• Bennett, A. F., , and Leslie L. M. , 1981: Statistical correction of the Australian Region Primitive Equation Model. Mon. Wea. Rev., 109, 453462.

    • Search Google Scholar
    • Export Citation
  • Coakley, J., , and Bretherton F. , 1982: Cloud cover from high resolution scanner data: Detecting and allowing for partially filled fields of view. J. Geophys. Res., 87 (C7), 49174932.

    • Search Google Scholar
    • Export Citation
  • Daniels, J., , Velden C. , , Busby W. , , and Irving A. , 2000: Status and development of operational GOES wind products. Proc. Fifth Int. Winds Workshop, Lorne, VIC, Australia, EUMETSAT P 28, 27–45.

  • Davies, T., , Cullen M. J. P. , , Malcolm A. J. , , Mawson M. H. , , Staniforth A. , , White A. A. , , and Wood N. , 2005: A new dynamical core for the Met Office’s global and regional modelling of the atmosphere. Quart. J. Roy. Meteor. Soc., 131, 17591782.

    • Search Google Scholar
    • Export Citation
  • Holmlund, K., 1998: The utilization of statistical properties of satellite-derived atmospheric motion vectors to derive quality indicators. Wea. Forecasting, 13, 10931104.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Pescod N. , , Seaman R. , , Mills G. , , and Stewart P. , 1994: An operational system for generating cloud drift winds in the Australian region and their impact on numerical weather prediction. Wea. Forecasting, 9, 361370.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Leslie L. M. , , and Bennett A. F. , 1996a: Tropical Cyclone Beti—An example of the benefits of assimilating hourly satellite wind data. Aust. Meteor. Mag., 45, 275279.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Spinoso C. , , Pescod N. R. , , and Mills G. A. , 1996b: Estimation and assimilation of hourly high spatial resolution wind vectors based on GMS-5 observations. Aust. Meteor. Mag., 45, 279284.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Leslie L. , , Morison R. , , Pescod N. , , Seecamp R. , , and Spinoso C. , 2000a: Recent developments in the continuous assimilation of satellite wind data for tropical cyclone track forecasting. Adv. Space Res., 25, 10771080.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Pescod N. , , Seecamp R. , , Rea A. , , Tingwell C. , , Ellis G. , , and Shi H. , 2000b: Recent advances in the quantitative generation and assimilation of high spatial and temporal resolution satellite winds. Proc. Fifth Int. Winds Workshop, Lorne, VIC, Australia, EUMETSAT P 28, 47–56.

  • Le Marshall, J. F., , Mills G. , , Pescod N. , , Seecamp R. , , Puri K. , , Stewart P. , , Leslie L. M. , , and Rea A. , 2002: The estimation of high density atmospheric motion vectors and their application to operational numerical weather prediction. Aust. Meteor. Mag., 51, 173180.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Rea A. , , Leslie L. , , Seecamp R. , , and Dunn M. , 2004a: Error characterization of atmospheric motion vectors. Aust. Meteor. Mag., 53, 123131.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Seecamp R. , , Daniels J. , , Velden C. , , Puri K. , , Bowen R. , , Rea A. , , and Dunn M. , 2004b: The contribution of GOES 9 to operational NWP forecast skill in the Australian region. Aust. Meteor. Mag., 53, 279283.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Seecamp R. , , Dunn M. , , Velden C. , , Wanzong S. , , Puri K. , , Bowen R. , , and Rea A. , 2008: The contribution of locally generated MTSAT-1R atmospheric motion vectors to operational meteorology in the Australian region. Aust. Meteor. Mag., 57, 359365.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Seecamp R. , , Xiao Y. , , Steinle P. , , Sims H. , , Skinner T. , , Jung J. , , and Le T. , 2011: The generation and assimilation of continuous AMVs with 4DVar. Aust. Meteor. Oceanogr. J., 61, 117123.

    • Search Google Scholar
    • Export Citation
  • Leslie, L. M., , Le Marshall J. F. , , Morison R. P. , , Spinoso C. , , Purser R. J. , , Pescod N. , , and Seecamp R. , 1998: Improved hurricane track forecasting from the continuous assimilation of high quality satellite wind data. Mon. Wea. Rev., 126, 12481258.

    • Search Google Scholar
    • Export Citation
  • Puri, K., , Dietachmayer G. , , Mills G. A. , , Davidson N. E. , , Bowen R. M. , , and Logan L. W. , 1998: The new BMRC Limited Area Prediction System, LAPS. Aust. Meteor. Mag., 47, 203223.

    • Search Google Scholar
    • Export Citation
  • Rea, A., 2004: The determination and accuracy of atmospheric motion derived from sequential satellite observations in the Australian region. Ph.D. thesis, Royal Melbourne Institute of Technology, 206 pp.

  • Smith, W. L., , Harrison F. W. , , Revercomb H. E. , , Bingham G. E. , , Huang H. L. , , and Le Marshall J. F. , 2000: The Geostationary Imaging Fourier Transform Spectrometer. Proc. 11th Int. TOVS Study Conf., Budapest, Hungary, International TOVS Working Group, 391–398.

  • Stephens, G. L., and Coauthors, 2002: The CloudSat mission and the A-Train: A new dimension of space-based observations of clouds and precipitation. Bull. Amer. Meteor. Soc., 83, 17711790.

    • Search Google Scholar
    • Export Citation
  • Teweles, S., , and Wobus H. , 1954: Verification of prognostic charts. Bull. Amer. Meteor. Soc., 35, 455463.

