SED navigation bar go to SED home page go to Extreme Winds page go to NIST home page SED Home Page SED Staff SED Projects Extreme Winds Publications Search Wind Pages

Standardized extreme wind speed database for the United States)

1. Introduction
Introduction This document contains information on the initial development of a standardized extreme wind speed database for the United States, including a description of the database, how the relevant wind data are extracted, processed and quality controlled and finally how the data are standardized. This document also contains links to relevant documents pertaining to this database, as well as the data files that were developed by the methodology described herein.
2. Description of the Database
National Climatic Data Center (NCDC) Data The wind speed data are available from the National Climatic Data Center (NCDC) via ftp link. The data used for this project were the Integrated Surface Hourly (ISH) Data Set 3505. Data Set 3505 contains a large amount of meteorological information at approximately 20,000 stations worldwide reported at hourly time intervals. For further details on this dataset please see the Format Documentation report (PDF format). Data Set 3505 is very similar to Data Set 9956, a format used in the formulation of ASOS-WX and used in Lombardo et al. (2009) (PDF format).
File Download For the purpose of this work, ISH 3505 data was downloaded from stations in the United States via the ftp site. The stations used in this document contained 5 or more years of wind data for analysis and are listed in stations_analyzed.csv (csv file). The dates of the data for the US stations typically range from the early 1970's to the present day. The methodologies and techniques described here can be used for any available station throughout the world depending on data availability and recording methods at each station. The entire list of stations in the United States and the world as well as dates of data availability is given in ish_history_all.csv (csv file). In addition to dates of data availability, this file contains information on a number of important variables, including station identification number, which is used as a defining characteristic for the standardized data discussed in Section 5.
3. Data Extraction, Processing and Quality Control
Data Extracted Using ASOS-WX Once the data have been downloaded via the ftp site described in Section 2, all pertinent wind data and related information were extracted using ASOS-WX. Some modifications to the original ASOS-WX program have been made, but are not currently available for download on the internet. Interested parties may contact or for inquiries about the updated ASOS-WX, which also is able to download data of interest as is discussed in Section 2.

ISH 3505 data sets come in specified archived format which is accounted for in the program. This format is available in the Format Documentation report (PDF format). Except for the location, the date and time of the observation in the ISH 3505 format (in positions 16-27), the format is the same as described on the website (ASOS-WX Documentation) using Data Set 9956.

Peak Wind Speed and Direction Data Two types of peak wind speed designations ('PK WND' and 'OC') are available and were extracted in the ISH 3505 files, a departure from the previous analysis of Lombardo et al. (2009) (PDF format) where only 'PK WND' values were extracted. Figure 1 shows an example of how the wind speed reports appear in the ISH 3505 files.

If both designations occurred at the same time on a report line (time is in UTC), the maximum wind speed was used and the time recorded. The 'OC' wind speed reports associated with the 'PK WND' reports were always equal to or less than 'PK WND' reports and an 'OC' report was almost always associated with a 'PK WND' report. This is likely due to the 'OC' reports corresponding to the time of the archived data line (usually hourly in the ISH database). The occurrence of both peak wind speed designations is shown in Figure 1. The 'OC' value would be associated with the time on the beginning of the line. In this case the time is July 11, 1993 at 2200 UTC ('19930711200') and a value of 33.4 m/s. The 'PK WND' data is taken from any point of time over the previous hour or reporting interval (Figure 1), and has a specific time associated with it, in this case 2205 UTC with a value of 88 kt. The ISH reports are typically recorded hourly, but reports were separated occasionally by less than an hour, especially at times of significant weather at the station. This would cause the 'OC' values to underestimate or "miss" actual peak wind speeds especially from small-scale events such as thunderstorms. Therefore 'OC' reports not associated with a 'PK WND' report were disregarded for the development of the standardized wind speed databases discussed here. In addition, some stations had only 'OC' reports, but those stations were not used in the analysis due to their reporting of wind speeds not always associated with the peak wind speed. Regardless, the data extraction program is able to extract these 'OC' reports as well. These two peak wind speed reports are not differentiated explicitly in the available processed data files (Section 5), however the raw wind speeds of the 'OC' data typically contain decimals as they are reported in m/s and then converted to knots, while the raw 'PK WND' reports are originally in knots. All winds when standardized are converted to miles per hour (mph) as discussed in Section 5.

