• Andersson, E., , and H. Järvinen, 1999: Variational quality control. Quart. J. Roy. Meteor. Soc., 125B , 697722.

  • Andersson, E., and Coauthors. 1998: The ECMWF implementation of three-dimensional variational assimilation (3D-Var). III: Experimental results. Quart. J. Roy. Meteor. Soc., 124B , 18311860.

    • Search Google Scholar
    • Export Citation
  • Atlas, R., , and R. N. Hoffman, 2000: The use of satellite surface wind data to improve weather analysis and forecasting. Satellites, Oceanography and Society, D. Halpern, Ed., Elsevier Oceanography Series, Vol. 63, Elsevier, 57–78.

    • Search Google Scholar
    • Export Citation
  • Atlas, R., , S. C. Bloom, , R. N. Hoffman, , E. Brin, , J. Ardizzone, , J. Terry, , D. Bungato, , and J. C. Jusem, 1999: Geophysical validation of NSCAT winds using atmospheric data and analyses. J. Geophys. Res., 104 (C5) 1140511424.

    • Search Google Scholar
    • Export Citation
  • Atlas, R., and Coauthers, 2001: The effects of marine winds from scatterometer data on weather analysis and forecasting. Bull. Amer. Meteor. Soc., 82 , 19651990.

    • Search Google Scholar
    • Export Citation
  • Brown, R. A., , and L. Zeng, 1994: Estimating central pressures of oceanic midlatitude cyclones. J. Appl. Meteor., 33 , 10881095.

  • Chelton, D. B., , M. H. Freilich, , and S. K. Esbensen, 2000: Satellite observations of the wind jets off the Pacific coast of Central America. Part I: Case studies and statistical characteristics. Mon. Wea. Rev., 128 , 19932018.

    • Search Google Scholar
    • Export Citation
  • Courtier, P., , J-N. Thépaut, , and A. Hollingsworth, 1994: A strategy for operational implementation of 4D-Var, using an incremental approach. Quart. J. Roy. Meteor. Soc., 120 , 13671387.

    • Search Google Scholar
    • Export Citation
  • Ebuchi, N., , and H. C. Graber, 1998: Directivity of wind vectors derived from the ERS-1/AMI scatterometer. J. Geophys. Res., 103 (C4) 77877797.

    • Search Google Scholar
    • Export Citation
  • Figa, J., , and A. Stoffelen, 2000: On the assimilation of Ku-band scatterometer winds for weather analysis and forecasting. IEEE Trans. Geosci. Remote Sens., 38 , 18931902.

    • Search Google Scholar
    • Export Citation
  • Francis, R., and Coauthors. 1991: The ERS-1 spacecraft and its payload. ESA Bulletin, Vol. 65, ESA, 27–48.

  • Freilich, M. H., , and R. S. Dunbar, 1999: The accuracy of the NSCAT 1 vector winds: Comparisons with National Data Buoy Center buoys. J. Geophys. Res., 104 (C5) 1123111246.

    • Search Google Scholar
    • Export Citation
  • Grantham, W. L., , E. M. Bracalente, , W. L. Jones, , and J. W. Johnson, 1977: The Seasat-A satellite scatterometer. IEEE J. Oceanic Eng., OE-2 , 200206.

    • Search Google Scholar
    • Export Citation
  • Grassotti, C., , S. M. Leidner, , J-F. Louis, , and R. N. Hoffman, 1999: Development and application of a visible-infrared rain flag for scatterometer data. J. Appl. Meteor., 38 , 665676.

    • Search Google Scholar
    • Export Citation
  • Hoffman, R. N., , Z. Liu, , J-F. Louis, , and C. Grassotti, 1995: Distortion representation of forecast errors. Mon. Wea. Rev., 123 , 27582770.

    • Search Google Scholar
    • Export Citation
  • Isaksen, L., 1997: Impact of ERS scatterometer data in the ECMWF 4D-Var assimilation system. Preliminary studies. Space at the Service of Our Environment, Florence, Italy, ESTEC, European Space Agency SP-414, 1829–1851.

    • Search Google Scholar
    • Export Citation
  • Isaksen, L., , and A. Stoffelen, 2000: ERS scatterometer wind data impact on ECMWF's tropical cyclone forecasts. IEEE Trans. Geosci. Remote Sens., 38 , 18851892.

    • Search Google Scholar
    • Export Citation
  • Järvinen, H., , E. Andersson, , and F. Bouttier, 1999: Variational assimilation of time sequences of surface observations with serially correlated errors. Tellus, 51A , 469488.

    • Search Google Scholar
    • Export Citation
  • Katsaros, K. B., , E. B. Forde, , P. Chang, , and W. T. Liu, 2001: QuikSCAT's SeaWinds facilitates early identification of tropical depressions in 1999 hurricane season. Geophys. Res. Lett., 28 , 10431046.

    • Search Google Scholar
    • Export Citation
  • Klinker, E., , F. Rabier, , G. Kelly, , and J-F. Mahfouf, 2000: The ECMWF operational implementation of four-dimensional variational assimilation. III: Experimental results and diagnostics with operational configuration. Quart. J. Roy. Meteor. Soc., 126A , 11911215.

    • Search Google Scholar
    • Export Citation
  • Lander, M. A., , E. J. Trehubenko, , and C. P. Guard, 1999: Eastern Hemisphere tropical cyclones of 1996. Mon. Wea. Rev., 127 , 12741300.

  • Mahfouf, J-F., , and F. Rabier, 2000: The ECMWF operational implementation of four-dimensional variational assimilation. II: Experimental results with improved physics. Quart. J. Roy. Meteor. Soc., 126A , 11711190.

    • Search Google Scholar
    • Export Citation
  • Naderi, F. M., , M. H. Freilich, , and D. G. Long, 1991: Spaceborne radar measurement of wind velocity over the ocean—An overview of the NSCAT scatterometer system. Proc. IEEE, 79 , 850866.

    • Search Google Scholar
    • Export Citation
  • Pasch, R. J., , and L. A. Avila, 1999: Atlantic hurricanes of 1996. Mon. Wea. Rev., 127 , 581610.

  • Perry, K. L., 2001: QuikSCAT science data product user's manual, overview and geophysical data products. Version 2.2. Jet Propulsion Laboratory JPL-D18053, Pasadena, CA, 89 pp.

    • Search Google Scholar
    • Export Citation
  • Price, J. C., 1976: The nature of multiple solutions for surface wind speed over the oceans from scatterometer measurements. Remote Sens. Environ., 5 , 4754.

    • Search Google Scholar
    • Export Citation
  • Rabier, F., , H. Järvinen, , E. Klinker, , J-F. Mahfouf, , and A. Simmons, 2000: The ECMWF operational implementation of four-dimensional variational assimilation. I: Experimental results with simplified physics. Quart. J. Roy. Meteor. Soc., 126A , 11431170.

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

    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., , and D. Anderson, 1997: Ambiguity removal and assimilation of scatterometer data. Quart. J. Roy. Meteor. Soc., 123B , 491518.

    • Search Google Scholar
    • Export Citation
  • Thépaut, J-N., , R. N. Hoffman, , and P. Courtier, 1993: Interactions of dynamics and observations in a four-dimensional variational assimilation. Mon. Wea. Rev., 121 , 33933414.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    (top) Plot of maximum likelihood estimator (MLE) for one wind vector cell as a function of u- and υ-wind components. (middle) Ambiguous NSCAT winds corresponding to the top panel. (bottom) Two-wind cost function used in 4DVAR

  • View in gallery

    Two-dimensional histogram of NSCAT vs ECMWF first guess winds for the period 0000 UTC 14 Oct–1200 UTC 20 Oct 1996. The NSCAT ambiguity closest to the first guess field is used for the comparison

  • View in gallery

    Histograms of wind speed departures of observations from (top) ECMWF first guess winds (ob) and (bottom) ECMWF analyzed winds (oa). These departures are for NSCAT data in the Northern Hemisphere extratropics only (20°–90°N) for the period 2100 UTC 15 Oct–0900 UTC 16 Oct 1996 (from two assimilation cycles)

  • View in gallery

    Time series of 5-day forecast scores for NoSCAT and NSCAT and Southern Hemispheres, the scores are 500-hPa height anomaly correlation. In the Tropics, the scores are 850-hPa vector wind rms errors

  • View in gallery

    Track of Hurricane Lili from 1200 UTC 14 Oct 1996 to 0000 UTC 29 Oct 1996. Observed positions at 0000 and 1200 UTC from the National Hurricane Center (NHC) are plotted as diamond–plus symbols along the track. The 6-h NSCAT data coverage beginning at 2100 UTC 19 Oct 1996 is shown by fine dots. The A marks the position of Lili at the time of the case study presented, 0000 UTC 26 Oct 1996

  • View in gallery

    Analysis of Hurricane Lili 0000 UTC 26 Oct 1996. (top) First guess 10-m wind (kt) and MSLP (hPa). (middle) The analyzed wind and MSLP. (bottom) Analysis increments of wind (times 10) and MSLP. The observed central pressure and position from the NHC at 0600 UTC 19 Oct was 975 hPa at 38.1°N, 41.0°W

  • View in gallery

    NSCAT winds in the vicinity of Hurricane Lili at 0051 UTC 26 Oct 1996. The NSCAT ambiguity closest to the analyzed wind direction is plotted. MSLP analysis from NSCAT experiment valid at 0000 UTC 26 Oct 1996. Winds rejected by the variational quality control are darker

  • View in gallery

    Non-NSCAT data presented to the analysis at 0000 UTC 26 Oct 1996: (top) satellite data (TOVS retrievals, crosses; satellite cloud motion winds, inverted triangles) and (bottom) synoptic and aircraft reports (synop/ship reports, circles; aircraft reports, diamonds). Hurricane Lili's observed position is marked in both panels. Each panel also has a cross-section locator for Figs. 9–12

