1. Introduction
Sea surface wind vectors have been estimated with active remote sensing instruments, such as the Quick Scatterometer (QuikSCAT; Yu and McPherson 1984), and with passive polarimetric microwave radiometery, such as WindSat (Bettenhausen et al. 2006). They have been shown to have a positive impact on the National Centers for Environmental Prediction (NCEP) global forecasts (Le Marshall et al. 2006; Zapotocny et al. 2008; Bi et al. 2010). The latest remotely sensed surface wind-measuring instrument is the Advanced Scatterometer (ASCAT), which is detailed online (http://oiswww.eumetsat.org/WEBOPS/eps-pg/ASCAT/ASCAT-PG-0TOC.htm). ASCAT is one of several instruments on the Meteorological Operational (MetOp) satellite program polar-orbiting satellite launched by the European Space Agency (ESA) and operated by the European Organization for the Exploitation of Meteorological Satellites (EUMESAT). It is the first in a series of such instruments dedicated to provide routine surface wind observations over the global oceans. ASCAT is an active microwave sensor designed to retrieve ocean surface vector winds. The mission of ASCAT is to enhance the spatial and temporal resolution of surface wind observations at global and regional scales thereby allowing better characterization of the air–sea interaction process as well as ocean wind forcing.
ASCAT surface wind data are currently used in daily weather forecast operations at the European Centre for Medium-Range Weather Forecasts (ECMWF; Hersbach and Janssen 2007a), the United Kingdom’s National Weather Service (Met Office, see online at http://research.metoffice.gov.uk/research/interproj/nwpsaf/scatter_report/ascat.html), the National Weather Service of France (Meteo-France, see online at http://cimss.ssec.wisc.edu/iwwg/iww10/talks/payan.pdf), and Environment Canada. Assimilation experiments of ASCAT surface wind vectors in the ECMWF analysis and forecast system have shown positive effects on forecast skill over the Southern Hemisphere (Hersbach and Janssen 2007b).
In this study, two seasons of ASCAT data have been assimilated into the NCEP Global Data Assimilation/Global Forecast System (GDAS/GFS), and their forecast impact assessed. This was accomplished by comparing the forecast results up to a range of 168 h with and without assimilating the ASCAT winds for the months of August 2008 and January 2009. Quality control procedures required to assimilate the surface winds are discussed. The geographical distribution of the anomaly correlations and forecast impacts are also presented.
The paper is structured as follows. Section 2 briefly describes background information related to the ASCAT instrument and the NCEP GDAS/GFS version used for this study. Section 3 discusses the diagnostic tools used to evaluate the forecast impacts. Section 4 presents the results for this forecast impact study through 7 days of model forecasts. Section 5 summarizes the overall results of these experiments.
2. Background
a. ASCAT
ASCAT is an aperture radar operating at 5.3 GHz (C band) using vertically polarized antennas. It is on the MetOP-A satellite that was launched on 19 October 2006 (Figa-Saldana et al. 2002). MetOp-A is the first in series of three satellites developed to provide meteorological data until 2020. MetOp is in a circular polar orbit (near-sun-synchronous orbit) with a period of about 101 min, at an inclination of 98.59° at a nominal height of about 800 km (817 km on average) with a 29-day repeat cycle. The ascending equatorial times occur approximately at 0930 local time.
Two sets of three antennas are used to generate radar beams, looking 45° forward, sideway, and 45° backward in azimuth with respect to the satellite’s flight direction, on both sides of the satellite ground track. These beams illuminate two 550-km-wide swaths (separated by about 700 km) as the satellite moves along its orbit, each provides measurements of radar backscatter from the sea surface sampling of 25 or 12.5 km. The ASCAT beams measure normalized radar cross sections with vertical polarization, which are a dimensionless property of the surface, describing the ratio of the effective echoing area per unit area illuminated. The result of three independent backscatter measurements is obtained for each wind vector cell using the three different viewing directions separated by a short time delay. Backscatter depends on the sea surface roughness, which is a function of the wind speed and direction at the ocean surface. It is possible to calculate the surface wind speed and direction by using these “triplets” within a mathematical model (as explained online at http://www.knmi.nl/scatterometer/publications/pdf/ASCAT_Product_Manual.pdf).
ASCAT wind retrievals used in this study were provided by the National Environmental Satellite, Data, and Information Service (NESDIS), which runs the ASCAT wind processing software developed by KNMI. Two versions of the ASCAT product are available from NESDIS, a 25 and 12.5 km. The 25-km version was chosen for these experiments, which has a spatial resolution of about 50 km.
b. Global Data Assimilation System
The December 2007 version of NCEP’s GDAS/GFS was used for these observing system experiments. Consistent with the operational GDAS/GFS, a model resolution of T382L64 (i.e., spectral triangular truncation 382 with 64 layers) was used through 168 forecast hours. The analyses and forecasts are centered at 0000, 0600, 1200, and 1800 UTC. Because of the limited computer resources, one forecast a day was run in this study. Only the 0000 UTC GFS forecasts were run out to 168 h. The 0000 UTC GFS forecast was chosen to be consistent with NCEP’s internal and external forecast verification techniques.
