics for the Ref (open boxes), RefM (solid triangles), M01 (solid line) and MA01 (dashed line) experiments for four levels are shown. Experiment A01 is left out of this comparison since it showed no positive impact previously. For almost all levels at analysis time (t = 0), the temperature RMS for the Ref and RefM runs is smaller than for the MA01 run (the exception is level 400 hPa). The smaller RMS is not surprising, since these observations are assimilated in these runs. The temperature bias, however, is smaller for the hourly runs (except 875 hPa). Furthermore, the temperature bias for MA01 is more constant over the forecast window than M01, but the difference is not very large. At t = 3, the temperature RMS of MA01 is close to the RMS of the 3-hourly runs, with RefM being the smallest on all levels, although the difference at levels higher than 875 hPa is very small. Apart from the 400-hPa level, the MA01 run has a smaller RMS over the whole forecast window.

Fig. 12.
Fig. 12.

Statistics from the comparison of the model forecasts against radiosonde observations from De Bilt. Thick lines represent the RMS (highest values); thin lines (lowest values) denote the bias between the model and observations.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

The statistics for wind speed show a clear impact of Mode-S at the lowest level (Fig. 12, middle row). At the analysis time the 3-h runs have the smallest wind speed RMS at all levels. In the top three levels the hourly runs have a smaller wind speed RMS at t = 3 than do the 3-hourly runs. In the bottom level, MA01 has the smallest RMS over the whole period (including t = 3). The wind speed biases differ for the lowest levels. The MA01 runs are shown to have the smallest bias, apart from the first hours at the 875-hPa level.

For wind direction, at the analysis time, the RefM clearly shows the smallest RMS for the highest levels (Fig. 12, bottom row). The biases for the different experiments are close except in the lowest level, where the hourly runs have the smallest bias. At this level the RMS of MA01 is clearly better and more constant than M01. At t = 3, RefM and MA01 are close, while Ref and M01 have much larger wind direction RMSs.

In conclusion, the largest positive impact on wind speed and direction RMS was observed in the lowest levels with experiment MA01. The Ref analysis dependency on the radiosonde profiles is visible in Fig. 12. It is remarkable that, apart from the top level, RefM compares better to radiosondes at t = 0 than does Ref. This in itself confirms that a large consistency of the Mode-S observations and radiosonde observations exists.

d. Nowcasting quality

In the previous sections it was shown that assimilating Mode-S data in an hourly cycle has a positive impact on wind and temperature forecasts, when compared to Mode-S observations and radiosonde observations. However, after a few hours this positive impact has disappeared.

For operational purposes that require timeliness, the quality of the forecast in “real time” is essential. Real-time applications include nowcasting (forecast ranges up to a few hours) and supplying meteorological information for air traffic control to guide continuous descent approaches of aircraft. As will be shown next, real-time applications will benefit from rapid updates by an NWP system when Mode-S observations are assimilated and the update frequency is 1 h. Using an hourly cycle, the forecast length used for nowcasting applications will be between 1 and 2 h, while for a 3-hourly cycle the forecasts lengths will be between 2 and 5 h. On the other hand, more analyses have to be run and some types of observations arrive too late to be included in the assimilation (e.g., radiosondes, some satellite data).

In Fig. 13 the statistics from experiments Ref (solid line) and MA01 (dashed line) are shown when compared to Mode-S observations in real time. The real-time comparison of experiment M01 is almost identical to MA01, because differences between these experiments only show up at longer forecast times. In line with the above results, the best statistics are obtained when both Mode-S and AMDAR are assimilated. Also shown in Fig. 13 is an indication of the significance of the statistics. The gray areas indicate the minimum and maximum of the bias and RMS obtained from 20 subsets created from the observation dataset. The impact on the wind speed and wind direction RMS is significant (except for wind direction at 1000 hPa, where a neutral impact is observed). The wind direction bias is improved for levels higher than 1000 hPa. The temperature bias is improved throughout the whole vertical range.

