NASAICR-20 10-216289

Peak Wind Tool for General Forecasting Phase II

Joe H Barrett III ENSCO, Inc., Cocoa Beach, Florida NASA Applied Meteorology Unit, Kennedy Space Center, Florida

September 2010 NASA STI Program ... in Profile

Since its founding, NASA has been dedicated • CONFERENCE PUBLICATION. to the advancement of aeronautics and space Coll ected papers from scienti fic and science. The NASA scientific and technical technical conferences, symposia, information (STI) program plays a key part in seminars, or other meetings sponsored hel ping NASA maintain this important role. or co-sponsored by NASA.

The NASA STI program operates under the • SPECIAL PUBLI CATION. Scientific, auspices of the Agency Chi ef Information technical, or hi stori cal information from Officer. It coll ects, organi zes, provides for NASA programs, projects, and missions, archiving, and disseminates NASA's STI. The often concerned with subjects having NASA STI program provides access to the substantial public interest. NASA Aeronautics and Space Database and its public interface, the NASA Technical Report • TECHNICAL TRANSLATION. Server, thus providing one of the largest English-language translations of fore ign coll ecti ons of aeronautical and space science scientific and technical material pertinent to STI in the world. Results are published in both NASA's mission. non-NASA channels and by NASA in the NASA STI Report Series, which includes the Specialized services also include organizing following report types: and publishing research results, distributing special ized research announcements and feeds, • TECHNICAL PUBLICATION. Reports of prov iding help desk and personal search com pleted research or a major significant support , and enabling data exchange services. phase of research that present the results of NASA Programs and inc lude extensive data For more information about the NASA STI or th eoretical analysis. Includes compila­ program, see the following: tions of signifi cant scientific and technical data and information deemed to be of • Access the NASA STI program home page continuing reference value. NASA counter­ at http://www.sli.nasa.gov part of peer-reviewed formal professional papers but has less stringent limitations on • E-mail your question via the Internet to manuscript length and extent of graphic [email protected] presentations. • Fax your question to th e NASA STI Help • TECHNICAL MEMORANDUM. Desk at 443-757-5803 Scientific and technical findings that are preliminary or of speciali zed interest, • Phone the NASA STI Help Desk at e.g., quick release reports, work ing 443-757-5802 papers, and bibliographies that contain minimal annotation. Does not contain • Write to: extensive analysis. NASA STI Help Desk NASA Center for AeroSpace Information • CONTRACTOR REPORT. Scientific and 71 15 Standard Drive technical findings by NASA-sponsored Hanover, MD 21076-1320 contractors and grantees. NASN CR-20 10-216289

Peak Wind Tool for General Forecasting Phase II

Joe H. Barrel/ Ill ENSCO, Inc., Cocoa Beach, Florida NASA Applied Meteorology Unit, Kennedy Space Center, Florida

National Aeronautics and Space Admini strati on

Kennedy Space Center Kennedy Space Center, FL 32899-0001

September 2010 Acknowledgments

The author thanks Mr. Mark Wheeler of the NASA Applied Meteorology Unit for providing expenise necessary to develop the MIDDS version of the tool. Personnel from the 45th Weather Squadron provided valuable feedback on the tool's user interface.

Available from:

NASA Center for AeroSpace Information 71 15 Standard Drive Hanover, MD 21076-1320 443-757-5802

This repon is also available in electronic form at hltp:/lscience.ksc.nasa.gov/amul Executive Summary The expected peak wind speed of the day is an imponant fo recast element in the 45th Weather Squadron's (45 WS) daily 24-Hour and Weekly Planning Forecasts. The forecasts are used for ground and space launch operations at the Kennedy Space Center (KSC) and Cape Canaveral Air Force Station (CCAFS). The 45 WS also issues wind advisori es for KSC/CCAFS when they expect wind gusts to meet or exceed 25 kt, 35 kt and 50 kt thresholds at any level from the surface to 300 ft. The 45 WS forecasters have indicated peak wind speeds are challenging to forecast, panicularly in the cool season months of October - April. In Phase I of this task, the Applied Meteorology Uni t (AMU) developed a tool to help the 45 WS forecast non-convective winds at KSC/CCAFS for the 24-hour period of 0800 to 0800 local time. The tool was delivered as a Microsoft Excel graphical user interface (GUI). The GUI displayed the forecast of peak wind speed, 5-m inute average wi nd speed at the time of the peak wind, timing of the peak wi nd and probability the peak speed would meet or exceed 25 kt, 35 kt and 50 kt. For the current task (Phase II ), the 45 WS requested additional observations be used for the creation of the forecast equations by expanding the period of record (POR). Additional parameters were evaluated as predictors, including wind speeds between 500 ft and 3000 ft, static stability classification, Bulk Richardson Number, mixing depth, venical wind shear, temperature inversion strength and depth and wind direction. Us ing a verification data set, the AMU compared the performance of the Phase I and II prediction methods. Just as in Phase I, the tool was delivered as a Microsoft Excel GU I. The 45 WS requested the tool also be avail able in the Meteorological Interactive Data Di splay System (MIDDS). The AMU first expanded the POR by two years by adding tower observations, surface observations and CCAFS (XMR) soundings for the cool season months of March 2007 to April 2009. The POR was expanded again by six years, from October 1996 to April 2002, by interpolating 1000-ft sounding data to 100-ft increments. The Phase II developmental data set included observations for the cool season months of October 1996 to February 2007. The AMU calculated 68 candidate predictors from the XMR soundings, to include 19 stability parameters, 48 wind speed parameters and one wind shear parameter. Each day in the data set was stratified by sy noptic weather pattern, low-level wind direction, precipitation and Richardson Number, fo r a total of 60 stratification methods. Linear regression equations, using the 68 predictors and 60 stratification methods, were created for the tool's three forecast parameters: the highest peak wind speed of the day (PWSD), 5-minute average speed at the same time (A WSD), and timing of th e PWSD. For PWSD and A WSD, 30 Phase II methods were selected for evaluation in the verification data set. For timing of the PWSD, 12 Phase \I methods were selected for evaluation. The verificati on data set contained observati ons for the cool season months of March 2007 to April 2009. The data set was used to compare the Phase I and II forecast methods to climatology, model forecast winds and wind advisori es issued by the 45 WS. The model forecast winds were derived from the 0000 and 1200 UTC runs of the 12-km Nonh American Mesoscale (MesoNAM) model. The forecast methods that performed the best in the verification data set were selected for the Phase II version of the tool. For PWSD and A WSD, linear regression equations based on MesoNAM forecasts performed significantly bener than the Phase I and II methods. For timing of th e PWSD, none of the methods performed significantly bener than cl imatology. The AMU then developed the Microsoft Excel and MIDDS GU ls. The GUls di splay the forecasts for PWSD, A WSD and the probability the PWSD will meet or exceed 25 kt, 35 kt and 50 kt. Since none of the prediction methods for timing of the PWSD performed significantly better than climatology, the tool no longer displays this predictand. The Excel and MIDDS GU ls display forecasts for Day-I to Day-3 and Day-I to Day-5, respectively. The Excel GU I uses MesoNAM forecasts as input, while the MIDDS GU I uses input from th e MesoNAM and Global Forecast System model. Based on feedback from the 45 WS, the AMU added the daily average wind speed from 30 ft to 60 ft to the tool, which is one of the parameters in the 24- Hour and Weekly Planning Forecasts issued by the 45 WS. In addition, the AMU expanded the MIDDS GUI to include forecasts out to Day-7. Table of Contents Executive Summary ...... I

Introduction ...... 7

2 Data ...... 8 2. 1 Tower Data ...... 8 2.2 Surface Observations ...... 9 2.3 Sounding Data ...... 9 3 Development of Phase II Prediction Equati ons ...... 14 3. 1 Synoptic Weather Pattern ...... 14 3.2 Phase II Predictors ...... 14 3.3 Prediction Equations fo r PWSD and A WSD ...... 18 3.4 Timing of PWSD ...... 20 4 Independent Verification of Prediction Methods ...... 22 4.1 Peak and Average Wind Speed ...... 27 4.1.1 Comparing Phase 1 and JJ Methods to Climatology ...... 27 4.1.2 Comparing Phase 1 and JJ Methods 10 Model Forecast Winds ...... 31 4.1.3 Comparison of Phase I and JJ Peak Wind Speed Predictions to 45 WS Wind Advisories ...... 35 4.2 Timing of the Peak Wind ...... 35 5 Peak Wind Tool GUls ...... 38 5. 1 Microsoft Excel GU I ...... 38 5.1.1 Equation Development ...... 38 5.1.2 Using the Excel GUI ...... 42 5.2 MIDDS GUI ...... 45 5.2.1 Equation Development ...... 46 5.2.2 Point Model Data ...... 47 5.2. 3 Using the MIDDS GUJ ...... 47 6 Summary and Future Work ...... 49 6.1 Summary ...... 49 6.2 Future Work ...... 50 References ...... 5 I

List of Acronyms ...... 52

2 List of Figures Figure I. A map showing the towers used in the task. Except for tower 0300 (lower-right, in blue), only the yellow- and red-colored towers we re used ...... 9 Figure 2. Comparison of I OO-ft sounding data on I January 2003 to interpolated data using Method I (left) and Method 2 (right) ...... 10 Figure 3. Monthly averages of height versus u-wind component, using interpolated (dashed) and 100-ft data (solid). The interpolated data used Method I (left) and Method 2 (right). The data are from the cool season months in the period October 2002 - April 2008 ...... I I Figure 4. Monthly averages of height versus MAE in u-wind component, using interpolated and 100-ft data. The data were interpolated using Method I (left) and Method 2 (right) ...... I I Figure 5. The December plots of height versus MAE in u-wind component, using interpolated and 100- ft data. The data were interpolated using Method I (left) and Method 2 (right). The 2002- 2007 December average is displayed by the black lines ...... 12 Figure 6. Height versus MAE in u- (left) and v-wind (right) components, using Method I (red) and Method 2 (green) interpolated data. The interpolated data were averaged over all months in the POR ...... 12 Figure 7. Height versus MAE in temperature (left) and dew point (right), using Method I (red) and Method 2 (green) interpolated data. The interpolated data were averaged over all months in the POR ...... 13 Figure 8. MAE values for least-squares si ngle linear regression equations for PWSD (left) and A WSD (right). The MAE is shown for the stability parameters (S I -S 19), wind speed parameters (W 1- W48) and wind shear parameter (H I). Mean MAE (blue) is the average of all stratification methods, wh ile Minimum MAE (red) is the lowest of all stratification categories ...... 18 Figure 9. MAE values for robust single linear regression equations for PWSD (left) and A WSD (right). The MAE is shown for the stability parameters (S I -S 19), wind speed parameters (WI -W48) and wind shear parameter (H I). Mean MAE (blue) is the average of all stratification categories, while Minimum MAE (red) is the lowest of all stratification categories ...... 18 Figure 10. Mean MAE values for least-sq uares (blue) and robust (red) single linear regression equations for PWSD (left) and A WSD (right). The MA E is shown for the 60 stratifications (x-axis) .. . 19 Figure I I. Mean MAE values for least-squares (blue), robust (red) and least-trimmed squares (green) multiple linear regression equations for PWSD (left) and A WSD (right). The MAE is shown for the 60 stratifications (x-axis) ...... 20 Figure 12. MAE values for least-squares (left) and robust (right) si ngle linear regression equations for timing of the PWSD. The values are shown for the stability parameters (S I -S I 9), wind speed parameters (W 1-W48) and wind shear parameter (H I). Mean MAE (blue) is the average of all stratification methods. Minimum MAE (red) is the lowest of all stratification categories.20 Figure 13 . Mean MA E values for sin gle (left) and multiple (right) linear regression equations for timing of the peak wind. The least squares and robust regressions are in blue and red, wh il e the least­ trimmed regressions are in green. The values are shown for the 60 stratifications (x-axis) ... 2 I Figure 14 . Mean Error for PWSD (left) and A WSD (right) on non-precipitation days For the Phase II methods, points I -6 (on x-axis) correspond to stratification methods SM I -SM6, respectively. The legend describes the colors corresponding to the Phase II linear regression types. Refer to

3 Table 3 for a description of the stratification methods and linear regression types. The cl imatology methods are plotted on points 1-4. The Phase I method is plotted on point I. .. . 27 Figure 15. Mean Error for PWSD (left) and A WSD (right) on precipitation days, for the climatology and Phase I and II methods. Refer to Figure 14 for a description of the x-axis ...... 28 Figure 16. MAE fo r PWSD (left) and A WSD (right) on non-precipitation days, for the climatology and Phase I and II methods. Refer to Figure 14 for a description of the x-axis ...... 28 Figure 17. MAE for PWSD (left) and A WSD (ri ght) on precipitation days, for the climatology and Phase I and II methods. Refer to Figure 14 for a description of the x-axis ...... 29 Figure 18. Mean Error for PWSD on days in which the forecast 's absolute error is within 5 kt (left) and greater than 5 kt (right). Refer to Figure 14 for a description of the x-axis ...... 30 Figure 19. Mean Error for A WSD on days in whi ch the forecast' s absolute error is within 5 kt (left) and greater than 5 kt (right). Refer to Figure 14 fo r a description of the x-axis ...... 30 Figure 20. Average difference between the PWSD and the tower-average peak speed, on days in which the forecast's absolute error is within 5 kt (left) and greater than 5 kt (right). Refer to Figure 14 for a description of the x-axis ...... 3 I Figure 2 1. Mean Error for PWSD and A WSD. The strongest 24-hour 0000 UTe MesoNAM forecast wi nds are plotted on points 1-20, in black. The 0000 UTe MesoNAM linear regression forecasts are plotted on points 1-20, in light blue. The climatology methods are plotted on points 1-4, in dark blue. The Phase I method is plotted on point I (i n red), and the Phase II methods are pl otted on points 1-30 (in yellow). Refer to Table 3 for a description of the Phase II methods ...... 32 Figure 22. MAE for PWSD and A WSD. Refer to Figure 21 for a description of the x-axis ...... 33 Figure 23. Mean Error for PWSD on precipitation (left) and non-precipitation days (right). Refer to Figure 2 1 for a description of the x-axis ...... 33 Figure 24. Mean Error for A WSD on precipitation (left) and non-precipitation days (right). Refer to Figure 2 1 for a description of the x-axis ...... 33 Figure 25. MAE for PWSD on precipitation (left) and non-precipitation days (right). Refer to Figure 21 for a description of the x-axis ...... 34 Figure 26. MAE for A WSD on precipitation (left) and non-precipitation days (right). Refer to Figure 2 1 for a description of the data points ...... 34 Figure 27. MAE fo r PWSD (left) and AWSD (right), fo r 0000 and 1200 UTe MesoNAM regression fo recasts. Points 1-1 7 depict model levels 2-18, and points 18-20 depict the strongest winds in the lowest 1000-, 2000-, and 3000-ft of the model. The legend shows the color corresponding to each method ...... 34 Figure 28. Mean Error (diamonds) and MAE (circles) for timing of the peak wind. The Phase I method is plotted on point I, Phase II methods on points 1-12, climatology methods on points 1-4 and 0000 UTe MesoNAM winds on points 1-20 ...... 37 Figure 29. The Mean Error (diamonds) and MAE (triangles) fo r th e timing of the peak wind by the 0000 and 1200 UTe MesoNAM winds ...... 37 Figure 30. MAE for PWSD on precipitation (left) and non-precipitation (right) days, for Day-I to Day-3 forecasts. Forecasts are from li near regression equati ons usi ng the 0000 UTe MesoNAM forecast winds. Points 1-1 7 depict model levels 2-18, and points 18-20 depict the strongest winds in the lowest 1000-,2000- and 3000-ft of the mode!...... 39

4 Figure 31. MAE for PWSD on precipitation (left) and non-precipitation (ri ght) days, for Day-I to Day-3 forecasts. Forecasts are ITom linear regression equati ons usi ng the 1200 UTe MesoNAM forecast winds. Points 1-17 depict model levels 2-18, and points 18-20 depict the strongest winds in the lowest 1000-, 2000- and 3000-ft of th e model...... 39 Figure 32. MAE for A WSD on prec ipitation (left) and non-precipitation (right) days, for Day-I to Day -3 forecasts. Forecasts are from linear regression equations using the 0000 UTe MesoNAM forecast winds. Points 1-1 7 depict model levels 2-1 8, and points 18-20 depict the strongest winds in the lowest 1000-, 2000- and 3000-ft of the model...... 40 Figure 33. MAE for A WSD on precipitation (left) and non-precipitation (ri ght) days, for Day-I to Day-3 forecasts. Forecasts are from linear regression equati ons usi ng the 1200 UTe MesoNAM forecast wi nds. Points 1-1 7 depict model levels 2-18, and points 18-20 depict the strongest winds in the lowest 1000-,2000- and 3000-ft of the model...... 40 Figure 34. Intro worksheet to start the Excel version of the Peak Wind Tool...... 42 Figure 35. Dialog box used to select a MesoNAM input file ...... 43 Figure 36. Dialog box used to select the fo recast day ...... 44 Figure 37. GU I displaying the output from the Excel version of the Peak Wind tool. The top section shows the predicted PWSD, the middle section shows the predicted A WSD, and the booom section shows the probability the PWSD will meet or exceed 25 kt, 35 kt and 50 kt...... 45 Figure 38. MAE for daily average speed for forecasts from 0000 (left) and 1200 UTe (right) MesoNAM. Forecasts for precipitation days are plooed on poi nts I, 4 and 7 of the x-axis. Forecasts for non-precipitation days are plooed on points 2, 5 and 8, and forecasts for all days are plooed on points 3, 6 and 9 ...... 47 Figure 39. MIDDS version of th e Peak Wind Tool showing forecasts from th e 7 July 2010 model runs. Forecasts for precipitation days have the header "Precip", while non-precipitation days have the header "NoPrecip". The header "All" includes days with and without precipitation ...... 48

