Annual Peak Flow Record Extension: Use Most-Recent Years vs. Attempt to Match Skew

Overview

By definition, the Maintenance of Variance Extension (MOVE) record extension methods attempt to maintain the mean and variance of the extended time series (observed plus extended record) using ordinary least squares regression. However, the skew of a annual peak flood series is highly variable and will change based on the years of record that are used for record extension. In HEC-SSP, the Record Extension | MOVE.3 (Bulletin 17C) computational method provides users with two Years Options: 1) Use Most-Recent Years and 2) Attempt to Match Skew, as shown in the following image.

Example

Little Lehigh Creek is a tributary of the Lehigh River. The Lehigh River is one of the three main tributaries to the Delaware River. Flows along the Lehigh River are regulated by 2 U.S. Army Corps of Engineers dams, as shown in the following image.

Annual peak flow data for 2 USGS gages (USGS 01451500 - Little Lehigh Creek Near Allentown, PA and USGS 01451650 - Little Lehigh Creek at Tenth St. Br. at Allentown) are shown in the figure below. The Little Lehigh Creek is a tributary of the Lehigh River, which is one of the three main tributaries to the Delaware River. The Allentown gage recorded 77 annual peak flows (WY1935-2022) and the Tenth Street Bridge gage recorded 36 annual peak flows (WY1987-2022). 

The linear correlation of the concurrent record is 0.96 which produces an n_e value of 29 years (n_e is the number of years of extended record). If the Use Most-Recent Years option is selected, the extended record is computed for WY1958-1986 (shown in yellow in the figure below). If the Attempt to Match Skew option is selected, the extended record is computed for WY1949-1977 (shown in purple in the figure below). Another possible 29-year n_e period is shown in green.

The blue annual peaks in the figure above were obtained by extending the Tenth Street Bridge gage record to the length of the Allentown gage (long record) site. In other words, the extended record length equals n₂, or the long record site length. 

Do not select this option in a MOVE.3 Record Extension for use a in a Bulletin 17C analysis as it underestimates the quantile variance and the amount of uncertainty. The entire record was extended here for illustration purposes only.

If the Attempt to Match Skew option is selected, the resulting time-series will contain a broken record (i.e. there will be a number of years without peak flow data). A Perception Threshold of [inf, inf] should be entered during the missing years to not violate the allowable number of extended years. A Perception Threshold of [inf, inf] provides no additional information content during the missing period. As a result, the Equivalent Record Length (ERL) is not affected by the missing years. The extended time-series contains 65 systematic events and spans a historical period of 74 years, but the ERL is equivalent to the systematic record.

Log Pearson-III distributions were fit to the two datasets (Match Skew and Most Recent) using the Expected Moments Algorithm. The resulting skew values are shown in the table below. The desired skew coefficient is from the dataset that is extended through all n₂ years of record and has a value of 0.343. The Match Skew Years Options attempts to approximate this skew value with the number of allowable extended years (n_e). Notice that the skew obtained from the Match Skew extended time series is much closer to 0.343 than the skew from the Most Recent Years extended time series.

What does Bulletin 17C say?

The MOVE.3 (Bulletin 17C) procedure is documented in Appendix 8 of Bulletin 17C. Step 5 of the Summary of Procedure section states the following:

"Check to make sure the extension is reasonable. As an example, if a short extension of 10 years included the first or second largest floods in the much longer x-series, that short extension may represent the likely skewness coefficient for the short-record site. In such cases, the n_e x-observations used to generate the extension might be selected to be a sequence of years that generated a more typical skewness coefficient (considering the different ne extensions that could be made) for the n₁ + n_e record."

where x-site is the long record site, n_e is the number of years of extended record, and n₁ is the length of the short record. The opposite may also be true. The most recent n_e years may contain relatively small annual peaks whereas the period selected to match the skew may contain larger annual peaks that result in a more typical skew value.

Performing a sensitivity analysis to determine the impact of Match Skew vs. Use Most Recent Years on the predicted flood quantiles is good practice.