The associated wind direction at the time of the report data is logged in degrees clockwise from true north and is recorded in 10 degree increments (10, 20, ..., 360) for 'PK WND' reports only. For Figure 1, the wind direction is read as 40 degrees '04', the first two digits following a 'PK WNDn.

0488/2205 PRESRR
Figure 1. Example of a line containing both a 'PK WND' and 'OC' wind gust report (in red text).
Thresholds For the purposes of this work, the ISH 3505 data contains peak wind gust data as described in Lombardo et al. (2009) (PDF format). This peak wind gust data ('PK WND') has a built-in threshold. For all years after the reporting of peak wind speeds at the stations were made automated (ASOS; see Averaging Time), the threshold was set at 25 knots. In the years previous to ASOS commissioning, the threshold varies depending on the station and the year. The ASOS-WX program and this document detail how the peak wind gust values are extracted from the data at large. Once the peak wind gust data are extracted, a time series plot of wind gusts for any station will help the user determine what the threshold of the data was at a particular time. Other wind speed thresholds, corresponding to extreme value analysis methods (Simiu and Heckert, 1996) (PDF format) can be chosen in post-processing depending on user needs. The gust wind reports denoted as 'OC' have no associated threshold value and are typically included with every hourly report, however a 25 kt threshold was set for all 'OC' data, if as stated before, it occurs on the same line as 'PK WND' report.
Anemometer Elevation The majority of ISH stations currently have an anemometer which is placed at the standard height of 10 m. However, if a station, at any point in time has an anemometer at a different height this information is available and can be accessed in the files described in Section 5. The conversion of wind speed values to a standard 10 m height is discussed in Section 4, and a detailed example is shown at the end of this document. If the anemometer elevation was unknown, it was assumed to be 10 m.
Averaging Time The averaging time of the ISH peak wind speed and direction data is determined based on the specific dates of anemometer changes. Typically three distinct periods are available in the ISH files. Information on these periods can be found in the processed files described in Section 5.

The first period was prior the station reporting becoming automated (ASOS). This period typically ranged from the early 1970's (for the ISH data) to the mid-1990's. Wind speeds were collected via anemometer/chart recorder system. The resultant wind speeds pre-ASOS had no digital filtering and the averaging time was dependent on the mechanical properties of the anemometer and the wind speed (McKee et al., 1996; Miller, 2007). For higher wind speeds, these effective averaging times were approximately 1-3 s. For conservatism, these values in database discussed in Section 5 remained the same as recorded for use in this database.

The second period, after ASOS commissioning, which typically ranged from the mid-1990's to the mid 2000's, the data were digitally sampled at 1 sample per second (1 Hz) and a 5-second block average (average of 5 consecutive samples) was used along with a cup anemometer. The wind data recorded within these periods are modified as discussed in Section 4.

In the third and final period, beginning in the mid to late 2000's, sonic anemometers were installed at a number of the ISH stations. The anemometers remain at the stations. The data recorded by sonic anemometers are also sampled at 1 Hz and digitally output a 3-s moving average peak wind gust as per WMO standards (WMO, 2008). Wind speeds recorded after the switch to sonic anemometers were also left unmodified. The conversion of wind speed values from all three periods to a standard 3 s gust value is discussed in Section 4, and information of the period of anemometer changes (i.e., averaging time changes) are available in the processed data files discussed in Section 5.

If no information on anemometer changes (averaging time) was found, the data remained the same as recorded in the raw file.

Terrain Roughness Terrain roughness for all stations for preliminary analysis was assumed to be in flat, open terrain exposure (a roughness length, zo, equal to 0.03 m). A number of methodologies are available for converting wind speeds at stations that may not have the characteristics of open terrain exposure. Two of these methodologies are described in Sections 4 and 6 as well as Example 1 at the end of this document.
Storm Type Separation The importance of separating by storm type has been illustrated in a number of publications (Gomes and Vickery, 1977; Lombardo et al. (2009) (PDF format). Different storm types have different statistical distributions, and combining all wind speeds together when performing an extreme value analysis may lead to wrong conclusions about an extreme wind climate. In addition, it may be best to account for all storm type distributions independently in a "mixed" distribution (see Simiu, 2011, p. 143).

The ISH 3505 data set has a number of codes that denote a thunderstorm in progress or in the general vicinity of the station as noted by observers. The ASOS-WX program looks for three specific sets of codes when attempting to determine whether a wind speed was generated by a thunderstorm. The first two sets, discussed in detail in Lombardo et al. (2009) (PDF format), involve specific times of thunderstorm beginning and ending ('TSB' and 'TSE') and manual weather reports ('MW') associated with the time of the ISH report. The times of these reports were then 'windowed' by 1.5 hr on either side. Any wind speeds within this window were classified as a wind speed generated from a thunderstorm.