5 List of Tables Table I ...... 16 Table 2 ...... 17 Table 3. Phase II methods verified for PWSD, A WSD and timing of the PWSD. Methods in red were not verified for the timing of the PWSD ...... 23 Table 4. Stratification categories used in Phase II methods. Wind direction was based on the vector average wind in the lowest 300 ft of XMR sounding. The stratification methods are described in Table 3 ...... 24 Table 5. Stratification category and predictors for each PWSD equation using single linear regression. The stratification categories are described in Table 4 ...... 25 Table 6. Stratification category and predictors for each PWSD equation using multiple linear regression. The stratificati on categories are described in Table 4 ...... 26 Table 7. Average heights and standard deviations of the MesoNAM venicallevels used in the verification data ...... 32 Table 8. Verification for days in which the highest 45 WS wind warning/advisory was 25-34 kt (left) and 35-49 kt (right). Phase II methods are shown in blue ...... 35 Table 9. Climatology methods for predicting timing of the peak wind, based on the mean peak winds in a previous AMU task (Lambert 2002). The "average wind speed" is the average of all the towers used in the previous AMU task, and the "maximum wind speed" is the maximum of all the towers ...... 36 Table 10. Prediction equations for PWSD, based on linear regressions in which the predictor is the strongest 24-hour wind speed at the MesoNAM model level, and the predictand is the PWSD. Refer to Table 7 for the height of each modelleve!...... 41 Table II . Prediction equations for A WSD, based on linear regressions in which the predictor is the strongest 24-hour wind speed at the MesoNAM model level, and the predictand is the A WSD. Refer to Table 7 for the height of each model leve!...... 41 Table 12 . MAE (kt) for daily average speed. "All Days" is al l days in the verification data set. "W. Avg" is the weighted average of precipitation and non-precipitation days. "00 UTC" is the 0000 UTC MesoNAM and "12 UTC" is the 1200 UTC MesoNAM ...... 47

6 1 Introduction The expected peak wind speed of the day is an important element in the 45th Weather Squadron' s (45 WS) daily 24,Hour and Weekly Planning Forecasts. The forecasts are used for ground and space launch operations at the Kennedy Space Center (KSC) and Cape Canaveral Air Force Station (CCAFS). The 24-Hour Forecast is valid from 0800 to 0800 local time the next day, and it is broken into six 4-hour time blocks. The Weekly Forecast is for the next six days starting with the next day and is divided into 12-hour time blocks. The 45 WS al so issues wind advisories for KSC/CCAFS and Patrick Air Force Base (PAFB) when they expect peak gusts to meet or exceed 25 kt, 35 kt and 50 kt thresholds at any level from the surface to 300 fl. The 45 WS forecasters have indicated peak wind speeds are challenging to forecast, particularly in the cool season months of October - April. In Phase I of this task (Barrett and Short 2008), the Applied Meteorology Unit (AMU) developed a tool to help them forecast non-convective winds during the cool season for the 24-Hour Forecast. The tool forecasts the highest peak wind speed, the timing of the peak speed and the 5-minute average wind speed at the time of the peak wind from the surface to 300 fl on KSC/CCAFS. In addition, it provides the probability the peak speed will meet or exceed the wind advisoty thresholds. The tool calculates its forecasts based on the morning CCAFS sounding (XMR), the synoptic weather pattern and whether or not precipitation is expected over KSC/CCAFS during the 24- hour period. For Phase II, the 45 WS requested additional observations be used in the creation of the forecast equations by expanding the period of record (POR). Additional parameters were evaluated, including wind speeds between 500 ft and 3000 fl, static stability classification, Bulk Richardson Number, mixing depth, vertical wind shear, inversion strength and depth and wind direction. Using a verification data set, the AMU compared the Phase I and II prediction methods to climatology and model forecasts. The methods that performed the best in the verification data set were selected for the Phase II tool. As in Phase I, the tool was delivered as a Microsoft Excel graphical user interface (GUI). The 45 WS requested the AMU also make the tool available in the Meteorological Interactive Data Display System (MIDDS), their main weather display system, to provide the ability to produce 5-day forecasts quickly.

7 2 Data In Phase I and Phase II, three types of observations were used: 5-minute observations from the KSC/CCAFS tower network, hourly and special surface observations from the Shuttle Landing Facility (SLF) and XMR soundings. The tower observations and soundings we re obtained from th e Range Technical Services Contractor, Computer Sciences Ray theon (CSR). Surface observati ons were obtained from the 14th Weather Squadron Strategic Cli matic Information Service. In Phase I, the data set in cl uded observations from the cool season months of October 2002 to February 2007. In Phase II , cool season data from March 2007 to April 2009 were added to the existing data set. Days in which east-central Florida was under the influence of a tropical cyclone were removed from the data set, si nce onl y cool­ season weather phenomena were of interest. The prediction equations in the Phase I tool required sounding data in 100-ft increments. To increase the size of the Phase II data set even further, the AMU evaluated whether data prior to October 2002 could be added by interpolating 1000-ft sounding data down to 100-ft increments. Sounding data between October 1996 and April 2002 are only available in 1000-ft increments, while data between October 2002 and April 2009 are available in both 1000-ft and 100-ft increments. The AMU interpolated the 1000-ft data from October 1996 to April 2008 to 100-ft increments, and then compared the interpolated data to the 100-ft data for the period October 2002 to April 2008. Since no significant differences were found between the interpolated and 100-ft data, the interpolated data from October 1996 to April 2002 were included in the Phase II data set. The 100-ft data were used for the period October 2002 to April 2009. The tool predicts th e peak wind in the 24-hour period 0800 to 0800 local time. The tool's prediction eq uati ons used observati ons fro m the 24-hour period 1300 to 1300 UTC, since the majority of days in the cool season were in Eastern Standard Time (EST).

2. 1 Tower Data Data were collected from all 32 towers used to verify wind advisories issued by the 45 WS. Their locati ons are shown in Figure I. Tower 0300, which is not used to ve rify wind advisories, was also used in ord er to provide better coverage of th e southern portion of KSC. The meteorological variables in the tower data set included temperature and relative humidity, 5-minute average and peak wi nd speeds, 5- minute average and peak wind directions, and the standard deviation of the 5-m inute average wind direction over a 30-minute period. Wind speed and direction data were sampled every second. The peak wind was the maximum I-second speed in the 5-minute period. Case and Bauman (2004) provides a detailed descripti on of the KSC/CCAFS tower network instrumentation. Since 45 WS wind advisories apply to winds up to 300 ft, only tower wind observations up to 300 ft were used in this task. Lambert (2002) describes the automated quality control (QC) algori thms and associated Fortran software to flag bad tower data prior to analysis. The AMU rewrote the QC software in th e Java programming language to make it more portable and easier to maintain. For example, Java software does not need to be recompiled in order to ru n on multiple computer operating systems. The Java version uses a configuration text fil e to store values previously hard-coded in the software, such as the years in the POR and the li st of wind tower identifiers. Thi s all ows the user to easily change the parameters by modifying the configurati on file . Previously, the user had to edit the source code and recompile the software. The AMU also fixed a bug in the Fortran software, wh ich did not properly handle the case in which a sensor's height changed during the POR. In addition, a supplementary QC algorithm was added to check for repeating observations that can occur during a sensor outage. After the QC software was run, the peak wind speed of the day (PWSD) was determ ined from the 33 towers at all levels up to 300 ft. The 5-minute average wind speed at the tower reporting the PWSD was also recorded. The AMU manually examined each PWSD of 60 kt or greater to check for erroneous outliers. If the observations from the tower appeared erroneous, then the highest daily peak wind speed from one of th e other towers was used. 8 Woo Tower locations

KSC WomingiAdvisoly Verfic-.

IIl22 .6. CCAFS _glActMory VOIIicalion .6. Non-_glActMory Towers .. " .... -'"

.6.'0'2

.6."" ~""

,- .6."" .6.2202

Figure I. A map showing the towers used in the task. Except for tower 0300 (lower-right, in blue), only the yellow- and red-colored towers were used.

2.2 Surface Observations The SLF hourl y and special observations were used to determine if precipitation occurred at or within 5 NM (the vici ni ty) of the SLF at least once during each 24-hour period. Each day was then classified as a precipitation- or non-precipitation day .

2.3 Sounding Data The XMR soundings were used to create the candidate predictors fo r PWSD, timing of the PWSD and th e 5-minute average wind at the time of the PWSD (herealler referred to as A WSD). The soundings included wind speed and direction, pressure, temperature and dew point. Only soundings between 0930 UTe and 1230 UTe were used, since soundings outside this time period may be unrepresentative due to diurnal changes in temperature, humidity and wind . If a sounding did not extend to at least 15 ,000 II mean sea level (MSL), the sounding was elim inated from the data set. If there were multiple soundings between 0930 UTe and 1230 UTe, the latest one was used. All of the sounding data needed to be in 100-11 in crements to calculate the predictors. While data from October 2002 to April 2009 were already avai lable in 100-11 increments, data between October 1996 and April 2002 were only available in 1000-11 increments. If the 1000-11 sounding data could be interpolated to 100-11 increments without a significant loss in accuracy, the interpolated sounding data would be used in the Phase II data set. The 1000-11 sounding data from October 1996 to April 2008 were interpolated to 100-11 increments up to 15,000 II MSL, using two different methods. In Method I, data at the 1000-11, sign ificant and

9 mandatory levels were linearly interpolated to 100-ft increments. In Method 2, only the significant and mandatory level data were linearly interpolated to 100-ft increments. A significant le vel occurs when there is a significant change in temperature or wind wi th height. Therefore, the number and heights of the significant levels were variable. Up through 15,000 ft MSL, mandatory levels in the XMR soundings were at 1000 mb, 950 mb, 900 mb, 850 mb, 800 mb, 750 mb, 700 mb, 650 mb and 600 mb. In July 2002, another mandatory level was added at 925 mb. The AMU compared the interpolated data to the 100-ft sounding data for the period October 2002 to April 2008. First, the AM U compared individual soundings from a small sample of seven days in the 2002/2003 cool season (October 2002 to April 2003). As an example, Figure 2 compares the 100-ft data in the I January 2003 XMR sounding to the interpolated data usi ng Method I and 2. The Method I and 2 wind speeds are nearly identical, except around 6000 ft MSL. The following conclusions were reached after analyzing data from the seven soundings: • The 100-ft data were usually similar to the interpolated data, • The interpolated data were usually closer to each other than to the 100-ft data, • The differences between the Method I and Method 2 interpolated data were largest at multiples of 1000 ft MSL, • At multiples of 1000 ft MSL, the Method I interpolated and 100-ft data were usually exactly the same, especially for temperature, dew point, and relative humidity, • Due to the hi gh vertical variability in dew point, there were occasionally large (on the order of 5 0c) differences between the interpolated and I OO-fi data, and • There were occasionall y large differences between the interpolated and 100-fi data in wind speed and direction (on the order of 5 kt and 90°).

~~- 1/1/2003 XMR Sounding, Methodl 1/1/2003 XMR Sounding, Method2 70 1,-4 70 ~60 '\. lr' I -so~ 60 r 1 1 I I r I :;so -g 40 1 " ~ 40 ...~ 30 - ~ 30 -jJ t- 1-+ l-I- ~ 20 'E 20 , i I I I 1 ~ 10 1 1 ~ 10 v.--l 0 0 I "'" ~ ~ ~ ~ ~ ~ !2 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ... 001 ...... Altitude inlt (MSLI ~ lDO-ft data Altitude in It (MSL) ~ lDO-ft data ~ Method2 ~ Met hodl I Figure 2. Comparison of 100-ft sounding data on I January 2003 to interpolated data using Method I (left) and Method 2 (right). Next, the monthly averages fo r the interpolated and 100-ft data were compared for the cool season months in the period October 2002 to April 2008. The averages were practically equal for each month, indicating no significant biases in the interpolated data. As an example, Figure 3 compares the u- wind component in the I OO-ft and interpolated data for each month. The figure shows no significant differences between the interpolated (dashed lines) and 100-ft (solid lines) data in the monthly averages.

10 All data (100-1t solid, Method1 dashed) All data (100-1t solid, Method2 dashed) 40 40 c·- r - Ja n-IOO Tn r - Jan-IOO - Feb-IOO / - Feb-IOO 30 30 B: Mar-IOO B: Mar-lOO ~ - ~1 ·100 v V' :::::1 - Apr-loo c 1: iii v. ;- - ..c 20 Oct-IOO ..c 20 Oct-IOO &. &. E Nov-IOO E t:t; ':f / Nov-lOO 0 0 ;4;:f~~:r ·· .' ..,u 10 Dec-IOO ..,u 10 Dec-lOO c ... _- Jan-Ml c v.:/v .- ' " .' " ----- Jan-M2 . ~ . ~ I I :y ".- " ,.. - - --_ .. Feb-Ml " ----- Feb-M2 • " 0 • 0 .' .- ... _- Mar-Ml m------Mar-M2 ____ . Apr-Ml i'\T: I I ____ A Apr-M2 -10 .J ._--- Oct-M1 -10 -._-- Oct-M2 IDOOOOOOOOOOOOOOO ~gg g88 888 g88 gg gg ..... Nov-Ml "':5:5:5:5:5:5:5:5:5 0:5:5:5:5:5 -- --. Nov·M2 """NM~II\ID""OOcnO"'NM~U'!I .... N M oot U'!IID"" GO en Dec-M2 ... """ """...-I", """ ----- Oe c-Ml ~::=:::: ~ ~ ~ ... __ L -- --- Figure 3. Monthly averages of height versus u-wind component, using interpolated (dashed) and 100-ft data (solid). The interpolated data used Method I (left) and Method 2 (right). The data are from the cool season months in the period October 2002 - Apri l 2008. The AMU compared the interpolated and 100-ft data usi ng mean absolute error (MAE). The MAE is the mean of the absolute value of the differences between the interpolated and 100-ft data. The MAE monthly averages for wind varied by height, but were generall y between 0.5 and 1.0 kt. As an example, Figure 4 shows the MAE monthly averages of the u-wind component versus height. The fi gure shows very little scatter in the monthly averages, and the MAE of the Meth od I data was smaller than the Method 2 data. The MA E minimums correlated to the mandatory pressure levels. There were also secondary MAE min imums for Method I at multiples of 1000-ft MSL. While the monthly averages of MAE contained little scatter, MAE for ind ividual months had more scatter. For example, Figure 5 shows a fai rly large amount of scatter in the MAE for the u-wind component in December.

MAE of u-wind compone nt in Methodl MAE of u-wind component in Method2 2.5 2,5 I I I I I I I I I - Jan I - Jan I I I I 2 2 - feb r----- ,I --+-t • -,---, - Feb Mar g 1.5 - Mar I w \; - Apr .c-- ~ >/A - Apr 1 1A-tJi!+-+--~-hif.. ~ 1 8. Oct ~~M Oct 0.5 0.5 - Nov ~ - Nov o - Dec ~ooooooooooooooo.... ggggggggggggggg .... NM'ltU'l\.D,...COC'lC.-4Nrt'I'ltU'l .-4.-4.-4.-4.-4.-4 Figure 4. Monthl y averages of height versus MAE in u-wind component, using interpolated and 100-ft data. The data were interpolated using Method I (left) and Method 2 (right).

I I December MAE of u-wind component, Methodl December MAE of u-wind component, Method2 2.5 2.5 - Oec2002 - Oea002 2 H+-H++H~-~-I--1 --Oec2003 2 - Oea003 ~ B:1.S Oec2004 _ 1.5 Dec2004 "' "' ~ 1 - Oec2005 ":Ii 1 , - DeaOOS 0.5 Oec2006 0.5 Oea006 - Oec2007 o ,-+++-H++-H++-H+-i --Oec2007 o ~ '-'-'--'-'---.L.C-'---L.L--'---'-'--L..J ... 0 0 0 0 0 0 0 0 0 0 0 0 0 0 o-OecAil .... 000000000000000 ~gg ggggg g ggg gggg -OecAll 000000000000000 000000000000000 .... NMOIt"'''O,...OOcnO .... NMOIt''' .... NM'ltIl'l"O,...OOcnO .... NMOIt'" ...... -1 ......

Figure 5. The December plots of height versus MAE in u-wind component, using interpolated and 100-ft data. The data were interpolated using Method I (left) and Method 2 (right). The 2002-2007 December average is displayed by the black li nes. When averaging the MAE over the cool-season months from 2002 to 2008, comparing the MAE of Method I and 2 showed a clearer pattern. Figure 6 compares the MA E of the u- and v-wind components using the Method I and 2 interpolated data. The MAE pattern was practically the same between the u­ wi nd (left) and v-wind (right) components, and the MAE of Method I was lower than Method 2. At the mandatory levels the MA E was the same in the two interpolation methods, which was expected since both methods used mandatory level data. The difference in MAE between the two methods was maximized near multiples of 1000-ft MSL, unless a mandatory level happened to be nearby. Figure 7 compares the MAE of the temperature and dew point data. Method I had the lowest MA E for temperature and dew point. The MA E values were small for Method I and 2, on the order of 0.1 °C and 0.2-0.6 °C for temperature and dew point, respectively.