  • Velden, C., and Coauthors, 2005: Recent innovations in deriving tropospheric winds from meteorological satellites. Bull. Amer. Meteor. Soc., 86, 205223.

    • Search Google Scholar
    • Export Citation
  • Zapotocny, T. H., , Jung J. A. , , Le Marshall J. F. , , and Treadon R. E. , 2008: A two-season impact study of four satellite data types and rawinsonde data in the NCEP Global Data Assimilation System. Wea. Forecasting, 23, 80100.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    (a) MTSAT-2 IR and VIS AMVs generated around 0032 UTC 9 Sep 2012. Magenta denotes upper-level tropospheric vectors, and yellow shows lower-level tropospheric vectors. (b) MTSAT-2 IR and VIS AMVs over the Tasman Sea around 0000 UTC 5 Apr 2011.

  • View in gallery

    (a) Measured error (m s−1) vs EE for low-level MTSAT-1R IR winds (1 Sep–9 Oct 2009). (b) Measured error (m s−1) vs EE for low-level MTSAT-1R IR winds (27 Jan–23 Feb 2011).

  • View in gallery

    Level of best fit to radiosonde data vs count for Southern Hemisphere MTSAT-1R upper-level (250–350 hPa) AMVs for November 2009.

  • View in gallery

    (a) Control 48-h forecast of 850-hPa geopotential height valid 0000 UTC 24 Apr 2011. (b) Experimental 48-h forecast of 850-hPa geopotential height (using continuous AMVs) valid 0000 UTC 24 Apr 2011. (c) Verifying analysis of 850-hPa geopotential height valid 0000 UTC 24 Apr 2011.

  • View in gallery

    (a) The rms difference between forecast and verifying analysis geopotential height at 24 h for ACCESS-R (blue) and ACCESS-R with AMVs (red) for the period 1 Sep–10 Oct 2009. (b) As in (a) but at 48 h. (c) As in (a) but for wind at 24 h. (d) The S1 difference between forecast and verifying analysis MSLP out to 48 h for the period 1 Sep–10 Oct 2009.

  • View in gallery

    (a) The rms difference between forecast and verifying analysis geopotential height at 24 h for ACCESS-R (blue) and ACCESS-R with AMVs (red) for the period 27 Jan–23 Feb 2011. (b) As in (a) but at 48 h. (c) As in (a) but for wind at 24 h. (d) The S1 difference between forecast and verifying analysis MSLP out to 48 h for the period 27 Jan–23 Feb 2011.

  • View in gallery

    Forty-eight-hour forecasts using AMVs at analysis time (green) and using hourly AMVs (red), along with the best track (black) for TC Dianne. The starting time for the forecasts is 1200 UTC 16 Feb 2011.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 40 40 9
PDF Downloads 8 8 0

The Operational Generation of Continuous Winds in the Australian Region and Their Assimilation with 4DVAR

View More View Less
  • 1 CAWCR, Bureau of Meteorology, Melbourne, Victoria, Australia
  • 2 Bureau of Meteorology, Melbourne, Victoria, Australia
  • 3 CAWCR, Bureau of Meteorology, Melbourne, Victoria, Australia
  • 4 JCSDA, Camp Springs, Maryland
  • 5 CAWCR, Bureau of Meteorology, Melbourne, Victoria, Australia
© Get Permissions
Full access

Abstract

Atmospheric motion vectors (AMVs) have been generated continuously from Multifunctional Transport Satellite 1 Replacement (MTSAT-1R) radiance data (imagery) since 2005, and more recently from MTSAT-2, which are operated by the Japan Meteorological Agency (JMA). These are the primary geostationary meteorological satellites observing the western Pacific, Asia, and the Australian region. The vectors are used operationally, for analysis in the Darwin Regional Forecast Office. The near-continuous AMVs have been stringently error characterized and used in near-real-time trials to gauge their impact on operational regional numerical weather prediction (NWP), using four-dimensional variational data assimilation (4DVAR). The use of these locally generated hourly vectors (the only hourly AMV source in the region at the time) and 4DVAR has resulted in both improved temporal and spatial data coverage in the operational regional forecast domain. The beneficial impact of these data on the Bureau of Meteorology’s (Bureau’s) current operational system is described. After these trials, the hourly MTSAT AMVs were introduced into the Bureau’s National Meteorological and Oceanographic Centre’s (NMOC) operational NWP suite for use by the operational Australian Community Climate Earth System Simulator (ACCESS) regional and global models, ACCESS-R and ACCESS-G, respectively. Examples of their positive impact on both midlatitude and tropical cyclone forecasts are presented.

Corresponding author address: John Le Marshall, Bureau of Meteorology, 700 Collins St., Docklands, Melbourne VIC 3008, Australia. E-mail: j.lemarshall@bom.gov.au

Abstract

Atmospheric motion vectors (AMVs) have been generated continuously from Multifunctional Transport Satellite 1 Replacement (MTSAT-1R) radiance data (imagery) since 2005, and more recently from MTSAT-2, which are operated by the Japan Meteorological Agency (JMA). These are the primary geostationary meteorological satellites observing the western Pacific, Asia, and the Australian region. The vectors are used operationally, for analysis in the Darwin Regional Forecast Office. The near-continuous AMVs have been stringently error characterized and used in near-real-time trials to gauge their impact on operational regional numerical weather prediction (NWP), using four-dimensional variational data assimilation (4DVAR). The use of these locally generated hourly vectors (the only hourly AMV source in the region at the time) and 4DVAR has resulted in both improved temporal and spatial data coverage in the operational regional forecast domain. The beneficial impact of these data on the Bureau of Meteorology’s (Bureau’s) current operational system is described. After these trials, the hourly MTSAT AMVs were introduced into the Bureau’s National Meteorological and Oceanographic Centre’s (NMOC) operational NWP suite for use by the operational Australian Community Climate Earth System Simulator (ACCESS) regional and global models, ACCESS-R and ACCESS-G, respectively. Examples of their positive impact on both midlatitude and tropical cyclone forecasts are presented.

