• Keyser, D., cited. 2001a: Code table for PREPBUFR report types used by the ETA/3DVAR. [Available online at http://www.emc.ncep.noaa.gov/mmb/papers/keyser/prepbufr.doc/table_4.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/papers/keyser/prepbufr.doc/table_19.htm.].

  • Keyser, D., cited. 2003: Observational data processing at NCEP. [Available online at http://www.emc.ncep.noaa.gov/mmb/papers/keyser/data_processing/.].

  • McNally, A. P., cited. 2002: The inclusion of NOAA-15 HIRS-3 and AMSU-A radiances in the global data assimilation system. [Available online at http://wwwt.emc.ncep.noaa.gov/gmb/gdas/research/jhtml/noaa15.html.].

  • NWS, cited. 2002a: EDAS. [Available online at http://www.emc.ncep.noaa.gov/mmb/research/eta.log.html.].

  • NWS, cited. 2002b: EDAS. [Available online at http://www.emc.ncep.noaa.gov/mmb/mmbpll/eta12tpb/.].

  • NWS, cited. 2000c: EDAS. [Available online at http://www.emc.ncep.noaa.gov/mmb/gcp/etarefs.html.].

  • Parrish, D., Purser J. , Rogers E. , and Lin Y. , 1996: The regional 3D variational analysis for the Eta Model. Preprints, 11th Conf. on Numerical Weather Prediction, Norfolk, VA, Amer. Meteor. Soc., 454–455.

  • Rogers, E., Parrish D. , Lin Y. , and DiMego G. , 1996: The NCEP Eta data assimilation system: Tests with regional 3-D variational analysis and cycling. Preprints. 11th Conf. on Numerical Weather Prediction, Norfolk, VA, Amer. Meteor. Soc., 105–106.

    • Search Google Scholar
    • Export Citation
  • Rogers, E., and Coauthors, 1997: Changes to the NCEP operational “early” Eta analysis/forecast system. NOAA/NWS Technical Procedures Bulletin 447, 16 pp. [Available from Office of Meteorology, National Weather Service, 1325 East-West Highway, Silver Spring, MD 20910.].

  • Zapotocny, T. H., Menzel W. P. , Nelson J. P. III, and Jung J. A. , 2002: An impact study of five remotely sensed and five in situ data types in the Eta data assimilation system. Wea. Forecasting, 17 , 263285.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Zapotocny, T. H., Menzel W. P. , Jung J. A. , and Nelson J. P. 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 , 161177.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • View in gallery

    Geographical distributions of the four-season, time-averaged 00-h sensitivity (%) for 850-hPa relative humidity from all (a) raob, (b) GOES, and (c) POES observations. The contour interval is 2%

  • View in gallery

    Geographical distributions of the four-season, time-averaged 00-h sensitivity (%) for 850-hPa relative humidity from raob (a) mass and (b) wind observations. The contour interval is 2%

  • View in gallery

    Geographical distributions of the four-season, time-averaged 00-h sensitivity (%) for 850-hPa relative humidity from (a) GOES mass and (b) GOES wind observations. The contour interval is 2%

  • View in gallery

    Geographical distributions of the four-season, time-averaged 00-h sensitivity (%) for 850-hPa relative humidity from (a) HIRS, (b) AMSU, and (c) MSU observations. The contour interval is 2%

  • View in gallery

    Geographical distributions of the four-season, time-averaged 00-h sensitivity (%) for 850-hPa relative humidity from (a) SSM/I wind and (b) SSM/I column total precipitable water observations. The contour interval is 2%

  • View in gallery

    Vertical profiles of 00-h sensitivity from the aggregate raob (thin solid), GOES (thick solid), and POES (dashed) denials. The three fields displayed are (a), (d), (g) temperature (K); (b), (e), (h) relative humidity (%); and (c), (f), (i) u component (m s−1). (a)–(c) are near a rawinsonde location northwest of Lake Winnipeg. (d)–(f) are from a location in the northwestern Gulf of Mexico. (g)–(i) are from a point southwest of the Baja Peninsula

  • View in gallery

    Geographical distributions of the four-season, time-averaged 24-h forecast impact (%) for 300-hPa u component from aggregate (a) raob, (b) GOES, and (c) POES observations. The zero contour has been suppressed for clarity

  • View in gallery

    Geographical distributions of the four-season, time-averaged 24-h forecast impact (%) for 300-hPa u component from (a) raob mass and (b) raob wind observations. The zero contour has been suppressed for clarity

  • View in gallery

    Geographical distributions of the four-season, time-averaged 24-h forecast impact (%) for 300-hPa u component from (a) GOES mass and (b) GOES wind observations. The zero contour has been suppressed for clarity

  • View in gallery

    Geographical distributions of the four-season, time-averaged 24-h forecast impact (%) for 300-hPa u component from (a) HIRS, (b) AMSU, and (c) MSU observations. The zero contour has been suppressed for clarity

  • View in gallery

    Geographical distributions of the four-season, time-averaged 24-h forecast impact (%) for 300-hPa u component from (a) SSM/I wind and (b) SSM/I column total precipitable water observations. The zero contour has been suppressed for clarity

  • View in gallery

    Scatter distributions of (a)–(c) 100-hPa temperature and (d)–(f) 850-hPa relative humidity from a 24-h forecast starting 1200 UTC 7 Nov 2001. The aggregate raob denial results are shown in (a) and (d). The aggregate GOES denial results are shown in (b) and (e). The aggregate POES denial results are shown in (c) and (f). In (a)–(f), the difference between the control simulation and the corresponding analysis is shown on the x axis, while the difference between the denied simulation and analysis is shown on the y axis

  • View in gallery

    The four-season summary of rms forecast impact (%) for mean sea level pressure after 24 h of Eta Model integration. Both the three aggregate denials (raob, GOES, and POES) and the nine individual denials are shown. The results are for the (a) 104 and (b) 212 grids

  • View in gallery

    The four-season summary of rms forecast impact (%) on the 104 grid for (a) temperature, (b) u component, and (c) relative humidity, after 24 h of Eta Model integration. Both the three aggregate denials (raob, GOES, and POES) and the nine individual denials are shown