Fig. 13.
Fig. 13.

“Real time” comparison of the different forecasts against Mode-S observations. For Ref the forecast length is minimal 2 h, while for MA01 it is maximal 2 h. The shaded area denotes the minimum and maximum of the bias and RMS computed from 20 subsets of the total dataset.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

The largest profit is obtained when the assimilation frequency is increased from once per three hours to once per hour. A small additional improvement is found when AMDAR information is assimilated as well (not shown). As stated before, adding AMDAR will improve the forecast over the whole forecast range, with the emphasis on t = 2 to 4, when compared to the hourly run without AMDAR. The RMS is reduced by 5% and the bias is closer to zero.

5. Results for selected cases

A number of cases have been selected that display differences between the Ref-run and the MA01 run. The selection criteria for the cases were based on 1) large differences in wind speed with either observations or between the models or 2) when large offsets of the Mode-S observations with the background were present. These criteria draw the attention to cases where the atmospheric conditions are uncertain and/or where Mode-S has a large impact—either beneficial or detrimental.

a. 24–25 February

A cold front passed over the Netherlands during the night of February (see Fig. 14). On the Dutch Wadden Islands (in the north), 10 mm of precipitation were observed.

Fig. 14.
Fig. 14.

Analysis from KNMI based on HIRLAM analysis at 0000 UTC 25 Feb 2008.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

Figure 14 shows the analysis of mean sea level pressure and 850-hPa wind at 0000 UTC 25 February based on the operational HIRLAM run. The cold front runs from the Bay of Biscay to the Baltic Sea and lies at the coast of the Netherlands. The wind speeds accompanying the cold front were moderate.

Figure 15 shows the observed Mode-S and AMDAR wind speed and direction (crosses) and radiosonde (circles) for three times. The 3-h wind forecasts from the Ref-run are depicted by the solid line and the 1-h forecasts are shown by the dashed line, both at corresponding verification times.

Fig. 15.
Fig. 15.

Profiles at Schiphol Airport showing wind speed and direction observations (Mode-S, denoted by crosses, and the De Bilt radiosonde, denoted by circles) and model runs for Ref (3-h forecast, solid line) and MA01 (1-h forecast, dashed line) for two different times.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

At 1800 UTC 24 February 2008 both models match the observations at the surface. Between 950 and 900 hPa, the MA01 run is clearly closer to the observations; both runs perform similarly for wind direction. At 0000 UTC 25 February 2008, in the proximity of the front, only wind observations from the De Bilt radiosonde were available (launched about 35 km away from Schiphol). The MA01 wind speed lies close to the radiosonde wind speed, while the NWP wind direction deviates from the observed, with MA01 closer to the radiosonde than Ref.

b. 4 March

On 4 March 2008, the general flow was from the North Sea bringing showers, some with snow. An occluded front has just left the Netherlands at 1200 UTC; on the North Sea, two troughs are heading toward the Netherlands. Figure 16 also shows the mean sea level pressure and wind at 850 hPa for 1200 UTC.

Fig. 16.
Fig. 16.

Surface pressure analysis and wind at 850 hPa based on HIRLAM valid at 1200 UTC 4 Mar 2008.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

Wind observations show a wind from north to northwest from 0600 to 2100 UTC (Fig. 17). The Ref forecast (t = 3) at 0600 UTC has a more westerly wind compared to the observations and the MA01 forecast (t = 1), where the latter matches the observations very well. At 1800 UTC, both wind speed forecasts are too high compared to the observations with MA01 closest, while the Ref wind direction forecast is better below 900 hPa. At 2100 UTC both forecasts underestimate the wind speed; the wind directions forecast from MA01 is clearly better.

Fig. 17.
Fig. 17.