  • View in gallery

    Cross sections of scalar wind speed valid at 0000 UTC 26 Oct 1996: (top) first guess wind speed (m s−1) from NoSCAT and NSCAT and (bottom) wind speed increments (m s−1) from NoSCAT and NSCAT. Cross-section location is shown in Fig. 8

  • View in gallery

    Same as in Fig. 9 but for vertical velocity (Pa s−1)

  • View in gallery

    Same as in Fig. 9 but for temperature (K and °C)

  • View in gallery

    Same as in Fig. 9 but for specific humidity (kg kg−1)

  • View in gallery

    Comparison of (top) NSCAT and (bottom) NoSCAT MSLP and 500-hPa temperature (°C) analyses valid at 0000 UTC 27 Oct 1996

  • View in gallery

    Analysis differences of MSLP (NSCAT − NoSCAT) for 6-hourly analyses for 0000 UTC 26 Oct–0600 UTC 27 Oct 1996. The contour interval is 3 hPa. Maximum differences are displayed (units, hPa)

  • View in gallery

    Forecasts of Hurricane Lili valid at 1200 UTC 28 Oct 1996: (middle) verifying analysis from the NoSCAT experiment and 2-day forecasts of Lili's position from (top) NSCAT and (bottom) NoSCAT. MSLP contour interval is 5 hPa, and wind speed is contoured every 2 m s−1 beginning at 20 m s−1. The observed central pressure from the NHC at 1200 UTC 28 Oct was 970 hPa

  • View in gallery

    Tracks of Typhoons Yates and Zane for the period 1200 UTC 17 Sep–0000 UTC 6 Oct 1996. Observed positions at 0000 and 1200 UTC from the JTWC are plotted as circle–plus (⊕) and box–plus symbols for Yates and Zane, respectively. The 6-h NSCAT data coverage beginning at 2100 UTC 27 Sep 1996 is shown in fine dots. The Y and Z mark the positions of Yates and Zane at 1200 UTC 26 Sep 1996, the time used as a case study in this paper

  • View in gallery

    Non-NSCAT data presented to the analysis valid at 1200 UTC 26 Sep 1996: (top) satellite data (satellite cloud motion winds, inverted triangles) and (bottom) synoptic and aircraft reports (synop/ship reports, circles; aircraft reports, diamonds; buoys, triangles; rawinsonde stations, small crossed squares). The positions of Typhoons Yates and Zane are marked in both panels. Each panel also has a cross-section locator for Figs. 24–27

  • View in gallery

    NSCAT data coverage in the region of Typhoons Yates and Zane. These plots show data used in analyses valid at (upper left) 1200 UTC 26 Sep, (upper right) 0000 UTC 27 Sep, (lower left) 1200 UTC 27 Sep, and (lower right) 0000 UTC 28 Sep 1996. Data are thinned for plotting purposes. The full resolution of the data may be seen by the small dots

  • View in gallery

    Analysis of Typhoons Yates and Zane at 1200 UTC 26 Sep 1996. (top) First guess 10-m wind (kt) and MSLP (hPa). (middle) Analyzed wind and MSLP. (bottom) Analysis increments of wind (times 10) and MSLP. At 1200 UTC 26 Sep, Yates was estimated to be a category 4 typhoon at 17.3°N, 142.4°E, and Zane was estimated to be a category 3 typhoon at 20.1°N, 127.7°E. MSLP contour intervals are 2 hPa for first guess and analysis panels, and 1 hPa for increment panels

  • View in gallery

    NSCAT winds in the vicinity of Typhoon Zane, at 1418 UTC 26 Sep 1996. The NSCAT ambiguity closest to the analyzed wind direction is plotted. MSLP analysis from NSCAT experiment valid at 1200 UTC 26 Sep 1996. Winds rejected by the variational quality control are darker

  • View in gallery

    Forecasts of Typhoons Yates and Zane valid at 0000 UTC 2 Oct 1996: (middle) verifying analysis from the NSCAT experiment and 5.5-day forecasts from (top) NSCAT and (bottom) NoSCAT. Observed or forecast 12-hourly positions of Yates and Zane are plotted in each panel as circle–plus and box–plus symbols, respectively. Forecast tracks are plotted with filled circles and squares. MSLP contour interval is 5 hPa

  • View in gallery

    Same as in Fig. 19 but for the T106 analyses

  • View in gallery

    NSCAT data rejected by 4DVAR quality control from the analysis for Typhoon Zane, at 1200 UTC 26 Sep 1996, from experiments (top) NSCAT and (bottom) NSCAT106. Three observations rejected in the NSCAT analysis but used in the NSCAT106 analysis are highlighted in the top panel. Winds rejected by the variational quality control in T63, but used in T106, are darker and located to the south and southeast

  • View in gallery

    Cross sections of scalar wind speed valid at 1200 UTC 26 Sep 1996 through Typhoon Zane: (top) first guess winds (m s−1) from NoSCAT106 and NSCAT106 and (bottom) wind speed increments (m s−1) from NoSCAT106 and NSCAT106. Cross-section location is shown in Fig. 17

  • View in gallery

    Same as in Fig. 24 but for vertical velocity (Pa s−1)

  • View in gallery

    Same as in Fig. 24 but for temperature (K and °C)

  • View in gallery

    Same as in Fig. 24 but for specific humidity (kg kg−1)

  • View in gallery

    Forecasts of Typhoons Yates and Zane valid at 0000 UTC 3 Oct 1996: (middle) verifying analysis and 5.5-day forecasts from (top) NSCAT106 and (bottom) NoSCAT106. Observed or forecast 12-hourly positions of Yates and Zane are plotted in each panel as circle–plus and box–plus symbols, respectively. Forecast tracks of Yates and Zane are plotted with filled circles and squares. MSLP contour interval is 5 hPa

  • View in gallery

    The 500-hPa anomaly correlation forecast scores for the North Pacific region. Results shown are an average of four cases in Sep 1996. (top) NoSCAT with NSCAT (T63 resolution) comparison, and (bottom) NoSCAT106 with NSCAT106 comparison. The large impact of NSCAT106 is due to the importance of correctly treating Typhoons Yates and Zane with the assimilation system. In general, a smaller impact of NSCAT data would be expected than is seen in this case study

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 39 39 5
PDF Downloads 17 17 7

Impact of NSCAT Winds on Tropical Cyclones in the ECMWF 4DVAR Assimilation System

View More View Less
  • 1 Atmospheric and Environmental Research, Inc., Lexington, Massachusetts
  • | 2 European Centre for Medium-Range Weather Forecasts, Reading, Berkshire, United Kingdom
  • | 3 Atmospheric and Environmental Research, Inc., Lexington, Massachusetts
© Get Permissions
Full access

Abstract

The impact of NASA Scatterometer (NSCAT) data on tropical cyclone forecasting on the European Centre for Medium-Range Weather Forecasts four-dimensional variational (4DVAR) data assimilation system is examined. Parallel runs with and without NSCAT data were conducted. The 4DVAR can use single-level data, such as scatterometer winds, to good advantage. The 4DVAR system uses data at appropriate times and has the potential to accurately resolve the ambiguity inherent in scatterometer data, by using a two-ambiguity cost function at each NSCAT location. Scatterometer data are shown to improve the depiction of the surface wind field in both tropical cyclones and extratropical lows, and can provide early detection of these features. Case studies of Hurricane Lili, and of Typhoons Yates and Zane (all in 1996), show major positive impacts of NSCAT data on forecasts of tropical cyclone intensity and position.

Corresponding author address: S. Mark Leidner, Atmospheric and Environmental Research, Inc., 131 Hartwell Ave., Lexington, MA 02421. Email: leidner@aer.com

Abstract

The impact of NASA Scatterometer (NSCAT) data on tropical cyclone forecasting on the European Centre for Medium-Range Weather Forecasts four-dimensional variational (4DVAR) data assimilation system is examined. Parallel runs with and without NSCAT data were conducted. The 4DVAR can use single-level data, such as scatterometer winds, to good advantage. The 4DVAR system uses data at appropriate times and has the potential to accurately resolve the ambiguity inherent in scatterometer data, by using a two-ambiguity cost function at each NSCAT location. Scatterometer data are shown to improve the depiction of the surface wind field in both tropical cyclones and extratropical lows, and can provide early detection of these features. Case studies of Hurricane Lili, and of Typhoons Yates and Zane (all in 1996), show major positive impacts of NSCAT data on forecasts of tropical cyclone intensity and position.

Corresponding author address: S. Mark Leidner, Atmospheric and Environmental Research, Inc., 131 Hartwell Ave., Lexington, MA 02421. Email: leidner@aer.com

1. Introduction

The National Aeronautics and Space Administration (NASA) Scatterometer (NSCAT), aboard the Advanced Earth Observing Satellite-1 (ADEOS-1), provided high quality ocean surface wind vector data during the 9-month ADEOS-1 mission beginning 16 August 1996 (Freilich and Dunbar 1999). The focus of this study is the impact of NSCAT data on tropical cyclone forecasting in the European Centre for Medium-Range Weather Forecasts (ECMWF) four-dimensional variational (4DVAR) data assimilation system. We conducted parallel assimilation experiments with and without NSCAT data. These experiments are denoted “NSCAT” and “NoSCAT” hereafter. In both NSCAT and NoSCAT experiments, winds from the European Remote Sensing-2 satellite (ERS-2) scatterometer were excluded to highlight the impact of NSCAT. Special Sensor Microwave Imagers (SSM/I) wind speed data were also not used, since these data were not available at ECMWF during the periods of the assimilation experiments (1996). Two principal goals motivated these experiments. First, we wanted to examine the impact of NSCAT in light of the previous study of ERS scatterometer data by Isaksen and Stoffelen (2000). Second, although the NSCAT mission ended prematurely, the follow-on SeaWinds is currently operational aboard the Quick Scatterometer (QuikSCAT) satellite and has many characteristics in common with NSCAT. Therefore, experience with NSCAT provides useful guidance for the future use of SeaWinds.