As a careful manual check revealed some storm misclassification from the original ASOS-WX program (which only separates thunderstorm from non-thunderstorm) , a third set of codes related to thunderstorms other than thunderstorm beginning and end reports or manual weather reports were employed. These codes included, for example, 'TSRA' – thunderstorm w/rain and 'CB' - cumulonimbus clouds. Manual checks of reports as well as archived radar data are able to further reduce errors in storm type classification and in high wind speeds. While it would beneficial to manually go through each data point to check for any errors, it would require prohibitive amounts of time to do so. Figure 2 shows an example of these codes in the ISH files. The time of report is again at the beginning of the line. The code 'MW295' stands for the second manual weather observation on the line and the '95' is a code that references a thunderstorm. According the observer at this particular station the thunderstorm began at 2225 UTC (hour of report line and minute of TSB or TSE code) and ended at 2258 UTC. The 'CB' code also occurs on this line.

199307112200.....9999MW1655MW2955OC102101REMAWY068BKN V SCT TSB25 TSE58 PCPN
Figure 2. Example of different types of thunderstorm codes/reports in ISH files

In addition, wind speeds generated from tropical cyclones were also extracted from the database. Initially, for this database, all wind speeds within 200 km of a tropical cyclone center and within a 12 hr time window on either side of the time listed by the HURDAT database were separated from the other wind speed data and classified as being generated from a tropical cyclone.

Information of all storm types are denoted in the processed files described in Section 5.

Reduced Statistical Dependence Once wind speed values are separated by storm type, the values can be separated by a time period that would reduce statistical dependence between individual values. Correlated values, when used in an extreme value analysis, can affect the results negatively. Therefore only the maximum wind speeds in “independent” events, within some time period are typically used in extreme value analysis.

The reduction of statistical dependence in wind speed values was done in accordance with Lombardo et al. (2009) (PDF format). This procedure is also explained in ASOS-WX documentation. For the data described in Section 5, non-thunderstorm wind speeds with reduced statistical dependence were separated by no less than 4 days and thunderstorm wind speeds with reduced statistical independence by no less than 0.5 days. The updated ASOS-WX program, when made available, will give the user the option to choose their own time periods to reduce statistical dependence.

Quality Control Methodologies While ASOS- WX is able to perform reasonable quality checks and the archived data has some quality control built-in, additional checks were needed as some of the extracted wind speeds appeared to be erroneous. Additional upgrading of the program to mitigate these issues included a manual check of all wind reports in excess of 60 knots (~70 mph) and a complete discarding of all wind speeds greater than 110 knots (~127 mph). As stated previously, when made available, the updated ASOS-WX program will enable the user to quality control any subset of data.

Stations that were determined to have regular reports of erroneous wind data were removed from the available data files (Section 5). In addition, stations that had good reports of wind speeds that were consistently of high magnitude or were already in a location of a "Special Wind Region" as determined by ASCE 7-10 (SEI, 2010) were placed in a "Special Wind Region" folder.

4. Data Standardization
Converting the Data using ASCE 7-10 The quality controlled data were then standardized using the methodology found in the ASCE 7-10 (SEI, 2010). For this database, wind speed data measured at elevation z over terrain with roughness length zo and averaged over time t were converted to standardized conditions z = 10 m, z0 = 0.03 m and t = 3 s. This methodology includes the use of the power law, with a constant power law exponent, alpha.

Assuming initially that all wind speeds occur in open terrain as discussed previously, the first conversion involves the modification of all wind speeds corresponding to a 10 meter height, if it isn’t already. This modification is performed using equation 1.

To convert from the 5-s (only wind speeds occurring at times between ASOS commissioning and sonic anemometer switch) to 3-s gust involves the use of the Durst curve. This factor, for open terrain, is 1.03.

Using the methodology in ASCE 7-10:

    V(3s,0.03m,10m) = V(5s,0.03m,h)×[(h/10)^(1/α)]^(-1)×1.03        (1)

alpha = 9.5 for open terrain (Exposure C) in ASCE 7-10 and h is the height of the anemometer as specified here.

