This paper presents the process used by the Marshall Space Flight Center Natural Environments Branch (EV44) to quality control (QC) data from the Kennedy Space Center’s 50-MHz Doppler radar wind profiler (DRWP) for use in vehicle wind loads and steering commands. The database has been built to mitigate limitations of using the currently archived databases from weather balloons. The DRWP database contains wind measurements from approximately 2.7- to 18.6-km altitude at roughly 5-min intervals for the August 1997–December 2009 period of record, and the extensive QC process was designed to remove spurious data from various forms of atmospheric and nonatmospheric artifacts. The QC process is largely based on DRWP literature, but two new algorithms have been developed to remove data contaminated by convection and excessive first-guess propagations from the median filter/first-guess algorithm. In addition to describing the automated and manual QC process in detail, this paper describes the extent of the data retained. Roughly 58% of all possible wind observations exist in the database, with approximately 100 times as many complete profile sets existing relative to the EV44 balloon databases. This increased sample of near-continuous wind profile measurements may help increase launch availability by reducing the uncertainty of wind changes during launch countdown.
The National Aeronautics and Space Administration (NASA; see Table 1 for a list of acronyms) has been designing, testing, and flying manned and unmanned space vehicles since the second half of the twentieth century. During this time, the Natural Environments Branch (EV44) at the Marshall Space Flight Center (MSFC) has helped ensure that supported launch vehicles can withstand the effects of the ascent wind environment over the central Florida region. EV44 accomplishes this task by developing meteorological databases and providing environmental definitions as inputs to a wide variety of discipline-specific engineering analyses supporting aerospace vehicle design, test, and operations activities.
EV44 has traditionally used observations made with weather balloons to statistically represent the ascent wind environment. However, balloon-based data archives have a number of inherent limitations for this application. Relative high cost makes high-frequency balloon sampling impractical, thereby limiting sample sizes and increasing statistical uncertainty. Downstream balloon drift can lead to potential misrepresentation of the ascent environment in a horizontally inhomogeneous wind field. Also, balloons have lengthy rise times, which prevent high temporal resolution assessments of ascent wind environments. This limitation reduces the ability to verify steering commands late in a launch countdown, possibly leading to larger-than-necessary persistence margins during rapid wind regime change events, such as frontal passages. The latter problem was recognized by NASA, and in 1988 a 50-MHz Doppler radar wind profiler (DRWP) was installed at Kennedy Space Center (KSC). During the 1990s, MSFC developed an algorithm to operationally quality control (QC) wind profiles from the DRWP on day of launch (DOL; Schumann et al. 1995), and various analyses using subsets of the DRWP archive were performed (Wilfong et al. 1993; Merceret 1997; Schumann et al. 1999). However, the Space Shuttle Program (SSP) did not certify the DRWP in the later stages of the program. Such a certification would have been very expensive and time-consuming, because a DRWP database did not exist during the shuttle’s design phase. EV44 highly desires to certify the DRWP for use in the design of future launch vehicles, and one of the first steps in certifying the instrument is to develop a climatological archive that can be used by the ascent loads and trajectory community.
EV44 has recently developed a QCed database of DRWP wind profiles covering the period of record (POR) from August 1997 to December 2009. This new dataset is expected to be widely applied to engineering analyses for future launch vehicle programs. The database contains a much larger sample size than those from balloons and has greater temporal coverage, providing flexibility in the prelaunch assessments. Given these two qualities, the DRWP provides the ability to increase confidence in our knowledge of a given wind environment, reducing unnecessary conservation in both design and operational margins, thereby potentially increasing launch availability. To develop the database, an extensive QC process was implemented on wind and spectral output from the DRWP. This paper documents the QC process and the database’s application. Following a brief description of the DRWP hardware and data, the QC process is outlined in detail. Then, the resulting database and its current applications are described followed by a summary and forward work plan.