MAE of u-wind, averaged over all months MAE of v-wind, averaged over all months 2.5 2.5 , I- Methodl l - Method~ 1 I I Method2 2 Method211 I I I B: 1.: ------1-1- B: 1.5 r"'1 "' 'if ,/ l 'j Ii "' r'l i f'o If'" '" r f\ ~ 1 ":Ii 1 '" '" .r\ (\ V V IV-"'V\ \J NY 'V 0.5 I~ v'V\. 1\. V W 0.5 ~ Vr--Y rw I I o I I I fr- 0 00 0 0 0 ... 0 0 g 0 0 0 0 0 g g g g g 0 g ... g g 00 g 0 0 0 ... 0 0 0 00 0 0 0 ... 00 00 0 0 0 0 §§§§§§ 0 0 0 0 00 0 0 0 0 00 o .... N "" 'It U"I ... N .... 00 i 8 ~ ~ g ~ N '" '" ...... '" ...... -----.. '" '" '" Figure 6. Height ve rsus MAE in u- (left) and v-wind (right) components, using Method I (red) and Method 2 (green) interpolated data. The interpolated data were averaged over all months in the POR.

12 MAE of temperature averaged over all months MAE of dew point averaged over all months j: 1 1 1 I I I I I I Methodl 1 0.8 ~ Methodl - Method2 v-.. 11'\ 1(\1 - Method2 J r, p- f'" ~O . 6 '\ r' /', ~O.l w r' (" r f\ :; y ~ 0.4 Irl i\ ~ I ~ 'A . ( ~ tJ rlI N IA ~ ~ ~ 0.2 IfV' VV IV 0 '---'- __ L- 0 I OD 0 0 0 000 0 0 0 OD 0 g 0 0 0 g 0 0 0 0 0 0 0 0 0 ~ 0 0 0 0 ~ 0 0 0 0 0 0 0 0 0 0 0 0 0 g 0 0 g g g 0 g 0 §§§§§§ 0000 0 0 0 0 0 0 0 0 0 0 0 .-I f'I'I OD ~ N ~ N r- oo o r4 N 1'1"1 o::t U"I N o::t r- oo 0 '" '" '" OD '" ...... -4 .... """ .... '" '" ~ ~ ~ '"~ '"~ '"~ Figure 7. Height versus MA E in temperature (left) and dew point (right), using Method I (red) and Method 2 (green) interpolated data. The interpolated data were averaged over all months in the POR. For temperature and dew point, the MAE of Method I was near 0 °C at mUltiples of 1000 ft MSL. On the other hand, for u- and v-wind components, the MAE of Method I was between 0.5 and 1.0 kt at multiples of 1000 ft MSL. The AMU investi gated the reasoning behind this. According to CSR staff, the 1000-ft winds are smoothed by averaging over a 1000-ft interval (Mr. Rick Kulow of CSR, personal communication). For example, the 3000 ft wi nd is derived by smoothing the raw wind data from 2500 ft to 3500 ft. The other meteorological variables, such as temperature or dew point, are not smoothed in the 1000-ft data. For all of the meteorological variables, raw data (i .e. not smoothed) are used for the mandatory, significant and 100-ft data. The AM U made the following conclusions after analyzing the monthly averages of MAE for the interpolated data: • For all of the meteorological variables except pressure, a minimum in MAE occurred at the mandatory levels and a maximum in MAE occurred halfway between mandatory levels, • A secondary minimum in MAE occurred in the Method I interpolated data at mUltiples of 1000 ft MSL, • For pressure, the maximum in MAE occurred at mandatory levels and the minimum occurred hal fway between the mandatory levels, and • For all meteorological variables, the Method I interpolated data's MAE was lower than the Method 2 interpolated data. Since no significant differences were found between the interpolated and 100-ft sounding data, the AMU used interpolated sounding data for the October 1996 to April 2002 period and 100-ft data for the October 2002 to April 2009 period. Method I was used to interpolate the sounding data since it had the lowest MA E values.

13 3 Development of Phase II Prediction Equations The QC'd observation data were used to develop prediction equations for each forecast parameter. The developmental data set, consisting of cool-season observations from October 1996 to February 2007, was used to develop the prediction equations. The verification data set, containing cool-season observations from March 2007 to April 2009, was later used to independently verify the performance of the Phase 11 equations.

3.1 Synoptic Weather Pattern The AMU stratified each day in the data set by synoptic weather pattern to develop separate prediction equati ons for each weather pattern. The surface synoptic weather pattern at 1200 UTC was analyzed for each day using surface charts obtained from the National Weather Service (NWS) Hydrometeorological Prediction Center's short-term and long-term archives (http://docs.lib.noaa.gov/rescue/dwm/datarescuedailyweathermaps.htm!). Each day was categori zed into one of the fo llowing: • High pressure over Florida, with light and variable winds across east-central Florida, • High pressure to the north or west of central Florida, with northwest, north, northeast, or east winds across east-central Florida, • High pressure to the south or east of central Florid a, with southeast, south, southwest, or west winds across east-central Florida, • Cold front over north Florida, • Cold or stati onary front over central Florida, • Cold or stationary front over south Florida or Florida Keys, • Tropical storm or hurricane affecting Florida, and • Surface weather map unavailable. On ly one day had a missing surface weather map. The days in which a tropical storm or hurricane affected Florida were removed from the data set to make the analysis more representative of the cool season.

3.2 Pbase II Predictors The AMU wrote and executed Perl scripts to calculate the candidate predictors from the XMR soundings. The predictors were evaluated for their skill in predicting th e PWSD, A WSD and timing of the PWSD. The predictors included 68 wind shear, stabi lity and wind speed parameters. There were 19 stability parameters (identifiers in parentheses): • Temperature inversion depth and strength (S I - S2), • Differences in temperature between 1000 ft and 16, 100, 200, 300,400 and 500 ft (S3 - S8), • Differences in temperature between 2000 ft and 16, 100, 200, 300, 400 and 500 ft (S9 - S 14), • A fternoon mixing height (S 15 - S 17), and • Gradient Richardson Number (S 18 - S 19). There were 48 wind speed parameters: • Maximum wind speeds from the surface to 500 ft, 1000 ft, 2000 ft and 3000 ft (W I - W4), • Maximum wind speed between 1000 ft and 2000 ft (W5), • Maximum wind speed between 2000 ft and 3000 ft (W6), • A verage wind speed from the surface to 500 ft and 1000 ft (W7 - W8),

14 • A verage wind speed between 500 ft and 1000 ft (W9), • A verage wind speed between 1000 ft and 2000 ft (W I 0), • A verage wind speed between 2000 ft and 3000 ft (W I I), • Wind speeds at 16 ft through 3000 ft, by 100-ft increments (W 12 - W42), • Wind speed at the afternoon mixing height (W43 - W45), and • Maximum wind speed from the surface to afternoon mixing height (W46 - W48). There was one wind shear parameter: wind shear between the surface and 1000 ft (H I). Table I describes the stability and wind shear parameters, while Table 2 describes the wind speed parameters.

15 Table I. Stability and wind shear parameters calculated from XMR soundings. Heights in MSL.

Parameter Description name

iov.depth Depth of the surface-based inversion in ft. The top of the inversion was defined as the (SI) first level in which the temperature did not increase with height. inv.str.C (S2) Strength of the surface-based inversion in °C, defined as the difference between the temperatures at the surface and top of the inversion.

XI6 ... IOOO to The difference between the temperature (0C) at 16 ft (or 100 ft , 200 ft, 300 ft, 400 ft, and X500 ... I OOO 500 ft) and 1000 ft. (S3 - SS)

XI6 ... 2000 to The difference between the temperature CC) at 16 ft (or 100 ft, 200 ft, 300 ft, 400 ft, and X500 ... 2000 500 ft) and 2000 ft. (S9 - S14)

mixhgtLI Morning mixing height (in ft) , defined as the height at which the temperature lapse rate (S IS) first meets or exceeds 4 °C/km . This was based on an average saturated adiabatic lapse rate of 4-5 °Clkm near the surface In temperature above freezing ( htl~: lIen. wi ki~ed i a.orglwiki / La~se rate).

mixhgtL2 Morning mi xing height (in ft) calculated by the same method as mixhghtLl, except data (S16) was in 200-ft increments instead of I ~O-ft.

mixhgtT Afternoon mixing height based on the forecast maximum temperature, in ft. The forecast (S17) maximum was derived by adding 2 degrees Kelv in to the average boundary-layer potential temperature.

brl (SIS) Bulk Richardson Number (BRN) for the lowest 1000 ft in the sounding. The BRN was calculated by averaging the gradient Richardson number between the surface and 1000 ft ( htl~ :i/amsglossa!), .al l~n~re ss . com/glossa!}' /searc h ? i d=bul k-richard son-num ber I ). brlmod (SI9) The BRN was calculated by the same method as brl, except values greater than 5.0 were set to 5.0 in order to improve the linear relationship to peak wind speed.

diff.sCc.1 k Difference between 500- I 000 ft and surface-SO~ ft layer-average wind speeds. (H I)

16 Table 2. Wind speed parameters calculated from XM R soundings. Heights in MSL.

Parameter Description name

max.sfc.Sh Maximum wind speed from the surface to 500 ft . (WI)

max.sfc.lk Maximum wind speed from the surface to 1000 ft. (W2)

max.sfc.2k Maximum wind speed from the surface to 2000 ft. (W3)

max.sfc.3k Maximum wind speed from the surface to 3000 ft. (W4)

max.lk.2k Maximum wind speed between 1000 ft and 2000 ft. (WS)

max.2k.3k Maximum win d speed between 2000 ft and 3000 ft. (W6)

ave.sfc.Sb A verage wind speed from the surface to 500 ft. (W7)

ave.Sh.lk A verage wind speed from 500 ft to 1000 ft. (WS)

ave.sfc.lk A verage wind speed from the surface to 1000 ft. (W9)

ave.1.2k A verage wind speed from 1000 ft to 2000 ft. (WIO)

ave.2.3k Average wind speed from 2000 ft to 3000 ft. (Wll)

Xl6.ft to Wind speeds at 16 ft through 3000 ft. (W12 - W42) X3000.ft

spdmixLl Wind speed at the height defined by mixhgtLl. (W43)

spdmixL2 Wind speed at the height defined by mixhgtL2. (W44)

spdmixT Wind speed at the height defined by mixhgtT. (W4S)

maxspdLl Maximum wind speed from the surface to the height defined by mixhgtLl. (W46)

maxspdL2 Maximum wind speed from the surface to the height defined by mixhgtL2. (W47)

maxspdT Maximum wind speed from the surface to the height defined by mixhgtT. (W4S)

The data set was stratified using the synoptic weather pattern, low-level wind direction, precipitation, Richardson Num ber and Gradient Richardson Number, for a total of 60 stratificati on methods. The total number of observation days in each stratification method varied due to missing or undefi ned data. For example, if the wind shear was zero, the Richardson and Gradient Richardson numbers could not be calcul ated. The AMU wrote scripts in the S-PLUS programming language to calculate the least-sq uares single linear regression equations for the PWSD, A WSD and the timing of the PWSD. Based on the lowest MAE of the equations, the best predictors were selected to create multiple linear regression equations. Single and multiple linear re gression equations were also created usin g robust functions in S-PLUS . Unlike least-squares functions, robust functions reduce the infl uence of data outl iers, while still providing a good fit to most of the data points (Insightful, 2007).

17 3.3 Prediction Eq uations fo r PWSD and A WSD Figure 8 shows the MAE values fo r predicting the PWSD and A WSD, usi ng least-squares single linear regression equations. The predictors shown on the x-axis incl ude the stability parameters (S I - S 19), wi nd speed parameters (W I - W48) and the wind shear parameter (H I). The mean MAE val ues (in blue) were calculated fo r each stratification method, and then all 60 strati fication methods were averaged together. The min imum MAE values (in red) were the lowest values fro m all strati fication methods. The wind speed parameters performed the best, since they had the lowest MAE values.

I Peak Speed MAE (kt ), least-squ ares regressi on Ave. Speed MAE (Id), least-squares regression 8 I111111111 8 7 7 6 11 1111111 111 1 . ~ 6 5 I 5 4 4 3 3 2 2 1 , 1 0 0 ~ ~ ~ ~ ~ ~g~ ~ ~ U ~ ~ ~ ~ ~ ~ ~ :z: ~~"'~~~~~ ~~~~ ~ g gg~~~~~ ~ :z: '" g g g .. ~ ~W~~~ b:~~N~~~~~ae ~ ~ ~OW~~ A b:~~N~~ ~A ~ 8 e '" • Mean MAE • Minimum MAE • Mean MAE • Minimum MA E

Figure 8. MAE val ues for least-squares single li near regression equations fo r PWSD (left) and A WSD (right). The MAE is shown for the stability parameters (S I-S I9), wind speed parameters (WI-W48) and wind shear parameter (H I). Mean MAE (blue) is the average of all strati fication methods, while Minimum MAE (red) is the lowest of all stratification categories. Figure 9 shows the MAE values for predicting PWSD and A WSD, using robust si ngle linear regression equations. The robust method is the "lmRobM M" functi on in the S-PLUS language. The ImRobMM function creates an equation with the fo ll owing features: • The data fit to the equation are minimally infl uenced by outliers and • The equation min imizes the maxim um possible bias of the coefficient estimate (i. e. the slope of the li ne). The win d speed parameters performed the best, with the lowest MAE values. The minimum MAE values (in red) were simil ar fo r PWSD and A WSD, but the mean MAE values (in blue) fo r A WSD were lower. ,- PeakSpeed MAE (kt), robust regressi on I Ave . Speed M AEtkt), robust regression 8 8 I 7 , 7 , 6 6 ...I. 5 5 I I ii' 4 4 ~ .... jIo'i 3 3 ~!m i, ~- 2 ~ , 2 1 , 1 0 0 ~ ~ ~ ~ ~ ~ ~g~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ i!i ~ :z: O WG'lIDA ...... g N g ..... AAA ~ ~ ~ ~ ~ ~ ~g ~ ~ ~ ~ ~ H g g ~ ~ ~ ~ ~ ~ CWG'I O Wa'lIDNVl OO ...... IoU .. O WO'I IDNVlOO ...... OWG'l '" I • Mean MAE • M inimum MAE • Mean MAE • Mini mum MAE

Figure 9. MAE values fo r robust si ngle linear regression eq uat ions fo r PWSD (left) and A WSD (ri ght). The MAE is shown for the stabi lity parameters (S I-S 19), wind speed parameters (W 1-W48) and wind shear parameter (H I). Mean MAE (blue) is th e average of all stratificati on categories, while Minimum MAE (red) is the lowest of all stratification categories.

18 The MAE value for each stratification method was calculated by averaging the MAE values from all 68 stability, wind speed and wind shear parameters. The MAE values were weighted by the size of the categories in the stratification methods. Figure 10 shows the least-squares method (in blue) performed better than the robust method (in red), with MAE values between 0.5 and 1.0 kt less than the robust method.

Peak Speed MAE (kt), single li near regression Ave. Speed MAE (kt), single linear regression 8 8 ~I I I I ,- 7 -- 7 I-- ~ - ---t- 6 I II 6 T ...... • . :. ~ ... ••••••• 4 ·. : _I; ·~:.:..; .' •• •+ :• •: :: •• •• •••• .. . .. I .1••• 5 -f,,~ ot!...! " . 5 ... ." .. • • .. " ...... ••• ••••• ...... 4 J 4 • • ." 3 I I 3 - I .. .. - 2 2 " - 1 1 1 I 1 I I 0 H-i- 0 0 ...... N N N ...... 0 ...... N N N ...... N 0 N 0 N ...... N N N ...... 0 .. 0 ...... 0 0 '" .. .. '" .. .. '" '" '" '" .. .. '" .. .. '" '" '" • MAE ~ast·sq u ares • MAE robust • MAE least-squares • MAE robust Figure 10. Mean MAE values for least-squares (blue) and robust (red) single linear regression equations for PWSD (left) and A WSD (right). The MAE is shown for the 60 stratifications (x-axis). Three different multiple linear regression methods were evaluated: least-squares, robust and least­ trimmed squares. The least-squares method uses the "step" function in S-PLUS. The step function performs a stepwise multiple linear regression to determine which predictors to use in the regression equation. Due to limitations of the S-PLUS software, only 26 predictors (out of68) were evaluated by the step fu nction: SI-S6, S9-S12, SI9, W2-W6, W8-WII, W22, W27, W32, W37, W48 and HI. The 26 predictors were chosen based on their performance in the single linear regression equations. The robust method combined the step function with the "1m Rob" function in S-PLUS. The ImRob function is similar to the ImRobMM function, and is available from the Robust library in S-PLUS. The S­ PL US 6 Robust Library User's Guide explains the use of the ImRob function (Insightful, 2002). The ImRob function only used 10 predictors (S I-S4, SI9, W4, W6, WI I, W37 and W48), due to the limitations of the S-PLUS software. The 10 predictors were chosen based on their performance in the single linear regression equations. The least-trimmed squares method is sim il ar to the least squares regression, and it is implemented by the "Itsreg" function in S-PLUS. The Itsreg function first removes observations corresponding to large errors to the linear fit. It then calculates a least squares regression equation using the remaining observations. The Itsreg function used the same 10 predictors as the robust method. Unlike the least­ squares and robust methods, the Itsreg function does not use stepwise regression to determine which predictors to use in the multiple regression equations. Instead, 10 predictors were used in each least­ trimmed squares regression equati on. Figure II shows the MAE values of PWSD and A WSD for the three multiple linear regression methods. The values from the least-squares (in blue) and least-trimmed squares (in green) methods were almost identical and overlap in the chart, while the robust (in red) method had MAE values higher than the other two methods.

19 ,--- Peak Speed MAE (kt), multiple linear regression Ave. Speed MAE (let), multiple linear regression 8 ,- , 8 7 I 7 6 --- -+ n 6 ., ...... • •• ,1. " .. ~ 5 ' .~ .. i...-'t'· .., • '. ... ,., ... .. ' • ••••••••• •• • • 5 • .• ••••• • "t'. 4 • 4 ...... " 3 _1 3 2 2 1 I I 1 o I I I 0 ., ...... A ~ ~ ~ N N N ~ W A A A ~ ~ ~ ...... , .. .,.. N ~ 0 A ~ N ~ 0 A ~ N ~ 0 " .... " ...... '".... '" '" " • MAE least·squares'" • MAE robust'" "• MAE leasHrimmed'" " • MAE least-sq uares • MAE robust • MAE least-trimmed

Figure II. Mean MAE values for least-squares (blue), robust (red) and least-trimmed squares (green) multiple linear regression equations for PWSD (left) and A WSD (right). The MA E is shown for the 60 stratifications (x-axis).