Corresponding author address: John Le Marshall, Bureau of Meteorology, 700 Collins St., Docklands, Melbourne VIC 3008, Australia. E-mail: j.lemarshall@bom.gov.au

1. Introduction

Australia’s position in the data-sparse great southern oceans has resulted in dependence on satellite remote sensing and satellite data assimilation for high quality analysis and NWP in the region. Atmospheric motion vectors (AMVs) are an important part of the satellite database, being one of the most important satellite observations for tropical cyclone track prediction (Zapotocny et al. 2008) and also being important for regional prediction (Le Marshall et al. 2008). To provide accurate, high spatial and temporal density winds in a timely fashion for operational NWP and for climate studies in the Australian region, AMVs have been operationally calculated locally, from sequential Geostationary Meteorological Satellite (GMS) images (Le Marshall et al. 1994, 2002), Geostationary Operational Environmental Satellite-9 (GOES-9) images (Le Marshall et al. 2004b), and now Multifunctional Transport Satellite 1 Replacement (MTSAT-1R) and MTSAT-2 images (Le Marshall et al. 2008). The operational processing and use of hourly AMVs was first introduced into the Australian operational regional forecast system in 1996 (Le Marshall et al. 1996b) using intermittent data assimilation. MTSAT-1R and MTSAT-2 have been used to provide AMVs at 15-min intervals 4 times daily, and at half-hourly or hourly intervals for the rest of the time. Winds have been generated using infrared (11 μm), visible (0.5 μm), and water vapor absorption (6.7 μm) band images. Here, we examine the generation, quality control, and application of winds generated from triplets of 15-min-interval infrared imagery every 6 h and from triplets of images separated by 0.5 or 1 h. The hourly generation of winds has allowed their use in four-dimensional variational data assimilation (4DVAR), where for example their utility for tropical cyclone prediction has previously been demonstrated (Le Marshall et al. 1996a, 2000a; Leslie et al. 1998).

2. The generation of MTSAT-1R and MTSAT-2 atmospheric motion vectors

In the Bureau of Meteorology (Bureau), the methods used to estimate AMVs from MTSAT-1R and MTSAT-2 data are described in Le Marshall et al. (2008). When MTSAT-1R replaced GOES-9 in 2005, methods similar to those employed at the National Oceanic and Atmospheric Administration (NOAA; see Daniels et al. 2000; Velden et al. 2005) and also at the Bureau (Le Marshall et al. 2000b) were used to generate AMVs from the MTSAT-1R high-resolution image data (HiRID) received at the Crib Point ground station, in Victoria, SE Australia. Later, MTSAT-1R and MTSAT-2 high-rate information transmission (HRIT) data were used to generate continuous AMVs in real time (Le Marshall et al. 2011). Three sequential images from MTSAT-1R and MTSAT-2 were navigated using land features to ensure that there was consistency between images used for estimating cloud displacement. In this system, target selection for infrared (11 μm) targets commenced with a search for tracers, using bidirectional brightness temperature gradients within 15 × 15 pixel boxes. As in the case of GOES-9 (Le Marshall et al. 2004b), targets with gradient features were subjected to a spatial coherence analysis (Coakley and Bretherton 1982) and tracked using lagged correlation.

Height assignment methods were similar to those employed for GOES-9 (Le Marshall et al. 2004b). As with GMS-5, the error characteristics of these vectors were determined and each vector was associated with error indicators such as the expected error (EE; Le Marshall et al. 2004a) and quality indicator (QI; Holmlund 1998), as well as with a correlated error and the length scale described in section 3.

Figure 1a is an example of the local MTSAT-2 visible (VIS) and infrared (IR) image based AMVs generated around 0632 UTC on 9 September 2012. Figure 1b is an example of the local MTSAT-2 VIS image based AMVs, generated around 0000 UTC on 5 April 2011. Here, low level means 700–959 hPa, middle means 400–699 hPa, and high is 150–399 hPa.

Fig. 1.
Fig. 1.

(a) MTSAT-2 IR and VIS AMVs generated around 0032 UTC 9 Sep 2012. Magenta denotes upper-level tropospheric vectors, and yellow shows lower-level tropospheric vectors. (b) MTSAT-2 IR and VIS AMVs over the Tasman Sea around 0000 UTC 5 Apr 2011.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00018.1

A representative part of the schedule for generating MTSAT-1R based winds in the Southern Hemisphere can be seen in Table 1a. The schedule shows that, for an 11-h period, 13 triplets of MTSAT-1R infrared channel 1 (IR1) images (indicated by a boldface Y in the IR1 column of Table 1a; e.g., 2230, 2330, and 0030 UTC images on 16 June 2008) are used to track features and generate 26 sets of AMVs for use in NWP and nowcasting applications. It also shows 12 high-resolution visible (HRV, 2-km resolution) triplets (indicated by the boldface Y in the HRV column of Table 1a) and 4 water vapor (WV) triplets (boldface Y in the WV column), which are used to track features and generate 24 and 8 sets of AMVs, respectively, over an 11-h period.

Table 1a.

A representative part of the schedule for Southern Hemisphere (SH) wind generation from MTSAT-1R. There are 26 infrared channel (IR1) based wind datasets, 24 HRV image based datasets, and 8 WV image based datasets from the full disk and SH images listed. Shown are image times (times 1–3), full disk image time in plain text, SH image time in boldface, and RT AMV calculation with a Y.