  • View in gallery

    Box-and-whisker displays of the forecast impact at 24 h for (a) temperature, (b) u component, and (c) relative humidity from the raob mass, raob wind, GOES mass, and GOES wind denials. The maximum and minimum values of the data make up the “whiskers” of the display, while the “box” comprises the median and upper and lower 25th and 90th percentiles, respectively. Note that in some instances, the vertical zero line obscures the median or percentiles within a given box. Furthermore, maximum and minimum values exceeding the scale of the x axis are enumerated on the “tails” of the whiskers. Values in this display are for the entire horizontal EDAS domain and are calculated from the four seasonal time periods used in this work

  • View in gallery

    Box-and-whisker displays of the forecast impact at 24 h for (a) temperature, (b) u component, and (c) relative humidity from the AMSU-A and -B, HIRS, MSU, and SSM/I PW denials. The maximum and minimum values of the data make up the “whiskers” of the display, while the “box” comprises the median and upper and lower 25th and 90th percentiles, respectively. Note that in some instances, the vertical zero line obscures the median or percentiles within a given box. Furthermore, maximum and minimum values exceeding the scale of the x axis are enumerated on the “tails” of the whiskers. Values in this display are for the entire horizontal EDAS domain and are calculated from the four seasonal time periods used in this work

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 57 26 2
PDF Downloads 22 12 0

A Four-Season Impact Study of Rawinsonde, GOES, and POES Data in the Eta Data Assimilation System. Part II: Contribution of the Components

View More View Less
  • 1 Cooperative Institute for Meteorological Satellite Studies, Madison, Wisconsin
  • | 2 Cooperative Institute for Meteorological Satellite Studies, and National Environmental Satellite, Data, and Information Service, Madison, Wisconsin
  • | 3 Cooperative Institute for Meteorological Satellite Studies, Madison, Wisconsin
Full access

Abstract

The impact of in situ rawinsonde observations (raob), remotely sensed Geostationary Operational Environmental Satellite (GOES), and Polar-Orbiting Operational Environmental Satellite (POES) observations routinely used in NCEP’s Eta Data Assimilation/Forecast System (EDAS) is studied for extended-length time periods during four seasons. This work examines the contribution of nine individual components of the total observing system. The nine data types examined include rawinsonde mass and wind observations, GOES mass and wind observations, POES observations from the Microwave Sounding Unit (MSU), the Advanced Microwave Sounding Unit (AMSU-A and AMSU-B), the High Resolution Infrared Radiation Sounder (HIRS), and column total precipitable water and low-level wind observations from the Special Sensor Microwave Imager (SSM/I). The results are relevant for users of the Eta Model trying to compare/contrast the overall forecast impact of traditional, largely land-based rawinsonde observations against remotely sensed satellite observations, which are available domainwide.

The case studies chosen consist of 15-day periods during fall 2001, winter 2001/02, spring 2002, and summer 2002. Throughout these periods, a November 2001 32-km version of the EDAS is run 10 times at both 0000 and 1200 UTC. The 10 runs include a control run, which utilizes all data types routinely used in the EDAS, and 9 experimental runs in which one of the component data types noted above is denied. Differences between the experimental and control runs are then accumulated over the 15-day periods and analyzed to demonstrate the 00-h sensitivity and 24-h forecast impact of these individual data types in the EDAS. The diagnostics are computed over the entire horizontal model domain and a subsection covering the continental United States (CONUS) and adjacent coastal waters on isobaric surfaces extending into the lower stratosphere.

The 24-h forecast impact results show that a positive forecast impact is achieved from most of the nine component data sources during all four time periods. HIRS, MSU, and SSM/I wind observations yield only a slight positive forecast impact to all fields. Rawinsonde and GOES wind observations have the largest positive forecast impact for temperature over both the entire model domain and the extended CONUS. The same data types also provide the largest forecast impact to the u component of the wind, while GOES wind observations provide the largest forecast impact to moisture.

Corresponding author address: Tom H. Zapotocny, CIMSS/SSEC, University of Wisconsin—Madison, 1225 West Dayton St., Madison, WI 53706-1265. Email: tomz@ssec.wisc.edu

Abstract

The impact of in situ rawinsonde observations (raob), remotely sensed Geostationary Operational Environmental Satellite (GOES), and Polar-Orbiting Operational Environmental Satellite (POES) observations routinely used in NCEP’s Eta Data Assimilation/Forecast System (EDAS) is studied for extended-length time periods during four seasons. This work examines the contribution of nine individual components of the total observing system. The nine data types examined include rawinsonde mass and wind observations, GOES mass and wind observations, POES observations from the Microwave Sounding Unit (MSU), the Advanced Microwave Sounding Unit (AMSU-A and AMSU-B), the High Resolution Infrared Radiation Sounder (HIRS), and column total precipitable water and low-level wind observations from the Special Sensor Microwave Imager (SSM/I). The results are relevant for users of the Eta Model trying to compare/contrast the overall forecast impact of traditional, largely land-based rawinsonde observations against remotely sensed satellite observations, which are available domainwide.

The case studies chosen consist of 15-day periods during fall 2001, winter 2001/02, spring 2002, and summer 2002. Throughout these periods, a November 2001 32-km version of the EDAS is run 10 times at both 0000 and 1200 UTC. The 10 runs include a control run, which utilizes all data types routinely used in the EDAS, and 9 experimental runs in which one of the component data types noted above is denied. Differences between the experimental and control runs are then accumulated over the 15-day periods and analyzed to demonstrate the 00-h sensitivity and 24-h forecast impact of these individual data types in the EDAS. The diagnostics are computed over the entire horizontal model domain and a subsection covering the continental United States (CONUS) and adjacent coastal waters on isobaric surfaces extending into the lower stratosphere.

The 24-h forecast impact results show that a positive forecast impact is achieved from most of the nine component data sources during all four time periods. HIRS, MSU, and SSM/I wind observations yield only a slight positive forecast impact to all fields. Rawinsonde and GOES wind observations have the largest positive forecast impact for temperature over both the entire model domain and the extended CONUS. The same data types also provide the largest forecast impact to the u component of the wind, while GOES wind observations provide the largest forecast impact to moisture.

Corresponding author address: Tom H. Zapotocny, CIMSS/SSEC, University of Wisconsin—Madison, 1225 West Dayton St., Madison, WI 53706-1265. Email: tomz@ssec.wisc.edu

1. Introduction

The role of data collected from the Geostationary Operational Environmental Satellite (GOES) and the Polar-Orbiting Operational Environmental Satellite (POES) in the atmospheric sciences is to provide a wealth of nearly continuous meteorological observations. These remotely sensed observations must be of good quality and available to users in regions where in situ observations have limited temporal and/or spatial coverage. The National Environmental Satellite, Data, and Information Service (NESDIS) obtains such remotely sensed data hourly from both the GOES and POES platforms. Both of these data streams are made available to users around the world and to the modeling branch within the National Centers for Environmental Prediction (NCEP), which then uses them operationally in the Eta Data Assimilation/Forecast System (EDAS) and other numerical weather prediction models.