3.4 Timing of PWSD The prediction equations for the timing of the PWSD predict the number of hours elapsed since the beginning of the forecast peri od at 1300 UTe on the current day. The number of hours that have elapsed can then be converted to clock time, in UTe. For example, if an equation predicts 17 hours have elapsed, then the PWSD is expected to occur at 0600 UTe (or 0 I 00 EST). Figure 12 shows the MAE (in hours) for predicting the timing of the PWSD, using least-squares (left) and robust (right) single linear regression equati ons. The mean MAE values (in blue) were the average from all stratification methods, and the minimum MAE values (i n red) were the lowest values from all stratifications. There was little difference in MAE among the 68 predictors. The mean coefficient of determination (R' ) values were below 0.1 , indicating the regression equations provided little ski ll in predicting the timing of the PWSD (figures not shown). - Timing of Peak Wind MAE (hr), least-sq. reg. TIming of Peak Wind MAE (h r), robust 7 - 7 111111 I 1 1 6 6 , , ~ 11 5 ~ II ~ 5 4 4 3 1 3 2 2 1 !1 1 1 1 1 1 , 1 I I , o 0 ~ \:: ~ ~ ~ ~ ~i!l:li ~ :Ii :Ii :Ii :Ii :Ii :Ii :Ii :Ii :Ii ~ :Ii :Ii :Ii % ~ \:: ~ ~ ~ ~ ~i!l:li ~ :Ii :Ii :Ii :Ii :Ii :Ii :Ii :Ii :Ii :Ii :Ii :Ii :Ii % ow~~~ ~~~~NNNWW A_A~ ow~~. ~~~~NNNWWW.A .. ~ OWO'IIDNVlO:O ...... OWO'l OWO'IIDNllloo ...... O'"'" .. • Mean MAE • Minimum MAE • Mean MAE • Minimum MAE Figure 12 . MAE values fo r least-squares (left) and robust (right) single linear regression eq uations for timing of the PWSD. The values are shown for the stability parameters (S I-S 19), wind speed parameters (W 1-W48) and wind shear parameter (H I). Mean MAE (blue) is the average of all stratification methods. Minimum MAE (red) is the lowest of all stratification categories. Figure 13 shows the MAE values for the 60 stratification methods, for single (left) and multiple (right) linear regressions. All of the linear regressions performed about the same, with MAE values mostly between 5.5 and 6.0 hours.

20 - Timing of Peak Wind MAE (hr), single linear reg. Timing of Peak Wind MAE (hr), multiple lin. reg. 7 7 ~ ~ :II T1Tr- , 6 6 r'" IIIIIIt.. ft' "li:1 mll.1 ft. , " ,.If 5 5 f-+---+-,,'t""" 4 ~~+l;~~~t~ ~i 4 f--- -'---'-I- .-.-f-. ~- 3 T 3 i 2 2 I I ' I I ! I 1 1 I I i o I 1 o 0 co ...... 0 co ...... 0 0 ...... 0 .. co ....'" '" ..0 .. ..co ....'" '" '"0 ...... co ....'" '" ...... co ....'" '" '"0 • MAE'" least-squares • '"MAE robust '" • MAE least-squares'" • MAE robust'" • MAE least-trimmed'" Figure 13. Mean MAE values fo r single (left ) and mul tiple (right) linear regression equati ons for ti mi ng of the peak wind. The least squares and robust regressions are in bl ue and red, while th e least­ tr immed regressions are in green. The val ues are shown fo r the 60 stratifications (x-axis).

21 4 Independent Verification of Prediction Methods The Phase II developmental data set used observations from October 1996 to February 2007, whi le the Phase I data set used observations from October 2002 to February 2007. An independent verification data set was used to compare the Phase I and II prediction methods to climatology, model forecast winds and wind advisories issued by th e 45 WS . The verification data set used observations from March 2007 to April 2009. The prediction methods showing th e greatest skill in the verification data set were chosen for the Phase II version of the tool. Several Phase II prediction methods for the three forecast parameters (PWSD, A WSD and timing of the PWSD) were selected for the independent verification based on the lowest MA E val ues in the developmental data set. The difficulty or subjectivity in how the predictors were calculated also affected th e selection, to make it easier to automate the tool. For example, the prediction methods usi ng the synoptic weather pattern performed well, however the classification of the weather pattern is somewhat subjective. Two trained, experienced fo recasters could classifY the weather pattern on a given day differently. In addition, stratification methods with more than 10 categories were not selected in ord er for each category to have a sufficie ntly large sample size. The stratification methods fo r precipitation and low-level wind direction performed well in the Phase II methods, so they were se lected for the verificati on. The stratificati on '·none", whi ch included all days in the Phase II data set, was used as a control in the verification. Table 3 shows the 30 Phase II predicti on methods selected for the verification of PWSD, A WSD and timing of the PWSD, based on five linear regression ty pes and six stratification methods. For predicting the timing of the PWSD, the multiple linear regression equations did not perform significantly better than th e single li near regression equati ons. Therefore, the mUltiple linear regressions were not independently verified for the timing of the PWSD. These regressions are indicated in red in the left column of Table 3. Table 4 describes the stratification categori es for each stratification method listed in Table 3. Table 5 li sts the predictor for each single li near regression equation for PWSD, and Table 6 lists the predictors for the multiple linear regression equations.

22 Table 3. Phase II methods verified for PWSD. A WSD and timing of the PWSD. Methods in red were not verified for the timing of the PWSD.

Method Name Regression Type Stratification Method precipt.best Least-squares single linear By precipitation and wind direction: NE. SE. SW and NW (SMl) precipl.best Least-squares single linear By precipitation and wind direction: N. E. Sand W (SM2) preeip3.best Least-squares single linear By precipitation and wind direction: NE. SE. SW. NW and wind speed ~ 6.0 kt (SM3) precip4.best Least-squares single linear By precipitation and wi nd direction: N. E. S. Wand wind speed < 6.0 kt (SM4) none.best Least-squares single linear No stratification (SM5) precipcon.best Least-squares single linear Mean of the first four stratifications (SM6) ~ precipl.bestrob Robust single linear SMI precipl.bestrob Robust single li near SM2 precip3.bestrob Robust single linear SM3 preci04.bestrob Robust single linear SM4 none. best rob Robust single linear SM5 preeipeon.bestrob Robust single linear SM6 precip1.step Stepwise least-squares multiple linear SMI preeipl.step Stepwise least-squares multiple linear SM2 precip}.step Stepwise least-squares multiple linear SM3 preci04.step Stepwise least-squares multiple linear SM4 none.step Stepwise least-squares multiple li near SM5 preeipeon.step Stepwise least-squares multiple linear SM6 precipt.steprob Stepwise robust multiple li near SM I precipl.steprob Stepwise robust multiple linear SM2 precip3.steprob Stepwise robust multiple li near SM3 preeip4.steprob Stepwise robust multiple linear SM4 Done.steprob Stepwise robust multiple li near SM5 precipeon.steprob Stepwise robust multiple linear SM6 preeipl.ItsrOl! Least-trimmed multiple linear SMI precipl.1tsrOl! Least-trimmed mu lt iple li near SM2 precip3.1tsre2 Least-trimmed mu ltiple linear SM3 preeip4.1tsreg Least-trimmed multiple linear SM4 none.ltsrOl! Least-trimmed multiple linear SM5 precipeon.ltsre2 Least-trimmed multiple li near SM6

23 Table 4. Stratification categories used in Phase II methods. Wind direction was based on the vector average wind in the lowest 300 ft of XMR sounding. The stratification methods are described in Table 3.

Category / Category / Category Description Category Description Stratification Stratification CO / SMS all days in data set C I8 I SM3 precipitation, SE wind direction, wind speed > 6 kt CI / SMI precipitation, NE wind direction CI9 I SM3 precipitation, SW wind direction, wind speed > 6 kt C2 / SMI precipitation, SE wind direction C20 / SM3 precipitation, NW wind direction, wind speed > 6 kt C3 / SMI precipitation, SW wind direction C2 11 precipitation, wind speed ~ 6 kt SM3&SM4 C4 I SM I precipitation, NW wind direction C22 I SM3 no precipitation, NE wind direction, wind speed > 6 kt CS / SMI no precipitation, NE wind C23 I SM3 no precipitation, SE wind direction direction, wind speed > 6 kt C6 / SMI no precipitation, SE wind direction C24 I SM3 no precipitation, SW wind direction, wind speed > 6 kt C7 / SMI no precipitation, SW wind C2S I SM3 no precipitation, NW wind direction direction, wind speed > 6 kt C8 / SMI no precipitation, NW wind C26 I no precipitation, wi nd speed ~ 6 kt direction SM3&SM4 C9 / SM2 precipitation, N wind direction C27 /SM4 precipitation, N wind direction, wind speed > 6 kt C IO I SM2 precipitation, E wind direction C28 I SM4 precipitation, E wind direction, wind speed > 6 kt CII I SM2 precipitation, S wind direction C29 I SM4 precipitation, S wind direction, wind speed > 6 kt CI2 I SM2 precipitation, W wind direction C30 I SM4 precipitation, W wind direction, wind speed > 6 kt CI3 / SM2 no precipitation, N wind direction C3 1 I SM4 no precipitation, N wind direction, wind speed > 6 kt CI4 I SM2 no precipitation, E wind direction C32 I SM4 no precipitation, E wind direction, wind speed > 6 kt CIS I SM2 no precipitation, S wind direction C33 I SM4 no precipitation, S wind direction, wind speed > 6 kt C I6 I SM2 no precipitation, W wind direction C34 I SM4 no precipitation, W wind di rection, wind speed > 6 kt CI7 / SM3 precipitation, NE wind direction, wind speed > 6 kt

24 Table 5. Stratification category and predictors for each PWSD equation using single linear regression. The stratification categories are described in Table 4. Stratification Category / Stratification Category / Method Nome Method Name Predictor Predictor precip I.best CII W42 precip I.bestrob CI / W42 precip I.best C2 / W39 precip I.bestrob C2 / W38 preci pl. best C3 1 WI8 precip I.bestrob C3 1 SI9 precip I.best C4 1 W4 precip I.bestrob C4 1 W3 precip I.best C5 / W25 precip I.bestrob C5 / W25 precip I.best C6 / W24 precip I.bestrob C6 1 W4 precip I.best C7 / W39 precip I.bestrob C7 / W35 precip I.best C8 1 W9 precip I.bestrob C8 / W3 precip2.best C9 / WI2 precip2.bestrob C9 / WI2 precip2.best CIO 1 W42 precip2.bestrob C IO/ W42 precip2.best CII I W48 precip2.bestrob C II I WI8 precip2.best CI2 / W37 precip2.bestrob C 12 / W23 prec i p2. best CI3 / W23 precip2.bestrob C I3 1 W3 precip2.best C14 / W25 precip2.bestrob C I4 1 W24 precip2.best C15 / W4 precip2.bestrob CI5 1 WIO precip2.best C16 / W48 precip2.bestrob C I6 1 WIO precip3.best C I7 1 W7 precip3.bestrob C I7 1 WI3 precip3.best C18 / W5 precip3.bestrob C I8 1 W2 prec ip3. best C 19 / SI8 precip3.bestrob CI9 1S19 precip3.best C20 / W30 precip3.bestrob C20 1 W3 prec ip3. best C21 / W37 precip3.bestrob C21 1 W38 ]Jrec ij>3. best C22 / W25 precip3.bestrob C22 / W25 precip3.best C23 1 W24 precip3.bestrob C23 1 W4 precip3.best C24 1 W38 precip3 .bestrob C24 / W38 precip3.best C25 1 W9 precip3.bestrob C25 / W9 precip3.best C26 1 W3 precip3.bestrob C26 / W28 precip4.best C27 / W48 precip4.bestrob C27 1 W3 precip4.best C28 / W4 2 precip4.bestrob C28 1 W13 prec ip4. best C29 1 WI precip4.bestrob C29 / WI8 prec ip4. best C30 / W42 precip4.bestrob C30 1 W42 precip4.best C2 11 W37 precip4.bestrob C2 1 1 W38 precip4.best C32 I W8 precip4.bestrob C32 1 W8 prec i p4. best C33 / W25 precip4.bestrob C33 1 W9 precip4.best C34 / W24 precip4.bestrob C34 1 W24 precip4.best C35 1 W48 . Qfecio4.bestrob C35 / W48 precip4.best C26 1 W3 orecip4.bestrob C26 / W28 none.best CO I W4 none.bestrob CO l W3

25 Table 6. Stratification category and predictors fo r each PWSD equati on using multiple li near regression. The stratification categories are described in Table 4.

Stratification Category / Stratification Category / Method Name Method Name Predictor(s) . Predictor(s) precip i.step C I / S2,W4 W5 , WI I precip I.steprob CI / SI,W4, WI I precip i.step C2 / W6 precip I.steprob C2 / SI , W37 precip i.step C3 / S19, W9 precip i.steprob C3 I S3 , S4, S19, W6, WII precip i.step C4 I W37 precip i.steprob C4 / S19, W4 precip i.step C5 I WIO precip I.steprob C5 / S19 , WI I precip i.step C6 / S9, W8 prec ip I.steprob C6 / W4 , W6, WII precip i.step C7 / SII,WII precip i.sleprob C7 I S2, S3, W II precip i.slep C8 / S10, W9, W48 precip i.sleprob C8 I S3, W6, W48 precip2.step C9 I S5, W3, W48 precip2.sleprob C9 / SI,SI9,W4 precip2.slep C IO I S2, WII precip2.sleprob C IO/ SI , WII precip2.slep C I I I SI9, W9, W22, W48 precip2.sleprob C II / S I9,W48 precip2.slep CI2 I W37 precip2.steprob CI2 I S2, S3 , S4, W6 precip2.slep CI3 / S4, SI9, W22 precip2.steprob CI3 I SI S2, S3, W48 precip2.slep C I4 I HI , S12, S19, W2, W5, precip2.sleprob CI4 I S4, S19, W4, W6, WI I, W22, W32, W37 W37 precip2.step CI5 I S2, W4, W9 precip2.steprob CI5 / S2, S3 , SI9, W4, W6, WI I precip2.step C I6 1 W9, W48 precip2.sleprob C 16 / S4 , W6, W37, W48 precip3.slep CI7 l SI , W2, W3, W4 precip3.sleprob CI7 / S4,519, W4 precip3.slep CI8 / W5 precip3.sleprob C I8 / SI , W4, W6, W37 precip3.slep CI9 I W48 precip3 .sleprob C I9 I 53, S4 S19, W6 precip3 .slep C20 I S4, S5 , W37 precip3.sleprob C20 / S4 , S19, W4, W6, WI I precip3.step C21 1519 , WIO, W37 precip3.sleprob C2 1 I S2, WII , W37 precip3.slep C22 I W5 precip3.sleprob C22 / SI , 52,S3, WI I, W48 j)recip3.sle~ C23 I W8 precip3.steprob C23 I W4 W6, WI I precip3.step C24 I S12, WII . jlrecij)3.steprob C24 I S2, S3 , WI I precip3.step C25 I S9, W6, W9, W48 precip3.steprob C25 l SI, S3 , W6, W48 precip3 .step C26 I S5, S 12, W I I precip3.steprob C26 l SI , S3 S4, W4, W6 W37 prec ip4.step C27 I S5, S19, W48 precip4.sleprob C27 I S4, S19, W48 precip4.step C28 I S4, S 10, W II precip4 .steprob C28 I S4, W6, WI I, W48 precip4.step C29 I W9, W22, W48 precip4.sleprob C29 I S2, S3 W48 precip4.slep C30 I S9, W32, W37 precip4.sleprob C30 I S2, S3, S4, W4 precip4.step C3 1 1519, WIO, W37 precip4.sleprob C31 I S2 , WI I, W37 precip4.step C32 I S9, W3, W6, W8 precip4 .steprob C32 I S3, W37, W48 precip4.slep C33 I W2, W3 , W9, W32, W3 7 precip4.sleprob C33 l SI , S2, S3 , S19, W4, W6, WI I, W37 precip4.step C34 I S2, S12, W2 precip4.steprob C34 I S2, S3 , S19, W4 precip4.slep C35 I SI I, W48 J)reci~.steprob C3S / S I, W48 precip4.step C36 / S5,512, WII precip4.steprob C36 I S 1, 53,54, W4, W6, W37 none. step CO I SIO, S12, W4, W6, W9, W37 none.sleprob CO l SI , S2, S3 , W4, W6, WI I

26 4.1 Peak and Average Wind Speed The Phase I and II prediction methods for PWSD and A WSD were compared 10 climatology and the 0000 and 1200 UTC runs of the 12-km Nonh American Mesoscale (MesoNAM) model. Comparisons were done for precipitation and non-precipitati on days in the verification data set. Then, the Phase I and II prediction methods for PWSD were compared to wind advisories issued by the 45 WS. 4.1.1 Comparing Phase I and II Methods to Climatology The Phase I and II prediction methods for PWSD and A WSD were first compared to four climatology methods. The prediction methods for PWSD were compared to the climatological PWSD (surface to 300 ft ), while the prediction methods for A WSD were compared to the climatological A WSD (surface to 300 ft ). The climatological peak/average wind speeds were also based on the mean peak/average wind speeds at 54 ft, 90 ft and 204 ft calculated in a previous AMU task (Lamben 2002). Figure 14 shows the mean error for PWSD (left) and A WSD (right) for days without precipitation. The Phase II climatological winds (point 4, in dark blue) had a positive bias around 2 kt, whi le the climatological winds at 54 ft, 90 ft and 204 ft (points 1-3 , in dark blue) had large negative biases. The Phase I and II methods generally had a slight negative bias. Figure 15 shows the mean error for PWSD (left) and A WSD (right) for days with precipitation. There was a negative bias in the Phase I and II methods, indicating the methods under-predict the PWSD and A WSD on precipitation days.