A short-term (6-h forecast) was run to obtain a first guess for data assimilation, which uses a larger ±3 h data cutoff window. This “late” analysis and 6-h forecast is typically called the GDAS. The GDAS analysis is run several hours later to include data that was not available to the previous “early” cycle.
The vertical domain of the GDAS/GFS forecast model ranges from the surface to approximately 0.27 hPa and is divided into 64 unequally spaced sigma layers with enhanced resolution near the bottom and top of the model domain. NCEP’s GDAS/GFS forecast model has been summarized by Kanamitsu (1989), Kalnay et al. (1990), and Kanamitsu et al. (1991). (More recent updates to the GDAS/GFS forecast model can be found online at http://www.emc.ncep.noaa.gov/gmb/STATS/html/model_changes.html.) For the most recent information about the GDAS/GFS model see Environmental Modeling Center (2003). (A summary of GDAS/GFS changes and references up to and past the dates of this study are available in an “updates” log of changes online at http://www.emc.ncep.noaa.gov/gmb/moorthi/gam.html.)
The GDAS/GFS utilizes a three-dimensional variational data assimilation (3D-VAR) scheme to obtain an estimate of the initial state that best fits the available observations and a short-range forecast (background) in an appropriate statistical sense. In this study, we use NCEP’s Gridpoint Statistical Interpolation (GSI; Derber et al. 1991; Wu et al. 2002; Derber et al. 2003; Kleist et al. 2009a,b) 3D-VAR system. The analysis becomes a 3D retrieval of mass, momentum, and moisture fields derived from all available data including the radiances (Caplan et al. 1997).
In the GSI 3DVAR system, the observation operator links the ASCAT wind vector to the model variables. If the observation is below the lowest layer of the model, a linear interpolation is used from the four nearest grid points. A variant of the Monin–Obukhov length approximation is then used to move the model value to the height of the observation. The atmospheric stability, derived from the model, is used to adjust the wind speed but does not seem to improve the quality of the analysis from using a neutral atmosphere. If the observation is above the lowest level of the model, a linear interpolation of the nearest eight grid points is used. The cost function is proportional to the squared norm of the vector wind difference between ASCAT and the model 10-m wind. We assigned the same observation error to ASCAT as is used by QuikSCAT, and WindSat. The estimated error for these observations in the GDAS is 3.5 m s−1.
3. Experimental design
The complete NCEP operational database of conventional and satellite data was used for this experiment. This included using the real-time data cutoff constraints for the early and late assimilation cycles produced at NCEP. The conventional data used in these experiments are summarized in Table 1. The satellite observations used in these experiments are summarized in Table 2. Satellite-derived surface wind vector and wind speed used in these experiments includes: Defense Meteorological Satellite Program (DMSP) Special Sensor Microwave Imager (SSM/I) surface wind speed (Alishouse et al. 1990), derived surface winds from QuikSCAT (Yu and McPherson 1984), derived surface winds from WindSat (Bettenhausen et al. 2006), and atmospheric motion vectors from geostationary satellites (Velden et al. 1997; Menzel et al. 1998). Keyser (2001a, 2001b, 2003) provides an overview of data types provided to NCEP on a daily basis and used operationally for the experiments of this study.
In situ data used within the NCEP GDAS for this study. Mass observations (temperature and moisture) are shown in the left column and wind observations are shown in the right column.
Satellite data used within the NCEP GDAS for this study.
Two seasons were used for this study: summer 2008 (1 July–31 August 2008) and winter 2008/09 (1 December 2008 to 31 January 2009). The first four weeks of each time period were removed from these results to allow the assimilation system and forecast model to adjust to the new data.
a. Quality control and data thinning
Most of the quality control (QC) of the ASCAT data was accomplished in the retrieval process. Observations that failed the retrieval process or were flagged for rain, land, or sea ice contamination were omitted from the observations (as outlined in http://www.knmi.nl/scatterometer/publications/pdf/ASCAT_Product_Manual.pdf).
A thinning technique was used for ASCAT instead of the superob technique used operationally for QuikSCAT and WindSat. WindSat are assimilated using 1° superob boxes (Bi et al. 2010). QuikSCAT are assimilated using 0.5° superob boxes. The orientation of the thinning boxes is north–south and east–west and is 100 km square. The wind closest to the center of the box is used and the latitude–longitude of the chosen observation is used. NCEP uses a “normalizing factor,” which attempts of give each observation type about the same weight per unit area. This keeps high density observations from dominating the system. We feel the difference in the reduction of the departure (O − B vs O − A) between the various data types is due to how well the observations fit with all the other data and the model background and not necessarily the observation density of a specific data type.
The preliminary statistical results from a short-term 20-day test experiment indicated that there were problems in the Antarctica regions due to contamination by sea ice. This suggested there were still some quality control problems with the observations after the retrieval process procedures were used. Based on the preliminary statistical results, a series of additional quality control procedures were added within GDAS/GFS. An SST check (SST < 273 K) was used as a criterion to remove observations suspected of still containing sea ice. This routine rejects the observations during the thinning process to allow vectors in warmer regions of the thinning box to be used.