A particular concern when using scatterometer wind data is that multiple winds or ambiguities are retrieved at each location (Price 1976). The 4DVAR system has the potential to be an accurate ambiguity removal method. Also, 4DVAR uses scatterometer data at the time of each observation, thereby avoiding errors caused by assuming the data are observed at standard synoptic times. Such errors can be significant for quickly evolving surface features, which may be well observed by the scatterometer. The 4DVAR system utilizes two NSCAT wind vector ambiguities at each observation location as described in section 4. Previously, the 4DVAR system has been shown to make good use of European Remote Sensing-1 satellite (ERS-1) scatterometer data for forecasting tropical cyclones (Isaksen and Stoffelen 2000). The experiments described here are the first to use NSCAT data in an operational 4DVAR assimilation system. Major positive impacts of NSCAT data on tropical cyclone forecasts are seen in these experiments.

The plan of this paper is the following. First we describe scatterometry from satellites focusing on the data characteristics and the effects of rain. Then we present an evaluation of the NSCAT data quality, and a discussion on the impact on global forecast skill scores. In order to explain the lack of impact on these scores, we go into more detail in the succeeding sections and describe the 4DVAR assimilation methodology, our experimental design, and case studies of Hurricane Lili and Typhoons Yates and Zane. This paper ends with a discussion of current problems and future opportunities, and some concluding remarks.

2. Scatterometry from space

The first satellite to carry a scatterometer was Seasat in 1978 (Grantham et al. 1977). When NSCAT was launched, it was the first Ku-band (∼14 GHz) satellite scatterometer since Seasat (Naderi et al. 1991). Since 1991, the European Space Agency (ESA) has flown ERS-1 and ERS-2, which carry an active microwave instrument (AMI) functioning both as a C-band (∼5 GHz) scatterometer and as a synthetic aperture radar (Francis et al. 1991). The ERS data have been used operationally at ECMWF since January 1996, first in 3DVAR as described by Stoffelen and Anderson (1997) and currently in 4DVAR as described by Isaksen and Stoffelen (2000).

Scatterometers measure the backscatter due to the ocean surface roughness via the Bragg scattering mechanism and, thus, are most sensitive to centimeter-scale waves, which are usually in equilibrium with the local wind field. Several measurements of backscatter at a single earth location from different azimuth angles allow the retrieval of near-surface wind vectors. The wind retrieval effectively inverts the geophysical model function that relates backscatter to 10-m neutral stability wind and direction, and to the viewing geometry (e.g., Naderi et al. 1991). It is not possible to determine a unique solution. Typically two, three, or four ambiguous winds will be retrieved (see Fig. 1, middle panel). ADEOS-1 and the other scatterometer-carrying satellites mentioned all have/had similar near-polar, sun-synchronous orbits (∼800 km altitude, ∼100 min period). NSCAT had three antennas on each side of the spacecraft, providing coverage in two simultaneous 600-km-wide swaths, separated by a 350-km-wide nadir gap. All NSCAT observations are available at roughly 0600 and 1800 local time (LT). The fore and aft NSCAT antennas were vertically polarized with incidence angles ranging approximately from 22° to 60°, and the mid-NSCAT antenna was vertically and horizontally polarized with incidence angles ranging approximately from 18° to 52°. With this geometry, the backscatter values at a single location on the surface are observed from four different azimuth angles within a time span of approximately 70–200 s, increasing with incidence angle. See the appendix of Chelton et al. (2000) for more NSCAT details.

There are several important differences between the wind datasets produced by ERS-1/2 and by NSCAT, which may have a bearing on the impact results. First, the C-band radar signal is unaffected by rain, while the Ku-band radar signal is both absorbed and scattered by rain, increasingly as the mean drop size grows (Grassotti et al. 1999; Figa and Stoffelen 2000). Second, ERS-1/2 have antennas only on the right side of the spacecraft. Thus, NSCAT collects measurements over an area roughly twice that of ERS-1 or ERS-2 in the same amount of time. Although the data sampling is 25 km for ERS-1/2 and 25 or 50 km for NSCAT, the resolution of all these instruments is effectively 50 km. The 50-km NSCAT science data product is used in this study. Third, ERS-1/2 usually retrieves exactly two ambiguities due to its symmetric antenna geometry. Additional wind ambiguities (i.e., more than two) for ERS-1/2 are ignored because the correct ambiguity is almost always the most likely or the next most likely (Stoffelen and Anderson 1997). But winds retrieved from NSCAT often have three or four ambiguities, with a primary pair of solutions opposed by 180° and a secondary pair also opposed by 180°. In the case of NSCAT the third or fourth ambiguity is closest to the true wind approximately 5% of the time.

Other authors including Freilich and Dunbar (1999), Atlas et al. (1999), Stoffelen (1998), and Ebuchi and Graber (1998) have compared scatterometer winds to ships, buoys, other satellite instruments, and to other forecasts and analyses.

We compared NSCAT winds to ECMWF forecasts and analyses. Figure 2 shows that NSCAT winds are in generally good agreement with ECMWF first guess winds. (First guess winds are the product of a 6-h forecast and are available at each forecast hour.) This figure shows collocations between ECMWF first guess winds and the closest NSCAT ambiguity over a 6-day period (more than 750 000 collocations). There are two notable features, however: a small wind speed bias and a small rain contamination signal. ECMWF first guess winds are about 0.63 m s−1 slower than the NSCAT winds for nearly all wind speeds. There are four factors that contribute to the wind speed bias. First, the scatterometer observes smaller-scale features than are represented by the forecast model at ECMWF (see section 4 for a discussion of the forecast model configuration used in this study). This is especially true for high winds and can be seen in Fig. 2 as an increasing bias for winds greater than 15 m s−1. Second, the NSCAT winds used in this study are the result of wind retrievals using the NSCAT-2 model function, which was derived in part using ECMWF surface winds. So one would expect the NSCAT winds statistically to agree more closely with the ECMWF model winds because of their dependence. However, the ECMWF forecast model is improved continuously, and the mean surface wind speed in the model changes slightly with time. The ECMWF surface winds are somewhat slower today than in 1996, due to a modified convective parameterization and introduction of a coupled wave model. Thus, NSCAT winds appear to be too fast relative to today's operational system. Since ECMWF surface wind speeds today are closer to buoy observations than in 1996, NSCAT winds may be slightly high. Third, there is some evidence of rain contamination in the NSCAT data in the 0–10 m s−1 wind speed range from Fig. 2. A small increase in the density of points far above the 45° line indicates high NSCAT wind speeds that may be rain contaminated. Finally, retrieved NSCAT winds assume neutral stability in the boundary layer, and this is a very good approximation over much of the global oceans. But where the boundary layer is stably stratified, the true winds will be less than those in neutral conditions. Therefore, NSCAT winds will overestimate the true wind speed by as much as 15% where the boundary layer is stable. The wind speed bias and its causes are not addressed in this study. We conducted several experiments with wind speed bias corrected NSCAT data (not shown). We found the bias correction had no effect on overall forecast skill or forecast skill with respect to tropical cyclones. Because of NSCAT's shortened mission, methods for detecting rain in NSCAT data were not fully developed and were not available to apply as additional quality controls (QC).

Figure 3 demonstrates that 4DVAR is very effective at drawing the analysis close to the NSCAT data. This typical example shows histograms of wind speed departures from the first guess (ob, top panel) and the analysis (oa, bottom panel). Departures are shown for a 12-h period in the Northern Hemisphere extratropics only (20°–90°N). The mean departure is reduced by almost half (0.818 → 0.469 m s−1) and the rms of the departures is reduced from 2.22 → 1.61 m s−1. Thus, the NSCAT winds are in good agreement with ECMWF short-range forecasts, and 4DVAR creates analyses that fit the data very well.

3. Overall evaluation of scatterometer data for NWP

There has been a long history of research on the use of scatterometer data for numerical weather prediction (NWP), which has focused on evaluating the impact of the data and on developing methods to optimize the use of the data (Atlas et al. 2001). Since NSCAT data were never available operationally due to the short lifetime of the satellite, the operational centers have not focused on these data. The only published NSCAT impact studies to date are those conducted by the Data Assimilation Office (DAO) (Atlas et al. 1999; Atlas and Hoffman 2000).

In our experiments the mean 500-hPa geopotential height anomaly correlation scores for NSCAT in the Northern Hemisphere extratropics are neutral and are slightly negative in the Southern Hemisphere extratropics for forecasts beyond 96 h (not shown). We expect to see the largest (positive) impact in the Southern Hemisphere extratropics where fewer observations are routinely available. Similar investigations using ERS-2 data (Isaksen 1997) and SeaWinds/QuikSCAT data (H. Hersbach 2001, personal communication) in the ECMWF 4DVAR system show a neutral Northern Hemisphere extratropics impact and a (small) positive impact in the Southern Hemisphere extratropics.

To illustrate the variability in forecast skill from day to day, Fig. 4 shows time series of forecast skill for 5-day forecasts in the Northern and Southern Hemisphere extratropics and the Tropics. Anomaly correlation of 500-hPa geopotential height is used for the extratropics and 850-hPa rms vector wind error is used in the Tropics. The periods of late September 1996 (Typhoons Yates and Zane) and late October 1996 (Hurricane Lili) are shown. The mean forecast skill difference between the NSCAT and NoSCAT experiments is small compared to the day-to-day variability in forecast skill for either experiment. Thus NSCAT does not result in a significant positive impact over these large regions, in spite of the fact that other satellite surface winds data have been withheld, and in spite of the fact that an advanced 4DVAR data assimilation system has been used. Please note that our experiments only cover three relatively short periods of data assimilation chosen to evaluate the impact on the specific tropical cyclones. It was not our purpose to evaluate the global impact of NSCAT, this would have required longer assimilation periods, closer attention to speed biases, and closer attention to rain contamination issues. Also we will show later (sections 6 and 7) that the spatial scale of tropical-cyclone-related improvements from using scatterometer data is often too small to be captured by global forecast scores. In Fig. 4 results are shown for our lower-resolution experiments. The results from the higher-resolution experiment are summarized later (in Fig. 29) and show an improvement over the lower-resolution experiments, but this is only available for four cases.