The primary goal of research reported herein is to examine the impact this remotely sensed data has on forecast quality in the EDAS. This work is a continuation of Zapotocny et al. (2005), which examined the aggregate impact of GOES and POES data in the EDAS. In that work (hereafter Part I), the total impact of these remotely sensed data were compared and contrasted against the impact of conventional rawinsonde observations for extended-length time periods during each season. This work uses the same 15-day periods, but expands the previous study by examining the impact of the nine individual data types producing the aggregate impacts of Part I.

The nine component data types examined are 1) rawinsonde mass observations, 2) rawinsonde wind observations (raob), 3) mass, and 4) wind observations from the GOES satellites positioned on the eastern and western coasts of North America (GOES-8 and GOES-10 during the time frame of this study, respectively), POES observations from the 5) High Resolution Infrared Radiation Sounder (HIRS), 6) the Microwave Sounding Unit (MSU), and 7) the Advanced Microwave Sounding Unit (AMSU-A and AMSU-B) from the three National Oceanic and Atmospheric Administration (NOAA) POES satellites operational at the time of this study (NOAA-14, -15, and -16), as well as 8) the Defense Meteorological Satellite Program (DMSP) Special Sensor Microwave Imager (SSM/I) low-level wind, and 9) column total precipitable water observations.

Examining this particular set of data types provides a means whereby both mass and wind impacts can be compared and contrasted from rawinsonde and GOES data sources. A breakdown of GOES observations from both the imager and sounder is also possible, as is an evaluation of impacts from most of the radiance information available from the GOES and POES platforms. The two SSM/I data types are included primarily for completeness, since they also are obtained from polar-orbiting satellites. Limited computer resources prohibit experiments involving a more in-depth partitioning of the remotely sensed data types and the inclusion of additional in situ denials.

The paper is structured as follows. Section 2 elaborates on the experimental design and provides details of the data types studied. Section 3 presents a comparison of the 00-h sensitivities and 24-h forecast impacts by showing time-averaged results, geographical and vertical distributions, and scatter distributions for a substantial cross section of the overall data. The results are discussed further in section 4 and summarized in section 5.

2. Experimental design

The rawinsonde denials examined in this study include the removal of all (aggregate) rawinsonde data, as well as the separate removal of both the mass and wind components. The GOES data types examined include the aggregate removal of both mass and wind observations, as well as the separate removal of GOES marine infrared cloud drift winds, marine cloud-top water vapor winds, and the infrared cloud picture triplet winds. The latter three GOES observations are from the imager and combine to yield a data type referred to as GOES wind (GOESW) observations. The GOES land clear-air three-layer precipitable water and GOES marine clear-air radiances compose the GOES mass (GOESM) removal and are from the sounder. Keyser (2001a, 2003) describes all of the above data types in greater detail.

The partitioning of POES denials from NOAA satellites in this study includes the aggregate POES removal, as well as the separate removal of satellite radiances from several different instruments. They are the HIRS data from NOAA-14, -15, and -16, the MSU data from NOAA-14 and -15, and the AMSU-A and AMSU-B data from NOAA-15 and -16. These datasets are described in greater detail by Keyser (2001b). NOAA-17 information was not operational during the time frame of these experiments.

Two DMSP POES data types are also investigated in this work. They include marine low-level superobed wind observations and column total precipitable water observations in a marine environment, from DMSP-13 and -15. As described in Part I, other remotely sensed data not produced by the GOES, NOAA POES, or DMSP POES platforms remained in the experiments of this study. These data types include, but are not limited to, the Meteosat and Geostationary Meteorological Satellite (GMS) products. The Meteosat and GMS products, while important in a global model, are confined to the extreme eastern and western portions of the EDAS model domain, respectively. As such, it is felt that their importance would be minimal in a study of this type.

Table 1 presents a summary of where each data type is used in this version of the EDAS. As seen, except for GOES three-layer precipitable water data, all remotely sensed data are only used within marine environments. Note that rawinsonde observations are listed as being used in both land and marine environments. However, rawinsonde data availability within marine environments is in general very sparse.

During the 2001/02 time frame of these experiments, the operational EDAS was executing at either 22 km/50 layers or 12 km/60 layers (NWS 2002a, 2002b,c). However, initial model conditions in the experiments reported here were obtained from a 32 km/60 layer EDAS, which was also being executed at NCEP twice-daily “parallel” to the operational EDAS. All EDAS runs performed in this work were also undertaken at the 32-km/60-layer resolution. Computational limitations preclude impact studies of this magnitude at the operational resolutions. Section 2 of Part I describes in greater detail how the data were postprocessed and plotted.

The absolute difference (S, hereafter also called sensitivity) of the EDAS assimilation to the denied data type is only evaluated at 00 h, meaning after a complete 12-h assimilation cycle. The root-mean-square (rms) sensitivity S is defined as
i1520-0434-20-2-178-e1
where C is the control assimilation containing all data types, D is the assimilation containing all but the denied data type, and N is the total number of grid points on the isobaric surface being evaluated. Since (1) only contains two analyses and does not contain an independent verification, the sensitivity diagnosed by (1) does not indicate whether the 00-h analysis is better or worse with the data type removed. As such the sensitivity statistics are only presented at 00 h.
Similar to Part I, forecast impact diagnostics are presented as the positive/negative impact provided by a particular data type. The rms forecast impact (FI) of an individual data type is evaluated as
i1520-0434-20-2-178-e2
In (2), N is the same as in (1). The variables C and D are the 24-h control and denied forecasts, respectively, and A is the 00-h EDAS control analysis containing all data types valid 24 h after the forecast began. The first of the two right-hand-side terms enclosed by parentheses can be considered the error in the denied forecast. The second term enclosed by the parentheses can be considered the error in the control forecast. Dividing the difference by the error in the control forecast and multiplying by 100 normalizes the results and provides a percent improvement with respect to the rms error of the control forecast. A positive forecast impact means the forecast compares more favorably to the corresponding analysis with the data type included than with it denied. (While only one denial per experiment is used in the discussion of (1) and (2), note that the aggregate studies used in this work and Part I contain as many as five denials for a single experiment.)