2. DRWP database description
Although detailed descriptions of the DRWP’s hardware and data processing algorithm have been previously documented (Schumann et al. 1999), an overview is provided here for context. The attributes of the data stream used in the QC process are also described in this section. The DRWP is located just east of the Shuttle Landing Facility at KSC. Physically, the DRWP consists of an irregular octagon-shaped antenna field spread across 15 600 m2. Coaxial collinear elements are set 1.5 m above the ground plane made of copper wire. These elements are arranged to send electronic pulses at 49.25 MHz through three beams. One beam is pointed vertically, and two oblique beams are pointed 15° off zenith at azimuths of 45° and 135° east from north. All beams have a 3° beamwidth (B. Gober 2010, personal communication). An adjacent equipment trailer houses the radar instrumentation and data acquisition electronics. Figure 1 provides a photograph and schematic of the DRWP.
To measure wind velocities, the DRWP transmits radio pulses in three beam directions sequentially and measures the return signal backscattered by temperature and humidity fluctuations in the atmosphere with scale lengths of about 3 m (Rinehart 2004). The signals are converted to Doppler power spectra by applying a fast Fourier transform over 256 frequency bins at each range gate. In the beginning of the POR, 112 gates were set from 2011 to 18 661 m every 150 m. After an upgrade during July–August 2004 (Pinter et al. 2006), 111 gates have been set from 2666 to 18 616 m every 145 m. Also before this upgrade, profiles were generated every 5 min. After the upgrade, profiles have been generated every 3 min. Once the Doppler spectra are obtained for each beam, horizontal velocities are computed using the median filter/first-guess (MFFG) algorithm (Schumann et al. 1999), which is applied to the oblique beams only because the vertical beam is not used to calculate horizontal winds (Wilfong et al. 1993; Schumann et al. 1999).
Although the MFFG algorithm has many advantages over the traditional consensus averaging technique used on other wind profilers (Schumann et al. 1999), using the algorithm does not completely prevent acquiring erroneous data. In many instances, the FG simply associates itself with spectral peaks that do not represent the real wind. If a strong, nonatmospheric signal persists within the FG window, then radial velocities would contain contributions from nonatmospheric effects. Other instances of suspect data occur when the signal is too weak to calculate a radial velocity. If the signal-to-noise ratio (SNR) is less than −15 dB, then the measured radial velocity is replaced by the FG. This process of FG “propagating” continues for an individual beam until the SNR reaches −15 dB. Although the radial velocity profiles are smoothed after at least four first-guess propagations (FGPs), using previously recorded measurements for an extended period can introduce errors in the radial velocity estimate, especially if a nonatmospheric signal exists near the signal associated with the wind (Wilfong et al. 1993). Heavy rain could cause spurious wind output because the DRWP can track the velocity of the raindrops instead of the air. Also, large vertical motions can violate the assumption of a homogeneous atmosphere used in the horizontal wind computation.
EV44 has archived DRWP data output, which was used to develop the QCed database. In addition to the computed horizontal wind speed (m s−1), wind direction (°), and altitude (m), spectral width (SW; m s−1), signal power (dB), noise power (dB), vertical velocity (w; m s−1), number of FGPs (dimensionless), and an internal shear value (s−1) at each gate and beam have been archived. Vertical velocity is the radial velocity of the vertical beam, with positive values indicating downward motion. The number of FGPs is the number of times the FG was propagated for the particular gate and beam, and the internal shear is the change of the radial velocities per unit altitude. These fields exist for each day from August 1997 through December 2009, but have not been regularly QCed. The QC process used to develop this database differs from the DOL QC process. The DOL process (Schumann et al. 1995) requires near-real-time examination of the Doppler spectra, which are not available in the data used for the current QC process. The current QC process will be described in detail in section 4.