Past scatterometer impact studies have been motivated in part by the belief that ocean surface wind data, properly used, must be critically important to NWP. The surface wind is the main factor controlling the exchanges of heat, moisture, and momentum between the ocean and atmosphere, physical processes that are critical to the energetics of both the ocean and atmosphere. Scatterometer data should be especially useful in 4DVAR systems, which can use single-level data to good advantage. For example, Thépaut et al. (1993) showed that 4DVAR systems can be extremely sensitive to a single surface wind observation, as well as to scatterometer data. Similar results were found by Järvinen et al. (1999). Scatterometer data can improve analyses and forecasts of extreme weather events by enhancing the depiction of the surface wind field for both tropical cyclones and extratropical lows and can provide early detection of these features. For example, Brown and Zeng (1994) showed that surface pressure analyses derived from ERS-1 winds capture many details not present in ECMWF operational analyses. Also, Katsaros et al. (2001) demonstrated how QuikSCAT's ability to see through upper-level cloud facilitated early identification of tropical depressions in the 1999 North Atlantic hurricane season. These examples indicate potential capabilities of scatterometer data for improving NWP in exceptional cases.

In spite of the above arguments suggesting that scatterometer data should significantly impact NWP, the operational centers have been slow to adopt these data. At ECMWF, the operational system is so highly optimized to current data that demonstrating a statistically significant positive impact of a new high quality observing system or an improved assimilation technique requires substantial work (e.g., Andersson et al. 1998). Therefore, our regional skill scores may be an effect of the particular sample of cases used. For this reason and because we expect the greatest impact of NSCAT data to be on tropical cyclone forecasting, we focus on case studies of particular tropical cyclones. In particular, we examine the impact of assimilation system resolution on the use of NSCAT data and its affect on the analysis and forecast performance.

4. 4DVAR assimilation method

The 4DVAR method seeks a best fit between the analysis and observations (ao) and between the analysis and background (ab) assuming certain observation and background errors and constrained by the atmospheric model to produce balanced, meteorologically consistent increment fields. The data assimilation system used in this study is the Integrated Forecast System (IFS) cycle 19r1 at ECMWF (Rabier et al. 2000; Mahfouf and Rabier 2000; Klinker et al. 2000). Only aspects of the system of particular relevance to this study will be detailed here. In the IFS, an incremental 4DVAR approach (Coutier et al. 1994) is used to produce the analysis by using different spectral resolutions to optimize the use of computer resources.

The assimilation time window for 4DVAR in this study is 6 h, ±3 h around the standard analysis times (0000, 0600, 1200, and 1800 UTC). All experiments are performed with 31 vertical levels (model top at 10 hPa). The first trajectory is simply a 6-h forecast from the first guess (background) field. The 6-h forecast is run at high resolution (T319) using the full nonlinear model of the atmosphere. Observation-background differences (innovations) are computed at the appropriate time for all observations presented to 4DVAR. At this stage, gross quality controls are applied to the observations to ensure that unreasonable values are not used during the minimization. NSCAT winds at an observation point are removed from the analysis if any of the following conditions is true:

  • Background check: If the background wind speed or the NSCAT wind speed is greater than 25 m s−1.
  • Sea ice check: The model sea surface temperature is less than 275 K at the observation point.
  • Dual-winds quality control: If the directions of the two NSCAT wind ambiguities are less than 135° apart. Retrieved winds failing this test are believed to be anomalous.

During the latter part of the minimization, variational QC is applied (Andersson and Järvinen 1999), which rejects observations whose oa departures are too large. We found variational QC to be very useful for rejecting occasional scatterometer wind pairs that are turned 90° to be background flow (see upper-right corner of Fig. 20 for an example). Horizontal observation error correlations are not explicitly taken into account. We take account of correlated observations error in a crude way by using the 50-km product instead of the 25-km product, and by increasing the observation error to 2 m s−1 for each wind component (despite the fact that the ob statistics would justify a smaller observation error). Minimizations are performed at lower resolution (T63 or T106) because of the high computational cost of repeated model integrations over a 6-h period.

After the first step of the minimization, the analysis increments (ab) from the T63 (or T106) fields are added to the T319 initial conditions. Then, a second high-resolution 6-h forecast is run from these updated initial conditions to provide the basis for a second linearized minimization. The second minimization uses more complete physics than the first and is thus computationally more expensive. Finally, the new increments are again added to the initial conditions, and a final 6-h forecast is run from these updated initial conditions to calculate observation-analysis departures (see the example Fig. 3, bottom panel). Fields at the center of the assimilation window (the “analysis”) are saved and used as initial conditions for a 10-day forecast.

In the IFS, NSCAT winds are assimilated differently than the other sources of wind data such as synoptic observations (SYNOPs), aircraft weather reports (AIREPs), and buoys, because winds retrieved from NSCAT data almost always have multiple solutions (ambiguities) at one location. Figure 1 shows an example of the NSCAT retrieval residual as a function of the u and υ wind components (top panel), and the corresponding NSCAT ambiguous winds (middle panel). Each wind solution has a known likelihood of being the true wind. These likelihoods are a by-product of the wind retrieval and are present in the data. In the experiments presented here, the most likely wind solution and the solution most opposite in wind direction are presented to the assimilation system. These two winds are used in a two-minima observation operator for scatterometer winds (Stoffelen and Anderson 1997):
i1520-0493-131-1-3-eq1
with
i1520-0493-131-1-3-eq2
where u and υ are the analyzed wind components; ui and υi are the corresponding scatterometer observations; σu and συ are component observation errors for scatterometer measurements, taken to be 2 m s−1 for both components; and p is an empirical exponent set to 4. The bottom panel of Fig. 1 shows an example of the two-wind cost function used in this study. This cost function allows the variational system to resolve the directional ambiguity between two wind vectors in a dynamic way during the assimilation process. Note that when the analyzed wind is close to one of the ambiguous winds, Jscato is effectively identical to a traditional quadratic cost function, Ji.

5. Design of the NSCAT experiments

Our goal in these experiments is to assess the impact of NSCAT wind data in 4DVAR. Although this is the first time NSCAT data have been used in the IFS, previous work using ERS scatterometer data has shown significant impacts on the analysis and forecast of tropical cyclones (Isaksen and Stoffelen 2000). Therefore, we focused our study on periods when intense tropical cyclones were active.

Two sets of experiments were performed for the latter halves of October and September 1996. The first experiments are for the period 14–29 October 1996 when Hurricane Lili was present in the Caribbean and Atlantic basins. The Lili experiment stops on 20 October and resumes 26 October, because no NSCAT data were available. The second experiments are for the period 20 September–3 October 1996 when Typhoons Yates and Zane were present in the western tropical Pacific. NSCAT level 2.0 science data (50-km resolution, and based on the NSCAT-2 model function) were used in all experiments. ERS-2 and SSM/I wind speed data were left out of the assimilations in order to make a cleaner investigation of the impact of NSCAT winds. Otherwise, the standard set of observations including radiance data from Television and Infrared Observing Satellite (TIROS) Operational Vertical Sounders (TOVS) and winds derived from cloud motions as observed by geostationary satellites were used.

The experiments are described in Table 1. Only two factors are varied in the experiments: the assimilation of NSCAT data, and the resolution of the tangent linear and adjoint model (T63, T106). This makes it possible to clearly investigate the impact of NSCAT data on one hand, and higher-resolution increments on the other. The effect of higher-resolution increments was examined to test the hypothesis that higher resolution during the minimization would allow a better match of the spatial scales of analysis and observation increments. Also, the scales of motion between Yates and Zane (interacting cyclones) were small enough to require higher resolution to resolve important features of their effects on each other.

6. Hurricane Lili

a. Synoptic description

Lili was the most intense Atlantic hurricane during the lifetime of NSCAT. Figure 5 shows the track of Hurricane Lili, 14–29 October 1996, and NSCAT coverage for a 6-h period. After forming in the Caribbean Sea, Lili moved north over Cuba, intensified as it went through the Bahamas on 19 October, and reached maximum intensity on 20 October (51 m s−1, 960 hPa) east of the Bahamas. Then Lili moved slowly toward the northeast and east during the next 4 days, gradually becoming weaker and stalling for a period at about 55°W. But on 24 October 1996, it started to intensify again and move faster toward Europe under the influence of extratropical weather systems. On 25 October it still had hurricane strength and Lili remained a tropical cyclone until 27 October 1996. It then moved to the east until crossing the coast of Ireland on 28 October. For a more detailed synoptic description see Pasch and Avila (1999).

b. Analysis and forecast impacts

By October 26, Lili was in the central Atlantic and making the transition from a hurricane to an extratropical low. Figure 6 shows the first guess and analysis fields of Lili for experiments NoSCAT and NSCAT. The NoSCAT and NSCAT experiments resumed at this time after an NSCAT data blackout, so the first guess fields are the same for both experiments. Lili is much too weak in both analyses, because the T319/T63 assimilation system is not able to represent the intense small-scale features of Hurricane Lili. But it is clear that the NSCAT analysis demonstrates a substantial improvement in both position and intensity of Lili. In this dramatic case, increments in mean sea level pressure (MSLP) of 12 hPa and wind increments of 17 m s−1 can be seen. The impact in this case is unusually large compared to typical impacts around cyclones, because of a large error in position in the first guess (uncommon for the Atlantic) and good coverage of Lili's circulation by NSCAT (see Fig. 7). Consistent with the tropical to extratropical transition is the hint of a front in the strong wind shifts extending to the northeast of Lili's center, a feature clearly observed by NSCAT. Figure 8 shows the availability of other observations in the vicinity of Lili. Cloud contamination eliminates TOVS data in the area of interest, the complex flow pattern results in few cloud motion winds near the center of Lili, and there are only two ships and some aircraft data close by. The NoSCAT analysis (Fig. 6, left, middle panel) clearly demonstrates that these data were too sparse to significantly improve Lili's position or intensity.