All time-averaged 00-h sensitivities and 24-h forecast impacts exclude the first day of each seasonal time period. This delay in evaluating the statistics allows more time for the impact of the denied data to be removed from the first guess and reduces the 16-day seasonal windows to 15 days diagnostically. Finally, it is important to note that all four 15-day periods used in this study were run with the EDAS operating in “full cycling” mode. Full cycling involves denying the particular data type(s) for consecutive model runs and carrying the results of each 12-h assimilation cycle forward to the starting point for the next 12-h assimilation.

In these experiments, data were assimilated via a three-dimensional variational data assimilation (3DVAR) technique at T−12, T−9, T−6, T−3, and T−0 using 3-h Eta Model forecasts as the background between each assimilation step (Parrish et al. 1996; Rogers et al. 1996, 1997). The result of the 12-h assimilation cycle provides a complete analysis, which is used to begin the Eta Model’s free forecast. Differences between the various experimental analyses (forecasts) and the control analysis (forecast) provide a measure of the sensitivity (forecast impact) of each data type in the EDAS. [It is noteworthy that all fields presented in Figs. 15 and 711 have been smoothed with two passes of a (1,4,1) symmetric smoother. The values within two grid points of the lateral model edges have also not been plotted. Moreover, underground grid points are not used in the evaluations.]

3. Results

a. The 00-h sensitivities

Figure 1 presents the 850-hPa 00-h sensitivity for relative humidity from the (a) aggregate raob, (b) GOES, and (c) POES data. These results are the four-season average, composing a total of 120 individual model runs. Relative humidity is selected as the variable diagnosed in this subsection for the following reasons. First, its values are bounded between 0% and 100% everywhere, making assessment of its change easier for the reader to identify. Second, by virtue of the very tight vertical and horizontal gradients that develop, moisture-related fields have historically been the most difficult to forecast. This remains true even in modern-day, high-resolution models.

From Fig. 1a, the impact of individual rawinsonde observations is clearly identifiable, with sensitivities ranging from approximately 4% in oceanic regions far removed from station locations to greater than 40% in close proximity to the Canadian rawinsonde stations. Large sensitivities are also diagnosed near the Hawaiian Islands and a few of the Caribbean islands. Overall, these 32-km 00-h rawinsonde sensitivity results are quite similar to the 80-km 00-h sensitivity results of Zapotocny et al. (2002); see their Fig. 8).

Figures 1b and 1c present the aggregate four-season 00-h sensitivities of GOES and POES data, respectively. The largest sensitivities of 850-hPa relative humidity to remotely sensed observations are almost always found in maritime environments. This is expected, since for the most part this is where the data are assimilated in the EDAS. In fact, the only remotely sensed moisture observations used over land in the EDAS are GOES three-layer precipitable water observations. The GOES sensitivity is largest over the Gulf of Mexico, western Atlantic Ocean, and eastern Pacific Ocean, where sensitivities reach upward of 25%. On the other hand, POES data have their largest 00-h sensitivities at low latitudes over the eastern Pacific Ocean. In this region, the four-season aggregate average sensitivity reaches 41.4%.

Much larger single-season relative humidity sensitivities are realized for both GOES and POES data during the Northern Hemisphere summer and fall seasons, with sensitivities approaching 60% during these two seasons (not shown). By contrast, large seasonal changes in sensitivity are not realized for rawinsonde data; they display more uniform season-to-season 00-h relative humidity sensitivities (also not shown).

Figures 25 present the 00-h sensitivity for the nine components composing the three aggregate data types displayed in Fig. 1. Rawinsonde mass (labeled RAOBM on the plots) and rawinsonde wind (RAOBW) sensitivities are shown in Figs. 2a and 2b, respectively. GOES mass (GOESM) and GOES wind (GOESW) sensitivities are shown and Figs. 3a and 3b, respectively. Sensitivity from HIRS, the combination of AMSU-A and AMSU–B, and MSU data are shown in Figs. 4a–c, respectively. Finally, Figs. 5a and 5b show the 00-h sensitivity from SSM/I low-level wind and column total precipitable water, respectively.

From Fig. 2, it is apparent that much of the 850-hPa 00-h relative humidity sensitivity over land is due to raob mass observations. Values from this component of the rawinsonde data range from just slightly greater than 0% to 36.2%, with individual rawinsonde locations clearly distinguishable in the North American network as well as the Hawaiian Islands. The rawinsonde wind contribution (Fig. 2b) is much smaller and largely confined to lower-latitude oceanic regions. A secondary sensitivity maximum from rawinsonde winds, with values as large or larger than the low-latitude oceanic maxima, is diagnosed in the Arctic Ocean north of Alaska. Note that all of the largest 00-h sensitivities to 850-hPa moisture from rawinsonde winds are far removed from their continental data source. Contrary to moisture, an inspection of the 00-h 850-hPa temperature and wind sensitivities from raob wind observations reveals that the largest contribution from this data type to these fields is located over North America.

Unlike the two raob components presented in Fig. 2, which showed large differences between the two data sources, the two components of GOES data (Fig. 3) demonstrate nearly equal contributions to the 00-h 850-hPa relative humidity sensitivity. The four-season average sensitivity from GOES mass observations reaches approximately 21% in the western Atlantic basin, while the GOES wind sensitivities reach approximately 32% in the eastern Pacific basin. Values are much smaller over the North American continent, where the only GOES information assimilated into the EDAS is the three-layer precipitable water data. Only a small contribution from this data type is expected, since it historically receives a relatively small weight in the 3DVAR analysis and is located over data-rich North America.

The HIRS, combined AMSU, and MSU 00-h 850-hPa relative humidity sensitivities are shown in Figs. 4a–c, respectively. The largest overall contribution from each of these datasets is found in oceanic regions, close to the observations. Four-season time-averaged sensitivities range from approximately 9% for HIRS and MSU data to 16% for AMSU data. The largest HIRS and MSU sensitivities are positioned northeast of the Bahamas, while the largest AMSU sensitivity is found in middle and high latitudes.

The final two data types making up the POES aggregate denial are SSM/I low-level wind and column total precipitable water. The four-season average 850-hPa 00-h sensitivity from these data types is shown in Fig. 5. Both these data types produce their largest sensitivity in oceanic regions, with SSM/I wind maxima diagnosed in middle latitudes and precipitable water maxima positioned in lower latitudes southwest of the Baja Peninsula. A final point of interest that is common to most panels of Figs. 25 is an increase of rawinsonde sensitivity and a decrease of remotely sensed sensitivity near the Hawaiian Islands. This result is consistent with the land/ocean and in situ/remotely sensed partitioning of data used in the EDAS.