Table 1a.

The blank elements in the case of WV AMVs in Tables 1a and 1b indicate where time constraints currently restrict real-time operational estimation and application of vectors. The high-resolution visible vectors currently are calculated at half resolution (2 km) due to time constraints but are calculated at full resolution (1 km) across 3000 × 3000 pixel areas around named tropical cyclone positions.

Table 1b.

As in Table 1a, but for Northern Hemisphere wind generation. Here, there are 24 IR1 based wind datasets, 22 HRV image based datasets, and 4 WV image based datasets from the full disk and Northern Hemisphere images listed.

Table 1b.

A representative part of the schedule for generating MTSAT-1R based winds in the Northern Hemisphere may be seen in Table 1b.

3. Error characterization and quality control

Accurate error characterization and thorough quality control (QC) ensure the AMVs have a beneficial impact on NWP (Le Marshall et al. 2004a). The error characterization employed here has used the Bureau’s initial error flagging procedure (ERR), which involved a number of basic checks including the departure from a first guess provided by the Bureau’s operational NWP model, a vector pair acceleration check, and a tracer constancy check—the QI (Holmlund 1998) and the more recent EE (Le Marshall et al. 2004a).

The AMVs are systematically thinned using these error indicators to reduce the volume of data while maintaining good data coverage with average separations consistent with the length scale of the correlated error (see below). The thinning methodology has also ensured that the average errors are generally no larger than the analysis background error field of the forecast model as measured at radiosonde sites. The approach is detailed in Le Marshall et al. (2004a). The EE now has several components: the total root-mean-square (rms) error (m s−1) currently used operationally, the meridional and the zonal error components (m s−1), and the AMV height error (hPa). The total rms error component of the EE has been used in this study. A typical comparison of the EE with the actual error for low-level MTSAT-1R and MTSAT-2 IR winds is seen in Figs. 2a and 2b. Low-level plots are provided as the EE is used to thin and quality control operational low-level AMVs for operational NWP. The time periods involved correspond to the two data assimilation periods studied.

Fig. 2.
Fig. 2.

(a) Measured error (m s−1) vs EE for low-level MTSAT-1R IR winds (1 Sep–9 Oct 2009). (b) Measured error (m s−1) vs EE for low-level MTSAT-1R IR winds (27 Jan–23 Feb 2011).

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00018.1

Here, the root-mean-square error component of the EE has been compared to the actual rms error determined using contemporaneous radiosonde data within 150 km of the AMVs and it can be seen to be an effective tool for selecting high quality AMVs. Table 2a shows statistics for the real-time ACCESS-G based AMV system in 2009. Table 2b shows statistics for the 2011 operational ACCESS-G based AMV system. Quality control of AMVs to provide low-, middle-, and high level-data in the Bureau system has used all three error indicators (EE, QI, ERR). The typical level of accuracy of the MTSAT-1R AMVs available for NWP is given in Table 2a, which shows the mean magnitude of the vector difference (MMVD) and the root-mean-square vector difference (RMSVD), between MTSAT-1R AMVs used in part of this assimilation study and radiosonde winds in the Australian region, for 1 September–9 October 2009.

Table 2a.

MMVD and RMSVD between MTSAT-1R IR1 AMVs, forecast model first-guess winds, and radiosonde winds for the period 1 Sep–9 Oct 2009.

Table 2a.
Table 2b.

As in Table 2a but for MTSAT-2 for the period 27 Jan–23 Feb 2011.

Table 2b.

Table 2b shows the MMVD and RMSVD for MTSAT-2 AMVs also used in part of the assimilation study (which are reported upon later), as well as radiosonde winds in the Australian region, for 27 January–23 February 2011. It may be noted that for lower-level AMVs the statistics show that the maximum separation between the radiosonde and the AMVs needs to be reduced to 30 km, to demonstrate that the AMVs are more accurate at radiosonde sites than is the model first-guess wind field. This is due to real gradients in the wind vector field. It should also be noted that the EE had to be used to thin and quality control the AMVs to this level of accuracy.

The accuracy of the height assignment of the AMVs has also been estimated using two methods. In the first, a match dataset, containing collocated contemporaneous AMVs, operational NWP fields, and radiosonde data, is used. In the other, a match dataset, containing collocated contemporaneous AMVs, operational NWP fields, and observations from the A-Train satellite members, CloudSat and Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO; Stephens et al. 2002), is used.

The first of these matched datasets, namely that containing collocated contemporaneous AMVs, operational NWP fields, and radiosonde data, has been used to compare the assigned height of the AMVs to the level where the atmospheric motion vectors best fit in the contemporaneous collocated radiosonde wind profile (Rea 2004). Typical results from this comparison are shown in Fig. 3, which shows the fit of infrared (IR) 11-μm-image-based upper-level MTSAT-1R AMVs, with heights assigned between 250 and 350 hPa, to radiosonde data for November 2009.

Fig. 3.
Fig. 3.

Level of best fit to radiosonde data vs count for Southern Hemisphere MTSAT-1R upper-level (250–350 hPa) AMVs for November 2009.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00018.1

The fit dataset appears to be good and represents a height bias (radiosonde − AMV height) of around +10 hPa and an rms difference of around 30 hPa. In addition to using the radiosonde level of the best fit to assess the height assignment, statistics have also been calculated using cloud-height error deduced from the cloud heights available from the A-Train members CloudSat and CALIPSO. The CloudSatCALIPSO cloud altitude data have been compared with AMV heights in cases where there is only a single level of cloud and the CloudSat and CALIPSO data are consistent. The results from this comparison typically indicate the fit of the IR 11-μm image based upper-level MTSAT-1R AMVs to the CloudSat data is good and shows a height bias of around +23 hPa and an rms difference of approximately 25 hPa for time and space separations of 5 min and 10 km, respectively.