3. Data display system
A graphical user interface (GUI) was developed to implement the current QC process. The GUI contains various functions to perform all QC procedures and save the desired output (Fig. 2). Each DRWP data file contains data for a single year, month, and day. Files are read sequentially and time–height sections of meteorological parameters are initially examined for potentially spurious features. In Fig. 2, for example, anomalies at very low altitudes appear to correspond to ground clutter, and streaks of enhanced meridional wind (υ) from 10 to 12 km at 1500–1700 UTC seem incompatible with the surrounding environment. The automated QC is then run and a new time–height cross section of the given variable is displayed. To perform the manual QC, a box surrounding the data in question is drawn and data that are flagged by the threshold are removed. An “undo” function exists to protect against operator error during the manual QC process. Once the QC process is complete, the QCed file and manual QC logs are saved. In addition, comparisons between profiles from low-resolution (LR) weather balloons and DRWP profiles at a desired time can be performed and images can be saved as desired. The LR balloon database consists of rawinsondes prior to October 2002 and the low-resolution flight element (Leahy and Overbey 2004) after October 2002.
4. QC process
In addition to methodologies documented elsewhere, the process used here to QC the DRWP data contains some unique attributes based on data examination. A number of distinct steps are performed sequentially, with flagged data being removed before the next step is implemented. Indicators are assigned to each gate to denote if data passed all checks or failed a particular check. This section describes in detail the QC sequence.
a. Automated QC process
The automated QC process contains initial procedures to fill data gaps and screen the vertical beam. The first step in the automated QC process fills data gaps. If greater than 6 min existed between adjacent time stamps in the original data, then time stamps were inserted at 5-min intervals in the data gap, with all variables in the profile containing the missing data flag. This procedure ensured a data record at least once every 6 min throughout the POR. The second step in the automated QC process evaluates vertical beam measurements. Because the vertical beam is not used to calculate horizontal winds, a valid wind calculation could coincide with an erroneous vertical beam measurement and be falsely flagged. Data from the vertical beam were thus removed if a signal or noise power report were missing, if the vertical beam’s SW exceeded 3.0 m s−1, or if a systematic error occurred when the Doppler shift from the vertical beam was near zero (F. Merceret and B. Gober 2009, personal communication). This error appeared when abnormally high |w| coincided with relatively low SNR.
The automated process then performs threshold checks and flags data possibly influenced by convection. Table 2 presents each check and its threshold in order. The process consists of checking for unrealistic wind reports and isolated data, performing a small median test, and applying thresholds to oblique beam SW, DRWP internally computed shear, w, FGP, oblique beam signal power, and convection. The rationale for each check is described below, with thresholds for checks other than the FGP, small median, convection, and isolated datum presented in the table.
Several automated checks were based on thresholds, which were derived from Carr et al. (1995) and Merceret (1997) and modified, if necessary, based on data examination. After detecting physically unrealistic wind reports, a check was implemented on the oblique beam SW so that the homogeneity assumption used to calculate the winds would not be violated due to excessive turbulence. The DRWP internal shear is useful for detecting large objects in the air, such as airplanes (Merceret 1997). Over Florida, |w| is generally very small, so any large perturbation in w indicates that there is some anomaly in the airflow or that the DRWP is measuring the velocity of raindrops instead of the air. The threshold selected here is more restrictive than that in Merceret (1997) to flag additional convective situations, especially after August 2004. The meteorological shear check serves the same purpose as the DRWP shear check, but it applies to the zonal (u) and meridional (υ) wind components. Missing signal power indicated that the DRWP did not receive a signal at that gate. Note that no check exists for the oblique beam noise power. An analysis was performed showing that missing noise values corresponded to some erroneous vertical beam SW and velocity reports. However, no such effect existed when examining oblique beam SW and radial velocities.
The small median check (Carr et al. 1995) flags observations that significantly differ from their nearest neighbors. The check compares a wind speed observation at a given time and altitude to the eight observations surrounding it and was only performed if the wind speed of interest and at least three neighboring observations existed. Thresholds were applied following Merceret (1997). Once all of the automated checks were performed on data for the day, gates with no surrounding measurements were removed to enhance the continuity of the database.
Two additional QC algorithms were developed specifically for the DRWP database, and differ significantly from the literature. These checks involved testing for convection and developing a criterion for the FGP threshold, and are presented in greater detail in the subsections below.
