Historical event information (along with the weighted moments adjustment procedures contained within Bulletin 17B) can be used to both add important information and place the systematic information into a longer context. Three historical events occurring in 1897, 1919, and 1926 are available for the Big Sandy River at Bruceton, TN gaging station. This information can be incorporated within HEC-SSP.
Create a New Bulletin 17 Analysis
To begin, select Analysis | New | Bulletin 17 Analysis.
Name the new analysis “BigSandyRiver_Historical_B17B” and add an adequate description.
Select the “BigSandyRiver” data set.
Select “17B Methods” within the Method for Computing Statistics and Confidence Limits panel.
Utilize the default selections of Use Station Skew, Single Grubbs-Beck low outlier test, and Median plotting position.
Add Additional Options
Move to the Options tab.
Check the option to Use Historic Data.
Enter a Start Year of 1897.
The 1897, 1919, and 1926 events are estimated to have peak flow rates of 25000, 21000, and 18500 cfs, respectively. Enter the necessary information into the Historic Events table.
The Options tab should look similar to the following figure.
Click Compute.
Analyze Results
Click Plot Curve. This will result in the computed curve, 5- and 95-percent confidence limits, and observed events being plotted.
Close the computed curve window.
Move to the Tabular Results Note the computed curve, 5-, and 95-percent confidence limits for all of the desired frequency ordinates, the moments/parameters of the Log Pearson Type III fit to the data, and other data related to the analysis.
Question: How did the inclusion of additional historical information affect the Log Pearson Type III parameters (mean, standard deviation, and skew)? How does this affect the 1-percent chance exceedance flow rate?
The mean, standard deviation, and skew are all larger when historical information is included and the statistics are computed using Bulletin 17B procedures. The BigSandyRiver_Historical_B17B 1-percent chance exceedance flow rate is approximately 4400 cfs larger than the BigSandyRiver_Systematic_B17B results, as shown in the following figure.
Question: How do the Log Pearson Type III parameters (mean, standard deviation, and skew) compare between the BigSandyRiver_Historical_B17B and BigSandyRiver_Historical_B17C analyses? Why are they different? How do these differences affect the 1-percent chance exceedance flow rate?
The mean and standard deviation are essentially the same between the two analyses. However, the computed at-site skew is different. Specifically, the at-site skew (and adopted skew) is more negative within the BigSandyRiver_Historical_B17C analysis which affects the flow rates for smaller exceedance probabilities. The BigSandyRiver_Historical_B17C computed 1-percent chance exceedance flow rate is approximately 450 cfs less than the corresponding B17B analysis, as shown in the following figure.
Question: In your opinion, how well do Bulletin 17B procedures (i.e. Single Grubbs-Beck test, conditional probability adjustment, etc.) compare against Bulletin 17C procedures (i.e. Expected Moments Algorithm, Multiple Grubbs-Beck test, flow ranges, etc.) when using all of the available systematic and historical data?
This is a subjective question and meant to incite discussion. Bulletin 17C contains numerous enhancements over Bulletin 17B including:
adoption of a generalized representation of flood data that allows for interval and censored data types,
a new method, called the Expected Moments Algorithm (Cohn et al., 1997 and Cohn et al., 2001), which extends the method of moments so that it can accommodate interval data,
a generalized approach to identification of low outliers in flood data (Cohn et al., 2013) and,
an improved method for computing confidence intervals.