The variation in the rms difference probably reflects the considerable variability among the match criteria used to calculate these differences; for example, distance and required degree of fit to the radiosonde wind for acceptance in the case of radiosondes, and for thresholding, to ensure similar cloud elements are being compared in the case of CloudSat and CALIPSO. It should be noted, in using these comparison numbers for improving cloud-height estimation, that absolute values are not as important as the relative use of the measures to monitor and improve the cloud-height assignment.

As with earlier Australian geostationary satellite AMV systems, the correlated error has been analyzed for the Bureau-produced MTSAT winds (Le Marshall et al. 2008). The correlated error and its spatial variation (length scale) were determined using the second-order autoregressive (SOAR) function:
e1
where R(r) is the error correlation, R0 and R00 are the fitting parameters (greater than 0), L is the length scale, and r is the separation of the correlates. Thus, the difference between AMV and the radiosonde winds (error) has been separated into correlated and noncorrelated parts. The parameters of the SOAR function that best fit the data are shown in Table 3.
Table 3.

Parameters of the SOAR function (see text) that best model the measured error correlations for the MTSAT-1R AMVs for two EE ranges.

Table 3.

4. AMV assimilation trials

Before local MTSAT-1R data were introduced into the Bureau’s 2007–10 real-time ACCESS NWP suite, real-time assimilation trials using MTSAT-1R IR1 AMVs were undertaken using the operational 2007 Local Analysis and Prediction System (LAPS; Puri et al. 1998), configured to run at 0.375° horizontal resolution with 61 levels in the vertical. The assimilation system and methodology are described in Le Marshall et al. (2011). The results from the trial were similar to, and consistent with, earlier results from the assimilation of GMS-5 and GOES-9 local AMVs (e.g., Le Marshall et al. 2002; Le Marshall et al. 2004b), showing positive forecast improvement through the troposphere. In effect, the trial showed that real-time MTSAT-1R IR and WV image based AMVs were of an accuracy that could benefit operational NWP in the Australian region. Addition of the vectors to the operational regional forecast system, which already contained some Japan Meteorological Agency (JMA) AMVs, provided both improved data coverage for the region and modest average forecast improvement (Le Marshall et al. 2008). These results and their consistency with previous studies led to the operational use of these local MTSAT-1R AMVs.

After the introduction of the ACCESS system at the Bureau, several short assimilation experiments were undertaken using local MTSAT-1R vectors. The impact of these time continuous local vectors, particularly when used with 4DVAR, was noted. An example of the differences recorded in 48-h forecasts from incorporation of continuous local vectors into the ACCESS system is seen in Fig. 4 opposite.

Fig. 4.
Fig. 4.

(a) Control 48-h forecast of 850-hPa geopotential height valid 0000 UTC 24 Apr 2011. (b) Experimental 48-h forecast of 850-hPa geopotential height (using continuous AMVs) valid 0000 UTC 24 Apr 2011. (c) Verifying analysis of 850-hPa geopotential height valid 0000 UTC 24 Apr 2011.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00018.1

It is clear in this case that the location of the complex low pressure system over New Zealand is improved and the secondary low in the Tasman Sea and that just east of New Zealand’s South Island are better defined by the use of continuous winds and 4DVAR.

5. The September–October 2009 and January–February 2011 ACCESS-based assimilation trials

In 2009, the Bureau introduced the real-time ACCESS regional NWP system (ACCESS-R), which subsequently became operational in August 2010. To test the utility of assimilating continuous wind observations using 4DVAR in the operational ACCESS-R system, AMVs generated using MTSAT half-hourly, hourly, and 4 times per day (0000, 0600, 1200, and 1800 UTC) 15-min interval images were assimilated. The periods studied were 1 September –10 October 2009 and 27 January–23 February 2011. The ACCESS-R system had a 0.375° horizontal resolution with 50 levels in the vertical. The assimilation system and methodology are described below. The results from the trials, as with earlier results from the assimilation of GMS-5, GOES-9, and MTSAT-1R local vectors (e.g., Le Marshall et al. 2002; Le Marshall et al. 2004b; Le Marshall et al. 2008 respectively), showed improved forecasts in the lower, middle, and upper troposphere from use of the continuous wind data.

a. The assimilation system

The ACCESS-R system is a regional implementation of the Met Office Unified Model (UKUM; see, e.g., Davies et al. 2005). The assimilation system used was similar to the real-time operational National Meteorological and Oceanographic Centre (NMOC) regional ACCESS-R system, which used all available data, including all available JMA AMVs. The analysis methodology on which the forecasts reported here were based was 4DVAR. The analyses were performed around 0000, 0600, 1200, and 1800 UTC and the time window for the analyses was +3 to −3 h. The forecast system was warm run and forecasts were nested in fields from the most recent ACCESS-G global forecast. The model configuration consisted of 220 × 320 grid points at 0.375° spacing in the horizontal and 50 levels in the vertical. The analysis was undertaken on a 110 × 160 grid with 0.75° spacing in the horizontal. The model and analysis time steps were both 15 min. Lateral boundary conditions for the forecast model were derived from the N144L50 ACCESS-G global forecast model. A summary of the characteristics of the ACCESS-R forecast system is presented in Table 4.

Table 4.