The convection algorithm is derived from previous work using the 915-MHz DRWP network at KSC. Lambert et al. (2003) developed a discriminant function based on w and SNR, which had two classes—convection and no convection—and was effective on the 915-MHz DRWP data. However, because the DRWP is much less sensitive to rain than the 915-MHz DRWP, and w differs from the boundary layer to the free atmosphere, this discriminant function was determined to not suit the DRWP QC process. In addition, a given parameter at an individual gate might have the same output in different situations throughout the year. Therefore, the convection algorithm’s parameters were derived for each month.
Both SW and w were used to determine if convection existed at a particular gate. Figure 3 shows w and SW for 20 August 2001, a day with typical summertime convective activity over the KSC region. Rain gauge data were obtained from the TRMM website (http://trmm.ksc.nasa.gov/trmm/rain/2001/08/DAILY_RPT_AUG20.HTM). The rain gauge at the field mill closest to the DRWP recorded 0.28 in. of rain during 0300–0500 UTC and 1.42 in. of rain during 2100–2200 UTC. Concurrent SW and w show clear signs the convection could be occurring: SW increased from just less than 1.0 to near 2.0 m s−1, and |w| increased from near 0.0 to approximately 1.5–2.0 m s−1. Other meteorological variables did not vary as significantly, so they were not used to discriminate between convective and nonconvective cases.
SW and w were used in a supervised classification technique to determine if convection existed. First, the classes “convective,” “possibly convective,” and “not convective” were chosen to classify the convective environment of each gate. Next, training samples representing instances of each class were selected for each month across the POR. Instances where large SW corresponded with large |w| across extensive vertical regions were selected as convective, and instances of small SW and |w| were selected as not convective. Training samples for the possibly convective class were selected to help mitigate a false positive convective classification. A range of 1638–6428, 11 546–32 265, and 7528–19 935 training samples existed per month for the convective, possibly convective, and not convective cases, respectively. These sample sizes are of one–two orders of magnitude over what is considered highly desirable in the literature (Richards 1993).
The training samples were used to develop a discriminant function, which classifies a pixel as being either convective or not convective. The discriminant function is a quadratic surface described by
where K, L, and Q are coefficients corresponding to the covariance of the training samples for each class combination and month. SWe, SWn, and SWυ are SW from the east, north, and vertical beams, respectively. The training samples were provided as input to the MATLAB discriminant function routine (http://www.mathworks.com/help/toolbox/stats/classify.html) along with the data to be classified (i.e., the DRWP data for the day of interest). The routine returned the class of each gate, the coefficients of the discriminant function for each class combination, and the posterior probabilities of the gate belonging to its determined class. If DF were positive for the convective/possibly convective combination and the convective/not convective combination, with a posterior probability of at least 0.95, then the gate would be classified as convective.
Plots showing gates flagged by the convection algorithm were examined and data were removed manually based on the extent of the flagged gates and the characteristics of the corresponding wind field. In Fig. 4, flagged data over extensive vertical regions that corresponded to anomalies in υ around 0400 and from 2030 to 2200 UTC were considered to be convective and were thus removed. However, some flagged gates neither span extensive horizontal or temporal regions nor seem to correspond to anomalies in the wind field. These gates were not removed because convection was determined to not affect the continuity of the winds on a large enough scale.
2) First-guess propagation
A unique FGP check was developed to suit the database’s multiple applications and to determine how propagating the FG velocity affects the wind estimate. Using radial velocity estimates with one FGP is basically equivalent to using estimates that are 5 min old. In situations when the wind changes little, an FGP should not significantly affect the wind estimate. However, in dynamic conditions, even a small FGP could lead to an inaccurate wind estimate. Previous research has thus used limits on the number of FGPs for their respective analyses. Merceret (1997) used a limit of six FGPs; however, no rationale was given for this threshold other than using it to prevent wind estimates greater than 30 min old from being incorporated into the analysis. Schumann et al. (1999) used a limit of two FGPs to relate to the capability to distribute DRWP data every 15 min to the end user. Using the latter criterion, data with greater than two FGPs would contain wind estimates at least 15 min old, and the end user would not be provided with a new wind estimate. This rationale implies that the FGP threshold should be selected based on the end user’s application. However, it was desired to have a single FGP threshold for the current DRWP database for three reasons: First, the FGP threshold significantly affects the number of available profiles. Second, the current DRWP database can be used for applications with varying time separations, which are currently unknown. Last, the exact relationship between a number of FG propagations and resulting measurement errors is unknown. Therefore, the following analysis was performed to better understand the effect of FG propagation on the output of the MFFG algorithm.