Peak Speed Mean Error (let) . No Precip Average Speed Mun Error (kt), No Preclp 4 o Climatolo8Y 4] 1 0 I0 elimatol08Y • o ~ i X Phase I ~ X Ph.se I • -: ; --- j--..-_- "'.- ----.:j,c----'==---..

.. Least-squares sln81e o - Least·squares sinlle .... o rei . o -8 - Stepwise least-squares -8 0 o - Stepwise least-squares o multiple res. multiple re,.

-12 - Robust sin lie res. -12 I - Robust sinlle ,el.

-16 - Least-trimmed squares , - leist-trimmed squares -16 multiple rea . t- ~ multiple rea.

-20 - Stepwise robust - 20 ~ ....L_~_~ - Stepwise robust 6 multlple,el· 6 multiple,e• . 1 2 3 4 5 1 2 3 4 5 Figure 14 . Mean Error for PWSD (left) and A WSD (right) on non-precipitation days For the Phase II meth ods, points 1-6 (on x-axis) correspond to stratification methods SM I-SM6, respectively_ The legend describes the colors corresponding to the Phase II linear regression types. Refer to Table 3 for a description of the stratification methods and linear regression types. The climatology methods are plotted on points 1-4_ The Phase I method is plotted on point I.

27 Peak Speed Mean Error Ikt), Predp Average Speed Mean Error lid). Precip o Climatology 1-- • • I ~ 0 Climatology

X Phase l I 0 ,---- 0 X Phasel • T ~ .. i ~ - " ! " Least-squaressingle -4 1< least-squares single i "'g. 0 -i 0 - -4 1 ',". I - Stepwise least-squares ·8 .. ---- - Stepwise least-squares muttlple reg. multiple reg . I 1 0 - Robust single reg . ·12 ·12 0 I - Robu st single reg.

- Least-trimmed squares ·1. T- ·1. - Least-trimmed squares 0

Pe ak Spee d MAE lid), No Preclp Average Speed MAE No Precip 0 Climatology o Climatology lid), 20 20 ~ -" I 18 --I. X Phase I X Phase I I. ",. f i ,. L_ least-squares single I least-sq uares single reg. I. reg.

12 12 +- I - Stepwise least-squares - Stepwise least-squares multiple reg. multiple reg . 10 10 -l t- o. 0 CD - Robu st si ngle reg. J - Robust single reg. 8 8 +- 0 I 0 • 0 - least-trimmed squares • - Least-trimmed squares multiple reB· 11 0 multiple reg. J • • • ; • • •I • • I .* ! - Stepwise robust - Stepwise robust 2 I 2 multiple reg. multiple res. 2 3 4 5 , IL 1 2 3 • 5 • Figure 16. MAE for PWSD (left) and A WSD (right) on non-precipitation days, for the climatology and Phase I and \I methods. Refer to Figure 14 for a description of the x-axis.

28 Peak Speed MAE (letl, Precip Average Speed MAE (ktl, Precip 0 r I o Climatology ~_ C unato ogy 20 r 120 1 I 0 18 118 1 x Phase I I --t X Phase I 0 0 16 '6 11 Lea st-squaresslngle ,. I tRast-sq uares single " 'eg. 'eg.

- Robust single reg. j. - Rob\.lst single res. 8 I 0 8 - --~ -~ ~ 6)( • 6 I 0 • = •- tRut-trlmmed sq\.lares • - tRast-trimmed squares multiple res. ~ • • • • multiple res. • I • 1 - Stepwise robust [ 2 , 2 - Stepwise robust multiple reg. multiple reg. 1 2 3 • 5 6 1 2 3 • 5 6 Figure 17. MAE for PWSD (left) and A WSD (right) on precipitation days, for the climatology and Phase I and II methods. Refer to Figure 14 for a description of the x-axis. The relationship between mean error, MAE and precipitati on was investigated further. Most of the days with large absolu te errors had precipitation (figures not shown). The large errors could be due to convecti ve wi nd gusts, since the Wind Tower QC program did not filter out all of the convective wind gusts. Some of the non-precipitation days with large errors in PWSD and A WSD mi ght actually have had rainfall over the forecast area. A day with precipitation was defined as having at least one observation at SLF with precipitation, thu nder, or a shower within 5 NM. It is possible some days had precipitation across KSC or CCAFS, but not within 5 NM of the SLF. Figure 18 shows the mean error when the absolute error for PWSD was within 5 kt (left) and greater than 5 kt (ri ght). The figure shows a bias near zero when the absolute error was ~ 5 kt, but large negati ve bias when the absolute error was > 5 kt. This indicates th e negative bias in PWSD becomes greater as the absolute error increases. There was also a much greater spread between the Phase I and II methods when the absol ute error was greater than 5 kt. Figure 19 shows the mean error when the absolute error for A WSD was ~ 5 kt (left) and > 5 kt (ri ght). Similar to Figure 18, the negative bias in A WSD was greater when the absolute error was> 5 kt. However, the spread between the Phase I and II methods was not as large. The negative bias was even greater for PWSD and A WSD when the absolute error was > 8 kt (figure not shown).

29 Peak Wind, Mean Error Ikt) when Abs. Error S 5 kt ~- .-,~-- "-;;.:..~ 4 I o Climatology ':*1-

O ~i --~I~--.i~-i.-~; X Phase I o 0 ... X Phase I o o o -4 _ : J. - Lust-squares single Least-squares single i ....

-8 - Stepwise least-squares -8 - Stepwise least-squares multiple reg . multiple reg.

·12 . - Robu st slnsle reg. -12 t- - Robust single rea . o o ·16 . _ LeasHrimmed squares -16 CD) - ~ast-trimmed squares multiple reg . rTKJltiple reg.

- Stepwise robust -20 1 - Stepwise robust 1'20 1 multiple res . , nwltiplereg. 2 3 4 5 6 [ 2 3 4 5 6 -~ Figure 18. Mean Error for PWSD on days in which the forecast 's absol ute error is within 5 kt (left) and greater than 5 kt (right). Refer to Figure 14 for a description of the x-axis.

Averall! Wind, Mean Error (Id) when Abs. Error S S kt Averas'!! Wind, Mean Error (Id) when Abs. Error> 5 kt 4 4 I - 0 Climato'oBY o Climatology o o ~~ r"---f--~ X Ph ... , X Phasel -4 f -4 ' Least-squares sinele Least-squares single reg. ! reg. -8 - - Stepwise least-squares J - Stepwise least-squares multiple rea. o multiple reg .

-12 +- - Robust single reg. -12 l r 1-Oobu ...',.'".,. -16 f - least-trimmed squares -16 ----1 - Least·trimmed squares r multlplereg. r--- I multiple reg .

-20 I - Stepwise robust .20 L - Stepwise robust multiple reg . multiple reg . 1 2 3 4 5 6 1 2 3 4 5 6

~~-- Figure 19. Mean Error for A WSD on days in which the forecast's absolute error is within 5 kt (left) and greater than 5 kt (right). Refer to Figure 14 for a description of the x-axis. A relationship was found between the absolute error of the Phase I and II equations for PWSD, and the difference between PWSD and the tower-average peak speed. As noted in Section 2.1 , the PWSD was the highest peak speed (surface to 300 ft) of all the KSC/CCAFS towers used in the task, during a 24-hour period. The tower-average peak speed is the average of the 24-hour peak speeds (surface to 300 ft) for all of the towers. Figure 20 shows the average difference between the PWSD and the tower-average peak speed, for days in which the absolute error was :s 5 kt (left) and > 5 kt (right). As the absolute error increased, the difference between the PWSD and the tower-average peak speed increased. The increase continued when the absolute error was > 8 kt (figure not shown). In other words, the difference between the PWSD and the tower-average peak speed was related to the predictability of the PWSD. The cause of

30 this effect may be due to wind gusts that only affect a portion of the KSC/CCAFS tower network. Measurement errors in the tower network could also contribute to the difference between the PWSD and tower-average peak speed. It is possible the absolute error also increases with the magnitude of the PWSD, but this was not investigated.

Peak Speed, Ava. Twr Diff. lid) when Abs. Error:S 5 kt I\ :~ak Speed, Avg. Twr Dltt. (kt) :hen Abs. Error> 5 kt 10 I o Climatology o Climatology • • X Phase I XPhasel 8 I 8

_ - Least-squaresslngle 7 --,- - least-squares single 7 +-- reg, : reg . II t • - 6

2 L --L-J. _ - Stepwise robust 2 - Stepwise robust multiple reg. multiple reg . 1 2 3 4 5 6 1 2 3 4 5 6 Figure 20. Average difference between the PWSD and the tower-average peak speed, on days- in which the forecast's absolute error is within 5 kt (left) and greater than 5 kt (right). Refer to Figure 14 for a description of the x-axis. 4.1.2 Comparing Phase 1 and II Methods to Model Forecast Winds The Phase I and II methods for predicting PWSD and A WSD were then compared to model forecast winds. The model forecast winds were derived from 0000 and 1200 UTC runs of the MesoNAM. The MesoNAM contained hourly forecasts out to 84 hours, although the independent verification only used the Day-I (1300 UTC to 1300 UTC) forecasts. The comparison used the MesoNAM wind data for the grid point closest to the XMR sounding. Levels 2to 18 of the MesoNAM were evaluated, along with the strongest winds in the lowest 1000-, 2000- and 3000-ft of the model. The height of each model level varied by forecast hour and model run, due to changes in surface pressure and temperatures aloft. Table 7 shows the average and standard deviation of the heights for the MesoNAM levels 2 - 18, based on the verification data set. Two sets of MesoNAM forecasts were used in the comparison. The first set included the strongest wind at each model level during the 24-hour period. The second set used least-squares single linear regression equations, in which the predictor was the model level's highest 24-hour peak speed and the predictand was th e PWSD or A WSD.

31 Table 7. Average heights and standard deviations of the MesoNAM venicallevels used in the verification data.

Average Standard Average Standard Level Height Deviation of Level Height Deviation of (MSL) Hei£ht (MSLJ Hei£ht 2 207 ft 4ft 11 1571 ft 23 ft 3 344 ft 5ft 12 175 1 ft 25 ft 4 483 ft 7ft 13 1940 ft 28 ft 5 626 ft 9ft 14 2 140 ft 31 ft 6 772 ft 12 ft 15 2352 ft 34 ft 7 922 ft 14 ft 16 2583 ft 37 ft 8 1076 ft 16 ft 17 2842 ft 41ft 9 1234 ft 18 ft 18 3145 ft 45 ft 10 1399 ft 20 ft Figure 2 1 shows the mean error for PWSD (left) and A WSD (ri ght) of the Phase I and II , climatology, and MesoNAM fo recasts. Only the 0000 UTe MesoNAM is shown, since there were only minor differences between the 0000 and 1200 UTe runs of the model. The Phase I and II methods generally had a weak negative bias. The strongest 24-hour MesoNAM forecast winds at each level are shown in black, and their bias varied by model level. The MesoNAM linear regression forecasts are shown in li ght blue, and their bias was zero. Figure 22 shows the MAE for PWSD (left) and A WSD (right). The MesoNAM linear regression forecasts were the most accurate, especially at the lower model levels.

Mean Error (ktl - Peak Speed Mean Error (ktJ - Avera,!!! Speed 8 8

0 •'2 f' -+------+- r--- 46 .------t 0 QG_oDQ 0 ~ 0 o~0 -..,.-- 2 00 9 - o lB-oae-Elso-ooO·se-ee e-o·e-SO-EJrc------0 ~ 8 o-§-ee-e·e-e·e-o-e-a-eGO e-e-eSC

~ ~ I OOOOIlJDOOCGOOo o ~ ~ ==:::---_--jl I -6 000 , . r.o~c"'lim"".C7to"IO"8-Y ----, I -6 0 0- o Cl1matolo,y -8 f aD O Phasel -8 0 C Phasel Phase II ·1. 1 Phlsell 1 • -12 ~0 0 0002 MesoNAM ~ '12 [J DOZ MesoNAM -14 0 0 OOZ MesoNAM Unear Relt . -14 ______0 002 MesoNAM Unear ReI: . • 5 ,. 15 2. 25 30 • 5 ,. 15 2. 25 30 Figure 21. Mean Error fo r PWSD and A WSD. The strongest 24-hour 0000 UTe MesoNAM forecast winds are plotted on points 1-20, in black. The 0000 UTe MesoNAM linear regression forecasts are plotted on points 1-20, in light blue. The climatology methods are pl otted on points 1-4, in dark blue. The Phase I method is plotted on point I (i n red), and the Phase II methods are plotted on points 1-30 (in yellow). Refer to Table 3 for a description of the Phase II methods

32 --rl------~ MAE (Ilt) · Pellk Speed MAE (kt)· Ave~le Speed

o o ClimatololY 14 r o CllmitololY 12 0 0 o Phase I 12 O Phasel Phue ll Phase II o OOZ MesoNAM ,. - __ DOO2MesoNAM ~ ,. 10 o OOZ MesoNAM Une., Rei: . o D OOZ MesoNAM Une., Rea .

• 0 • 0 0 o . \ 0 80 0 0 - 0000000000 0 00 I • :I COOOOOOO"' j oooeeo"'Eloe -

• 5 ,. 15 2. 2S 30 • 5 ,. 15 20 2S 30 Figure 22. MAE for PWSD and A WSD. Refer to Figure 2 I for a description of the x-axis. Figure 23 shows the mean error for PWSD for days with (left) and without precipitation (ri ght). For the MesoNAM data, separate regression equations were not developed for precipitation and non­ precipitation days. The MesoNAM regression fo recasts (i n li ght blue) had a negative bias around 2 kt on precipitation days and a positive bias around I kt on non-precipitation days. Figure 24 shows the mean error for A WSD for days with (left) and without (right) preci pitati on. Similar to PWSD, the MesoNAM regression forecasts had a negative bias on precipitati on days and a positive bias on non-precipitation days.

Meln Error (kt) · Peak Speed. Predp Mean Error (Ilt) · Pelk Speed, No Predp 4 ~ ~ - 2 ~OE 'EE ooaOO~90c?OCO~~ . ~ I +--- • 0 ~ 1 8 000000000000000000 0 -2 0 000_0000000000 ~I ·4 0000000 a .. 000 -6 0 ocoo ~ 0 .. 0 0 t -. 't 0 000 - - .- -8 o 0 -1. 0 """=='=:-__--., -1. o .12 0 D CllmatololY ·12 o ClimatololY O Phue l ·14 -14 o Phase I Phuell Ph.se II -1' ·1. o 0 OOOZMHONAM _ 0 OOZ MesoNAM -18 C ·18 a OOZ MesoNAM Unur Rei. a OOZ MesoNAM Unear R•• . -20 ·20 • 5 ,. 15 2. 2S 30 • 5 ,. 15 20 2S 30 Figure 23. Mean Error for PWSD on precipitati on (left) and non-precipitation days (ri ght). Refer to Figure 2 I for a description of the x-axis.

Mun Error (kt) · Avera,e SpHd, Predp Mean Error (kt) - Avel1lle $pHd, No Predp • o • 6 ' ooooe oo 0 • 4 ~ 0000 0 . 2 0 0 ] , o 0 0 a ---l-..- +---_ ::J I .2 BOoooooooooooooooooo -2

·4 0 , - I .. .. "o.,c"";::...= ,;:.;:,.=,,:-----., .. o ° C ClllnltololV ·8 0 Phlse I -I ° C Phase I -1. Ph ase II -1. Phllsell ·12 o ° o OOZ Me50NAM -12 COOl Me50NAM o OOZ M esoNAM Une., RH. C OOl MHONAM Unur ReI: . ·14 .... o -14 • 5 ,. 15 20 2S 30 • 5 ,. 15 20 2S 30 Figure 24. Mean Error for AWSD on precipitation (left) and non-precipitation days (right). Refer to Figure 2 I for a description of the x-axis. Figure 25 and Figure 26 show the MAE fo r PWSD and A WSD for days with (left) and without precipitation (ri ght). The MesoNAM regression fo recasts performed better than the Phase I and II methods, especially for precipitation days.

33 MAE (ktl- Peak Speed, Precl p MAE (ktJ· Peak Speed, No Pre<:ip 2. 2. 18 i 0 o Ctlmatology ,. _ 0 Climatology ,, Cl " o Ph ase I o Phase I ,. ,. Phase II Ph ase II ,. a OOZ MesoNAM ._ 0 OOZ MesoNAM 0 002 MesoNAM Untar Rei. o OOZMesoNAM Unear ReI: . 12 Cl 12" Cl ,. ClCl 10 0 t- Cl • o 00 a Cl Cl 1 1 = ~ OOOOOO~OO 00 : 0 00 t------00 ----. • 00000000000000000000 _ ~ooooooooooo 00 • • OClClClOClOOClOOClClClClaaOCl B 2 2 • 5 ,. 15 2. 2S 30 • 5 ,. 15 2. 2S 3. Figure 25. MAE for PWSD on precipitation (left) and non-precipitation days (right). Refer to Figure 2 1 for a description of the x-axis.