Characteristics of the ACCESS-R forecast system. Abbreviations used are AIRS = Atmospheric Infrared Sounder, ATOVS = Advanced Television and Infrared Observation Satellite (TIROS) Operational Vertical Sounder, SYNOP = surface synoptic observation, AMDAR = Aircraft Meteorological Data Relay, AIREP = aircraft report.

Table 4.

b. Assimilation method: 2009 trial

In the case of the September–October 2009 trial, the assimilation method employed was similar to that used operationally in NMOC in the regional ACCESS-R system (described above). The control forecast used the full real-time database, including all available JMA AMVs. The experimental system was similar but had local MTSAT-1R AMVs added to the operational, or control, database, which again included all available JMA AMVs.

The local MTSAT-1R AMVs employed were 15-min, ½-hourly, and 1-hourly IR and VIS image based AMVs. The AMVs were generated in real time using 11-μm (channel IR1), 0.55–0.9-μm (channel VIS), and 6.7-μm (channel IR3) imagery. The wind generation system used the real-time ACCESS-G system for a number of tasks including tracking, height assignment, and quality control. The characteristics of the real-time IR1 image based AMVs generated from 15-min interval images, available every 6 h, are shown in Table 2a. The characteristics of the IR1 image based AMVs generated from ½- and 1-hourly imagery were similar, but the ½- and 1-hourly images produced fewer vectors (Le Marshall et al. 2000b). The visible image based winds also appeared to have characteristics similar to the IR1 winds for the periods studied. Again, local quality control methods were used to provide vectors with a coverage and accuracy consistent with the length scale of the correlated error, the analysis background error, and the resolution of the data assimilation system.

For the control and experimental forecasts, the rms errors in wind, MSLP, temperature, and geopotential height and the S1 skill scores (Teweles and Wobus 1954) were calculated. The S1 skill scores were calculated on the NMOC operational verification grid using 0000 and 1200 UTC analyses. The verification grid consists of 58 points within the domain 15°–55°S, 90°–170°E. The exact grid is seen in Bennett and Leslie (1981).

c. Assimilation method: 2011 trial

In the case of the January–February 2011 trial, the assimilation system used was similar to the real-time operational NMOC regional ACCESS-R system of 2011, which used the full real-time database. The experimental system was similar but had local MTSAT-2 rather than MTSAT-1R AMVs added to the operational/control database. The analysis methodology, on which the forecasts reported here were based, was again 4DVAR. The analyses were again performed around 0000, 0600, 1200, and 1800 UTC and the time window for the analyses was +3 to −3 h. The forecast system was warm run and again forecasts were nested in fields from the most recent ACCESS-G forecast. In effect, this was a repeat of the 2009 trial using MTSAT-2 data.

d. Results

The rms difference between forecast and verifying analysis geopotential heights at 24 and 48 h for the control ACCESS-R system (operational database) and the experimental ACCESS-R system (operational database plus local continuous AMVs) is shown in Figs. 5a and 5b, for the period 1 September–10 October 2009. The rms differences between the forecast and verifying analysis wind field at 24 h and the S1 skill score for the forecast MSLP out to 48 h for the control ACCESS-R system and the experimental ACCESS-R system for the period 1 September–10 October 2009 are also shown in Figs. 5c and 5d.

Fig. 5.
Fig. 5.

(a) The rms difference between forecast and verifying analysis geopotential height at 24 h for ACCESS-R (blue) and ACCESS-R with AMVs (red) for the period 1 Sep–10 Oct 2009. (b) As in (a) but at 48 h. (c) As in (a) but for wind at 24 h. (d) The S1 difference between forecast and verifying analysis MSLP out to 48 h for the period 1 Sep–10 Oct 2009.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00018.1

It can be seen that by using 4DVAR with the ACCESS-R system, the addition of continuous AMVs to the ACCESS system has provided a consistent improvement to forecasts during the period 1 September–10 October 2009.

In terms of the impact of hourly winds, the statistics are consistent with those recorded in earlier impact studies with GMS-5 (e.g., Le Marshall et al. 1996b), where a modest positive impact through the troposphere was recorded. The results have shown that real-time hourly MTSAT-1R IR, VIS, and WV image-based AMVs assimilated using 4DVAR are of an accuracy that can benefit operational NWP in the Australian region. The addition of the vectors to the operational ACCESS-R regional forecast system, which already contains some JMA AMVs, provided both improved spatial and temporal data coverage of the region and a consistent average forecast improvement.

To further document the utility of assimilating continuous wind observations using 4DVAR in the Bureau’s new operational ACCESS-R system, AMVs generated using half-hourly, hourly, and 15-min MTSAT-2 image triplets 4 times per day (0000, 0600, 1200, and 1800 UTC) were used in an assimilation trial for a later period, namely 27 January–23 February 2011. The operational ACCESS-R system of 2011, which again had a 0.375° horizontal resolution with 50 levels in the vertical, was used. The results from the trial were once more consistent with earlier results from the assimilation of GMS-5, GOES-9, and MTSAT-1R local vectors (e.g., Le Marshall et al. 2002; Le Marshall et al. 2004b; Le Marshall et al. 2008, respectively) and showed improved forecasts in the lower, middle, and upper troposphere.

The rms difference between forecast and verifying analysis geopotential heights at 24 and 48 h for the control ACCESS-R system (operational database) and the experimental ACCESS-R system (operational database plus local hourly AMVs) for the period 27 January–23 February 2011 can be seen in Figs. 6a and 6b. In addition, the RMS difference between the forecast and verifying analysis wind at 24 h is provided in Fig. 6c.