An innovation vector difference test (observation minus background) was developed and incorporated into the ASCAT assimilation procedures. Through various trial and error tests it was determined that innovation differences greater than 5 m s−1, for the U or V component were degrading the forecast and were rejected. Observations near coastlines were also not used.
Several methods were tested to remove the ASCAT bias including speed, u and υ, and height estimation parameters. In most cases the penalty increased and the analyses fit to observations were worse, suggesting a poorer analysis. These methods typically generated worse forecasts. Thus, no bias correction is performed for the ASCAT wind.
Tests were run to determine whether 150-, 100-, or 50-km thinning boxes were more effective for ASCAT wind vector assimilation. The same observation error is assigned to QuikSCAT, WindSat, and ASCAT. The estimated observation error in the GDAS is 3.5 m s−1 for these instruments. It was found that overall the 100-km thinning box gave the best forecast results in terms of the anomaly correlation scores.
b. Ambiguity QC
In addition to the aforementioned QC procedures, an ASCAT ambiguity QC procedure was also developed and tested in these experiments. The purpose was to identify those ASCAT vectors that are pointed in the opposite direction compared to the GDAS/GFS background field. The ambiguity QC is based on comparing the vector difference of both retrieved wind vectors to the GFS model 6-h forecast and determining which direction had the smallest vector error.
Two ambiguity QC scenarios were tested to determine which one produced the better forecast. The first scenario removed observations where the vector difference of the originally selected vector is larger than its pair. The second scenario used the same criteria, but replaces the originally selected vector with its pair. Removing the suspect observations generally produced a better forecast. Less than 2% of the observations were identified as being suspect by this ambiguity check.
c. Diagnostics
Several diagnostics were performed using the Control and ASCAT experiment analyses and forecasts. The anomaly correlation (AC) statistics were recorded using the traditional NCEP algorithms (NWS 2006), which are commonly used by NWP centers worldwide. The computation of all anomaly correlations for forecasts produced by the GFS are completed using code developed and maintained at NCEP. NCEP (NWS 2006) provides a description of the method of computation while Lahoz (1999) provides an interpretation for the anomaly. The reanalysis fields from the NCEP–National Center for Atmospheric Research (NCAR; Kistler et al. 2001) 50-yr Reanalysis are used for the climate component of the anomaly correlations. This reanalysis was run at a resolution of T62L28. The output grids were reduced to 2.5° by 2.5° horizontal resolution and to rawinsonde mandatory levels. To calculate anomaly correlations the output grids from both the control and experiment were reduced to this 2.5° by 2.5° horizontal resolution using NCEP’s GFS post processor. The fields being evaluated using anomaly correlations are truncated to only include spectral wavenumbers 1–20. These fields are also limited to the zonal bands of 20°–80° in each hemisphere and to a tropical belt within 20° of the equator (20°N–20°S).
The first term on the right enclosed by parentheses in (2) can be considered the error in the Control experiment. The second term enclosed by parentheses in (2) can be considered the error in the ASCAT experiment forecast. Dividing by the error of the ASCAT experiment forecast normalizes the results. Multiplying by 100 provides a percent improvement/degradation with respect to the RMS error of the Control forecast. A positive forecast impact means the ASCAT forecast compares more favorably to the corresponding analysis than the control.
All FI diagnostics were computed from grids generated by NCEP’s post processing package. These grids have a 1° × 1° horizontal resolution and have 26 vertical isobaric surfaces. None of the fields were smoothed during plotting.
4. Results
Assimilation experiments were conducted to test and compare the attributes of using the ASCAT data. The Control contains all the operational data used during the period and includes all of the real-time data cutoff requirements. For this paper, forecast impact comparisons will be presented from assimilating the ASCAT data to a benchmark or Control experiment. The impact of assimilating the ASCAT data on the quality of forecasts made by the GFS for two time periods are explored in detail. The selection of these time periods enables the diagnostics to sample two seasons in each hemisphere. The fields diagnosed in this paper consist of geopotential height, temperature, and wind speed. Grid points determined to be underground were not included in the evaluations.
a. Analysis statistics
Figure 1 shows the monthly mean RMS difference of observation minus background (O − B) and observation minus analysis (O − A) from the ASCAT experiment and the Control for various types of marine observations used in operational NCEP GDAS/GFS for the month of August 2008 (Figs. 1a,c) and January 2009 (Figs. 1b,d). Ship data typically have the worst fit to the model background and analysis, WindSat, QuikSCAT, and ASCAT have a very similar fit among these various types marine observations. Comparing Figs. 1a,c, there is a small but sure reduction in the QuikSCAT O − B statistics for ASCAT experiment. The first-guess improvement at the QuikSCAT location is the result of a slightly improved analysis that had used ASCAT data 6 or 18 h before at a similar location. Comparing QuikSCAT, WindSat, and ASCAT wind vectors, ASCAT typically has a better fit to the model background but does not fit the model analysis quite as well as QuikSCAT for both seasons. The RMS fits for QuikSCAT and WindSat to the model background and analysis are generally the same in the Control and ASCAT experiment.