The large impact of NSCAT data on Lili near the surface is propagated by 4DVAR with substantial effect in the vertical. Figures 9–12 show northwest–southeast cross sections through Lili. The cross-section location is shown in Fig. 8. Wind speeds in the NSCAT experiment (Fig. 9) are increased through nearly the whole troposphere, with a maximum scalar wind speed increment of greater than 10 m s−1 along this cross section between 700 and 800 hPa. This corresponds to a ∼50% increase in midlevel wind speeds near the center of Lili. There is even evidence of an eye in the NSCAT wind speed increments at (39°N, 40°W), because smaller increments were produced close to the center of circulation. The NoSCAT wind speed increments demonstrate that some aircraft data were used around the 300-hPa level, but had little effect on Lili's intensity or position.

Increments of vertical velocity through the troposphere (Fig. 10) show that NSCAT data have increased the updrafts in Lili by 5 Pa s−1 at the 600-hPa level and increased sinking motion in the environment around Lili, consistent with a tropcial cyclone circulation. Temperature increments for the NoSCAT experiment (Fig. 11) show an isolated impact on the analysis from aircraft observations (−1.0° to +1.5°C) at the 300-hPa level. Temperature increments for the NSCAT experiment, however, are larger (−1.5° to +2.5°C), and temperature is increased by 1.5°–2.0°C throughout nearly the whole troposphere in the vicinity of Lili. Lili is still a warm core system, but note that the temperature increments in Fig. 11 are tilted in the vertical, a feature normally associated with a developing baroclinic system.

Because 4DVAR optimizes the fit between the model and observations over a 6-h period, a surface wind observation may be just as effective as a temperature or moisture observation for modifying an analysis. Mass and wind fields are coupled together by the dynamics and thermodynamics of the model, and it is clear from the cross sections that scatterometer data are useful for much more than improving the surface winds when used in 4DVAR. For example, cross sections of specific humidity (Fig. 12) show that the NSCAT data have produced increments as large as 3.5 g kg−1 at 800 hPa. This corresponds to a ∼50% increase in moisture at midlevels.

Lili's increasingly baroclinic character is also clear from a plan view. Figure 13 shows analyses of 500-hPa temperature and MSLP 1 day later at 0000 UTC October 27 for NoSCAT and NSCAT. These fields are no longer aligned in the vertical, as they would be for a tropical cyclone. The use of NSCAT data has clearly increased Lili's intensity. Also, just northwest of the center of Lili between 45° and 50°N, the 500-hPa horizontal temperature gradient is nearly twice as strong in NSCAT as in NoSCAT. As Lili continued to move northeastward, it quickly encountered a strong baroclinic region farther to the north. The ensuing interaction strengthened and accelerated Lili toward its eventual landfall in Ireland 2 days later.

The analysis impact of NSCAT on Lili persisted for the next 36 h. Figure 14 shows NSCAT − NoSCAT MSLP differences for six consecutive analysis times. These plots show that the NSCAT MSLP analyses are 9–15 hPa lower than NoSCAT in the vicinity of Lili. This result is surprising so near Europe, and shows that scatterometer data can make significant impacts even in areas traditionally thought to be well observed.

Lili made landfall in Ireland at 1200 UTC 28 October, 60 h after the analysis time examined in detail above. Figure 15 shows the 2-day forecast impact for experiments NoSCAT and NSCAT. The NoSCAT position error is smaller than the NSCAT error because the NSCAT forecast is lagging a few hours behind, but it is on the right track. Both position errors are quite small compared to typical 48-h position errors. The minimum central pressure in the NSCAT experiment is 5 hPa lower than in NoSCAT and closer to the verifying analysis. The wind speeds are clearly more intense in the NSCAT experiment, especially in the region south of the low where 25 m s−1 winds are forecast. In the NoSCAT experiment only 20 m s−1 winds are forecast. It is clear that a forecaster would issue a much more severe and correct warning from the NSCAT forecast than from the NoSCAT forecast.

7. Typhoons Yates and Zane

a. Synoptic description

Yates and Zane were northwestern Pacific tropical cyclones. Yates was active from 17 September until 3 October 1996. It had typhoon strength winds from 23 September until 1 October and even supertyphoon strength (67 m s−1 maximum wind speed) from 24 to 29 September. Zane was active from 30 September until 6 October 1996. It had typhoon strength (57 m s−1 maximum wind speed) from 25 September until 3 October 1996. Yates formed at 7°N near the date line, whereas Zane originated farther west near (6°N, 155°E). Figure 16 shows their tracks. There are clear signs of tropical cyclone interaction between Yates and Zane. For example, during the period 25 September to 1 October when both Yates and Zane have reached typhoon strength, the tracks follow a tandem linked movement northward, and then northeastward. Lander et al. (1999) have a comprehensive discussion regarding the interaction between Yates and Zane, and their Fig. 18 shows the interaction between Yates and Zane during the recurving period. Based on satellite images (not shown), it can be seen that the inflow region for Yates is affected by Zane and vice versa.

b. Analysis and forecast impacts

The benefits of NSCAT data to analyses and forecasts demonstrated for Lili in section 6b are emphasized further by the results for Yates and Zane. Here we focus on the analysis and forecast performance on 26 September 1996, 6 days into the experiment.

In coarse-scale assimilation systems, errors in representing the interaction and boundaries of the flow in the vicinity of two interacting tropical cyclones can often lead to misinterpretation of observations. This difficulty is intensified for the ambiguous scatterometer data. Therefore to investigate the importance of resolution on this case, in addition to the normal resolution assimilation system (T319/T63) used for the Lili cases described above in section 6b, we performed an additional assimilation with a higher-resolution analysis system (T319/T106). The normal-resolution assimilation experiments were performed for the period 20 September–3 October, and the higher-resolution experiments were performed for the period 26–29 September 1996 (see Table 1).

1) Assimilation experiments at nominal resolution (T319/T63)

The northwestern Pacific region generally has few conventional observations. In specific cases like this, even fewer conventional observations are available since ships and aircraft try to avoid typhoons. Figure 17 (top panel) shows the few active conventional observations available in the vicinity of both Yates and Zane for the 1200 UTC 26 September assimilation. Within a radius of 500 km, only high-level aircraft reports were available. The bottom panel of Fig. 17 shows that no TOVS data were available in the region for the 1200 UTC 26 September assimilation. Only cloud-motion-derived winds from geostationary satellites were available, and none were near Yates and Zane due to difficulties in height assignment and the complex flow in the area.

As in the Lili experiments, ERS-2 scatterometer and SSM/I winds were not used in these assimilation experiments so the only remaining source for near-surface satellite winds is NSCAT data. Figure 18 shows the NSCAT data available to the assimilation system from 1200 UTC 26 September until 0000 UTC 28 September 1996. Due to the satellite orbit, NSCAT data are available at 0000 and 1200 UTC in this region, so NSCAT observes both Yates and Zane well every 12 h during this period. The plots show the positions of Yates and Zane and the NSCAT winds in the region available to the assimilation. The dots show the full 50-km resolution of the NSCAT data used.

An assimilation experiment similar to the Lili case described in section 6b was performed at T63 analysis resolution. Figure 19 shows the first guess, analysis, and analysis increment of MSLP pressure and 10-m wind fields for NoSCAT and NSCAT, valid 1200 UTC 26 September 1996. The NSCAT data locations are marked as small dots. The first guess fields near Zane (20°N, 128°E) are similar, but the analyzed intensity for Zane is significantly better in NSCAT, with an MSLP analysis increment of −6 hPa at the right location. Figure 20 shows NSCAT winds near Typhoon Zane at 1200 UTC 26 September 1996. The MSLP analysis is overlaid, and the Joint Typhoon Warning Center (JTWC, Hawaii) observed center location is marked with a cross square. The NoSCAT experiment only has a −2.7 hPa increment to the northeast of Zane resulting in an elongated tropical cyclone. The intensity of Yates (17°N, 142°E) is left almost unchanged in both analyses. The position is improved slightly in the NSCAT assimilation, whereas the NoSCAT analysis leaves Yates unchanged. The observed intensity of Yates is a category 4 typhoon (120 mi h−1 wind speed) and Zane is a category 3 typhoon (100 mi h−1 wind speed) at 1200 UTC 26 September. The assimilation system is not capable of representing such intense tropical cyclones properly (see discussion in section 6b), but it is able to represent a tropical-cyclone-like feature at the resolution of the assimilation system. In this situation, the NSCAT assimilation is able to improve the analysis of Zane but not of Yates. This results in an overall worse analysis, since the relative intensities of the cyclone pair have been changed, with Zane much stronger than Yates. In fact, a comparison of the forecast evolution from the NoSCAT and NSCAT analyses shows a somewhat poorer medium-range NSCAT forecast. Figure 21 shows the NoSCAT and NSCAT 132-h forecasts and the verifying analysis. In the NSCAT forecast Zane is too intense (relative to Yates), overtakes and swallows Yates, and produces a larger-scale intense low in the wrong position. On the other hand, in the NoSCAT forecast, Zane is too weak and lags behind the observed track. Even though the NoSCAT forecast is not a good forecast, it still may have been better to leave Zane unchanged in the NSCAT experiment in this case, rather than change the relative strengths of the two typhoons.