Fig. 6.
Fig. 6.

(a) The rms difference between forecast and verifying analysis geopotential height at 24 h for ACCESS-R (blue) and ACCESS-R with AMVs (red) for the period 27 Jan–23 Feb 2011. (b) As in (a) but at 48 h. (c) As in (a) but for wind at 24 h. (d) The S1 difference between forecast and verifying analysis MSLP out to 48 h for the period 27 Jan–23 Feb 2011.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00018.1

The S1 skill score for MSLP forecasts out to 48 h for the control ACCESS-R system and the experimental ACCESS-R system for the period 27 January–23 February 2011 is also shown in Fig. 6d. Overall, it can be seen that for the period studied, forecasts from the experimental ACCESS-R system show a consistent general improvement throughout the troposphere when compared to the control ACCESS-R system.

In summary, the verification statistics are not surprising given those already recorded in earlier impact studies with GMS-5 and GOES-9 (e.g., Le Marshall et al. 1996b), where a modest positive impact through the troposphere was recorded. The results here have demonstrated that real-time near-continuous MTSAT-1R and MTSAT-2, IR, VIS, and WV image based AMVs assimilated using 4DVAR are of an accuracy that can benefit operational NWP in the Australian region. Addition of the hourly vectors to the operational ACCESS-R regional forecast system, even when it already contained some JMA AMVs, provided both improved temporal and spatial data coverage of the region and a consistent average forecast improvement. These results and those reported previously for September and October 2009, and their consistency with previous studies, have led to the operational use of these local AMVs.

6. The future

Development work is currently under way in several important areas. One is the further documentation of the utility of continuous MTSAT AMVs for tropical cyclone forecasting within the current operational ACCESS forecast suite. The utility of continuous winds for tropical cyclone forecasting has been documented in a number of earlier works (e.g., Leslie et al. 1998).

Recently, these high-density continuous (hourly) visible and infrared image-based winds have been used in a series of experiments examining the motion of tropical cyclones (TCs; e.g., TC Dianne; see Fig. 7). The forecasts shown here were undertaken using the operational ACCESS-R system at 37.5-km resolution employing 4DVAR with a 6-h time window. Forty-eight-hour forecasts are shown in Fig. 7 for the control (green) using the operational database, which contained vectors only at the analysis time (i.e., 1200 UTC 16 February 2011) and for the experimental run (red), which employed 4DVAR and hourly vectors. The benefit of what was far more extensive data coverage in both time and space in the case of TC Dianne is clearly seen in the forecasts. In this case a poor operational forecast was significantly improved upon by using high spatial and temporal resolution high-quality AMVs to specify the initial state. The verification statistics gathered from the limited assimilation periods reported here are currently being expanded and will be subsequently documented. What is already clear, however, is that the use of continuous winds with 4DVAR has the potential to beneficially modify tropical cyclone track prediction.

Fig. 7.
Fig. 7.

Forty-eight-hour forecasts using AMVs at analysis time (green) and using hourly AMVs (red), along with the best track (black) for TC Dianne. The starting time for the forecasts is 1200 UTC 16 Feb 2011.

Citation: Weather and Forecasting 28, 2; 10.1175/WAF-D-12-00018.1

7. Summary and conclusions

The local estimation of real-time operational MTSAT-1R and MTSAT-2 AMVs and their impact on operational Australian region NWP has been examined.

The recent ACCESS-R trials have used continuous (at least hourly) local real-time operational MTSAT-1R and MTSAT-2 IR and VIS AMVs, and their beneficial impact on operational regional NWP using 4DVAR has been recorded. The use of 4DVAR and higher spatial and temporal resolution imagery has provided the opportunity to exploit a more spatially and temporally uniform wind database and has resulted in considerable beneficial impacts from these data. These results were consistent with and represent an extension of past trials with GMS-5, GOES-9, and MTSAT-1R data and have resulted in the acceptance of these continuous local vectors for operational use.

It should be noted that the current system is still limited in a number of ways. The limited number of channels available for height assignment limits the accuracy for MTSAT, the cloud-free water vapor image data could be assimilated as clear radiances, and the tracking algorithm could be specifically tuned to be consistent with the length scale of the model.

As a result and looking forward, the continuing trend toward space-based observations with higher spatial, temporal, and spectral resolutions should enable improved estimation of atmospheric flow and result in further quantitative benefits to NWP. In the near future, the prospects of realizing benefits from the expanded use of sequential observations from MTSAT-2, the Chinese satellite Fengyun 2-07 (FY-2F), and also from next-generation GOES-R and Himawari-8 satellites, as well as instruments like the ultraspectral Geostationary Imaging Fourier Transform Spectrometer (Smith et al. 2000), are very good.

Acknowledgments

Thanks are due Terry Adair for his assistance in the preparation of this manuscript and to Bert Berzins, Ian Senior, and Weiqing Qu for assistance in data preparation. Thanks are also due to Stuart Young, Alain Protat, Michael Whimpey, and the National Aeronautics and Space Administration for access to CloudSatCALIPSO data during the study period.

REFERENCES

  • Bennett, A. F., , and Leslie L. M. , 1981: Statistical correction of the Australian Region Primitive Equation Model. Mon. Wea. Rev., 109, 453462.

    • Search Google Scholar
    • Export Citation
  • Coakley, J., , and Bretherton F. , 1982: Cloud cover from high resolution scanner data: Detecting and allowing for partially filled fields of view. J. Geophys. Res., 87 (C7), 49174932.