As in Fig. 15, but for 0600–2100 UTC 4 Mar 2008.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

c. 1200 UTC 10 March

The last case is from 10 March 2008. The general weather was governed by disturbances in a westerly flow from the Atlantic Ocean passing over the United Kingdom toward the North Sea, On 10 March an occluded front passed over the Netherlands; this front was accompanied by rain and strong winds. Figure 18 shows the mean sea level pressure and 850-hPa winds. The general flow was southerly.

Fig. 18.
Fig. 18.

Analysis from KNMI based on HIRLAM analysis of 1200 UTC 10 Mar 2008.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

At 1200 UTC, both forecasts verify very well with the observations. At 1500 UTC, the Ref-forecast underestimates the wind. Observations show a wind maximum at 800 hPa; the Ref-forecast has a lower maximum at a lower altitude, while MA01 matches this maximum, but slightly overestimates the wind speed above the maximum. The wind direction forecast below 800 hPa from Ref is closer to the observations than MA01; above 700 hPa the opposite is true (Fig. 19).

Fig. 19.
Fig. 19.

As in Fig. 15, but for 1200 and 1500 UTC 10 Mar 2008. Note that the Ref forecast is now t = 4 for 1300 UTC. The MA01 forecasts are still t = 1.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

In general, the MA01 forecasts (t = 1) compare better to the observations than Ref (t = 2 to 4). This is consistent with the results presented earlier.

6. Conclusions

The impact of timely, high spatial and temporal resolution aircraft observations available around Schiphol Airport (de Haan 2011) is tested on the nowcasting time scale in the HIRLAM numerical weather prediction (NWP) system (Undén et al. 2002) and is found to be generally beneficial.

Upper-air observations, especially wind observations, are important for short-range weather forecasting of extreme weather and for meeting new requirements in aviation meteorology. To this end, in this paper, a novel method of measuring wind and temperature from aircraft is discussed and assimilation in HIRLAM is carried out. All aircraft up to a range of 270 km from the Mode-Selective (Mode-S) surveillance radar at Schiphol are polled every 4 s to extract observations of wind and temperature. The observations are accumulated in batches of 10 min and are available with almost no latency.

The impact is assessed by performing NWP experiments with and without the new data in HIRLAM, configured to be close to the settings, boundaries, and inputs used operationally at KNMI. The reported impacts are thus representative for KNMI’s operational practice.

An hourly assimilation cycle is applied to exploit the high resolution (in space and time) of these new observations. Given the relatively small domain of the new observations, we ran the experiments on a small HIRLAM domain after verifying negligible forecast impacts from domain size over the forecast range of interest of 4 h.

The general weather situation during the experiment from 1 February to 10 March 2008 was quite mixed with at times a well-developed jet, a meandering flow, and also a blocked flow. As a result the impact statistics vary over time, but generally show a beneficial impact in the most variable (and uncertain) weather. Note however that the period used here was a winter period; summer periods for example with severe convection are not addressed in the impact experiment.

Verification of the forecasts with independent Mode-S aircraft observations shows clear and beneficial analysis and short-range forecast impacts. When the Mode-S observations, only available in a rather limited domain, are complemented with AMDAR aircraft observations, available over a large region, then a clear synergetic effect emerges from both data sources. On the other hand, the AMDAR observations on their own do not improve the analyses and forecasts. This is a clear indication that the availability of more airport surveillance radar aircraft observations would further improve the analyses and forecasts in our area of interest (×11). A remaining puzzling result is the generally detrimental impact of AMDAR observations (without Mode-S) in the experiments. This could be due to the presence of observations near the domain boundary and/or a combination of poor vertical coverage over the domain, especially around the Netherlands, and suboptimal error covariances.

Verification with radiosondes at the De Bilt station confirms the above results. Although the analyses use the radiosonde information (at degraded resolution), it is revealing to see that the assimilation of Mode-S observations generally results in an improved fit of the analysis with the radiosonde observations. Radiosonde and Mode-S observations thus act in a complementary fashion in the data assimilation system.