MAE {ktJ - Averalfl Speed, Preclp MAE (kt) · Average Spe~ . No Predp "Cl"C~II~mo='t~ol~o'~V~C===-,l ,. o Climatology 12 0 o Phase I "r 12 o Phase I Pha se II Ph.sell a a a OOZ Me50NAM a DOl MesoNAM 1. o OOZ MesoNAM Unear ReI: . 10 Cl o OOZ MesoNAM Unur Re . • Cl o 6 0 0 00 0 0000000 0 ClCl " • B 80~ - f - ~ ~ ~ -~- - 0 8 0 , 0000 0000000000 2 ~---+----~------~------~ • 5 ,. 15 2. 2S 30 o 5 10 15 20 2S 30 Figure 26. MAE for A WSD on precipitation (left) and non-precipitation days (right). Refer to Figure 21 for a description of the data points. Figure 27 shows the MAE for PWSD (left) and A WSD (right) from the MesoNAM regression forecasts. On precipitation days, the 1200 UTe model runs (in purple) perform ed just sl ightly better than the 0000 UTe model runs (in red). On non-precipitati on days, there was virtually no difference between the 0000 UTe (i n light blue) and 1200 UTe (in orange) model runs.

MAE (ktJ - Peak Speed MAE (ktl' Averale Speed

• o DOl MesoNAM Unear Reg ., Precip • O OOlMesoNAM Unear Rei., Precip 0 122 MesoNAM Linear Rei., Precip o 12ZMesoNAM Unear Rei., Precip o OOZ MesoNAM linear Reg ., No Precip D OOlMesoNAM Unear Rei" No Precip • I 122 MesoNAM Unear Rei., No Precip • 12l MesoNAM Unear Rei., No Preclp ~CI§CI CI !l !l , !l 8 Ia § EI 8 !l III CI EI CI a 4 I _ - - . - - - - - = = • Cl Cl CIllC Cl§§§§CI ~ ~68C1~E1E1E1 1 - - -.... f ---- 2 -1 -- - 1 2 - - -I "_ ~ r ",

o ~~ 0 0 • • 12 1. 20 0 • • 12 ,. 20 Figure 27. MAE for PWSD (left) and AWSD (right), for 0000 and 1200 UTe MesoNAM regression forecasts. Points 1-17 depict model levels 2-18, and points 18-20 depict the strongest winds in the lowest 1000-, 2000-, and 3000-ft of the model. The legend shows the color corresponding to each method. The MesoNAM linear regression forecasts were clearly th e most accurate in the verificati on data set. The differences between the 0000 and 1200 UTe model runs were not significant. The Phase I and II

34 forecasts were similar, although the best Phase II methods were slightly more accurate than the Phase I forecasts. The climatology forecasts performed the worst. 4.1.3 Comparison of Phase I and /I Peak Wind Speed Predictions to 45 WS Wind Advisories The Phase I and II forecasts of PWSD were compared to the 45 WS wind advisories on days in which the 45 WS issued at least one non-convective wind advisory. In addition to KSC and CCAFS, the comparison also included wind advisories for PAFS. There were three types of wind advisories for non­ convective peak winds: 25-34 kt, 35-49 kt and 50 kt or greater. The valid time periods for the wind advisories were variable in length, ranging from around one hour to close to two days. Some of the wind advisories extended across multiple 24-hour forecast periods (0800 to 0800 local time). Multiple wind advisories could al so be iss ued during one 24-hour forecast period. For this comparison, if a wind advisory extended across two 24-hour forecast periods, then the comparison analyzed the two days separately. If a 24-hour forecast period included multiple wind advisories, then they were counted as one wind advisory. The comparison used the strongest wind advisory for the forecast period. Wind advisories of the same type and forecast period, but different areas (such as KSC and CCAFS), were merged for the comparison. Also, the comparison did not include wind advisories with a time period of less than four hours, in order to avoid small-scale weather systems. Table 8 shows the comparison for days in which the strongest wind advisory was for 25-34 kt (left) and 35-49 kt (right). The wind advisories for winds of 50 kt or greater are not shown, since only two warnings meeting the criteria for this comparison (see previous paragraph) were issued during the verification period. A "hit" was defined as an observed PWSD in the correct forecast interval (25-34 kt or 35-49 kt). An "over-forecast" was defined as an observed PWSD weaker than the forecast interval. An "under-forecast" was defined as an observed PWSD stronger than the forecast interval. Table 8 shows the 45 WS out-performed the Phase I and II methods, because the 45 WS had the most hits. On days in which the 45 WS issued a wind advisory for 35-49 kt, they tended to over-forecast more often than under­ forecast, while the Phase I and II methods under-forecast more often than they over-forecast.

Table 8. Verification for days in which the highest 45 WS wind warning/advisory was 25-34 kt (left) and 35-49 kt (right). Phase II methods are shown in blue.

25 - 34 kt 35 - 49 kt Method Under- Over- Under- Over- Hits Hits forecast forecast forecast forecast

least-squares single regression 35 13 4 35 29 3 robust single regression 32 19 I 29 36 I stepwise least-squares regression 36 13 4 33 30 3 stepwise robust regression 34 15 3 34 30 3 least-trimmed squares regression 31 17 4 31 33 3 Phase I 34 16 2 34 31 I 45WS 45 6 2 41 2 28

4.2 Timing of the Peak Wind As described in the first paragraph of Section 3.4, the timing of the peak wind was defined as the number of hours elapsed since the beginning of the fo recast period (1300 UTC). The Phase I and II methods for predicting the timing of the peak wind were compared to six climatology values. The first three climatology values were based on the mean peak winds at 54-, 90- and 204-ft (Lambert 2002). For

35 each of the three levels, the cool season mean peak wi nd speed was calculated for each ho ur of the day. The tim ing was then defined as the hour in which the hi ghest mean peak wi nd speed occurred. Table 9 shows the climatological timing of the peak wind. The hours used for this comparison are highlighted in yellow.

Table 9. Climatology methods for predicting timing of the peak wi nd, based on the mean peak winds in a previous AMU task (Lambert 2002). The "average wi nd speed" is the average of all the towers used in the previous AMU task, and the "maximum wind speed" is the maximum of all the towers. Hour 54 ft / 60ft 54 ft / 60ft 90ft 204ft 204ft 90 ft average in average wind maximum maximum average wind maximum wind speed UTC speed wind speed wind weed speed wind speed 0000 11.03 12.69 11.71 11.73 14 .14 14.1 9 0100 10.94 12.53 11.66 11.68 14. 13 14.18 0200 10.88 12.45 11.62 11 .63 14. 11 14.18 0300 10.76 12.28 11. 56 11 .57 13.98 14.06 0400 10.62 12.07 11.5 1 11.53 13.91 14.03 0500 10.47 11.87 11.42 11.43 13.71 13.85 0600 10.39 11.76 11.33 11.33 13 .6 1 13.75 0700 10.24 11.57 11.18 11.18 13.41 13.56 0800 10.1 4 11.50 11.05 11.06 13.22 13.39 0900 10.16 11.58 10.98 10.99 13.19 13.37 1000 10.14 11.58 10.94 10.96 13.18 13.37 1100 10. 10 11.51 10.88 10 .89 13.07 13.26 1200 10.36 11.71 10.96 10.98 13 .13 13.36 1300 11.21 12.40 11 .48 11.48 13.25 13.42 1400 12.3 1 13.42 12.37 12.38 13 .67 13.79 1500 13.18 14.38 13 .08 13. 11 14.26 14.33 1600 13.75 15.07 13.47 13.51 14.74 14.79 1700 14.09 15 .37 13 .77 13.8 1 15.07 15.09 1800 14 .24 15.50 13 .98 14.04 15.28 15.28 1900 14.10 15 .35 13.99 14.06 15.22 15.22 2000 13.76 15.11 13 .7 1 13.79 15.05 15.05 2100 13.06 14.51 13.20 13.26 14.65 14.66 2200 12.03 13 .63 12.30 12.35 14.26 14.28 2300 11.27 12.97 11.77 11. 80 14.09 14.12 The last three climatology values were based on the average timing of the peak wind in the verification data set and the Phase 1 and 11 developmental data sets. The average timing for the verification data set was 9.8 hours, while the Phase 1 and 11 data sets had an average timing of 9.7 hours. The MesoNAM forecasts for the timing of the peak wind were deri ved by determining when the peak wind occurred at model levels 2-18, as well as the strongest wind in the lowest 1000-, 2000- and 3000-ft of the model. Figure 28 compares the mean error (diamonds) and MAE (circles) of the timing of the peak wind, in hours. The Phase 1 (point 1) method had a bias near zero. The Phase 11 methods using least-squares single linear regressions (points 1-6, in yell ow) also had a bias near zero, while the robust single linear regressions (points 7- 12, in yellow) had a large negative bias, around 4 hours. The 0000 UTC MesoNAM winds (points 1-20, in blue) had a positive bias around 2 hours. The first three climatology values had a negative bias of 4-5 hours, while the last three climatology values had a bias near zero. None of the methods performed significantly bener than climatology. 36 Figure 29 compares the mean error (di amonds) and MA E (triangles) for the 0000 and 1200 UTe runs of the MesoNAM. There was little di ffe rence in MAE between the 0000 and 1200 UTe runs. All of the model winds had a positive bias, al though the bias was larger in the 1200 UTe model runs. Since none of the methods improved significantly upon cli matology, th e 45 WS fo recasters should use climatology for the tim ing of the peak wind unt il a more accurate method can be developed.

Mean Error and MAE(hr) for Peak Wi nd TI ming Mean Error and M AE (hr) for Peak Wind TIming 0000 and 1200 UTe Model Runs 8 8 ~ 6 [ ~ .. + ME-OOZ 4 6 A A ;..~ ... M E· Phase II ' ····r ~ MAE - Phase II 4 0 000 • ME - 12Z ; <> M E· Climatology ·2 2 .. ······t:·S···.·: M AE - Climatology :- 4 ~ o .... -4 [ 00.0 MAE- 12Z

T ~-- .. -6 () ME - OOZ model o 0 5 10 15 20 0 MAE - OOZ model o 5 10 15 20 I Figure 28. Mean Error (di amonds) and MAE Figure 29. The Mean Error (diamonds) and (circles) for timi ng of the peak wind. The Phase I MA E (triangles) for the tim ing of the peak wind meth od is plotted on poi nt I, Phase \I methods on by the 0000 and 1200 UTe MesoNAM winds. points 1- 12, cl imatology methods on points 1-4 and 0000 UTe MesoNAM wi nds on poi nts 1-20.

37 5 Peak Wind Tool GUls The AMU delivered the Phase II tool as a Microsoft Excel GUI. The 45 WS also requested a tool for MIDDS, their main weather display system. The Excel and MIDDS tools display the same forecast parameters: PWSD, A WSD, and the probability the peak wind will meet or exceed 25 kt, 35 kt and 50 kt. Unlike Phase I, the Phase II tool does not display the timing of the peak wind since none of the methods improved upon climatology. The Excel GUI uses the 0000 and 1200 UTC MesoNAM winds as input, therefore it can display predictions for the Day- I to Day-3 forecast periods. The MIDDS tool uses winds from the 0000 and 1200 UTC MesoNAM and Global Forecast System (GFS) models. The MIDDS tool can use the MesoNAM winds to display predictions for the Day-I to Day-3 forecast periods, and use the GFS winds to display predictions for the Day-I to Day-7 forecast periods.

5.1 Microsoft Excel GUI The Phase I version of the Excel GUI used the moming XMR sounding as input to the prediction equations. The 45 WS forecaster manually entered so unding data, and then the GU I calculated and displayed the predicted PWSD, A WSD, timing of the PWSD and the probability the peak speed will meet or exceed 35 kt, 50 kt and 60 kt. The 45 WS no longer issues wind advisories for peak winds of 60 kt or greater, therefore, the Phase II version does not display the probability the peak speed will meet or exceed 60 kt. However, the 45 WS did request the AMU add the probability the peak speed will meet or exceed 25 kt, therefore the Phase II version displays the probability the peak speed will meet or exceed 25 kt, 35 kt and 50 kt. Since the MesoNAM linear regression forecasts performed the best for PWSD and A WSD in the verification, the Phase II version uses MesoNAM forecast winds as input. MesoNAM forecasts are provided to the 45 WS by ACTA, Inc. and include hourly forecasts from 0 to 84 hours based on the model runs at 0000, 0600, 1200 and 1800 UTe. The 45 WS receives the MesoNAM forecasts via e-mail after each model run, and they can be stored on a computer hard drive as a text file. 5.1.1 Equation Development The AMU created single linear regression equations for the Day-I to Day-3 forecasts of PWSD and A WSD, using both the 0000 and 1200 UTC MesoNAM forecast winds. The equations were developed with MesoNAM forecasts from the same POR as the verification data set, March 2007 to April 2009. The data were first stratified by precipitation and non-precipitation days. Eq uations were then developed for each model level, from level 2 to level 18, as well as the lowest 1000-,2000- and 3000-ft. Figure 30 and Figure 31 show the MAE values for PWSD from the 0000 UTC and 1200 UTC MesoNAM regression forecasts, respectively. As expected, the Day-I forecasts were the most accurate, followed by the Day-2 and Day-3 forecasts. The forecasts were significantly more accurate on non­ precipitation days. On precipitation days, the 1200 UTC runs were more accurate than the 0000 UTC runs, especially for Day-3. On non-precipitation days, the forecasts from the 0000 and 1200 UTC runs were very similar.

38 MAE (kt) - Pe ak Speed, Precip MAE (kt) - Peak Speed, No Precip 0000 lITe MesoNAM 0000 UTe MesoNAM 6 6 ~ , 0000000 a Oay- l 0 a I 00 a Oay-2 aooomoo OJ O Oay-3 5 888g~ S-8 88 g9-BS P 8 g~ 5

4 4 - 1 a Oay-l 0 0 - t 0000 8 00 a Oay-2 [ 00.0000 88BB 8 a Oay-3 o .on 1'1 88 8 !jl:l 3 ---!---- - 3 a a 0 -0-

2 2 o 5 10 15 20 o 5 10 15 20 Figure 30. MAE fo r PWSD on precipitation (left) and non-precipitation (ri ght) days, for Day-I to Day-3 fo recasts. Forecasts are from linear regression equations using the 0000 UTe MesoNAM forecast wi nds. Poi nts 1-1 7 depict model levels 2- 18, and points 18-20 depict the strongest winds in the lowest 1000-, 2000- and 3000-ft of the model.

MAE (kt) - Peak Speed, Precip MAE (kt) - Peak Speed, No Preclp 1200 lITe M esoNAM 1200 lITe M esoNAM 6 -,-- 6 o Oay-l o Oay-2 5 o I I a Oay-3 9 Bg-g -to'C-c-h -Cl-C1j1=li a a IJ ~ 5 0oomoooo m ooo ~ 00 0

4 I a Oay-l 4 'P 00 oom oooo CI om o Oay-2 0 a a a m 0 ~ 8 B 8 8 B 8 8 ~ 8 R8 1 3 --+L-='-0 O~a~-~3 1 3 - 81l-Cl-s-a-B-"-'''-=--t--=--'---+---''Y I I 2 +0 ------~5 ------1~0------1~5 ------~2 0 II 2 ~0 ------5~------1~0 ------~1-5 ------~20 '----" _-----=-______....J Figure 3 1. MAE fo r PWS D on precipi tation (left) and non-precipitation (ri ght) days, for Day-I to Day-3 forecasts. Forecasts are fro m linear regression equations using the 1200 UTe MesoNAM forecast winds. Points 1-17 depict model levels 2- 18, and points 18-20 depict the strongest winds in the lowest 1000-, 2000- and 3000-ft of the model. Figure 32 and Figure 33 show the MA E values for A WSD. The Day- I and Day-2 forecasts had sim ilar MAE values. The Day-3 fo recasts were worse than the Day- I and Day-2 forecasts, except the Day-2 and Day -3 forecasts fro m the 0000 UTe runs we re similar on non-p recipitation days. On precipitati on days, the 1200 UTe fo recasts were sli ghtl y more accurate than the 0000 UTe forecasts.

39 MAE (kt) - Average Speed, Predp MAE (kt) - Average Speed, No Precip 0000 UTe MesoNAM 0000 UTe M esoNAM 6 6 O Oay-1 o Oay-1 O Oay-2 I o Day-2 5 O Oay-3 S +- o Oay-3 0000 0000 1il 0 0000 4 4 ------8-8 8 EI f- O OElEI 88-8 B8E1'-g-~ 00 ElCCI:! U§ ~ 8BBB 00 ~8 3 -- j- - - 3 -.D_0-g.J:L' 2 2 o 5 10 15 20 o 5 10 15 20 Figure 32_ MAE for A WSD on prec ipitation (left) and non- prec ipitation (ri ght) days, fo r Day-I to Day-3 fo recasts. Forecasts are from linear regression equati ons us ing the 0000 UTC MesoNAM forecast winds. Points 1- 17 depict model levels 2-18, and points 18-20 depict the strongest wi nds in the lowest 1000-, 2000- and 3000-ft of the model.