Monthly mean RMS difference of O − B and O − A from various types of marine observation from the ASCAT experiment and the Control. Results from ASCAT experiment from (a) August 2008 and (b) January 2009. (c),(d) Results from the Control from the same time period.
Citation: Monthly Weather Review 139, 11; 10.1175/2011MWR3391.1
In both the control and ASCAT experiment, QuikSCAT and WindSat surface wind measurements were used. Figure 2 shows the orbital coverage of ASCAT, WindSat, and QuikSCAT observation over 6-h synoptic window. There is considerable overlap between sensors, especially QuikSCAT and WindSat.
Typical coverage of ASCAT, WindSat, and QuikSCAT observations over 6-h synoptic window.
Citation: Monthly Weather Review 139, 11; 10.1175/2011MWR3391.1
Figure 3 displays a comparison of the bias, standard deviation, wind speed histogram, and wind speed innovation histogram for ASCAT. Figures 3a–d shows the analysis statistics for August and Figs. 3e–h shows the analysis statistics for January. The green curve represents the observation minus background (O − B) and red curve represents the observation minus analysis (O − A). While there are relatively large biases for ASCAT observed wind speeds less than 2 m s−1 during both seasons, these observations account for less than 5% of the total number of observations (Figs. 3c,g). Figures 3a,e show that there are also large biases and standard deviations for ASCAT-observed wind speeds greater than 20 m s−1 and in some cases the bias and standard deviation for O − A are even higher for O − B, again these observations account for less than 1% of the total number of observations (Figs. 3c,g) only. Figures 3c,g show that the majority of the counts are located within the wind speed range of 5–10 m s−1. In Figs. 3d,h, the O − B histograms are slightly skewed to the left for both seasons which suggests ASCAT speeds are a little slower than the model background. Although not shown, the U and V departures for ASCAT are minimal except at the higher speeds where the samples size is small.
A comparison of the bias, standard deviation, wind speed histogram, and ASCAT wind speed difference histogram for ASCAT data. Results from (a)–(d) August 2008 and (e)–(h) January 2009.
Citation: Monthly Weather Review 139, 11; 10.1175/2011MWR3391.1
Figure 4 represents the comparison of the wind speed biases by taking the average of O − B and O − A per given ASCAT-observed wind speed bin (Figs. 4a,d), model wind speed bin (Figs. 4b,e), and the average of ASCAT and model wind speed bin (Figs. 4c,f). Figures 4a–c show the results from August 2008 and Figs. 4d–f show the results from January 2009. Figures 4a,d show that large negative biases are noticed for light wind and smaller biases associated with strong winds for August 2008 (Fig. 4a). For the other season, only minimal biases associated with strong winds are observed (Fig. 4d). Results from the model wind speed bin statistics are different. Large positive biases are observed with light winds and negative biases are observed with strong winds for both seasons (Figs. 4b,e). Since neither ASCAT winds nor model winds are error free, attempts have been made to bin the information on the basis of the average of ASCAT and model wind (Figs. 4c,e). As previously mentioned, based on the fact that removing ASCAT bias leads to increased penalty and poorer analysis, no bias correction is performed for the ASCAT winds.
A comparison of the wind speed biases by taking the average of O − B and O − A per given (a),(d) ASCAT-observed wind speed bin vs (b),(e) model wind speed bin and the average of (c) ASCAT and (f) model wind speed bin. Results from (a)–(c) August 2008 and (d)–(f) January 2009.
Citation: Monthly Weather Review 139, 11; 10.1175/2011MWR3391.1
Figure 5 displays the geographic distribution of bias for wind speed at 10-m height for ASCAT, WindSat, and QuikSCAT for O − B (left) and O − A (right) from August 2008. The January 2009 results are not shown here since they are very similar. For ASCAT, the largest speed biases with respect to the background are found in the tropical western Pacific, and in bands extending west from the Mexican coast to the central Pacific and to the east-southeast south of Central America (Fig. 5a). The overall speed bias for ASCAT is generally negative and within the range of −0.5 to 0.5 m s−1. For WindSat, the overall bias is generally positive and within the range of −0.5 to 1.2 m s−1, the largest biases were found over the Pacific and Indian Ocean south of 30°N (Fig. 5b). The overall speed bias for QuikSCAT is also positive and is around 0 to 1.0 m s−1 with the largest biases found in the tropical western Pacific.
A comparison of the geographic distribution of bias (for both O − B and O − A) for ASCAT, WindSat, and QuikSCAT from August 2008. Results from (a)–(c) O − B statistics and (d)–(f) O − A statistics.
Citation: Monthly Weather Review 139, 11; 10.1175/2011MWR3391.1
The biases are significantly reduced in most regions after assimilating the data (Figs. 5d–f). ASCAT continues to have a negative bias albeit reduced, while WindSat continues to have positive biases in the Antarctic region and negative biases in Northern Pacific (Fig. 5e). The QuikSCAT bias becomes mostly neutral after the assimilation (Fig. 5f).