Figure 6 presents vertical profiles of the 00-h sensitivity of temperature, relative humidity, and u component every 25 hPa between 1000 and 100 hPa. Figures 6a–c are from a grid point located just northwest of Lake Winnipeg, Figs. 6d–f are located at a grid point in the west-central Gulf of Mexico, and Figs. 6g–i are located at a grid point southwest of the Baja Peninsula. These locations are in close proximity to a local maximum from raob, GOES, and POES data, respectively (see Fig. 1).

Several interesting features are noticed in this multipanel comparison. First, there is a distinct lower-tropospheric sensitivity maximum from rawinsonde data (thin solid line) to both temperature and relative humidity in the profile near Lake Winnipeg (Figs. 6a and 6b). Both of these maxima are located between 800 and 900 hPa and exceed 5 K for temperature and approximately 35% for relative humidity. The second feature is the large high-altitude temperature sensitivity observed in Figs. 6a, 6d and 6g for all three locations. Only the GOES sensitivity near Lake Winnipeg (Fig. 6a) and the raob sensitivity southwest of the Baja Peninsula (Fig. 6g) fail to show a marked increase in sensitivity above 200 hPa. (These two locations are in regions far removed from where raob and GOES data are typically available to the EDAS.) The third point of interest is the large sensitivity noticed from GOES data to the u component southwest of the Baja Peninsula (Fig. 6i). Although not shown, the 200-hPa sensitivity from GOES is a relatively large-scale feature, with values of 10 m s−1 or greater covering most of the eastern Pacific Ocean south of 50°N latitude. As such, GOES data are making significant adjustments to both the polar and subtropical jet streams in this oceanic region.

b. The 24-h forecast impacts

In this subsection, the u component of the wind near jet stream level is presented rather than 850-hPa relative humidity in an effort to present as many fields as possible in the diagnostics. Figures 711 present the 24-h forecast impact on the 300-hPa u component, time averaged over the 120 runs of the four seasons. Figure 7 displays the aggregate contribution from all components making up the rawinsonde, GOES, and POES denials, respectively. Figures 811 display the forecast impact contributions from the same nine data components as shown in Figs. 25. However, unlike section 3a, which showed 850-hPa 00-h relative humidity sensitivities, this section shows 300-hPa 24-h u-component forecast impacts. The primary purpose of switching from relative humidity to u component is to provide a larger selection of geographical distributions. For completeness, the contribution from all nine individual denials plus the three aggregate denials on all seven isobaric levels are presented below.

Figure 7 clearly shows two preferred regions for 24-h u-component positive forecast impact. The rawinsonde and POES impacts are maximized at high latitudes, while the GOES impacts are maximized at lower and middle latitudes. The largest overall 24-h forecast impact is from raob data, with a greater than 100% impact on the western shore of Greenland. Numerous other regions also show a forecast impact greater than 50% over northern reaches of the EDAS domain. GOES data display the second largest positive forecast impact, with one localized region southwest of the Baja Peninsula exceeding 80%. Other positive forecast impacts from GOES are widespread, but in general lie south of 35°N.

The largest POES contribution to 300-hPa u-component forecast impact (Fig. 7c) is over the Canadian Archipelago and Arctic Ocean, with much smaller impacts realized elsewhere. Two factors are noteworthy about this high-latitude POES contribution. First, the primary source of data in the Arctic Ocean is POES mass observations. There are SSM/I low-level winds, but they are only found near the surface. Second, the enhanced POES impact at high latitudes could be attributed to the increased frequency of polar-orbiter overpasses, compared to lower latitudes.

It is interesting to note that some of the smaller forecast impacts to 24-h 300-hPa u component are found over CONUS from all three aggregate data types (Figs. 7a–c). Specifically, several small regions of negative forecast impact are noticed over CONUS from all three aggregate data types. These geographic distributions display how some of the rawinsonde impact has been advected to the north and east of the CONUS, even in the relatively short time frame of 24 h. The distributions also show that the large sensitivities from POES data initially over the Pacific Ocean (not shown) have not yet propagated eastward into the interior of the CONUS by 24 h.

Figures 811 present the individual component forecast impacts from raob, GOES, and POES data, respectively. Figure 8 illustrates that the mass (Fig. 8a) and wind (Fig. 8b) components of rawinsondes are of similar importance. This figure also indicates that the sum of the component forecast impacts is not required to equal the total forecast impact. For example, the aggregate raob 24-h forecast impact is slightly greater than 100% on the western shore of Greenland (Fig. 7a), while in the same location both component forecast impacts are less than 30% (Figs. 8a and 8b).

Unlike the raob component contributions, which display comparable forecast impacts, the GOES (Fig. 9) components are more one sided. The GOESW component generates nearly as large a forecast impact as the total GOES contribution, while the GOESM contribution is much smaller. Finally, all five POES denial experiments displayed in Figs. 10 and 11 show small contributions, with the largest forecast impact being just slightly greater than 20% in the Arctic Ocean from AMSU data (Fig. 10b).

c. Scatter distributions

A problem identified in Part I and Figs. 6a, 6d and 6g herein is the large sensitivity and forecast impacts noticed from several of the data types in the lower stratosphere. Recall from Part I that NCEP personnel are aware of this problem and have planned to raise the top of the EDAS to 2 hPa in a future release (G. DiMego 2003, personal communication). As such, this subsection only briefly examines this high-altitude “spike” by presenting scatter distributions of the difference between the control forecast and the analysis and the denied forecast and the analysis. For clarity, concentration is directed at only one time period—the last time period of the fall 2001 season (1200 UTC 7 November 2001). The results are shown in Fig. 12. In all six panels of this figure, the difference of the control simulation and the corresponding analysis is shown on the x axis, while the difference of the denied forecast and analysis is shown on the y axis. On each of the six scatterplots, there is a one-to-one correspondence of dots to grid points on the isobaric surface presented.

