Example 3 illustrates the computation of a peak flow frequency curve using EMA and Bulletin 17C procedures with an annual maximum series comprised of a broken record of systematic flood peaks. For this example, USGS gage 01614000 Back Creek near Jones Springs, West Virginia is used. Back Creek is a tributary to the Potomac River; the 235 square mile watershed lies within the Valley and Ridge province in West Virginia (England, et al., 2018). Gage 01614000 has an annual peak record consisting of 56 peaks beginning in 1929 and ending in 2012. There are three "broken record" periods where the gage was discontinued: 1932-1937, 1976-1991, and 1999-2003. Thus, there are 28 years of missing data at this gage during the period 1929-2012. There is a historic flood that occurred on March 17, 1936, during one of the periods when the gage was discontinued. This flood is noted in the USGS Annual Water Data Report for this gage, available in the peak-flow file, and there is historical information available for this large flood (Grover, 1937).
The Back Creek near Jones Springs, WV stream gage has an annual peak record consisting of 56 annual peaks. Of the 56 annual peaks, the October 1942 flood slightly exceeds the March 1936 historic flood peak. Based on the historical flood information in Grover (1937) for the 1936 flood, and the large regional floods and historical floods described by Wiley and Atkins (2010) in West Virginia for the period 1888-1996, information from the March 1936 flood can be used as a perception threshold to represent the 28 years of missing information. This assumes that if a flood equal to or larger than the March 1936 event occurred when the gage was not operational, it would have been noted and recorded in some way. The annual maximum series is plotted in Figure 1 and tabulated in Table 1.

Figure 1. Back Creek near Jones Springs, WV Annual Peak Flow Record.
Table 1.  Back Creek near Jones Springs, WV Annual Peak Flow Record.

Date

Flow (cfs)

17 Apr 1929

8750

23 Oct 1929

15500

08 May 1931

4060

17 Mar 1936

22000

04 Feb 1939

6300

20 Apr 1940

3130

06 Apr 1941

4160

22 May 1942

6700

15 Oct 1942

22400

24 Mar 1944

3880

18 Sep 1945

8050

03 Jun 1946

4020

15 Mar 1947

1600

14 Apr 1948

4460

31 Dec 1948

4230

02 Feb 1950

3010

05 Dec 1950

9150

28 Apr 1952

5100

22 Nov 1952

9820

02 Mar 1954

6200

19 Aug 1955

10700

15 Mar 1956

3880

10 Feb 1957

3420

27 Mar 1958

3240

03 Jun 1959

6800

09 May 1960

3740

19 Feb 1961

4700

22 Mar 1962

4380

20 Mar 1963

5190

10 Jan 1964

3960

06 Mar 1965

5600

21 Sep 1966

4670

08 Mar 1967

7080

17 Mar 1968

4640

02 Feb 1969

536

10 Jul 1970

6680

13 Nov 1970

8360

22 Jun 1972

18700

09 Dec 1972

5210

27 Dec 1973

4680

20 Mar 1975

7940

05 Mar 1993

11800

08 May 1994

8730

16 Jan 1995

2300

19 Jan 1996

13900

09 Nov 1996

4190

21 Mar 1998

6370

29 Sep 2004

9460

29 Mar 2005

6560

30 Nov 2005

2000

16 Apr 2007

5040

21 Apr 2008

7670

05 May 2009

4830

14 Mar 2010

9070

17 Apr 2011

10300

01 Mar 2012

4650

A Bulletin 17 Analysis using EMA and Bulletin 17C procedures has been developed for this example. To open the analysis, either double-click on the analysis labeled "B17C Example 3" from the Study Explorer or from the Analysis menu select open, then select "B17C Example 3" from the list of available analyses. When "B17C Example 3" is selected, the Bulletin 17 analysis editor will appear as shown in Figure 2. As shown, the Skew option was set to use the Station Skew.

Figure 2. Bulletin 17 Analysis General Tab for B17C Example 3.
No changes to the Options tab are necessary. The EMA Data tab for this example is shown in Figure C-17. This example uses an annual maximum series consisting of both systematic data along with an historical event in March 1936. Also, the record is broken with multiple periods of missing with a complete record. Since 17C EMA requires a non zero – inf perception threshold for all periods of missing data, a total of four perception thresholds are required. In this case, the March 1936 event can be used to inform the perception thresholds for the periods of missing annual peak flow data. The use of a perception threshold of 21000 – inf for these periods of missing data implies that had a flood event occurred with a peak flow greater than 21,000 cfs, someone would have measured and recorded it. Once all four perception thresholds have been entered as shown in Figure 3, click the Apply Thresholds button to assign the complementary flow ranges for the periods of missing data. Finally, the March 1936 event should be set to the historical data type, as denoted by a USGS peak flow rate qualification code of "7", because it occurred during a period when the gage was discontinued.

Figure 3. Bulletin 17 Analysis EMA Data Tab for B17C Example 3.
Once all of the General and EMA Data tab settings are set or selected, the user can press the Compute button to perform the analysis. Once the computations have been completed, a message window will open stating Compute Complete. Close this window and then select the Tabular Results tab. The analysis window should resemble Figure 4.

Figure 4. Bulletin 17 Analysis Tabular Results Tab for B17C Example 3.
In addition to the tabular results, a graphical plot of the computed frequency curves can be obtained by pressing the Plot Curve button at the bottom of the analysis window. The Log Pearson Type III distribution fit using EMA to the input annual maximum flow data set, the 5% and 95% confidence limits, and the computed plotting positions are shown in Figure 5.

Figure 5. Plotted Frequency Curves for B17C Example 3.
As shown in Figure 4, the Multiple Grubbs-Beck Test identified a low outlier threshold (critical value) of 2000 cfs, which corresponds to the smallest peak that was retained. Two annual peak flows were identified and censored, as shown in Figure 6.  These annual peak flows were then recoded to have a flow interval of zero – 2000. The perception thresholds for the years in which an annual peak flow was recorded were also adjusted to correspond with the Multiple Grubbs-Beck Test low outlier threshold (critical value). Consequently, these perception thresholds were changed from zero – inf to 2000 – inf, as shown in Figure 7.

Figure 6. Report File for B17C Example 3 showing censored low outliers.
 
Figure 7. Report File for B17C Example 3 showing recoded perception thresholds.