MAE (kt) - Average Speed, Precip MAE (kt) - Average Speed, No Precip 1200 UTe MesoNAM 1200 UTe MesoNAM 6 6 I I o Day-1 o _'Oay-2 ~ o Day-2 5 o Oay-3 5 - o Oay-3 I I 4 0000 ';' 000 0 0000 4 - - [J°B°'ilBB CI - 00 00 gB B ~fBo 888 8 OOOCI 8 3 8"OF_r " 3 6-B-tI

2 2 o 5 10 15 20 o 5 10 15 20 Figure 33. MAE fo r A WSD on precipitation (left) and non-precipitation (ri ght) days, for Day-I to Day-3 fo recasts. Forecasts are from li near regression equati ons using the 1200 UTC MesoNAM fo recast winds_ Points 1-1 7 depict model levels 2- 18, and poi nts 18-20 depict the strongest wi nds in the lowest 1000-, 2000- and 3000-ft of the model. For each forecast parameter, the tool uses the model level wit h the most accurate equation, based on the lowest MAE values. Since the tool uses separate equations for precipitation and non-precipitation days, as well as each fo recast day, 12 prediction equat ions are used fo r both PWSD and A WSD. Table 10 and Table II describe the PWSD and A WSD equations used in the Excel GU L

40 Table 10. Prediction equati ons for PWSD, based on linear regressions in which the predictor is the strongest 24-hour wind speed at the MesoNAM model level, and the predictand is the PWSD. Refer to Table 7 for the height of each model level. Foreeasl Preeipilalion Model Run Model MAE Slope Intercept Day Occurrence (UTe) Level (kO Day-I Yes 0000 5 4.6 1 1.05 7.99 Day-I No 0000 2 2.76 1.08 5.55 Day-I Yes 1200 4 4.53 1.10 7.87 Day-I No 1200 2 2.73 1.13 5.41 Day-2 Yes 0000 4 4.78 1.08 8.76 Day-2 No 0000 3 3.07 0.98 6.72 Day-2 Yes 1200 18 4.56 0.58 16.24 Day-2 No 1200 4 2.88 0.99 5.84 Day-3 Yes 0000 15 5. 17 0.58 17.04 Day-3 No 0000 3 3.42 0.95 7.54 Day-3 Yes 1200 17 4.89 0.65 14.45 Day-3 No 1200 4 3.43 0.90 7.60

Table II . Prediction equations for A WSD, based on linear regressions in which the predictor is the strongest 24-hour wi nd speed at the MesoNAM model level , and the predictand is the A WSD. Refer to Table 7 for the he ight of each model level. Forecast Preeipilalion Model Model MAE Slope Intereepl Day Occurrence Run Level (kO Day-I Yes 0000 4 3.68 0.78 4.76 Day-I No 0000 4 3.02 0.75 2.25 Day-I Yes 1200 4 3.64 0.8 1 4.31 Day-I No 1200 4 2.98 0.76 2.59 Day-2 Yes 0000 4 3.61 0.79 4.76 Day-2 No 0000 5 3.36 0.63 4.20 Day-2 Yes 1200 2 3.5 1 0.82 5.37 Day-2 No 1200 5 3.16 0.67 3.53 Day-3 Yes 0000 15 4.0 1 0.40 11.62 Day-3 No 0000 5 3.30 0.64 4.28 Day-3 Yes 1200 5 3.94 0.71 5.9 1 Day-3 No 1200 4 3.36 0.66 4. 17 In Phase I, the 45 WS developed a statistical method for estimating the probability the PWSD wi ll meet or exceed the thresholds for wind advisories (Barrett and Short 2008). The method is based on the error bars of the linear regression equations. The equation for calculating the probabilities gives the area under the right-side of the Gaussian curve: 1-[0.5 * (I ± ~I- J2/".*((x - Y) / Zf))]

In the equation, x is the threshold value (25 , 35 and 50), Y is the predicted PWSD and z is the predicted sigma (estimated error of the linear regression equation). The + sign before the radical is used for y ~ x, and the - sign is used fo r y > x.

41 5. J.2 Using Ihe Excel GUI The AMU developed the Excel GUI using the Visual Basic for Applications programming language. To use the tool, the forecaster opens the Excel file to the " Intro" worksheet (Figure 34). The Intro worksheet contains instructions on how to use the tool. Common user questions are answered in the " FAQs" (Frequently Asked Questions) worksheet. To start the tool, the forecaster selects the "Start Cool­ Season Peak Wind Calculation" button. The tool then displays a "Browse" dialog box containing a list of files. The dialog box opens to the file directory last used. The forecaster needs to navigate to the directory containing the MesoNAM files and select one of the files (Figure 35). , 6 ' , •, ..,• I , P.. k WInd Tool lor Gene'" FOfK.. tIII9 . ION .. I; v ....!on 1 .• ,"'-Y 2010) 12 o...IopM .,. ~,,",~u... U ANd ...... Uta."""_ ...... IMI ...... I• .." .." ~ ~ofPMkWlndTooIfofco.-elFor.ea."", : 11 n.s ,s 'IP~ use. tnte1&ee IGlIjlOOlh, ~ Ihe U5e1loliMd anN:.TA MesoNAM leld" 22 In 0II1er 10 CIb.ftIe me 'Ipeed and mq d lie 0IIi,. (800 am tw.y to 8'00 ~ IOmOfTOfIj peak WIne! I 2J speed The 1001 can be used rOf !he Giy. 1, dlJ.2 Of OIJ.3 forecasts I" 2S IMtrvctIoft. on h_ to II •• the i00i: 21 I SIan Ihe I0OI by ckking on1he "Su.rI Peak WIIId C81a8tion" bll!0I\

~ 2 SeIed in ACT'" I.4esoNAAIIeId fI/IIlfom me &_ cbIog 00., by Ieft.cllcklng on. fl/elwme 1.., n. fileNrnes IIOITIlIII¥ SWtvdl"slr, net off' SeIed. 'IetOlll' 00 or 12 UTe modeIlIJI 1I After W'IkIng .1\IefWrIIt. (ido; on"OK"I000re....OI "CanteI" 10 SIOO .. IOOI 31 The I0OI ihenwnfles !he Me IS 110m. 00 or 12 UTe model l1l'i, .ndrtmllkes Sin !he file ~II\ u12 Ihe_ tomId IonnM f!he Me IS ntI mtie correa rom.~ IIf\ error message IS !i_~ and !he I0OI i$

1: 3 The "Petk Wind CaIOJanon" dialog bol lSeIIspllyed 5eIect. roreuSl dly II1d IhQn seied!he l6 "Calculet. P.. ~ W1r.r' bIAIorI

".. . The pl"edIc*I...ues Ire dt~yed ItI IM "Peak WInd Pf~ · GlI • dispNYS .,. dlifit ~ peak .ncI • ...... age 'Mfodspeed IMJOS:ti KSC 100 CCAFS (14110 300. NSl ~ lor

"U 5 5eIecI fle "Choose ~ ForK." Dey From Slme Model AI"II" bUIon 10 selecllIlOIhef .u forea,51 pertOd SeIed IhB "Pnnl 0\fpI.t' bt.CIonlo Pllnl1he GUI S8llcl1Ie "end rod' bUIorl tG slop . S hloo1 Figure 34. Intro worksheet to start th e Excel version of the Peak Wind Tool.

42 It r OW'ioC ? 'x

look~ , 10 Ap<2OO9NAMsn 1:' 1 ~ r.a X r:;" Ii] - sn 7.. 79"0.e32..Q9040 1-QOOJ sn" ,9<0.,32-0901lJ5-1200 sn7 .. mo.e32-09Q"10-0000 sn7-41910.e32..Q904 1"-I 200 Docuner

< I l ~ File [}MOe : 1 L.... I _of typo, IAI '1M (0.0) vi

Tools C«><~ 1 -I "" ~ l Figure 35 . Dialog box used to select a MesoNAM input fil e_ The tool veri ties the fil e chosen by the forecaster is in the correct format and is from a 0000 or 1200 UTC run of the MesoNAM. If the tile is invalid, the tool displays an error message and exits. Otherwise, the " Peak Wind Calculation" dialog box is displayed (Figure 36)_ The dialog box shows the date and time of the model run. In this example, it is the 0000 UTC run from I April 2009. The forecaster selects a forecast day and th en selects the "Calcul ate Peak Wind" bunon. The Peak Wind Prediction GU I with the desired output is then displayed (Figure 37). The GUI shows the forecasts for PWSD, A WSD and the probability the peak wind will meet or exceed 25 kt, 35 kt and 50 kt. The lef\Jright side of the GU I shows the forecasts for precipitation days/non-precipitation days. Since no methods for predicting the ti ming of the PWSD performed bener than climatology, the GUI contains the fo ll owing note: "The peak wind speed of the day usually occurs during the afternoon or evening. The climatological timing of the peak speed is 2248 UTe. Adjust the time of the peak wind, based on expected movement of fronts, wi nd surges, changes in pressure gradient, etc _" The forecast period is displayed at the bonom of the GU I. In thi s example, the forecast period is from I April 2009 (0800 local) to 2 April 2009 (0800 local). The forecaster can then select one of three options: "Print Output", "Choose Another Forecast Day From Same Model Run" and "End Tool".

43 Calculations use the 00 UTe IUl of I>CTA Meso'IAM on 4/1~ _t Fon!cast Day (void 13Z - 13Z) (0'00

r 0ay-2

r Oay-3

I calcula.. Peak WI"d I cancel I

Figure 36. Dialog box used to select the forecast day.

44 $ Peak WhI Speed (13Z-l3Z) fnm ~f_ to 3DO It Aa.

' '''IrF-"~l~UUfl Ul...UM"a:.,.;~ K5C If IJrt!UpllMlUI Uu~ IILI' uu;u­ 1 or CCAFS: across kSC or CCAFS:

Pl:I.ot WiaJ r::-:- knots Speed I 29.9 ~------~

•~--~~'~·'-----·-·:· If preciPtaUon ()t(U'S across KSC If preciPtation does not occur 1 I or CCAFS: across rsc or CCNS:

'_Wrd r-:;;"7 ''''aoeWrd ~ Speed I 193

• If precidtation occurs across rsc If PI ec:lpltatki i does not oca.r or CCAFS: a:ross KSC (I" CCAf'S:

25 knolS I 78.7 percarr. 251crml I 10.0 pe-cent ~ knots f19.9 percen: ~ kno:s f"OD percert 50 knos ~ pElfcen: so kro:s f"OD percert

1a-_~ForeoastDay I Print nrtput I From &ome Model Run EI1d Tool I

+ on days With thunderstorms, ttB tool wei ~tty l.a'KIerestimate the pe_ ,.;00 ..-I and the"..-ty the peak ..-I wi! meet ... """oed 2S kL 3S Irt. and 5Okt, $ The peak wInd..-l of the day usuaIy oca.n rumg the aft""""", ... -*'9. 1l1e tlmatoklgKal ~ of the pede speed ts 2248 UTe. ~t the tme of peak WInd, based on expected IOOYefTlE!f'It of frunh, wnt swges. changes In pre5SU'e gradent. etc. Based on day-l forecast fl"Oftl MesoNMt 00 UT e run on 4/1/2009

Valid forecast period; 4/ l/ 2009 (ill) to 4/ 2/ 2009 (13Z)

Figure 37. GUI displaying the output from the Excel version of the Peak Wind tool. The top section shows the predicted PWSD, the middle section shows the predicted A WSD, and the bonom section shows the probability the PWSD will meet or exceed 25 kt, 35 kt and 50 kt.

5.2 MIDDS GUI The AMU initially developed the MIDDS tool to use gridded model data as input because the ACTA MesoNAM files are not available in MIDDS. The tool was created using the Tool Command Languagerrool Kit (TcIlTk) programming language. It used the model data for the closest grid point to XMR. The MIDDS gridded data contains output from the 0000 and 1200 UTC model runs of the North American Mesoscale (NAM) and GFS models. Both models have a horizontal grid spacing of 80 km in

45 MIDDS. After the 45 WS was given access to receive point model data from Johnson Space Center, the tool was modified to use these data. 5.2.1 Equation Development While the MesoNAM contains hourly forecasts out to 84 hours, the NAM and GFS gridded data in MIDDS contain 6-hourly forecasts out to 60 hours and 240 hours, respectively. The MIDDS tool used NAM data to generate Day-I and Day-2 forecasts and GFS data to generate Day- I to Day-5 forecasts. The Day- I, Day-2 and Day-3 forecast equations developed for the Excel tool were also used for the Day- 1,2 and 3 forecasts in the MIDDS tool. The Day-3 forecast equations were used for the Day-4 and Day-5 forecasts. The forecasts based on the GFS gridded model data used the MesoNAM linear regression equations developed for the Excel tool, since the verification data set did not contain GFS model data. Since the MIDDS gridded model forecasts are only available every six hours, the AMU updated the linear regression equations to take the forecast interval into account. Otherwise, the MIDDS tool would have a low bias in predicting wind speeds. The MIDDS tool calculates the probability the PWSD will meet or exceed 25 kt, 35 kt and 50 kt, using the same statistical method as the Excel tool (Section 5.1.1). Based on feedback from the 45 WS, least-squares single linear regression equations were developed for the daily average wind speed at 30-60 ft, which is one of the forecast parameters in the 24-Hour and Weekly Planning Forecasts issued by the 45 WS. The predictands were calculated by averaging the 5- minute average wind speeds at 30 ft and 54/60 ft from all of the towers used in this task through a 24-hour period (1300 UTC to 1300 UTC). The predictors were calculated by averaging the hourly MesoNAM winds for the 24-hour period. The predictors were evaluated for model levels I and 2 of the MesoNAM, which have average heights of70 ft and 207 ft MSL, respectively. Figure 38 shows the MAE for daily average wind speed from the 0000 (left) and 1200 UTC (right) model runs. The level-I (in blue) and level-2 (i n red) forecasts are compared to clim atology (in green) for Day- I to Day-3, including precipitation days, non-precipitation days and "all days". The "all days" forecasts included precipitation and non-precipitation days. Separate climatology val ues were calculated for precipitation days (8.4 kt), non-precipitation days (6.8 kt) and "all days" (7.3 kt). The level-2 forecasts performed slightly better th an level-I forecasts and significantly better than climatology. As expected, the MAE values were lowest on non-precipitation days and highest on precipitation days. The MAE values increased slightly from Day- I to Day-3. There was not much difference between the 0000 and 1200 UTC runs, except the level-I and level-2 forecasts from the 1200 UTC runs did better than the 0000 UTC runs for Day-3. The forecasts for "all days" were then compared to the average of the precipitation and non­ precipitation days, weighted by the number of precipitation and non-precipitation days. Table 12 shows the differences between the "all days" forecasts and weighted average forecasts were not significant. Therefore, unlike for PWSD and A WSD, the prediction equations for daily average wind speed did not stratify by precipitation.

46 MAE (kll- Dally Average Speed MAE (kll- Daily Average Speed 0000 UTC M esoNAM 1200 UTC MesoNAM 2.5 0 ~ I f-t- I 2 1 1.5 Ii! 8 i r 1 '-- r o level-l o level.1 0.5 1 C- o level-Z 0.5 o Level-2 o Oimatology o dimatology o ------'--~- 0 ~ ·S ~ ·S ~ ·S ·S ~ ·S .,s. ~ ... "9- t- ... IlIw ,.,V' ~ If' ",'" ,~ ~ ;.. ~ ~ "v'" ~ t- o.' .. .:. .. o.' o.' o.' .:. o.' 4,,4;', ;t.. Q " 7> Q'" ;.. Qt- ~o Q'" 7> .-I <:J'fI t- :t t- ...:t t- t- Q 7> Q .:. Q " Q 7> Qt- Q'" Qt- " Qt-

Figure 38. MAE for daily average speed for forecasts from 0000 (left) and 1200 UTC (right) MesoNAM. Forecasts for precipitation days are plotted on points 1, 4 and 7 of the x-axis. Forecasts for non-precipitation days are plotted on points 2, 5 and 8, and forecasts fo r all days are plotted on points 3, 6 and 9.

Table 12. MAE (kt) for daily average speed. "All Days" is all days in the verification data set. "W. Avg" is the weighted average of precipitation and non-precipitation days. "00 UTC" is the 0000 UTC MesoNAM and " 12 UTC" is the 1200 UTC MesoNAM. Leve/-I Leve/-I Leve/-2 Leve/-2 Leve/-I Leve/-I Level-2 Leve/-2 Forecast 00 UTe 00 UTe 00 UTe 00 UTe 12 UTe 12 UTe 12 UTe 12 UTe Day All Days W. Av-,! All Days W. Av-,! All Days W. Av-,! All Days W. Av-,! Day-I 1.00 0.99 0.96 0.95 0.98 0.97 0.94 0.93 Day-2 1.1 5 1.14 1.11 1.10 1.08 1.07 1.05 1.04 Day-3 1.39 1.37 1.36 1.34 1.22 1.19 1.17 1.16 5.2.2 Point Mode/ Data The 45 WS MIDDS workstations were given access to point model data from a MIDDS server at Johnson Space Center. The MIDDS point model data are available for the MesoNAM and GFS models. The MIDDS point model data have a hi gher vertical resolution than the gridded model data. For example, the vertical levels in the gridded model data are primarily confined to the NWS mandatory sounding levels (surface, 1000 mb, 925 mb, 850 mb, etc.). On the other hand, the point model data include all of th e model' s vertical levels. The MesoNAM contains 60 vertical levels from the surface to 14.5 mb, while the GFS model contains 64 vertical levels from the surface to 0.29 mb. The point model data also have a higher temporal re solution. While MIDDS gridded model forecasts are available every six hours, point model forecasts are available every three hours from the GFS and every hour from the MesoNAM. In order to increase the accuracy of the tool, the AMU updated the tool to use the higher resolution point model data. The tool ' s source code and installation instructions were then delivered to the 45 WS for operational use. 5.2.3 Using the MIDDS GUJ To use the MIDDS tool, the forecaster first opens it from the MIDDS Weather Menu. The tool reads in MesoNAM and GFS data from the latest 0000 and 1200 UTC model runs. After the tool has finished reading the model data, it displays the message " Initialization Complete". The user selects a model name

47 (NAM or GFS) and model run ("OOl" or" 12l") and then selects the "Calculate Peak Wind" bUlton. The tool calculates and displays the forecasts for PWSD, A WSD, daily average wind speed, and the probability the peak wind will meet or exceed 25 kt, 35 kt and 50 kt. Except for the daily average wind speed, separate forecasts are made for days with and without precipitation. Based on feedback from the 45 WS, the tool was updated to provide forecasts out to Day-7. Therefore, the MesoNAM forecasts are for Day- I to Day-3, while the GFS forecasts are for Day-I to Day-7. Figure 39 shows the tool 's forecasts from the model runs on 7 July 20 I O. The forecaster has the option of printing out the displayed forecasts by selecting the "Print Display" bulton. The user closes the tool by selecting the "Exit Tool" bulton. - ...... : ...., HAM • GFS ...... rw.: " ooz •