The poor NSCAT forecast in this case is in stark contrast to the improved analyses and forecasts for the two Lili case studies in section 6b. The Lili experiments were, as described above, performed with the same inner loop resolution (T63), so they show that it is possible to obtain a positive impact using NSCAT data even with a coarse analysis resolution. Often the analyses of tropical cyclones are improved because the inflow in the surrounding area is described better by the NSCAT data and this helps to improve the general description of the cyclone. Even in the Lili cases in which we observed positive impacts, we typically saw much of the NSCAT data rejected by the analysis system near the center of the topcial cyclones. Therefore to investigate the importance of inner loop resolution on the use of the NSCAT data, we performed parallel assimilation experiments using T106 resolution. They will be referred to hereafter as NoSCAT106 and NSCAT106.

2) Assimilation experiments at higher resolution (T319/T106)

Figure 22 shows the T106 analyses for NoSCAT106 and NSCAT106 at 1200 UTC 26 September 1996, 12 h into the assimilation experiment. The two first guess fields are virtually identical in the area of interest. Zane has a central pressure of 1001 hPa, and Yates's is 1003 hPa. This is left almost unchanged by the NoSCAT106 analysis, whereas the NSCAT106 analysis intensifies Zane by 6 hPa and moves and intensifies Yates by 4 hPa. The resulting NSCAT106 analysis is the best of the four analyses presented here.

We will now compare the results of the T106 experiment with the T63 experiment around Zane. Figure 23 shows NSCAT data rejection around Zane in T63 and T106 analyses. In the NSCAT experiment, 20 NSCAT observations were removed from the analysis by 4DVAR QC. But only 17 were removed from the NSCAT106 analysis. The three additional observations accepted by NSCAT106 are just to the south and southeast of the center of Zane (highlighted in the upper panel of Fig. 23). High wind speeds and small spatial scales near the center of a tropical cyclone make these observations very important for improving placement and intensity in the analysis. These three observations alone move Zane ∼20 km west (closer to the observed position) and lower the central pressure by ∼1 hPa. While this is a modest improvement, it is only attributable to an increase in the resolution of the analysis and clearly shows an improved use of the data.

In section 6b, we showed that the 4DVAR assimilation system is capable of propagating information from surface observations throughout the troposphere in a dynamically consistent way. This enhances the impact of NSCAT surface winds. Figures 24–27 show southwest-to-northeast cross sections of first guess and analysis increment fields through Typhoon Zane at 1200 UTC September 1996. The cross-section location is shown in Fig. 17. Figure 24 shows the cross section of the scalar wind speed. The first guess fields are very similar because the assimilations started from identical background fields at 0000 UTC 26 September 1996. The use of NSCAT data introduces a much more intense and balanced tropical cyclone in the NSCAT106 analysis: to the southwest (left part of the figure) wind speed increments of 8 m s−1 are seen in NSCAT106 versus only 2 m s−1 in the NoSCAT106 assimilation. The center of the typhoon is also well defined in the NSCAT106 wind speed increments. 4DVAR's ability to propagate surface wind information throughout the troposphere is clearly seen in this example where the NSCAT106 increments refine the location of the center of the typhoon, maintaining a tropical-cyclone-like structure throughout the troposphere. Note the barotropic (untilted) pattern of the increments, which have the effect of moving the entire typhoon toward the northeast, very close to the correct location.

Because 4DVAR is a multivariate analysis with a dynamic forecast model constraint, surface wind observations produce analysis increments for other prognostic variables. In Fig. 25 we show the cross section of vertical velocity. The pattern in the first guess is amplified by NSCAT data, and only slightly changed in the NoSCAT106 analysis by rawinsonde and aircraft reports in the area (see Fig. 17). Figures 26 and 27 show cross sections of potential temperature and specific humidity increments, respectively. The potential temperature cross section shows how a warm core increment is introduced in the lower half of the troposphere near the center of the tropical cyclone. Other observations (in this case aircraft data near the 500-hPa level in Fig. 26, lower-right panel) are responsible for some of the temperature increments in the northeast region. The NSCAT106 increments that can be attributed to NSCAT data clearly show a “barotropic like” structure for the wind and temperature, as would be expected for a tropical cyclone still at 20°N. It is interesting to compare this barotropic-like cross-section structure for Zane (see Fig. 24) to the more “baroclinic” structure for Lili in Fig. 11, where Lili was becoming an extratropical disturbance in the baroclinic region near 40°N. The vertical cross sections for the equivalent T63 assimilation (not shown) share the same patterns but have, as one would expect, a broader structure than the T106 increments. This is also clear when comparing the Lili T63 case (Fig. 9 has a cross-sectional horizontal length of 2800 km) with the Zane T106 case (Fig. 24 has a cross-sectional length of 1700 km): the latter covers a much smaller horizontal area.

Figure 28 (top and bottom) shows the 132-h NSCAT106 and NoSCAT106 forecasts starting from the analyses of Fig. 22. The verifying analysis is shown in the middle panel. The NoSCAT forecast is very similar to the T63 forecast presented in Fig. 21 (bottom panel) with Zane much weaker than Yates, with too large a distance between the two storms. The NSCAT106 forecast is substantially better than the NSCAT forecast for this case—the forecast position and intensities of both Yates and Zane are improved. Recall, in the NSCAT forecast, that Yates and Zane merge. Here, in the NSCAT106 experiment, the separation and intensity of the storms are more correctly forecast. This case is also an example that shows the importance of treating tropical cyclones properly when they start to interact with the extratropical flow. In Fig. 29, the 500-hPa anomaly correlation scores for the North Pacific show how the general circulation of the region was dramatically improved during the first days of October (i.e., in the 4–5-day forecast range) in the NSCAT106 experiment compared to the NSCAT experiment. Notice that the NoSCAT106 experiment is also improved over the NoSCAT experiment, but not as much as the improvement from NSCAT to NSCAT106. The improvement in the NSCAT106 experiment can be attributed to higher-resolution analysis increments, which leads to improved use of the data. Only in a dramatic situation such as this would one expect to see such a marked impact of NSCAT data.

We can conclude that the complexity of the two interacting typhoons requires T106 analysis resolution to use the NSCAT data properly, and that higher-resolution analyses make better use of NSCAT data.

8. Current issues and the future

The first Sea Winds instrument is now flying on the QuikSCAT satellite. QuikSCAT data have been used experimentally at ECMWF since December 2000 and were implemented in operations at ECMWF January 2002. The second SeaWinds instrument is scheduled for launch on ADEOS-2 in December 2002. Like NSCAT, SeaWinds is also a Ku-band scatterometer, but utilizes a very different viewing geometry (Perry 2001). There are a number of issues that could improve the use of NSCAT data, and which have a bearing on the use of SeaWinds data for NWP. Anticipated improvements in the 4DVAR data assimilation system and/or improvements in the scatterometer data are expected to resolve many of the following issues in the next several years.

  • Higher spatial coverage due to wider swaths, or multiple instruments, increases the effect of scatterometer data. Data coverage increases approximately by a factor of 2 from ERS-2 to NSCAT to SeaWinds. Also, two SeaWinds instruments should be operational after the launch of ADEOS-2, and the double-swath C-band Advanced Scatterometer (ASCAT) instrument on METOP-1 (1 in a series of 3 European polar-orbiting meteorology satellites) will replace ERS-2. The wider swaths of SeaWinds should result in fewer circulation centers just beyond the swath edges such as Yates in the lower-right panel of Fig. 18. Wider and/or overlapping swaths should make it less likely that the scatterometer data are consistent with multiple analysis interpretations.
  • The Ku-band scatterometers are affected by rain. Special quality control (QC) or correction for rain effects may be important. In the present study we ignored rain, but the QC procedures in the data assimilation system eliminate the worst scatterometer wind observations. QuikSCAT data seem to be more sensitive to rain than NSCAT, probably due to the higher incidence angles used by QuikSCAT. However QuikSCAT has a number of QC flags that are useful for data screening. SeaWinds on ADEOS-2 will be accompanied by an Advanced Microwave Scanning Radiometer (AMSR), which will allow precise rain QC, and which may allow the SeaWinds backscatter observations to be corrected for rain effects.
  • Although 4DVAR makes use of a two-wind scatterometer cost function and all other a priori data when resolving the ambiguity in the scatterometer winds, if the location of an intense feature in the background has a substantial error, the wrong ambiguity will be in agreement with the background and may, therefore, be used. The methodology of Hoffman et al. (1995) could be used to align features present in the background and observations. However, it is not possible to avoid selection of the wrong ambiguity in all cases.
  • When the scatterometer winds are in error due to rain contamination or ambiguity removal errors, it is likely that patches of data will be affected. The 4DVAR could make use of improved quality information.
  • Higher (inner loop) resolution of the 4DVAR will allow the analysis of smaller scales of motion, scales present at full model resolution and in the scatterometer data, but which are currently filtered by 4DVAR minimization. The ECMWF inner-loop resolution increased to T159 in November 2000. An updated background error covariance matrix would also improve the use of scatterometer data on smaller scales.
  • A longer cycling window for the 4DVAR (say 12 h instead of 6 h) will allow for more physically consistent use of the scatterometer data, and greater influence on upper levels. This is partly because the analysis is forced to fit more scatterometer orbits in a consistent flow-dependent way. However, a longer cycling window may violate the 4DVAR assumption that model errors are negligible, compromising the quality of innovations and the model trajectory. The ECMWF cycling window was increased from 6 to 12 h in September 2000.
  • Refined observation cost functions for scatterometer data may enhance the impact of these data. For example it is possible to directly use the backscatter data (Thépaut et al. 1993). Alternatively, it should be possible to formulate cost functions in terms of the wind ambiguities, which more closely reflect the backscatter cost function used to retrieve the ambiguities.