Figure 6 shows the geographic distribution of wind speed RMS for ASCAT, WindSat, and QuikSCAT for both O − B and O − A during August 2008. Again, the January 2009 results are similar and are not shown. The wind speed RMS differences O − B distribution for ASCAT (Fig. 6a), WindSat (Fig. 6c), and for QuikSCAT (Fig. 6c) show similar patterns with the largest RMS differences noticed in tropical western Pacific, and in bands extending west from the Mexican coast to the central Pacific. After these surface winds are assimilated (Figs. 6d–f), the RMS differences are reduced in most of the regions. The magnitude of the wind speed RMS for O − A is around 1.0 m s−1 for all the surface winds vectors.
A comparison of the geographic distribution of RMS (for both O − B and O − A) for ASCAT and WindSat from August 2008. Results from (a),(d) ASCAT wind speed, (b),(e) WindSat wind speed, and (c),(f) QuikSCAT wind speed.
Citation: Monthly Weather Review 139, 11; 10.1175/2011MWR3391.1
b. Geographic distribution of ACs
Figure 7 is a bar chart of anomaly correlation scores for day 5 forecasts without ASCAT (control) and with ASCAT (ASCAT) data for 500- and 1000-hPa heights in the Northern and Southern Hemisphere during both seasons. During August 2008 (Fig. 7a), improvements in AC scores for 500- and 1000-hPa heights are noted in the Southern Hemisphere. The 1000- and 500-hPa AC scores for the day 5 forecast of geopotential heights increased from 0.80 to 0.815 and 0.825 to 0.841, respectively, by assimilated ASCAT winds. In the Northern Hemisphere, the AC scores for Control and ASCAT experiment are mostly neutral. For January 2009 (Fig. 7b), there are slight improvements for 500 and 1000 hPa in the Southern Hemisphere and neutral in the Northern Hemisphere.
ACCs for day 5 forecasts without ASCAT (Control) and with ASCAT (ASCAT) data for 500- and 1000-hPa heights in the Northern and Southern Hemisphere from August 2008 and January 2009.
Citation: Monthly Weather Review 139, 11; 10.1175/2011MWR3391.1
Here we introduce the geographic distribution of the AC scores in order to have better understanding of spatial representation of the skillfulness of a forecast relative to climatology beyond the commonly used latitudinal-averaged AC scores. The two heights chosen are 500 (Fig. 8) and 1000 hPa (Fig. 9) at day 5 for August 2008, which had the largest AC scores.
(a),(b) The geographic distribution of anomaly correlation of day 5 for 500-hPa geopotential height from the control experiment and ASCAT experiment from August 2008. (c) The difference between (a) and (b) using ASCAT minus Control. (d) The anomaly correlation for days 0–7 for 500-hPa geopotential height in the regions 20°–80° in the Southern Hemisphere. In (d), the results have been truncated to only show results for waves 1–20.
Citation: Monthly Weather Review 139, 11; 10.1175/2011MWR3391.1
(a),(b) The geographic distribution of anomaly correlation of day 5 for 1000-hPa geopotential height from the control experiment and ASCAT experiment from August 2008. (c) The difference between (a) and (b) using ASCAT minus control. (d) The anomaly correlation for days 0–7 for 1000-hPa geopotential height in the regions 20°–80° in the Southern Hemisphere. In (d), the results have been truncated to only show results for waves 1–20.
Citation: Monthly Weather Review 139, 11; 10.1175/2011MWR3391.1
Figures 8a,b show the global geographic distribution of 500-hPa geopotential height anomaly correlations for the Control and ASCAT experiment simulations during August 2008, respectively. The geographic distribution of the difference between the Control and ASCAT experiment anomaly correlation are presented in Fig. 8d. The geographic AC distribution for the Control and ASCAT experiments (Figs. 8a,b) have similar pattern: large AC (>0.8) are noticed over most of the continents and South Pacific, smallest AC (<0.4) are noticed off the west coast of the United States and near Antarctica. It is also noted that in Greenland, the AC score is quite large with values close to 0.9 for both the Control and ASCAT experiment. The anomaly correlation difference distribution for August has the largest negative differences over northern Canada and off the East Coast of the United States, the largest positive differences are located over Russia and the Antarctic (Fig. 8d). Figure 7c shows the anomaly correlation for days 0 to 7 for 500-hPa geopotential height in the regions 20°–80° in the Southern Hemisphere for August. The blue line is the Control simulation, which closely replicates NCEP operations and includes all data routinely used by the GDAS/GFS. The red line is the anomaly correlation diagnosed from the ASCAT simulation. The results indicate that the control simulation AC scores are close to the ASCAT experiment AC scores until day 3, then the ASCAT experiment shows greater forecast skill from days 4 through 7.