Statistics from the observations minus background and observations minus analysis indicate that Mode-S aircraft observations are weighted relatively highly with respect to the background field (see the appendix). This is in line with our expectations and due to the high spatial and temporal density of the observations. The high spatial and temporal density does moreover offer the possibility of computing the background error correlation structure based on the actual weather conditions. This will be elaborated upon in a future study with the Mode-S data. It is expected that such improved and weather-dependent background error covariances further improve the beneficial exploitation of high-density winds, such as from Mode-S. In this article, the HIRLAM background and Mode-S observation error statistics were not modified to cope with these large observation densities. Therefore, it is encouraging to see the high analysis impact in combination with the beneficial forecast impact of Mode-S observations in an hourly assimilation cycle.

Case studies confirm the beneficial impact of Mode-S and illustrate the benefits in particular weather conditions.

Comparisons with observations in real time of forecast fields show a decrease in wind speed and wind direction RMS of nearly 5% over the whole profile when the assimilation cycle is increased from once per three hours to once per hour. The improvement is significant. Implementation of an hourly cycle leads to a clear enhancement in forecast skill, which is beneficial for nowcasting applications. For example, forecast fields after Mode-S data have been assimilated provide improved forecast RMS landing times at Schiphol Airport.

Acknowledgments

The authors thank LVNL for provision of raw Mode-S data. The time and effort of Gerrit Burgers and Sander Tijm (both KNMI) spent reading and giving feedback on this article are highly appreciated.

This study has been carried out in close cooperation with the Knowledge and Development Centre Mainport Schiphol in the Netherlands (KDC, http://www.kdc-mainport.nl).

APPENDIX

Observation Minus Background and Analysis

In section 4, the impact of additional AMDAR and Mode-S assimilations was shown. By comparison of observations minus background and observations minus analysis statistics, the reduction in RMS can be compared to a theoretical expected value based on a single observation. The Mode-S observation minus background bias and RMS are shown in Fig. A1 for the two runs that assimilate Mode-S observations hourly. The bias and RMS in the u component of the wind are reduced when additional AMDAR observations are introduced in the assimilation. The RMS in the υ component also decreases; the bias shows a mixed impact but this is small indeed. For temperature no significant improvements are observed.

Fig. A1.
Fig. A1.

Observation minus background statistics for Mode-S observations.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

When the analysis statistics of the Mode-S observations are inspected (see Fig. A2), one can see that there are small improvements in wind RMS as well. Again, the bias in the u and υ components is negligible. As before, no significant impact on temperature is noted (not shown). In both experiments Mode-S observations are assimilated; the difference between the experiments lies in the use of the AMDAR observations. Since both the Mode-S analysis statistics as well as the background statistics show an improvement when AMDAR observations are assimilated, it can be concluded that the background is closer to the observations, resulting in a better use of the observations in the analysis.

Fig. A2.
Fig. A2.

Observation minus analysis statistics for Mode-S observations.

Citation: Weather and Forecasting 27, 4; 10.1175/WAF-D-11-00088.1

To obtain some notion on the relative weight of the background, we compare background errors with analysis errors and assume
ea1
where 〈·〉 represents the mean, which holds for independent observation errors (with standard deviation of σo) and background errors (σb). Further, when we assume equal observation and background error standard deviations (σ) in M01, then
ea2
When assimilating a single observation, we expect that
ea3
and thus
ea4
which is a factor of 0.87 lower than the (ob) RMS difference. Looking at the plots, Mode-S impact appears to be slightly larger at 30% (see Figs. A1 and A2). Using the above formulas for σb = 1.7σo would imply roughly a 30% impact. Rather than applying sparse high quality observations (in this single observation test), the large impact here is more probably caused by providing densely spaced and consistent observations.