Converting the Data using Theoretical Formulations

In general, known or estimated z0 values for the ISH stations (assumed to be 0.03 initially) can be coupled with information on anemometer response, relevant elements of filtering theory (Masters et al. (2010), and relations between mean wind speeds in different micrometeorological conditions (Simiu, 2011), to convert peak wind speeds to standardized conditions. This can be done using Equation 2.

    Uhat(3,10,0.03) = 
{Uhat(z,zo)*Uhat(3,z,zo)/(U(10,0.03)*U(z,zo))}        (2)

where U(t,z,zo) is the peak wind speed averaged over time t at height z over terrain with roughness length zo (i.e., a 'PK WND' report for the ISH data). As an example, the factor to convert from a 5-s block average gust (ASOS commissioning to sonic installation) and a 3-s moving average taking the information in to account described here is 1.065, greater than the 1.03 using the Durst Curve described above. The roughness length value, zo can be estimated by using the information in Section 6.

5. Standardized Wind Speed and Direction Data
Available Files Standardized wind speed and direction databases, with identifying information described in Sections 3 and 4 are available Contained within the files is the information described below. If no information was listed for a particular file header, that information was not available.
File Name The name of the file itself contains the six digit station code number. These numbers correspond to the codes in the file here. Within each of these files, header information is available on:
File Headers Station Description: Name or identification of station

Latitude/Longitude: Coordinates of station in decimal degrees

Changed to ASOS: Date of ASOS commissioning, change to 5-s block averaging time and possible change to 10 m anemometer elevation

Changed to Sonic: Date of change to sonic anemometer, and change to 3-s moving averaging time

Anemometer Height Change Date: Date of anemometer height changes

Anemometer Heights (ft): Value in feet, of new anemometer height changed at the corresponding date

File Body Date/Time: Date and time of peak wind report

Wind Speed (Raw, kt): Original wind speed value (in knots) contained in the files

Wind Speed (Standardized, mph): Standardized wind speed in mph

Wind Direction (deg): Wind direction in 10 degree increments

Non-Thunderstorm: The number "1" denotes an identified non-thunderstorm wind speed, while "0" denotes a wind speed that was not identified as non-thunderstorm

Independent Non-Thunderstorm: The number "1" denotes a non-thunderstorm wind speed that has reduced statistical dependence as was described in Section 3. The number "0" denotes a non-thunderstorm wind speed that did not meet these criteria.

Thunderstorm: The number "1" denotes an identified thunderstorm wind speed, while "0" denotes a wind speed that was not identified as thunderstorm

Independent Thunderstorm: The number "1" denotes a thunderstorm wind speed that has reduced statistical dependence as was described in Section 3. The number "0" denotes a thunderstorm wind speed that did not meet these criteria.

All Independent: The number "1" denotes a wind speed regardless of storm type that has reduced statistical dependence as was described in Section 3. The number "0" denotes a wind speed that did not meet these criteria.

Tropical: The number "1" denotes an identified tropical wind speed, while "0" denotes a wind speed that was not identified as thunderstorm or non-thunderstorm

Annual Max (NT): The number "1" denotes a maximum non-thunderstorm wind speed in a calendar year.

Annual Max (T): The number "1" denotes a maximum thunderstorm wind speed in a calendar year.

Annual Max (ALL): The number "1" denotes a maximum wind speed regardless of storm type in a calendar year.

6. Additional Roughness Estimations
Estimating Roughness from Aerial Photos Aerial photos showing terrain roughness in all directions were collected for the majority of the stations used in the analysis. The filenames contain both the station number and the station call letters (i.e., Wichita Mid-Continent Airport has the station number 724500 and call letters KICT). The database of aerial photos can be found here. Roughness length values (zo) can be estimated from aerial photos using only this relationship (Lettau, 1969):

    z(o)=0.5*H(ob)*(S(ob)/A(ob))       (3)
Example 1 For aerial photos, the average height of the obstacles (Hob), the average vertical area per obstruction, Sob, and the area occupied by obstacles (Aob) will have to be estimated using a reasonable estimate of building dimensions given the aerial photos. Plan (aerial) densities for available ISH/ASOS stations can be found in each of eight directional sectors (Excel file) to aid in estimation of Aob. Other methods to calculate zo using plan density can be found in Grimmond and Oke (1999).