Global distribution of 1000-hPa geopotential height anomaly correlations is presented in Fig. 9. Heights determined to be underground were not used. Again, the geographic AC distribution for the Control and ASCAT experiments (Figs. 9a,b) shows a similar pattern with small AC (0.5–0.7) realized in the Arctic, North Atlantic, and part of the North Pacific. The anomaly correlation difference distribution for August (Fig. 9d) has the largest negative differences over northern Canada, off the East Coast of the United States, and a band off the West Coast of the United States extends to the central Pacific. The largest positive differences for Fig. 8d are scattered mainly in the west Pacific and Arctic. The die off curve (Fig. 9c) indicates that the control simulation AC scores are close to ASCAT experiment AC until day 3, then the ASCAT experiment shows greater forecast skill from days 4 through 6. By day 7, the forecast skills become similar. Comparing Figs. 8d and 9d, large negative differences are noted near the Arctic, northwest of the Atlantic Ocean for both 500 and 1000 hPa. In the Southern Hemisphere, large positive difference patterns are seen southeast of Australia for both 500 and 1000 hPa. There are also several isolated small positive difference patterns along the Indian Ocean. The tropics look noisy for both 500 and 1000 hPa as expected.
To understand how the improvements in AC are realized on a day-to-day basis, instead of calculating the time mean of the AC, mean daily normalized AC scores have been calculated. The fraction of daily AC is defined as (AC_of_day_i)/(Average_AC_over_all_days). This day by day improvement versus degraded forecast assessment provided insight into whether the factors governing the average AC score are dominated by greatly improved specific forecast busts or small improvement on average each day. Figure 10 shows the mean daily normalized geopotential height AC scores for the Control and the ASCAT experiment at 500 hPa for day 5 forecast in the (Fig. 10a) Northern and (Fig. 10b) Southern Hemisphere. The normalized daily AC is calculated using the daily AC score normalized by the mean AC scores over the full period of evaluation. Both plots indicate that in most of the cases, the governing factor of the AC score is dominated by a small improvement on average each day.
Mean daily normalized geopotential height AC scores from 7 Jul to 30 Aug 2008 over (a) Northern Hemisphere height and (b) Southern Hemisphere for Control and ASCAT experiments at 500 hPa for day 5 forecasts.
Citation: Monthly Weather Review 139, 11; 10.1175/2011MWR3391.1
c. Geographic distributions of FIs
The FI is an attempt to measure the difference in the error growth between the two model runs. The definition of FI is the difference between RMS of the control and RMS of the experiment, which is then normalized by the RMS of the experiment. The RMS is typically used to measure the error growth in a model. For some variables (moisture) the error growth is very rapid, while for others (i.e., temperature) the error growth is much slower. The error typically asymptotes to a constant value with time. The initial rapid decrease in FI suggests that the error growth in the control and experiment is rapidly increasing relative to the analysis. A positive FI suggests the error growth of the experiment is slower than the control. A negative FI suggests the error growth of the experiment is faster than the control. As the experiment and control error asymptote to the same value, the FI will eventually become zero.
Figure 11 displays geographic distributions of the sigma level 0.9950 temperature field of average FI, determined from using (2), for August 2008 at forecast hours (Fig. 11a) 6, (Fig. 11b) 12, (Fig. 11c) 24, and (Fig. 11d) 48. Sigma level 0.9950 is the lowest level in the model. The range of FI is from −60% to 60%. The 6-h results (Fig. 11a) show the largest FIs of the temperature field are in the central Pacific, Indian Ocean, and Atlantic Ocean. By 12 h (Fig. 11b), the FIs in the central Pacific were reduced, with the largest FI still realized over the Atlantic Ocean and Indian Ocean. The results from January 2009 are not shown because of the similarity.
Geographic distribution of sigma level 0.9950 temperature FI (%) at forecast hours (a) 6, (b) 12, (c) 24, and (d) 48 for August 2008. The range of forecast impact is from −60% to +60%.
Citation: Monthly Weather Review 139, 11; 10.1175/2011MWR3391.1
The FIs for 10-m U and V component of the winds during August 2008 are shown in Fig. 12. The greatest FIs are realized in southern Africa, the Central Pacific, and South America for both U and V as shown by the 6-h forecasts (Figs. 12a,b). By 24 h (Figs. 12c,d), the FIs are reduced with small positive impacts still seen in southern Africa and South America. The FI of the U and V components have similar patterns. Although not shown, by 48 h, the FIs are mostly neutral with some small positive impacts remaining. Figures 11 and 12 suggest that the ASCAT data have positive impacts on the temperature and wind fields through the 24-h forecast.
Geographic distribution of 10-m u and υ component FI (%) at forecast hours (a) 12, (b) 12, (c) 24, and (d) 24 for August 2008. The range of forecast impact is from −60% to +60%.
Citation: Monthly Weather Review 139, 11; 10.1175/2011MWR3391.1
Figure 9d shows no noticeable difference between the Control and ASCAT experiments through the first 2 forecast days, yet the FI statistics in Figs. 11–12 shows the largest values for the 6-h forecast, with values rapidly decreasing with time through the first 2 forecast days. This is because anomaly correlation plots measure the change in an anomaly with time and will always start near 1 and decrease slowly. The FI measures the difference in the error growth between two experiments. The difference is typically larger in the earlier part of a forecast and will decay to zero with time.