~':;~;:dt;:r~o:~;!:=o:-~:r:..~:" unit

•• P; ,. i4 ~ hou.r peW<. i-•• cenci wind ep •• d. tt .... aw:fac:. to 300 tt AO!. 5-11i.n ,. S-..inu.t.e __ ",ag. rind .p•• d at th. tt... ot th. paN:. rind, f,,_ .urf~. to 300 ft AOL Dly.__ 24_hour --""!if& rind ap•• 6 hoa 30 ft to 60 't MIL INti.u .. in9 )IAJII d.au tr_ 00 l1l"C run .... at .i9"Oa a-l. II .•• l"itialicin'l"" ute froa 12 ~ run on at .i9*" l.ev.l. 2. Initial..iain'l )lAX data IT .... DO tn'C run on .t .:i.ga.. a-l. 3 .. :tnitiallain9 DX dee.. tr .... 12 l1T'C run on .t. lIill'l'a l.ev.l. 3 . I.n.itia1icing .,.. data tr .... 00 l1l"C .. un .... at .i9"Oa Lev.l. 4 .. . lnitial..1c>.J\9 IiUIX dau tr .... 12 OTC run O<'l .t .ill'l'. t.v.l. 4 .. . tN.t.i.u.inV MItJII dee. fr .... 00 UTe run on .t ai~ I.e.. al. 5 .. . lnitial.l.c:r.t\9 IIAX au fr .... 12 ore nan on .t .i9M< l..vel 5 .. . lnitial.1cin9 IO\X data fr .... DO UTC run on .t aip. laval. 1.5 .•. :tniti.u.it\9 )IAJII ela ta It .... 12 urc r\WI on .t .ipa 1.eval. 15 ... %nitial.l.I;:I.n9 JfAX ate fro .. DO UTe run on at .i9"Oa l.aval. 17 •.. Xn.i.tia.l..1O::I.n9 MAlI data froa 12 UTC run on at ai9"O. laval. 17. :1nitiilici0\9 MM choe. fr .... 00 UTe run on .t aipa l.ev.l. 18.. Xnitial..iC1ll9 IfAX data frca 12 vrc run on .t .i~ Lev.l 18 •.. :rn.itial..i.in9 Dr s dar.. tr.... 00 UTC run on at ai~ laval. 2 .. . lnitialiZ1n9 ars dau fro. 12 urc run on at. aip_ l.ev.l. 2 .. . :1nitial.iC1n9 01'1 dac.a tr .... DO UTC run on at. .ipa lav.l. 3 .. . :bI.itial.i"int O. S data fro& 12 uro c ...... 0>'> .t .i9"O. l. .... l. 3 .. Zrtitial.iKi"9 OI'S clae.. tco. 00 UTe c ...... on .t .ip_ l.ev.l. 4 .. . tr.itialio:inq 01'1 dae.. rc.... 12 UTe .t .i... lav.1. 4 . . . :1rI.itial.io:in9 OI'S clae.. teo. 00 U'1"O t'UI\ 0>'> .t .ipa l..,.1. 10 .. Znitialilli"9 01'1 data tco. 12 uro t'UI\ on ..t .:i.gaa 1.e.,.1. 10. :lI\itialiunq O1'S dae.. tc.... 00 vrc IWI on at. .igaa 1.8 ... e1. 11 .. . :Initialillil\9 01'1 clata. tco. 12 U'1"O nm 0>'> ..t dgaa 1. ....1. 11 .. . 1nitialio::u.9' 01'1 data. fco. 00 In'C IWI on at. .i..... 1.8v.1. 1 2 .. . Xftitial..1O:Lnf 01'1 d .. ta fco. 12 U'1"O l'W\ on at. ei,... 1.8v.1 12 .. , Irtitial.ill .. tl0n COIlJlLe t. IlIO/CeAI'l 24- 8our (BAJI _8AJ1) ronc a . t

HAX 00.1 nlft I'r.e ip 1f000ncip All pcacip HoPreeip Pr. cip lfoPceeip lIeqin Du1 Puk(kt.) ::'.i(i~f S~~~t.) S- lf.in Olt) Dly"". Olt) I'r: ~ ~i~t p~:~i:e~{ Prb. 35 kt. Prb. 35 kt I' rb. SO k t pcb. SO kt. Oy - l 07 / 07 07/0B 27. 4 22. 7 lB. 2 >S. , ... 65. 1_ IS. I-" 10. 3'- 0 . 01t 0 . 0_ 0 . 0_ Oy-2 07/OB 071051 25. 22. 3 17. 3 14 . 5 ' .3 55. 511t 24. 71t 7 . 0'- 0 . 01t 0 . 0'- 0 . 0 .. 11J' - 3 01/051 07/ 10 22. 2 U . S 15. 2 10 . 6 ... 34 . 2_ 3. 0_ 2. lilt 0 . 01t 0 . 01t 0 . 01t IfU 12.1 cun Pn e i p lI'oPr.cip All pc-c ip HoP rec ip rr.cip HoI' recip r.ak(kt.) ::'..k(i~f 5~~~t.) S-lf.in(k.t.) D1.yA.,. {kt.) pr:~:.;i~t. p~:~i;C~{ rcb. 35 kt Prb 3S kt Prb. 50 kt. r rb, SO kt. Oy-l ~;m 0f't88 25. ' 22. 3 17. 3 14. 8 5 3 . B_ 21, n 5 . BIt 0 . 01t 0 . 0'- 0 . 0_ 22, 7 .3 .., 35. 0_ 0 . 0 .. 0 . 0 .. Oy-2 0710B 011051 14. 7 11. 7 0 . 2_ 1 . 81t 0 . 0'- 16, 6 ... .., 26. 6 .. 1 . lit 0 . 0 .. Oy- 3 07/051 01/10 n , 13. 1 10. 7 2. ' • 0 . 0 .. 0 . 0 ..

OpS ooz cun 1'recip Ho1' r.oi f 1'r.ci~ No1'ncip AU rrec ip MoPr.cip P.ak(ltt. S - Ilin~ t ) 5-MinOlt) DlyAv. Olt) 1'r:~2Si~t 1'~:1'2Sc~~ Prb. JS kt P rb. 3S kt. P:b~~~P ltt 1'~:~~~o~t Oy- l ~;~1 O~gB "~~t) 19. 1 lS. 12. 6 ... :W . 21t 4 . 7_ 1. . Bit ,a. 0 . 01t 0 . 0 .. 11J'- 2 07108 071051 23 6 n . l >S. ' 12. 51 5 . ' 41. lit 6 . 5I1t 3 . n ,a. 0 . 0 1t 0 . 01t Oy- 3 071051 07/10 ZS . O 18. 3 17. 1 12. 1 5 . 5 451 . 81t 7 . 01t 6 . Bit 0 . 01t 0 . 01t 0 . 01t 0.,-4 07 / 10 07/11 ZS 5 110 . 51 17. 5 13. 6 5 . ' 5 2 . ,_ U . 4" 8 . O. ,a 0 . 01t 0 . 01t 0.,- 5 07/ 11 07112 25. 4 21 . 7 17. 4 14. 1 ' . 5 52. 4. 23. Sit 7 . S. 0 . 11t 0 . 01t 0 , 0. 0.,- ' 07112 07113 26. " 21. , lS. 4 14. 1 ,., 60. Bit 24. 11 .. 11. 5' 0 . 1' 0 . 01t 0 . 0' 0.,_7 07/ 13 07/ 14 2B . 5I 23. 6 151 . B 15. 6 .., 71. ,.- 37, 11'- 18. 4' 0 . 5' o . n 0 . 0' on 12'1 run rrec ip Ifol'r.c ip All 1'rec ip )JoPncip PnOlP NoPreoip lIagift End reak(kt.) ;::k(k~f S~~~~t.) S · lfin(kt.) DlyAv. {kt.) pr:~2Si~t. 1'~:.'2Sc~{ rrb. 35 kt. rcb. 35 Itt. pcb. SO kt. prb. SO kt Oy- l 01/07 0710B 23. 6 n . l >S . • 13. 5 5.' 41. n 4 . 2' 2 . B. 0 , 0' 0 . 0 ' 0 . 0'- Oy-2 07/oS 071051 20 . 0 >S. , 13. 1 3 3. o. a . 7' 1 . 6. ,a. 0 . 0. 0 , 0. ... 0 , 0. 0.,_3 07/051 07 / 10 15, 7 12. 3 10. 1 ' . 5 2a. e. 1 . a. 1. 4. 0 . 0. 0 . 0111 Oy-4 07110 07/ 11 '"21 . 5 17. 1 13. 6 11. 3 .., 2B . B. 4 . 3. 1 . 41t 0 . 0111 0 . 0 . 0 , Dill Oy- s 07/11 01/ 12 '" 18. 7 14 . , 12. 3 S . 3 31 . 1111 B. O. 1 . 6" ,a. 0 . 0' 0 . 0" 0.,-' 07112 07113 ".255 21. 1 16. 5 14. 0 •., 5 3 . 1 .. UI . 3. ' . 2' D. O. 0 . 0' 0.,- 1 07113 07114 26, 4 23. 11 18. 8 16. 1 ' . S .511 , lit 40, 4 .. 8 . 4111 '",o. 0 . 0111 0 . 0. Figure 39. MIDDS version of the Peak Wind Tool showing forecasts from the 7 July 2010 model runs. Forecasts for precipitation days have the header "Precip", while non-precipitation days have the header "NoPrecip". The header "A ll" includes days with and without precipitation.

48 6 Summary and Future Work The expected peak wind speed is an important element in the 45 WS daily 24-Hour and Weekly Planning Forecasts. The forecasts are used for ground and space launch operations at KSC and CCAFS. The 45 WS forecasters have indicated peak wind speeds are challenging to forecast, especially in the cool season months of October - April. In Phase I of this task (Barrett and Short 2008), the AMU developed a Microsoft Excel tool to help the 45 WS forecast non-convective winds at KSCICCAFS for the 24-hour period of 0800 to 0800 local time. The tool displayed forecasts for the PWSD, A WSD, timing of the PWSD and the probability the PWSD will meet or exceed the 45 WS wind advisory thresholds of 35 kt, 50 kt and 60 kt. The wind advisory threshold for 60 kt wind gusts was later dropped and a threshold for 25 kt added. For Phase II , the 45 WS requested the AMU expand the POR to increase the number of observations used in the creation of the prediction equations. In addition to an Excel tool, the 45 WS also requested a MIDDS tool to provide forecasts out to seven days.

6.1 Summary The AMU first expanded the POR by two years, by adding KSCICCAFS tower observations, SLF surface observations and XMR soundings for the cool season months of March 2007 to April 2009. The prediction equations required sounding data in 100-ft increments. The AMU evaluated whether the POR could be expanded even more by interpolating 1000-ft sounding data down to 100-ft increments. A comparison showed no significant differences between the interpolated and 100-ft sounding data. Therefore, the POR was expanded again by six years after adding interpolated sounding data for the cool season months of October 1996 to April 2002. The Phase II developmental data set contained tower observations, surface observations and soundings for the cool season months of October 1996 to February 2007. The AMU calculated candidate predictors from the XMR soundings. The predictors included 19 stability parameters, 48 wind speed parameters and one wind shear parameter. The data set was stratified by sy noptic weather pattern, low­ level wind direction, precipitation, Richardson Number and Gradient Richardson Number, for a total of 60 stratification methods. Linear regression equations, using the 68 predictors and 60 stratification methods, were created for the tool 's three forecast parameters (PWSD, A WSD, and timing of the PWSD). Instead of selecting the most accurate Phase II method for each forecast parameter, several well­ performing methods were selected for evaluation in the verification data set. The verification data set contained observations for the cool season months of March 2007 to April 2009. The verification data set was used to compare the Phase I and II prediction methods to climatology, model forecasts and wind advisories issued by the 45 WS . The comparison was first performed for PWSD and A WSD. The Phase II methods performed slightly bener than Phase I and significantly better than climatology. The Phase I and II methods were then compared to the 0000 and 1200 UTC runs of the MesoNAM. The MesoNAM data contained wind forecasts for model levels 2 - 18 (between 207 ft and 3145 ft MSL), as well as the strongest winds in the lowest 1000-, 2000- and 3000-ft of the model. Linear regression equations were created, in which the predictor was the 24-hour peak speed in the MesoNAM data and the predictand was the PWSD or A WSD. The MesoNAM linear regression equations performed significantly better than the Phase I and II methods. The Phase I and II forecasts of PWSD were compared to wind advisories issued by the 45 WS. On days in which the 45 WS issued at least one non-convective wind advisory, the 45 WS out-performed the Phase I and II methods. The verification data set was then used to evaluate the Phase I and 11 methods for the timing of the PWSD. The Phase I and II methods were compared to climatological winds and the 0000 and 1200 UTC runs of the MesoNAM. The MesoNAM forecasts were derived by determining when the 24-hour peak wind occurred at each model level. The Phase I and 11 methods, as well as the MesoNAM forecasts, did not perform significantly bener than climatology.

49 The Microsoft Excel and MIDDS versions of the tool were then developed by the AMU. The tool displays forecasts for the PWSD, A WSD and the probability the PWSD will meet or exceed 25 kt, 35 kt and 50 kt. The probabilities, based on the PWSD, were calculated using a statistical method developed by the 45 WS (Barren and Short 2008). For PWSD and A WSD, the tool uses linear regression equations based on the 0000 and 1200 UTe MesoNAM forecast winds. Since none of the prediction methods for the timing of the peak wind performed significantly bener than climatology, the tool does not provide forecasts for the timing of the peak wind. Instead, the forecaster should use climatology to forecast the timing of the peak wind. The forecaster can then adjust the timing based on the movement of weather features. The Excel and MIDDS GUls display forecasts for Day-I to Day-3 and Day-I to Day-7, respectively. The Excel GUI uses MesoNAM point forecasts as input, while the MIDDS GUI uses point forecasts from the MesoNAM and GFS models. The AMU also updated the Microsoft Excel and MIDDS tools to provide the daily average wind speed from 30 ft to 60 ft AGL, which is one of the forecast parameters in the 24-Hour and Weekly Planning Forecasts issued by the 45 WS.

6.2 Future Wo rk The 45 WS proposed a follow-on task (Phase III) to the Peak Wind tool. Instead of providing 24-hour forecasts from Day-I to Day-7, the Phase III tool would display 4-hour forecasts for Day- I, then 12-hour forecasts for Day-2 to Day-7. The Phase III tool would not provide forecasts for AWSD or timing of the PWSD. Instead, the tool would provide forecasts for PWSD, daily average wind speed and the probability the PWSD will meet or exceed 25 kt, 35 kt and 50 kt. Phase III would also expand the POR for the MesoNAM forecasts to October 2006 to April 2010.

50 References Barrett, J. H. and D. A. Short, 2008: Peak Wind Tool for General Forecasting Final Report. NASA Contractor Report CR-2008-214743, Kennedy Space Center, FL, 59 pp. [Available from ENSCO, Inc. , 1980 N. Atlantic Ave. , Suite 230, Cocoa Beach, FL 3293 I and http://science. ksc.nasa.gov/am u/fi nal .htm I) Case, J. L. and W. H. Bauman III , 2004: Tower Mesonetwork Climatology and Interactive Display Tool Final Report. NASA Contractor Report CR-2004-21 1526, Kennedy Space Center, FL, 80 pp. [Available from ENSCO, Inc. , 1980 N. Atlantic Ave., Suite 230, Cocoa Beach, FL 32931 and http://science.ksc.nasa.gov/am ulfinal .htm J) Insightful, 2002: S-PLUS 6 Robust Library User's Guide, Insightful Corporation, Seattle, WA, 188 pp. Insightful, 2007: S-PLUS 8 Guide to Statistics, Volume I, Insightful Corporation, Seattle, WA, 7 18 pp. Lambert, W. C., 2002: Statistical Short-Range Guidance for Peak Wind Speed Forecasts on Kennedy Space Center/Cape Canaveral Air Force Station: Phase I Results Final Report. NASA Contractor Report CR-2002-21 I 180, Kennedy Space Center, FL, 33 pp. [Available from ENSCO, Inc. , 1980 N. Atlantic Ave., Suite 230, Cocoa Beach, FL 3293 I and hnp:llscience.ksc.nasa.gov/amulfinaJ.html)

51 List of Acronyms Term Description 45 WS 45th Weather Squadron AGL Above Ground Level AMU Applied Meteorology Unit AWSD 5-minute average wind speed at the time of the PWSD CCAFS Cape Canaveral Air Force Station CSR Computer Sciences Raytheon EST Eastern Standard Time FAQ Frequently Asked Questions GFS Global Forecast System GUI Graphical User Interface KSC Kennedy Space Center MAE Mean Absolute Error MesoNAM 12-krn North American Mesoscale Model MIDDS Meteorological Interactive Data Display System MSL Mean Sea Level NAM 40-km or 80-km North American Mesoscale Model NWS National Weather Service PAFB Patrick Air Force Base POR Period Of Record PWSD Peak Wind Speed of the Day QC Quality Control SLF Shunle Landing Facility Tclrrk Tool Command Languagerrool Kit XMR CCAFS rawinsonde 3-lener identifier

52 NOTICE Mention of a copyrighted, trademarked or proprietary product, service, or document does not constitute endorsement thereof by the author, ENSCO Inc., the AMU, the National Aeronautics and Space Administration, or the United States Government. Any such mention is solely for the purpose of fully informing the reader of the resources used to conduct the work reported herein.

53