Figures 12a–c present the aggregate raob, GOES, and POES temperature scatter distributions at 100 hPa for the selected time period. These three aggregate denials show that most of the scatter is found in the upper-right-hand quadrant of the plot, with only a few points extending down the diagonal axis farther than −2 for any of the three aggregate denials. On the other hand, values extend well past the +3 value for all distributions. There is also a tendency for values to be asymmetrically scattered about the diagonal line, with raob values in general above the line by 1–2 K and POES values in general below the line by approximately the same amount. For the most part, GOES data are more symmetrically scattered about the diagonal than the other two data types. These results indicate that rawinsonde data are in general cooling the EDAS atmosphere at 100 hPa for this time period, whereas the POES data are warming the EDAS atmosphere.

While the scatter results of Figs. 12a–c show asymmetry for 100-hPa temperature, the same is not true for 850-hPa relative humidity (Figs. 12d–f). Although the relative humidity scatterplots show large differences between the control and denied experiments, the scatter appears more random about the diagonal line than the temperature results just discussed.

d. Forecast impact of all aggregate and component denials

This subsection presents bar chart results similar to Part I for the four-season averages. Fields shown include mean sea level pressure (Fig. 13) on both the 104 and 212 grid, and three conventional meteorological upper-air parameters on seven pressure levels spanning the troposphere (Fig. 14). The three upper-air parameters, which are only shown on the 104 grid, are temperature, u component of the wind, and relative humidity. Figures 13 and 14 both show the aggregate raob, GOES, and POES contributions, as well as the contribution from all nine individual components. In Fig. 13, the raob, GOES, and POES values equal the average of the four seasonal values shown in Table 3 of Part I. The raob, GOES, and POES values of Fig. 14 are reproduced from Fig. 5 of Part I and are presented here for easier comparison.

From Fig. 13, it is clear that the largest 24-h forecast impact to mean sea level pressure is from the aggregate raob data type. This is true on both the 104 and 212 grids. RAOBW is approximately 3 times larger than its mass counterpart (RAOBM). A similar result is obtained for the GOES data type, with the wind (GOESW) forecast impact being much larger than its mass counterpart (GOESM). The aggregate POES sea level pressure forecast impact is larger than the GOES forecast impact over the 104 grid, but smaller than the GOES contribution over the 212 grid. AMSU data are the largest contributors of the five POES components, with MSU contributions very small positive, and HIRS contributions very small negative.On the whole, the 212 grid results (Fig. 13b) nearly shadow the 104 grid results (Fig. 13a) for this particular field, set of denials, and accumulation of time periods.

A comparison of the upper-air results in Fig. 14 with the sea level pressure results of Fig. 13 indicates that many of the data types display similar characteristics between the two figures. For example, GOESM, HIRS, and MSU impacts are negligible for all fields and levels displayed. AMSU data are almost always the largest contributor to the entire POES aggregate. GOESW is very similar in size to the GOES aggregate, and RAOBW is larger than RAOBM. These statements are further substantiated by the seven-level average displayed near the top of each data box in Fig. 14. Finally, of all values presented in Figs. 13 and 14, only HIRS and SSM/I wind (SSM/IW) data provide any degradation to the 24-h forecast over the four-season average. Sea level pressure on both grids in Fig. 13 and the 200-hPa HIRS and SSM/IW temperature and u component in Figs. 14a and 14b are just slightly negative. The four-season summary of all 12 experiments on the 212 grid is not shown, since those results are quite similar to the 104 grid results just presented.

Similar to Part I, the bar charts just presented do not give a true appreciation of whether or not there is a significant difference between two values in Fig. 14. The following two box-and-whisker charts, plotted in identical fashion to Fig. 6 of Part I, attempt to address this concern for eight of the nine component distributions diagnosed in this study. Figure 15 portrays rawinsonde mass and wind as well as GOES mass and wind box-and-whisker charts for the four-season average over the 104 grid, while AMSU, HIRS, MSU, and SSM/I precipitable water (SSM/I PW) box-and-whisker diagrams are shown in Fig. 16. The SSM/I low-level wind box-and-whisker charts are not displayed in this paper.

Similar to Part I, the box-and-whisker displays presented in Figs. 15 and 16 show the maximum and minimum values of the field being diagnosed, with the value enumerated at the end of the tail if it exceeds the scale of the horizontal axis. The median and upper and lower 25th and 90th percentiles are also shown. A vertical zero line in each panel provides a reference for quick identification of whether the median is greater than or less than zero. Finally, note that the scale of these figures is identical to Fig. 6 of Part I, with values ranging from −60 to 100.

In agreement with the aggregate box-and-whisker displays of Part I, the tail associated with the maximum whisker is nearly always longer than the tail associated with the minimum whisker. The most notable exception in Fig. 15 is the 200-hPa rawinsonde mass contribution to the u component of the wind. In general, the median (centerline of the rectangle) is fairly close to zero for the vast majority of the fields shown in Fig. 15. GOES wind observations, which are shown as the bottom entry in each column of Fig. 15, are the largest exception to this statement. For this data type, nearly all levels for all three fields displayed (temperature, u component, and relative humidity) have the center of the rectangle to the right of the zero line.

No other data type displays a signal as strong as the GOES wind observations for the four POES data types shown in Fig. 16. Most of these data types display a rectangle whose bounds do not extend far from the zero line. In fact, data types such as HIRS and MSU have most of the rectangles located very close to the zero line.

Table 2 presents the lower and upper 95% confidence intervals for forecast impact from eight of the nine component data types presented in this work for temperature, u component, and relative humidity (SSM/I wind is not shown). For brevity, results in this table are truncated to only include three isobaric levels (100, 500, and 1000 hPa) rather than the eight levels shown elsewhere in this paper. Similar to Table 5 of Part I, most of the 95% confidence limits do not vary far from the median. The most notable exceptions are AMSU 100-hPa temperature, GOES wind relative humidity at 1000 hPa, 100- and 1000-hPa raob wind relative humidity, and raob wind 1000-hPa temperature, where there is greater than a 4% difference between the lower and upper confidence limits. Finally, a few negative forecast impacts are evident in this table, which are not diagnosed in Table 5 of Part I, which shows the total contribution.

4. Discussion

The results above raise several points for further discussion. First, has the nearly neutral forecast impact from HIRS been diagnosed by others? Second, why are the raob and GOES forecast impacts from wind observations generally larger than their mass counterparts (see Fig. 14)?

Concerning the issue of small HIRS forecast impacts, McNally (2002) undertook an experiment investigating the impact of HIRS data in NCEP’s Global Data Assimilation System (GDAS). That study, which explored the impact of HIRS data for a single time period, found that its forecast impact was relatively small and most pronounced at high altitudes. As such, even though they are limited to a single time period, the McNally results are similar to the HIRS impacts of this study in that HIRS observations generate a very small forecast impact for the time windows being examined.