All the FIs shown in this paper have been tested as differences between the control and experiment. The differences in the temperature and wind fields shown here are statistically significant at the 99% level for forecasts out to 36 h.
ASCAT surface vector winds in the ECMWF analysis and forecast system showed a positive effect on forecast skill over the Southern Hemisphere (Hersbach and Janssen 2007b). Our results also indicated a positive effect on forecast skill in the Southern Hemisphere at 500 and 1000 hPa.
5. Summary
Observing system experiments were conducted during two seasons to quantify the impacts of assimilating the surface wind retrievals from the ASCAT microwave scatterometer on the MetOp-A satellite. A December 2007 version of the NCEP GDAS/GFS was used for the assimilation system and forecast model. These experiments were conducted at the NCEP operational resolution of the time (T382 with 64 layers) and used the NCEP operational observation database.
A thinning technique was used for ASCAT instead of the superobs used operationally for QuikSCAT and WindSat. Tests were run to determine whether 150-, 100-, or 50-km thinning boxes were more effective. It was found that overall the 100-km thinning box gave the best forecast results in terms of the anomaly correlation scores. The QC procedures specific to the assimilation system were also necessary and include testing the absolute value of the wind components and rejecting those greater than 5 m s−1, and removing observations suspected of still containing sea ice using 273-K SST criterion.
Two ambiguity QC scenarios were tested to determine which one produced the better forecast. The first scenario removed observations where the vector difference with the background of the originally selected vector is larger than its pair. The second scenario used the same criteria but replaced the originally selected vector with its pair. Less than 2% of the observations were identified as being suspect by this ambiguity check. Removing the suspect observations (first scenario) generally produced a better forecast.
Several verification techniques were used to measure the impact the ASCAT wind retrievals had on the forecast. These techniques included assimilation statistics, anomaly correlations, and geographical FI. Decreases in the bias and standard deviation for both u and υ components of the 10-m wind are realized after the ASCAT data were assimilated.
The results from the anomaly correlation calculations show neutral to modest improvements in forecast skill in midlatitudes in both seasons for most of the cases. The AC scores are similar to the control until the day 4 forecasts, when the improvements in the ASCAT experiment become greater for both seasons and in most cases. The results from geographical FI of this study show that assimilating the surface wind retrievals from the ASCAT satellite improve the NCEP GDAS/GFS wind and temperature forecasts. Positive FI occurred in the wind and temperature fields through the 48-h forecast range.
Acknowledgments
The authors wish to thank Stephen Lord and John Derber of NCEP for providing the GFS/GDAS, computer resources, and taps space; John Derber and Lars Peter Riishojgaard (NASA) for giving insightful advice; and Zorana Jelenak for providing the ASCAT retrieval data. Funding and computer resources were provided by the Joint Center for Satellite Data Assimilation (JCSDA). This research was supported under National Oceanic and Atmospheric Administration (NOAA) Grant NA07EC0676, which supports JCSDA activities.
REFERENCES
Alishouse, J. C., S. Snyder, J. Vongsathorn, and R. R. Ferraro, 1990: Determination of oceanic total precipitable water from the SSM/I. IEEE Trans. Geosci. Remote Sens., 28, 811–816.
Bettenhausen, M. H., C. K. Smith, R. M. Bevilacqua, N. Wang, P. W. Gaiser, and S. Cox, 2006: A non-linear optimization algorithm for Wind Sat wind vector retrievals. IEEE Trans. Geosci. Remote Sens., 44, 597–610.
Bi, L., J. Jung, M. Morgan, and J. Le Marshall, 2010: A two-season impact study of the WindSat surface wind retrievals in the NCEP Global Data Assimilation System. Wea. Forecasting, 25, 931–949.
Caplan, P., J. C. Derber, W. Gemmill, S. Hong, H.-L. Pan, and D. F. Parrish, 1997: Changes to the 1995 NCEP operational medium-range forecast model analysis-forecast system. Wea. Forecasting, 12, 581–594.
Derber, J. C., D. F. Parrish, and S. J. Lord, 1991: The new global operational analysis system at the National Meteorological Center. Wea. Forecasting, 6, 538–547.
Derber, J. C., P. Van Delst, X. Su, X. Li, K. Okamoto, and R. Treadon, 2003: Enhanced use of radiance data in the NCEP data assimilation system. Proc. 13th Int. TOVS Study Conf., Ste. Adele, Canada, 52–59.
Environmental Modeling Center, 2003: The GFS Atmospheric Model. NCEP Office Note 442, 14 pp. [Available online at http://www.emc.ncep.noaa.gov/officenotes/newernotes/on442.pdf.]
Figa-Saldana, J., J. W. Wilson, E. Attema, R. Gelsthorpe, M. R. Drinkwater, and A. Stoffelen, 2002: The Advanced Scatterometer (ASCAT) on the Meteorological Operational (MetOp) platform: A follow on for European wind scatterometers. Can. J. Remote Sens., 28 (3), 404–412.