Spectra from 3 days obtained from the DRWP operations and maintenance contractor during 2009 were examined. Each day represented a weather regime common to eastern Florida:
21 August: light winds with an afternoon thunderstorm;
21 October: no rain, dynamic day with moderate winds; and
4 December: strong southwest winds.
Control wind components from each day were calculated using FG radial velocities derived from the Doppler spectra. First, a three-point median filter was applied to spectra with at least two valid time stamps. Radial velocities were then calculated for each oblique beam, with the calculated radial velocity profile being the FG for the next radial velocity profile. The first profile in the database and the first profiles after data gaps were computed. Radial velocities were replaced by the mean of the radial velocities from the adjacent gates if the shear criteria were violated (Taylor et al. 1993). Wind components were then computed from the radial velocities, and were compared to the wind components in the output data files for accuracy. If the magnitude of the difference between the calculated wind component and the wind component in the output file exceeds 2.0 m s−1, then the wind value from the output file replaces the calculated wind from the spectra.
The control wind components were then differenced from wind components calculated after propagating the FG velocity from each beam. To simulate propagation of the FG velocity, radial velocities were calculated using the spectra for the current time stamp with the FG radial velocity from the previous n time stamps, where n was incremented from 1 to 20. Modified wind components were then calculated using the previous n FG velocities from each beam. FGPs from each beam were simulated by cycling n for the north beam before incrementing the east beam (0 east FGPs/0 north FGPs, 0 east FGPs/1 north FGP, … , 20 east FGPs/19 north FGPs, 20 east FGPs/20 north FGPs). A median of 29 554 gates were used, varying from 28 310 to 32 820 gates, depending on the FGP combination. The vector differences between the modified and control winds were then computed at altitudes over 10 km because the FG is propagated more frequently at higher altitudes. The RMS of the vector changes for each FGP combination was then plotted versus the number of FGPs from the oblique beams (Fig. 5). Light (dark) grayscale shades represent large (small) RMS differences from the control wind. As expected, differences increase as the number of FGPs increase from either beam. An RMS vector error of m s−1 was selected as the threshold for this analysis to correspond to the RMS measurement error specification of 1.0 m s−1 for each wind component (Pinter et al. 2006). A quadratic fit was then applied to the maximum number of north beam FGPs, which yielded an RMS difference below the threshold for each east beam FGP. The fit is expressed as
where T is the threshold parameter and FGPn and FGPe are the FGPs from the north and east beams, respectively. If T were less than zero, then data at the gate were removed.
b. Manual QC process
Once the automated process was complete, data for each day were manually examined for temporal and spatial inconsistencies. This process involved examining multiple variables to see if a spurious output from one variable coincided with that from another variable. If spatial discontinuities in multiple variables occurred within the same time–height region, then greater evidence would be presented to remove the data in question. In addition, temporal changes in wind components at each altitude were examined to detect the edges of radar sidelobes and ground clutter. Each manual QC was logged for reference. Data judged to be contaminated by convection or ground clutter were assigned their own QC flags to be tracked separately. On occasion, extensive time–height regions of data were not flagged by the automated QC process but needed to be removed. In these cases, entire time–height boxes or profiles were removed manually.
An example of the manual QC process is presented here. Figure 6 shows before and after images of υ on 19 October 2008. The left panel shows υ from the original database. Note the barlike features that do not compare well with the surrounding environment at approximately 5.0 km during 0100–0300 UTC and at approximately 4.0 km during 0300–0400 UTC. These features are likely attributed to the MFFG algorithm tracking a signal from a sidelobe, and not the real wind between these altitudes. Examining the change in wind components over short time intervals, in addition to the wind field itself, indicates that ground clutter likely contributed to the signal around 0100, 0700, and 2000 UTC at very low altitudes. Thus, data were also removed from these regions. The QCed data are presented in the right panel.