Concerning the large difference between mass and wind forecast impacts from raob and GOES data, historically it is generally accepted that mass observations have a larger forecast impact within numerical weather prediction models than wind observations [a result substantiated for the EDAS by Zapotocny et al. (2002)]. This is especially true in middle and high latitudes, while wind observations may be more important at low latitudes. For the most part, just the opposite is diagnosed in this study. The wind observations from both raob and GOES data provide larger forecast impacts than their mass counterparts for virtually all fields examined (see Fig. 14).

One very important difference between these results and the Zapotocny et al. (2002) data impact results is the transition from assimilating marine GOES precipitable water retrievals in the earlier work to GOES radiances in the present work. This certainly has potential to influence the results of this work, particularly if the radiances are not being assimilated properly or are given an inadequate assimilation weight. The increase in horizontal resolution from 80 km in the Zapotocny et al. (2002) work to 32 km in this work also plays a vital role in the assimilation of both raob and GOES observations. Specifically, the issue of correctly modifying the correlation lengths (how far the impact of a datum extends from the observation location) for a given data type is an unresolved issue.

While both of these issues are very important and have the ability to alter EDAS forecasts, they are beyond the scope of this work. The main goal of this study has been to use operational software valid at the time the experiment began and report on the findings.

5. Summary

This work is a follow-on to Zapotocny et al. (2005), which examined the aggregate contribution of all raob, GOES, and POES data to forecast quality in the EDAS at 24 and 48 h. The present manuscript examines the 00-h sensitivity and 24-h forecast impact of nine of the components composing these three aggregate data types. Specifically, results for both the mass and wind components of rawinsonde data are presented, as are the mass and wind components of GOES data, three radiance data types from NOAA POES observations, and DMSP SSM/I low-level wind and precipitable water data. The time periods studied include 15 days during each of the four seasons, with results displayed on multiple isobaric surfaces extending from 1000 to 100 hPa. Computer limitations preclude the addition of in situ data types other than rawinsondes, as well as a study of data from GMS and/or Meteosat.

The combination of denials presented allows for a direct comparison of the mass and momentum contributions from both rawinsondes and GOES data. The comparison also examines the impact of many of the radiances available from GOES and POES data. While this work is closely related to the previous work of Zapotocny et al. (2002), it goes beyond that work by examining the impacts of these data sources in a 32-km, 60-layer model and investigates the impact of radiances rather than retrievals.

The 00-h sensitivity results are dominated by rawinsonde information over the North American continent. This large sensitivity extends throughout the atmospheric column, with largest sensitivities clearly positioned around the rawinsonde locations of the North American network, as well as stations on the Hawaiian and Caribbean islands. Raob contributions tend to be somewhat larger in Canada than the CONUS, where there is a smaller amount of Aircraft Communication, Addressing, and Reporting System (ACARS) and other in situ data to modify the model atmosphere during the 12-h assimilation cycle. GOES and POES data provide the largest initial sensitivities over maritime environments. The 00-h geographical distributions are very similar to the Zapotocny et al. (2002) results, in spite of the fact that this study examines radiance assimilation rather than retrieval assimilation and was completed in a much higher horizontal and vertical resolution model.

Similar to the Part I results, rawinsondes provide the largest forecast impacts to temperature when examined cumulatively over the four seasons. However, the remotely sensed data provide forecast impacts as large or larger than the rawinsonde data for several state variables and levels, most notably the u component and relative humidity results. In general, the GOES data provide a larger 24-h forecast impact than the POES data for most of the levels and seasons examined. The GOESW contribution is clearly larger than the GOESM component, in some instances by more than an order of magnitude. The only components of POES data showing appreciable forecast impact are the combined AMSU-A and AMSU-B data and the SSM/I PW data. The three remaining POES components (MSU, HIRS, and SSM/I wind) show little impact at 24 h.

As highlighted in the conclusions of Part I, a future study of this type in a global model should prove very interesting. A global study allows investigation of all the data types examined here, plus several other data types not available in the limited spatial domain of the EDAS. Such an experiment would also remove contamination from the lateral boundaries inherent in a regional model.

Acknowledgments

The authors wish to thank Eric Rogers, David Parrish, Dennis Keyser, and Tom Black of NCEP for providing the appropriate assimilation/forecast software and establishing the twice-daily data stream. The authors also wish to thank Timothy J. Schmit of NOAA/NESDIS/ORA, Ralph Petersen of NOAA, and Todd Schaack of SSEC for their enlightening scientific input. This research was supported under NOAA Grant NA67EC0100.

REFERENCES

  • Keyser, D., cited. 2001a: Code table for PREPBUFR report types used by the ETA/3DVAR. [Available online at http://www.emc.ncep.noaa.gov/mmb/papers/keyser/prepbufr.doc/table_4.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/papers/keyser/prepbufr.doc/table_19.htm.].

  • Keyser, D., cited. 2003: Observational data processing at NCEP. [Available online at http://www.emc.ncep.noaa.gov/mmb/papers/keyser/data_processing/.].

  • McNally, A. P., cited. 2002: The inclusion of NOAA-15 HIRS-3 and AMSU-A radiances in the global data assimilation system. [Available online at http://wwwt.emc.ncep.noaa.gov/gmb/gdas/research/jhtml/noaa15.html.].

  • NWS, cited. 2002a: EDAS. [Available online at http://www.emc.ncep.noaa.gov/mmb/research/eta.log.html.].

  • NWS, cited. 2002b: EDAS. [Available online at http://www.emc.ncep.noaa.gov/mmb/mmbpll/eta12tpb/.].

  • NWS, cited. 2000c: EDAS. [Available online at http://www.emc.ncep.noaa.gov/mmb/gcp/etarefs.html.].

  • Parrish, D., Purser J. , Rogers E. , and Lin Y. , 1996: The regional 3D variational analysis for the Eta Model. Preprints, 11th Conf. on Numerical Weather Prediction, Norfolk, VA, Amer. Meteor. Soc., 454–455.

  • Rogers, E., Parrish D. , Lin Y. , and DiMego G. , 1996: The NCEP Eta data assimilation system: Tests with regional 3-D variational analysis and cycling. Preprints. 11th Conf. on Numerical Weather Prediction, Norfolk, VA, Amer. Meteor. Soc., 105–106.