Hersbach, H., and P. Janssen, 2007a: Preparation for assimilation of surface-wind data from ASCAT at ECMWF. ECMWF Research Department Memo., ECMWF, Reading, United Kingdom.
Hersbach, H., and P. Janssen, 2007b: Operational assimilation of surface wind data from the MetOp ASCAT scatterometer at ECMWF. ECMWF Newsletter, No. 113, ECMWF, Reading, United Kingdom, 6–8. [Available online at http://www.ecmwf.int/publications/newsletters/pdf/113.pdf.]
Kalnay, E., M. Kanamitsu, and W. E. Baker, 1990: Global Numerical Weather Prediction at the National Meteorological Center. Bull. Amer. Meteor. Soc., 71, 1410–1428.
Kanamitsu, M., 1989: Description of the NMC Global Data Assimilation and Forecast System. Wea. Forecasting, 4, 335–342.
Kanamitsu, M., and Coauthors, 1991: Recent changes implemented into the Global Forecast System at NMC. Wea. Forecasting, 6, 425–435.
Keyser, D., cited 2001a: Code table for PREPBUFR report types used by the Global GFS and GDAS GSI analyses. [Available online at http://www.emc.ncep.noaa.gov/mmb/data_processing/prepbufr.doc/table_2.htm.]
Keyser, D., cited 2001b: Summary of the current NCEP analysis system usage of data types that do not pass through PREPBUFR processing. [Available online at http://www.emc.ncep.noaa.gov/mmb/data_processing/prepbufr.doc/table_18.htm.]
Keyser, D., cited 2003: Observational data processing at NCEP. [Available online at http://www.emc.ncep.noaa.gov/mmb/data_processing/data_processing/.]
Kistler, R., and Coauthors, 2001: The NCEP–NCAR 50-Year Reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc., 82, 247–267.
Kleist, D., D. F. Parrish, J. C. Derber, R. Treadon, R. M. Errico and R. Yang, 2009a: Improving incremental balance in the GSI 3DVAR analysis system. Mon. Wea. Rev., 137, 1046–1060.
Kleist, D., D. F. Parrish, J. C. Derber, R. Treadon, W. Wu, and S. Lord, 2009b: Introduction of the GSI into the NCEP Global Data Assimilation System. Wea. Forecasting, 24, 1691–1705.
Lahoz, W. A., 1999: Predictive skill of the UKMO Unified Model in the lower stratosphere. Quart. J. Roy. Meteor. Soc., 125, 2205–2238.
Le Marshall, J., L. Bi, J. Jung, T. Zapotocny, and M. Morgan, 2006: WindSat polarimetric microwave observations improve Southern Hemisphere Numerical Weather Prediction. Aust. Meteor. Mag., 56, 35–40.
Menzel, W. P., F. C. Holt, T. J. Schmit, R. M. Aune, A. J. Schreiner, G. S. Wade, and D. G. Gray, 1998: Application of GOES-8/9 soundings to weather forecasting and nowcasting. Bull. Amer. Meteor. Soc., 79, 2059–2077.
NWS, cited 2006: Model performance statistics. [Available online at http://www.emc.ncep.noaa.gov/gmb/STATS/STATS.html.]
Stoffelen, A., S. de Haan, Y. Quilfen, and H. Schyberg, 2002: ERS scatterometer ambiguity removal scheme comparison. OSI SAF Rep., 28 pp. [Available online at http://www.knmi.nl/publications/fulltexts/safosi_w_arcomparison.pdf.]
Velden, C. S., C. M. Hayden, S. J. Nieman, W. P. Menzel, S. Wanzong, and J. S. Goerss, 1997: Upper-tropospheric winds derived from geostationary satellite water vapor observations. Bull. Amer. Meteor. Soc., 78, 173–195.
Vogelzang, J., A. Stoffelen, A. Verhoef, J. de Vries, and H. Bonekamp, 2009: Validation of two-dimensional variational ambiguity removal on Sea Winds scatterometer data. J. Atmos. Oceanic Technol., 26, 1229–1245.
Wu, W.-S., D. F. Parrish, and R. J. Purser, 2002: Three-dimensional variational analysis with spatially inhomogeneous covariances. Mon. Wea. Rev., 130, 2905–2916.
Yu, T.-W., and R. D. McPherson, 1984: Global data assimilation experiments with scatterometer winds from Seasat-A. Mon. Wea. Rev., 112, 368–376.
Zapotocny, T., W. P. Menzel, J. A. Jung, and J. P. Nelson III, 2005: A four-season impact study of rawinsonde, GOES, and POES data in the Eta data assimilation system. Part I: The total contribution. Wea. Forecasting, 20, 161–177.
Zapotocny, T., J. A. Jung, J. F. Le Marshall, and R. E. Treadon, 2007: A two-season impact study of satellite and in situ data in the NCEP Global Data Assimilation System. Wea. Forecasting, 22, 887–909.
Zapotocny, T.,J. A. Jung, J. F. Le Marshall, and R. E. Treadon, 2008: A two-season impact study of four satellite data types and rawinsonde data in the NCEP Global Data Assimilation System. Wea. Forecasting, 23, 80–100.