Figure 5 shows an aerial photo from an airport station in the US. The location of the anemometer is noted by the text 'KBHM' and the photo extends approximately 1/2 mile (800 m) out from the anemometer location in every direction. All aerial photos in the database have an approximate radius of 800 m out from the anemometer location. Figure 5 has the south sector (wind directions 157.5 to 202.5 degrees) highlighted. Assuming that Hob = 6 m and Sob = 2000 m2 (6 buildings assuming to have a 6 m x 50 m frontal area, plus trees) and an estimated Aob of approximately 65,000 m2 [e.g. 18π80020.26]. The 0.26 value is a plan (aerial) density value for the highlighted taken from the file mentioned above.

Then using Eq. 3, the estimated roughness value is approximately 0.09 m. More detailed measurements of the obstacles included in the aerial photos can be made using other spatial analysis tools (e.g., Google Earth, ArcGIS).

    Aerial Photograph

Figure 5. Aerial photo (PNG format) for KBHM with 45 degree sector highlighted
Estimating Roughness form ASOS Data In addition to the method described above, a second method of zo estimation can be used and is outlined in Masters et al. (2010). Wind speed data that met stationary conditions and had a mean wind speed greater than 8 m/s (17 mph) were used to calculate an estimated zo for approximately 800 ISH stations that were available. The zo values were estimated over eight 45 degree sectors centered on 0, 45, ..., 315 degrees. Both average zo values as well as the entire time history of zo over the 8 sectors are available for the approximately 800 ISH stations at the link here. If a particular sector did not have 30 or more values within it, 'NaN' values were assigned to that sector. There were approximately 5300 sectors that had enough data to calculate a mean zo value. A cumulative distribution function of all zo values over the approximately 5300 sectors is shown in Figure 4. The zo value at the 50% percentile is approximately 0.06 m. For example in the same sector identified in Example 1, the calculated roughness length was estimated to be approximately 0.15 m.

    Aerial Photograph

Figure 4. Cumulative distribution function of zo values from approximately 5300 wind direction sectors (PNG format)
Converting zo to alpha Once an estimation of zo has been achieved, this value can be converted back to a power law exponent value, using the following relationship found in ASCE 7-10:

    alpha = c(1)*z(o)^(-0.133)

where c1= 5.65 if computing roughness length values in meters.

Example Conversion For example, if the zo value of 0.15 m is used from above, the analogous alpha value is approximately 7.3. This 7.3 value can be used in, for example, Eq. 1.
7. References
Grimmond, C. S. B., & Oke, T. R. (1999). Aerodynamic properties of urban areas derived from analysis of surface form. Journal of Applied Meteorology, 38(9), 1262-1292.

Gomes, L., & Vickery, B. J. (1977). On the prediction of extreme wind speeds from the parent distribution. Journal of Wind Engineering and Industrial Aerodynamics, 2(1), 21-36.

Lettau, H. (1969). Note on aerodynamic roughness-parameter estimation on the basis of roughness-element description. Research and Development Technical Report, 163.

Lombardo, F. T., Main, J. A., & Simiu, E. (2009). Automated extraction and classification of thunderstorm and non-thunderstorm wind data for extreme-value analysis. Journal of Wind Engineering and Industrial Aerodynamics, 97(3), 120-131.

McKee, T. B., Doesken, N. J., & Kleist, J. (1996). Climate Data Continuity with ASOS: Report for the Period September 1994-March 1996. Colorado Climate Center, Department of Atmospheric Science, Colorado State University.

Masters, F. J., Vickery, P. J., Bacon, P., & Rappaport, E. N. (2010). Toward objective, standardized intensity estimates from surface wind speed observations. Bulletin of the American Meteorological Society, 91(12), 1665-1681.

Miller, C. (2007). Defining the effective duration of a gust. Proceedings 12th International. Conference Wind Engineering.

Simiu, E., & Heckert, N. A. (1996). Extreme wind distribution tails: A "peaks over threshold" approach. Journal of Structural Engineering, 122(5), 539-547.

Simiu, E. (2011). Design of buildings for wind. John Wiley & Sons.

Structural Engineering Institute (SEI). (2010). Minimum design loads for buildings and other structures (Vol. 7). American Society of Civil Engineers.

World Meteorological Organization (WMO) (2008). Guide to meteorological instruments and methods of observation. 8th Ed., 681 pp.

SED Home |  Extreme Winds Home |  Previous |  Next ]

Privacy Policy/Security Notice
Disclaimer | FOIA

NIST is an agency of the U.S. Commerce Department.

Date created: 02/17/2012
Last updated: 10/03/2016
Please email comments on this WWW page to