9. Conclusions

We examined the impact of NSCAT data on tropical cyclone forecasting on the ECMWF 4DVAR data assimilation system. Although the NSCAT mission ended prematurely, the follow-on SeaWinds instrument aboard the QuikSCAT satellite is currently operational, and experience with NSCAT provides useful guidance for SeaWinds. We conducted parallel runs denoted NSCAT and NoSCAT, with and without NSCAT data, respectively. In the NSCAT experiments, 4DVAR is presented with two ambiguities and effectively resolves the ambiguity using a special cost function. This paper has concentrated on evaluating the impact of NSCAT data on tropical cyclones by highlighting a few typical examples. ERS-2 and SSM/I data were excluded from both NSCAT and NoSCAT experiments to isolate the impact of NSCAT. It would be interesting to examine the impact of NSCAT in a system with these other data sources, but this should be properly addressed by a paper focused on such a comparison. For the case studies that are the focus of this study, these other surface wind data would make interpretation of the results more difficult. In general NSCAT data are expected to improve the surface wind fields, especially in the Tropics and the Southern Hemisphere, leading to improved surface fluxes. These effects cannot be quantified in the relatively short assimilation experiments we have conducted. Similarly it is not possible to evaluate the impacts on overall skill scores because of the small sample. Even small improvements in forecast skill are hard to come by in current state-of-the-science data assimilation systems like the ECMWF 4DVAR even with the presumed advantage of excluding the ERS-2 and SSM/I data. However, case studies of Hurricane Lili, and of Typhoons Yates and Zane, show major positive impacts of NSCAT data on tropical cyclone forecasts of both intensity and position. This is a significant result but not unexpected: scatterometer data have previously been shown to improve the depiction of the surface wind field in both tropical cyclones and extratropical lows, and can provide early detection of these features. Additionally in the 4DVAR, scatterometer observations are assimilated properly in time and space, and the dynamics of 4DVAR allows the influence of surface observations to be propagated more correctly in the vertical.

Specific notable impacts include the following:

  • improved 2–5-day forecast tracks of tropical cyclones,
  • better analyzed thermodynamic structure of tropical cyclones via 4DVAR,
  • improved use of NSCAT winds using higher-resolution (T106) analysis increments,
  • improved representation of interacting tropical cyclone systems using higher-resolution (T106) analysis increments, and
  • consistent and remarkably large increments for all model variables attributable solely to NSCAT data. The largest increments linked to NSCAT data are of the same magnitude as the first guess field values for horizontal wind speed and vertical velocity. For temperature, increments of 1°–2°C are common throughout the troposphere, and for specific humidity, the increment magnitude is 10%–40% of the background field values.

Acknowledgments

This research was supported by the NSCAT and SeaWinds NASA scatterometer projects and ESA/ESTEC Project 111699/95/NL/CN. This work was carried out while the first author was a visiting scientist at the European Centre for Medium-Range Weather Forecasts. We thank ECMWF for generous facilities and administrative support. We thank the NASA Physical Oceanography Data Active Archive Center (PODACC) for supplying the NSCAT data.

REFERENCES

  • Andersson, E., , and H. Järvinen, 1999: Variational quality control. Quart. J. Roy. Meteor. Soc., 125B , 697722.

  • Andersson, E., and Coauthors. 1998: The ECMWF implementation of three-dimensional variational assimilation (3D-Var). III: Experimental results. Quart. J. Roy. Meteor. Soc., 124B , 18311860.

    • Search Google Scholar
    • Export Citation
  • Atlas, R., , and R. N. Hoffman, 2000: The use of satellite surface wind data to improve weather analysis and forecasting. Satellites, Oceanography and Society, D. Halpern, Ed., Elsevier Oceanography Series, Vol. 63, Elsevier, 57–78.

    • Search Google Scholar
    • Export Citation
  • Atlas, R., , S. C. Bloom, , R. N. Hoffman, , E. Brin, , J. Ardizzone, , J. Terry, , D. Bungato, , and J. C. Jusem, 1999: Geophysical validation of NSCAT winds using atmospheric data and analyses. J. Geophys. Res., 104 (C5) 1140511424.

    • Search Google Scholar
    • Export Citation
  • Atlas, R., and Coauthers, 2001: The effects of marine winds from scatterometer data on weather analysis and forecasting. Bull. Amer. Meteor. Soc., 82 , 19651990.

    • Search Google Scholar
    • Export Citation
  • Brown, R. A., , and L. Zeng, 1994: Estimating central pressures of oceanic midlatitude cyclones. J. Appl. Meteor., 33 , 10881095.

  • Chelton, D. B., , M. H. Freilich, , and S. K. Esbensen, 2000: Satellite observations of the wind jets off the Pacific coast of Central America. Part I: Case studies and statistical characteristics. Mon. Wea. Rev., 128 , 19932018.

    • Search Google Scholar
    • Export Citation
  • Courtier, P., , J-N. Thépaut, , and A. Hollingsworth, 1994: A strategy for operational implementation of 4D-Var, using an incremental approach. Quart. J. Roy. Meteor. Soc., 120 , 13671387.

    • Search Google Scholar
    • Export Citation
  • Ebuchi, N., , and H. C. Graber, 1998: Directivity of wind vectors derived from the ERS-1/AMI scatterometer. J. Geophys. Res., 103 (C4) 77877797.

    • Search Google Scholar
    • Export Citation
  • Figa, J., , and A. Stoffelen, 2000: On the assimilation of Ku-band scatterometer winds for weather analysis and forecasting. IEEE Trans. Geosci. Remote Sens., 38 , 18931902.

    • Search Google Scholar
    • Export Citation
  • Francis, R., and Coauthors. 1991: The ERS-1 spacecraft and its payload. ESA Bulletin, Vol. 65, ESA, 27–48.

  • Freilich, M. H., , and R. S. Dunbar, 1999: The accuracy of the NSCAT 1 vector winds: Comparisons with National Data Buoy Center buoys. J. Geophys. Res., 104 (C5) 1123111246.

    • Search Google Scholar
    • Export Citation
  • Grantham, W. L., , E. M. Bracalente, , W. L. Jones, , and J. W. Johnson, 1977: The Seasat-A satellite scatterometer. IEEE J. Oceanic Eng., OE-2 , 200206.

    • Search Google Scholar
    • Export Citation
  • Grassotti, C., , S. M. Leidner, , J-F. Louis, , and R. N. Hoffman, 1999: Development and application of a visible-infrared rain flag for scatterometer data. J. Appl. Meteor., 38 , 665676.

    • Search Google Scholar
    • Export Citation
  • Hoffman, R. N., , Z. Liu, , J-F. Louis, , and C. Grassotti, 1995: Distortion representation of forecast errors. Mon. Wea. Rev., 123 , 27582770.

    • Search Google Scholar
    • Export Citation
  • Isaksen, L., 1997: Impact of ERS scatterometer data in the ECMWF 4D-Var assimilation system. Preliminary studies. Space at the Service of Our Environment, Florence, Italy, ESTEC, European Space Agency SP-414, 1829–1851.

    • Search Google Scholar
    • Export Citation
  • Isaksen, L., , and A. Stoffelen, 2000: ERS scatterometer wind data impact on ECMWF's tropical cyclone forecasts. IEEE Trans. Geosci. Remote Sens., 38 , 18851892.

    • Search Google Scholar
    • Export Citation
  • Järvinen, H., , E. Andersson, , and F. Bouttier, 1999: Variational assimilation of time sequences of surface observations with serially correlated errors. Tellus, 51A , 469488.

    • Search Google Scholar
    • Export Citation
  • Katsaros, K. B., , E. B. Forde, , P. Chang, , and W. T. Liu, 2001: QuikSCAT's SeaWinds facilitates early identification of tropical depressions in 1999 hurricane season. Geophys. Res. Lett., 28 , 10431046.

    • Search Google Scholar
    • Export Citation
  • Klinker, E., , F. Rabier, , G. Kelly, , and J-F. Mahfouf, 2000: The ECMWF operational implementation of four-dimensional variational assimilation. III: Experimental results and diagnostics with operational configuration. Quart. J. Roy. Meteor. Soc., 126A , 11911215.

    • Search Google Scholar
    • Export Citation
  • Lander, M. A., , E. J. Trehubenko, , and C. P. Guard, 1999: Eastern Hemisphere tropical cyclones of 1996. Mon. Wea. Rev., 127 , 12741300.

  • Mahfouf, J-F., , and F. Rabier, 2000: The ECMWF operational implementation of four-dimensional variational assimilation. II: Experimental results with improved physics. Quart. J. Roy. Meteor. Soc., 126A , 11711190.

    • Search Google Scholar
    • Export Citation
  • Naderi, F. M., , M. H. Freilich, , and D. G. Long, 1991: Spaceborne radar measurement of wind velocity over the ocean—An overview of the NSCAT scatterometer system. Proc. IEEE, 79 , 850866.

    • Search Google Scholar
    • Export Citation
  • Pasch, R. J., , and L. A. Avila, 1999: Atlantic hurricanes of 1996. Mon. Wea. Rev., 127 , 581610.

  • Perry, K. L., 2001: QuikSCAT science data product user's manual, overview and geophysical data products. Version 2.2. Jet Propulsion Laboratory JPL-D18053, Pasadena, CA, 89 pp.

    • Search Google Scholar
    • Export Citation
  • Price, J. C., 1976: The nature of multiple solutions for surface wind speed over the oceans from scatterometer measurements. Remote Sens. Environ., 5 , 4754.

    • Search Google Scholar
    • Export Citation
  • Rabier, F., , H. Järvinen, , E. Klinker, , J-F. Mahfouf, , and A. Simmons, 2000: The ECMWF operational implementation of four-dimensional variational assimilation. I: Experimental results with simplified physics. Quart. J. Roy. Meteor. Soc., 126A , 11431170.