    • Search Google Scholar
    • Export Citation
  • Rogers, E., and Coauthors, 1997: Changes to the NCEP operational “early” Eta analysis/forecast system. NOAA/NWS Technical Procedures Bulletin 447, 16 pp. [Available from Office of Meteorology, National Weather Service, 1325 East-West Highway, Silver Spring, MD 20910.].

  • Zapotocny, T. H., Menzel W. P. , Nelson J. P. III, and Jung J. A. , 2002: An impact study of five remotely sensed and five in situ data types in the Eta data assimilation system. Wea. Forecasting, 17 , 263285.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Zapotocny, T. H., Menzel W. P. , Jung J. A. , and Nelson J. P. 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 , 161177.

    • Crossref
    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Geographical distributions of the four-season, time-averaged 00-h sensitivity (%) for 850-hPa relative humidity from all (a) raob, (b) GOES, and (c) POES observations. The contour interval is 2%

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 2.
Fig. 2.

Geographical distributions of the four-season, time-averaged 00-h sensitivity (%) for 850-hPa relative humidity from raob (a) mass and (b) wind observations. The contour interval is 2%

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 3.
Fig. 3.

Geographical distributions of the four-season, time-averaged 00-h sensitivity (%) for 850-hPa relative humidity from (a) GOES mass and (b) GOES wind observations. The contour interval is 2%

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 4.
Fig. 4.

Geographical distributions of the four-season, time-averaged 00-h sensitivity (%) for 850-hPa relative humidity from (a) HIRS, (b) AMSU, and (c) MSU observations. The contour interval is 2%

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 5.
Fig. 5.

Geographical distributions of the four-season, time-averaged 00-h sensitivity (%) for 850-hPa relative humidity from (a) SSM/I wind and (b) SSM/I column total precipitable water observations. The contour interval is 2%

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 6.
Fig. 6.

Vertical profiles of 00-h sensitivity from the aggregate raob (thin solid), GOES (thick solid), and POES (dashed) denials. The three fields displayed are (a), (d), (g) temperature (K); (b), (e), (h) relative humidity (%); and (c), (f), (i) u component (m s−1). (a)–(c) are near a rawinsonde location northwest of Lake Winnipeg. (d)–(f) are from a location in the northwestern Gulf of Mexico. (g)–(i) are from a point southwest of the Baja Peninsula

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 7.
Fig. 7.

Geographical distributions of the four-season, time-averaged 24-h forecast impact (%) for 300-hPa u component from aggregate (a) raob, (b) GOES, and (c) POES observations. The zero contour has been suppressed for clarity

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 8.
Fig. 8.

Geographical distributions of the four-season, time-averaged 24-h forecast impact (%) for 300-hPa u component from (a) raob mass and (b) raob wind observations. The zero contour has been suppressed for clarity

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 9.
Fig. 9.

Geographical distributions of the four-season, time-averaged 24-h forecast impact (%) for 300-hPa u component from (a) GOES mass and (b) GOES wind observations. The zero contour has been suppressed for clarity

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 10.
Fig. 10.

Geographical distributions of the four-season, time-averaged 24-h forecast impact (%) for 300-hPa u component from (a) HIRS, (b) AMSU, and (c) MSU observations. The zero contour has been suppressed for clarity

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 11.
Fig. 11.

Geographical distributions of the four-season, time-averaged 24-h forecast impact (%) for 300-hPa u component from (a) SSM/I wind and (b) SSM/I column total precipitable water observations. The zero contour has been suppressed for clarity

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 12.
Fig. 12.

Scatter distributions of (a)–(c) 100-hPa temperature and (d)–(f) 850-hPa relative humidity from a 24-h forecast starting 1200 UTC 7 Nov 2001. The aggregate raob denial results are shown in (a) and (d). The aggregate GOES denial results are shown in (b) and (e). The aggregate POES denial results are shown in (c) and (f). In (a)–(f), the difference between the control simulation and the corresponding analysis is shown on the x axis, while the difference between the denied simulation and analysis is shown on the y axis

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 13.
Fig. 13.

The four-season summary of rms forecast impact (%) for mean sea level pressure after 24 h of Eta Model integration. Both the three aggregate denials (raob, GOES, and POES) and the nine individual denials are shown. The results are for the (a) 104 and (b) 212 grids

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 14.
Fig. 14.

The four-season summary of rms forecast impact (%) on the 104 grid for (a) temperature, (b) u component, and (c) relative humidity, after 24 h of Eta Model integration. Both the three aggregate denials (raob, GOES, and POES) and the nine individual denials are shown

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 15.
Fig. 15.

Box-and-whisker displays of the forecast impact at 24 h for (a) temperature, (b) u component, and (c) relative humidity from the raob mass, raob wind, GOES mass, and GOES wind denials. The maximum and minimum values of the data make up the “whiskers” of the display, while the “box” comprises the median and upper and lower 25th and 90th percentiles, respectively. Note that in some instances, the vertical zero line obscures the median or percentiles within a given box. Furthermore, maximum and minimum values exceeding the scale of the x axis are enumerated on the “tails” of the whiskers. Values in this display are for the entire horizontal EDAS domain and are calculated from the four seasonal time periods used in this work

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Fig. 16.
Fig. 16.

Box-and-whisker displays of the forecast impact at 24 h for (a) temperature, (b) u component, and (c) relative humidity from the AMSU-A and -B, HIRS, MSU, and SSM/I PW denials. The maximum and minimum values of the data make up the “whiskers” of the display, while the “box” comprises the median and upper and lower 25th and 90th percentiles, respectively. Note that in some instances, the vertical zero line obscures the median or percentiles within a given box. Furthermore, maximum and minimum values exceeding the scale of the x axis are enumerated on the “tails” of the whiskers. Values in this display are for the entire horizontal EDAS domain and are calculated from the four seasonal time periods used in this work

Citation: Weather and Forecasting 20, 2; 10.1175/WAF838.1

Table 1.

Illustration of the environment within which remotely sensed and in situ data types are assimilated into the version of the EDAS used for this study

Table 1.
Table 2.

Distribution of the median and 95% confidence limits for each term and several of the isobaric levels presented in Figs. 15 and 16. In this display, the three grouped numbers within each cell from left to right are the lower confidence limit, median of the distribution, and upper confidence limit, respectively. Similar to Figs. 15 and 16, the numbers computed in this table use the entire horizontal EDAS domain and all 120 runs (30 per season)

Table 2.
Save