    • Search Google Scholar
    • Export Citation
  • Daniels, J., , Velden C. , , Busby W. , , and Irving A. , 2000: Status and development of operational GOES wind products. Proc. Fifth Int. Winds Workshop, Lorne, VIC, Australia, EUMETSAT P 28, 27–45.

  • Davies, T., , Cullen M. J. P. , , Malcolm A. J. , , Mawson M. H. , , Staniforth A. , , White A. A. , , and Wood N. , 2005: A new dynamical core for the Met Office’s global and regional modelling of the atmosphere. Quart. J. Roy. Meteor. Soc., 131, 17591782.

    • Search Google Scholar
    • Export Citation
  • Holmlund, K., 1998: The utilization of statistical properties of satellite-derived atmospheric motion vectors to derive quality indicators. Wea. Forecasting, 13, 10931104.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Pescod N. , , Seaman R. , , Mills G. , , and Stewart P. , 1994: An operational system for generating cloud drift winds in the Australian region and their impact on numerical weather prediction. Wea. Forecasting, 9, 361370.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Leslie L. M. , , and Bennett A. F. , 1996a: Tropical Cyclone Beti—An example of the benefits of assimilating hourly satellite wind data. Aust. Meteor. Mag., 45, 275279.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Spinoso C. , , Pescod N. R. , , and Mills G. A. , 1996b: Estimation and assimilation of hourly high spatial resolution wind vectors based on GMS-5 observations. Aust. Meteor. Mag., 45, 279284.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Leslie L. , , Morison R. , , Pescod N. , , Seecamp R. , , and Spinoso C. , 2000a: Recent developments in the continuous assimilation of satellite wind data for tropical cyclone track forecasting. Adv. Space Res., 25, 10771080.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Pescod N. , , Seecamp R. , , Rea A. , , Tingwell C. , , Ellis G. , , and Shi H. , 2000b: Recent advances in the quantitative generation and assimilation of high spatial and temporal resolution satellite winds. Proc. Fifth Int. Winds Workshop, Lorne, VIC, Australia, EUMETSAT P 28, 47–56.

  • Le Marshall, J. F., , Mills G. , , Pescod N. , , Seecamp R. , , Puri K. , , Stewart P. , , Leslie L. M. , , and Rea A. , 2002: The estimation of high density atmospheric motion vectors and their application to operational numerical weather prediction. Aust. Meteor. Mag., 51, 173180.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Rea A. , , Leslie L. , , Seecamp R. , , and Dunn M. , 2004a: Error characterization of atmospheric motion vectors. Aust. Meteor. Mag., 53, 123131.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Seecamp R. , , Daniels J. , , Velden C. , , Puri K. , , Bowen R. , , Rea A. , , and Dunn M. , 2004b: The contribution of GOES 9 to operational NWP forecast skill in the Australian region. Aust. Meteor. Mag., 53, 279283.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Seecamp R. , , Dunn M. , , Velden C. , , Wanzong S. , , Puri K. , , Bowen R. , , and Rea A. , 2008: The contribution of locally generated MTSAT-1R atmospheric motion vectors to operational meteorology in the Australian region. Aust. Meteor. Mag., 57, 359365.

    • Search Google Scholar
    • Export Citation
  • Le Marshall, J. F., , Seecamp R. , , Xiao Y. , , Steinle P. , , Sims H. , , Skinner T. , , Jung J. , , and Le T. , 2011: The generation and assimilation of continuous AMVs with 4DVar. Aust. Meteor. Oceanogr. J., 61, 117123.

    • Search Google Scholar
    • Export Citation
  • Leslie, L. M., , Le Marshall J. F. , , Morison R. P. , , Spinoso C. , , Purser R. J. , , Pescod N. , , and Seecamp R. , 1998: Improved hurricane track forecasting from the continuous assimilation of high quality satellite wind data. Mon. Wea. Rev., 126, 12481258.

    • Search Google Scholar
    • Export Citation
  • Puri, K., , Dietachmayer G. , , Mills G. A. , , Davidson N. E. , , Bowen R. M. , , and Logan L. W. , 1998: The new BMRC Limited Area Prediction System, LAPS. Aust. Meteor. Mag., 47, 203223.

    • Search Google Scholar
    • Export Citation
  • Rea, A., 2004: The determination and accuracy of atmospheric motion derived from sequential satellite observations in the Australian region. Ph.D. thesis, Royal Melbourne Institute of Technology, 206 pp.

  • Smith, W. L., , Harrison F. W. , , Revercomb H. E. , , Bingham G. E. , , Huang H. L. , , and Le Marshall J. F. , 2000: The Geostationary Imaging Fourier Transform Spectrometer. Proc. 11th Int. TOVS Study Conf., Budapest, Hungary, International TOVS Working Group, 391–398.

  • Stephens, G. L., and Coauthors, 2002: The CloudSat mission and the A-Train: A new dimension of space-based observations of clouds and precipitation. Bull. Amer. Meteor. Soc., 83, 17711790.

    • Search Google Scholar
    • Export Citation
  • Teweles, S., , and Wobus H. , 1954: Verification of prognostic charts. Bull. Amer. Meteor. Soc., 35, 455463.

  • Velden, C., and Coauthors, 2005: Recent innovations in deriving tropospheric winds from meteorological satellites. Bull. Amer. Meteor. Soc., 86, 205223.

    • Search Google Scholar
    • Export Citation
  • Zapotocny, T. H., , Jung J. A. , , Le Marshall J. F. , , and Treadon R. E. , 2008: A two-season impact study of four satellite data types and rawinsonde data in the NCEP Global Data Assimilation System. Wea. Forecasting, 23, 80100.

    • Search Google Scholar
    • Export Citation
Save