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

    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., , and D. Anderson, 1997: Ambiguity removal and assimilation of scatterometer data. Quart. J. Roy. Meteor. Soc., 123B , 491518.

    • Search Google Scholar
    • Export Citation
  • Thépaut, J-N., , R. N. Hoffman, , and P. Courtier, 1993: Interactions of dynamics and observations in a four-dimensional variational assimilation. Mon. Wea. Rev., 121 , 33933414.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

(top) Plot of maximum likelihood estimator (MLE) for one wind vector cell as a function of u- and υ-wind components. (middle) Ambiguous NSCAT winds corresponding to the top panel. (bottom) Two-wind cost function used in 4DVAR

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 2.
Fig. 2.

Two-dimensional histogram of NSCAT vs ECMWF first guess winds for the period 0000 UTC 14 Oct–1200 UTC 20 Oct 1996. The NSCAT ambiguity closest to the first guess field is used for the comparison

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 3.
Fig. 3.

Histograms of wind speed departures of observations from (top) ECMWF first guess winds (ob) and (bottom) ECMWF analyzed winds (oa). These departures are for NSCAT data in the Northern Hemisphere extratropics only (20°–90°N) for the period 2100 UTC 15 Oct–0900 UTC 16 Oct 1996 (from two assimilation cycles)

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 4.
Fig. 4.

Time series of 5-day forecast scores for NoSCAT and NSCAT and Southern Hemispheres, the scores are 500-hPa height anomaly correlation. In the Tropics, the scores are 850-hPa vector wind rms errors

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 5.
Fig. 5.

Track of Hurricane Lili from 1200 UTC 14 Oct 1996 to 0000 UTC 29 Oct 1996. Observed positions at 0000 and 1200 UTC from the National Hurricane Center (NHC) are plotted as diamond–plus symbols along the track. The 6-h NSCAT data coverage beginning at 2100 UTC 19 Oct 1996 is shown by fine dots. The A marks the position of Lili at the time of the case study presented, 0000 UTC 26 Oct 1996

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 6.
Fig. 6.

Analysis of Hurricane Lili 0000 UTC 26 Oct 1996. (top) First guess 10-m wind (kt) and MSLP (hPa). (middle) The analyzed wind and MSLP. (bottom) Analysis increments of wind (times 10) and MSLP. The observed central pressure and position from the NHC at 0600 UTC 19 Oct was 975 hPa at 38.1°N, 41.0°W

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 7.
Fig. 7.

NSCAT winds in the vicinity of Hurricane Lili at 0051 UTC 26 Oct 1996. The NSCAT ambiguity closest to the analyzed wind direction is plotted. MSLP analysis from NSCAT experiment valid at 0000 UTC 26 Oct 1996. Winds rejected by the variational quality control are darker

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 8.
Fig. 8.

Non-NSCAT data presented to the analysis at 0000 UTC 26 Oct 1996: (top) satellite data (TOVS retrievals, crosses; satellite cloud motion winds, inverted triangles) and (bottom) synoptic and aircraft reports (synop/ship reports, circles; aircraft reports, diamonds). Hurricane Lili's observed position is marked in both panels. Each panel also has a cross-section locator for Figs. 9–12

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 9.
Fig. 9.

Cross sections of scalar wind speed valid at 0000 UTC 26 Oct 1996: (top) first guess wind speed (m s−1) from NoSCAT and NSCAT and (bottom) wind speed increments (m s−1) from NoSCAT and NSCAT. Cross-section location is shown in Fig. 8

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 10.
Fig. 10.

Same as in Fig. 9 but for vertical velocity (Pa s−1)

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 11.
Fig. 11.

Same as in Fig. 9 but for temperature (K and °C)

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 12.
Fig. 12.

Same as in Fig. 9 but for specific humidity (kg kg−1)

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 13.
Fig. 13.

Comparison of (top) NSCAT and (bottom) NoSCAT MSLP and 500-hPa temperature (°C) analyses valid at 0000 UTC 27 Oct 1996

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 14.
Fig. 14.

Analysis differences of MSLP (NSCAT − NoSCAT) for 6-hourly analyses for 0000 UTC 26 Oct–0600 UTC 27 Oct 1996. The contour interval is 3 hPa. Maximum differences are displayed (units, hPa)

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 15.
Fig. 15.

Forecasts of Hurricane Lili valid at 1200 UTC 28 Oct 1996: (middle) verifying analysis from the NoSCAT experiment and 2-day forecasts of Lili's position from (top) NSCAT and (bottom) NoSCAT. MSLP contour interval is 5 hPa, and wind speed is contoured every 2 m s−1 beginning at 20 m s−1. The observed central pressure from the NHC at 1200 UTC 28 Oct was 970 hPa

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 16.
Fig. 16.

Tracks of Typhoons Yates and Zane for the period 1200 UTC 17 Sep–0000 UTC 6 Oct 1996. Observed positions at 0000 and 1200 UTC from the JTWC are plotted as circle–plus (⊕) and box–plus symbols for Yates and Zane, respectively. The 6-h NSCAT data coverage beginning at 2100 UTC 27 Sep 1996 is shown in fine dots. The Y and Z mark the positions of Yates and Zane at 1200 UTC 26 Sep 1996, the time used as a case study in this paper

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 17.
Fig. 17.

Non-NSCAT data presented to the analysis valid at 1200 UTC 26 Sep 1996: (top) satellite data (satellite cloud motion winds, inverted triangles) and (bottom) synoptic and aircraft reports (synop/ship reports, circles; aircraft reports, diamonds; buoys, triangles; rawinsonde stations, small crossed squares). The positions of Typhoons Yates and Zane are marked in both panels. Each panel also has a cross-section locator for Figs. 24–27

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 18.
Fig. 18.

NSCAT data coverage in the region of Typhoons Yates and Zane. These plots show data used in analyses valid at (upper left) 1200 UTC 26 Sep, (upper right) 0000 UTC 27 Sep, (lower left) 1200 UTC 27 Sep, and (lower right) 0000 UTC 28 Sep 1996. Data are thinned for plotting purposes. The full resolution of the data may be seen by the small dots

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 19.
Fig. 19.

Analysis of Typhoons Yates and Zane at 1200 UTC 26 Sep 1996. (top) First guess 10-m wind (kt) and MSLP (hPa). (middle) Analyzed wind and MSLP. (bottom) Analysis increments of wind (times 10) and MSLP. At 1200 UTC 26 Sep, Yates was estimated to be a category 4 typhoon at 17.3°N, 142.4°E, and Zane was estimated to be a category 3 typhoon at 20.1°N, 127.7°E. MSLP contour intervals are 2 hPa for first guess and analysis panels, and 1 hPa for increment panels

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 20.
Fig. 20.

NSCAT winds in the vicinity of Typhoon Zane, at 1418 UTC 26 Sep 1996. The NSCAT ambiguity closest to the analyzed wind direction is plotted. MSLP analysis from NSCAT experiment valid at 1200 UTC 26 Sep 1996. Winds rejected by the variational quality control are darker

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 21.
Fig. 21.

Forecasts of Typhoons Yates and Zane valid at 0000 UTC 2 Oct 1996: (middle) verifying analysis from the NSCAT experiment and 5.5-day forecasts from (top) NSCAT and (bottom) NoSCAT. Observed or forecast 12-hourly positions of Yates and Zane are plotted in each panel as circle–plus and box–plus symbols, respectively. Forecast tracks are plotted with filled circles and squares. MSLP contour interval is 5 hPa

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 22.
Fig. 22.

Same as in Fig. 19 but for the T106 analyses

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 23.
Fig. 23.

NSCAT data rejected by 4DVAR quality control from the analysis for Typhoon Zane, at 1200 UTC 26 Sep 1996, from experiments (top) NSCAT and (bottom) NSCAT106. Three observations rejected in the NSCAT analysis but used in the NSCAT106 analysis are highlighted in the top panel. Winds rejected by the variational quality control in T63, but used in T106, are darker and located to the south and southeast

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 24.
Fig. 24.

Cross sections of scalar wind speed valid at 1200 UTC 26 Sep 1996 through Typhoon Zane: (top) first guess winds (m s−1) from NoSCAT106 and NSCAT106 and (bottom) wind speed increments (m s−1) from NoSCAT106 and NSCAT106. Cross-section location is shown in Fig. 17

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 25.
Fig. 25.

Same as in Fig. 24 but for vertical velocity (Pa s−1)

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 26.
Fig. 26.

Same as in Fig. 24 but for temperature (K and °C)

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 27.
Fig. 27.

Same as in Fig. 24 but for specific humidity (kg kg−1)

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 28.
Fig. 28.

Forecasts of Typhoons Yates and Zane valid at 0000 UTC 3 Oct 1996: (middle) verifying analysis and 5.5-day forecasts from (top) NSCAT106 and (bottom) NoSCAT106. Observed or forecast 12-hourly positions of Yates and Zane are plotted in each panel as circle–plus and box–plus symbols, respectively. Forecast tracks of Yates and Zane are plotted with filled circles and squares. MSLP contour interval is 5 hPa

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Fig. 29.
Fig. 29.

The 500-hPa anomaly correlation forecast scores for the North Pacific region. Results shown are an average of four cases in Sep 1996. (top) NoSCAT with NSCAT (T63 resolution) comparison, and (bottom) NoSCAT106 with NSCAT106 comparison. The large impact of NSCAT106 is due to the importance of correctly treating Typhoons Yates and Zane with the assimilation system. In general, a smaller impact of NSCAT data would be expected than is seen in this case study

Citation: Monthly Weather Review 131, 1; 10.1175/1520-0493(2003)131<0003:IONWOT>2.0.CO;2

Table 1. 

Experimental design

Table 1. 
Save