Wind components from LR balloons and the DRWP could also be compared to determine if the DRWP measurements were acceptable to use on a given day. Balloon data were downloaded from the KSC Tropical Rainfall Measurement Mission website (ftp://trmm.ksc.nasa.gov/midds/sonde), and available data were used to visually examine the characteristics of the wind components from both sources. Following EV44 DOL procedures, wind components from the closest DRWP profile to 30 min after the balloon launch were examined to minimize errors in the comparison associated with the balloon’s rise rate. If the DRWP profile did not compare well with a balloon profile, which was considered acceptable, then the DRWP measurements would be removed around the time of the comparison. The left panel of Fig. 7 shows that on 11 February 2000 the 1415 UTC DRWP profile deviated from the 1345 UTC balloon profile above 7 km, with both wind components around 10–15 m s−1 above 12 km. Thus, DRWP measurements showing this measurement characteristic on this day were removed. Conversely, the profiles during 11 January 2001 (Fig. 7, right panel) show similar characteristics from both sources. Thus, the DRWP measurements were considered acceptable on this day.
Once the database was QCed, investigations were performed to examine the algorithm’s attributes. Missing data tended to exist throughout entire profiles. The ground clutter check, which is performed manually, mainly affected scattered gates at lower altitudes. However, larger clusters of data contaminated by ground clutter did exist. The vertical beam QC flagged either isolated or narrow vertical regions. The SW, small median, noise power, and isolated datum QC algorithms all flagged data at sporadic intervals with no general pattern. However, the SW checks flagged larger clusters of data compared to the other three similar checks. Conversely, the FGP check flagged adjacent gates at the same altitudes. Gates flagged by the convection algorithm were removed if data were flagged across extensive vertical regions. The shear checks tended to flag the boundaries of spurious data regions, with the inside of the regions being removed manually. The manual QC process also removed vertical discontinuities, sidelobes near ground clutter and convection, and any other unacceptable feature.
The number and percentage of gates affected by each QC process were tallied. Each process was assigned a flag, and the number of times an individual flag occurred in each month was recorded (Table 3). The entire POR contains 162.1 million gates, with a given month containing 12.2–14.3 million gates. Percentages of affected data herein are noted as %POR (the percent lowest month to the percent highest month). Missing data accounted for 35.4% (30.0%–41.0%) of all the possible data. The missing data flag was tallied most often because it tended to exist throughout an entire profile, and days in which no data existed were recorded as containing missing data at every gate and time stamp. The other QC processes combined removed an additional 6.5% (3.7%–10.5%) of the possible data. The manual QC process dominated these QC processes, removing 4.8% (1.8%–8.6%) of the data. The convection QC process removed 0.6% (0.1%–1.1%) of the available data. Note that the automated convection algorithm flagged 2.6% of the data for the POR, but only 0.6% of the data were removed, indicating the significance of removing flagged data manually. The other automated QC processes removed no more than 1.0% of the available data for a given month. The isolated datum check removed data at 4767 gates over the POR. The meteorological wind shear check removed data at only 570 gates throughout the POR because data that would have been flagged by this check were likely removed by the DRWP internal shear check. No observations existed that had unrealistic reports of wind speed or wind direction. The QCed database contains 58.1% (51.6%–64.5%) of the possible wind observations.
Retained complete profiles and pairs were also tallied to support the launch vehicle community’s interest in examining the vehicle’s entire ascent trajectory. Although the QCed database contains profiles that contain data removed by the QC process and can be used for any application involving winds aloft near KSC, the central focus of generating a DRWP winds database involved generating a larger sample of complete profiles and profile sets to be used in vehicle loads and trajectory analyses.