Nevertheless, when improves in MA01 by 10% with respect to M01, it indicates a background error that is improved by 20%. This would lead in this hypothetical case to an improvement in the analysis of 13% and a reduction of of 4%. Similarly, for a background error of 2σ, that is twice the observation error, the 10% improvement in would improve b by 13%, a by 6%, and (oa) by only 2.6%. This appears similar to our results, indicating the relatively large weight of the observations, likely due to their density. This leads to a clear beneficial impact in the Mode-S measurement cone. Detrimental effects due to Mode-S assimilation away from the Mode-S coverage are subsequently suppressed by the assimilation of AMDAR observations.

REFERENCES

  • Benjamin, S. G., Brewster K. , Brummer R. , Jewett B. F. , Schlatter T. W. , Smith T. L. , and Stamus P. A. , 1991: An isentropic three-hourly data assimilation system using ACARS aircraft observations. Mon. Wea. Rev., 119, 888906.

    • Search Google Scholar
    • Export Citation
  • Benjamin, S. G., Jamison B. D. , Moninger W. R. , Sahm S. R. , Schwartz B. E. , and Schlatter T. W. , 2010: Relative short-range forecast impact from aircraft, profiler, radiosonde, VAD, GPS-PW, METAR and mesonet observations via the RUC hourly assimilation cycle. Mon. Wea. Rev., 138, 13191343.

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

    • Search Google Scholar
    • Export Citation
  • de Haan, S., 2011: High-resolution wind and temperature observations from aircraft tracked by Mode-S air traffic control radar. J. Geophys. Res., 116, D10111, doi:10.1029/2010JD015264.

    • Search Google Scholar
    • Export Citation
  • Graham, R. J., Anderson S. R. , and Bader M. J. , 2000: The relative utility of current observation systems to global-scale NWP forecasts. Quart. J. Roy. Meteor. Soc., 126, 24352460.

    • Search Google Scholar
    • Export Citation
  • Moninger, W. R., Mamrosh R. D. , and Pauley P. M. , 2003: Automated meteorological reports from commercial aircraft. Bull. Amer. Meteor. Soc., 84, 203216.

    • Search Google Scholar
    • Export Citation
  • Moninger, W. R., Benjamin S. G. , Jamison B. D. , Schlatter T. W. , Smith T. L. , and Szoke E. J. , 2010: Evaluation of regional aircraft observations using TAMDAR. Wea. Forecasting, 25, 627645.

    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., Bonavita M. , Eyre J. , Goldberg M. , Järvinen H. , Serio C. , Thépaut J.-N. , and Wulfmeyer V. , 2006: Post-EPS developments on atmospheric sounding and wind profiling. EUMETSAT Position Paper, 36 pp. [Available online at http://www.eumetsat.int/groups/pps/documents/document/pdf_peps_pp_atmos_sound_wind.pdf.]

  • Undén, P., and Coauthors, 2002: HIRLAM-5 scientific documentation. HIRLAM Project Tech. Rep., 144 pp. [Available online at http://hirlam.org/index.php?option=com_content&view=article&id=64:general-description-of-the-hirlam-model&catid=48:synoptic-scale-model-hirlam&Itemid=101.]

  • WMO, 2003: Aircraft Meteorological Data Relay (AMDAR) reference manual. WMO-958, 84 pp. [Available online at http://www.wmo.int.]

  • WMO, 2004: Statement of guidance regarding how well satellites and in situ sensor capabilities meet WMO user requirements in several application areas. WMO/TD-1052, 52 pp. [Available online at http://www.wmo.int.]

  • WMO, 2006: Implementation plan for the evolution of space and surface-based sub-systems of the GOS. WMO/TD-1267, 20 pp. [Available online at http://www.wmo.int.]

  • WMO, cited 2011: Fourth WMO Workshop on the Impact of Various Observing Systems on NWP. WMO Tech. Rep. [Available online at http://www.wmo.ch/pages/prog/www/OSY/Reports/NWP-4_Geneva2008_index.html.]

Save