Table 4 depicts the number of complete DRWP profiles and pairs. Generally more profiles are retained from more recent years than earlier years. March 2000 was the only month over the POR where zero complete profiles were retained. In addition, the DRWP’s poor performance during individual periods can be inferred (e.g., February–March 2000). An average of 30 320 profiles per month exist, ranging from 27 436 (October) to 35 239 (March) profiles per month. No obvious trend in the number of complete profiles seemed to exist from month to month. The 2-h pairs have an average sample of 15 816 month−1, ranging from 12 352 (July) to 19 023 (March) pairs per month. Sample sizes for other time separations are on the same order of magnitude.
The DRWP database has three major advantages over balloon archives. First, the DRWP database contains on the order of 100 times as many profiles and pairs as the databases derived from balloon measurements, which would improve confidence in launch simulation results. Second, using the DRWP database provides the capability to examine time separations other than 2.0 and 3.5 h because the DRWP pairs archive is not driven by any program requirements. In addition, using the DRWP database enables launch vehicle personnel to perform assessments closer than 2.0 h to launch, which reduces the uncertainty of the wind profile loaded to the vehicle’s steering commands and the wind through which the vehicle will fly, potentially leading to a launch availability increase resulting from decreased loads knockdowns. Third, the DRWP database enables launch vehicle engineers to perform simulations with more than two profiles at a time. For example, launch −3.0 h, launch −1.0 h, and launch −0.0 h wind triplets can be used to simulate loading a trajectory at launch −3.0 h, making a go/no-go decision at launch −1.0 h, and then flying to the launch −0.0 h wind. This capability would allow more accurate simulations to be performed before launch vehicle requirements are written. Times before launch can also be examined to determine the most optimal DOL assessment sequence.
To improve the sample size of the EV44 winds database, QC algorithms were developed and implemented on DRWP data for the August 1997–December 2009 POR. A larger sample of wind measurements not only gives greater confidence in loads and trajectory assessments, but also provides flexibility to simulate different DOL situations. These features of the DRWP database should mitigate the limitations of the balloon databases used to support the SSP and other previous NASA flight vehicle programs.
In addition to increasing the sample size of the database used and providing more flexibility for DOL simulations in the vehicle design phase, the QCed DRWP database provides any upcoming launch vehicle program with the capability to utilize the DRWP profiles on DOL to compute vehicle steering commands, provided the automated and manual QC procedures developed are applied to new DRWP data on DOL. In the past, only balloons have been certified to do this. Although the current DRWP QC process on DOL could be enhanced by an automated QC process, manual intervention would still be needed to ensure only valid profiles are used. If high spatial resolution profiles are desired (such as the SSP’s desire for Jimsphere measurements), then high-frequency components could be randomly added to the DRWP profiles. The DRWP database provides lots of flexibility in how DOL simulations are performed, and the QC algorithms provided in this paper will hopefully benefit the aerospace and atmospheric communities which are interested in utilizing the DRWP.
7. Forward work
Despite the benefits of utilizing the DRWP database, it does contain a limitation in that only measurements above 2.7 km are provided for the entire POR. Currently, no QCed database exists that contains the sample size of the DRWP database and measurements from near the surface to 2.7 km. EV44 is thus performing a similar QC to that presented in this paper to data from the 915-MHz DRWP network at KSC. If an adequate sample exists, complete profiles from both sources will be combined to generate an extensive database of DRWP profiles extending from approximately 0.13 to 18.6 km. The QC algorithms presented in this report, and any others that are developed for the 915-MHz DRWP, will then be evaluated for operational use during DOL.
The author would like to thank the colleagues who provided assistance during this extensive project. Conversations with Dr. Frank Merceret at the KSC Weather Office helped determine many facets of the QC algorithm. Bill Gober (KSC/Modern Technologies Corporation) provided the Doppler spectra used in the FGP algorithm development, and gave a personal tour of the DRWP describing how the hardware worked. Richard Leach (MSFC/EV44/MEI Technologies, Inc.) provided the raw DRWP data. Collaborations with numerous other EV44 colleagues greatly improved the quality of this work. Much appreciation also goes to the reviewers of this paper, which was written under NASA Contract MSFC-NNM05AB50C. Mention of a proprietary product or service does not constitute an endorsement thereof by the author, Jacobs, or the National Aeronautics and Space Administration.