User Notice for PeakFQ 7.5 (Beta)
PeakFQv7.5 (beta) includes changes to the regional skew calculation to resolve computational issues observed in earlier versions. Because weighted skew values are computed iteratively within the Expected Moments Algorithm (EMA), the algorithm used by PeakFQv7.4.1 and earlier gave too much weight to the regional skew whenever censored data were present in an analysis. The new default skew weighting algorithm uses an adjustment factor developed by USGS to reduce the weight given to the regional skew when censored data are present. When no censored data are present, results are identical to those from PeakFQ 7.4.1. The old skew weighting algorithm from PeakFQ 7.4.1 can also be used in PeakFQ 7.5 (beta) from a user option in the specifications file, but is no longer the default. The U.S. Army Corps of Engineers implemented an algorithm with a different correction factor from PeakFQ in HEC-SSP 2.3. This algorithm can also be used in PeakFQ 7.5 (beta) based on an option in the specifications file. In most cases, the differences in results between the three algorithms are of no practical significance. However, in cases where the regional skew and at-site skew have a large difference and a large fraction of the data are censored, differences in results due to choice of skew weighting algorithm can become larger. PeakFQ 7.5 will be considered a beta software release until USGS publishes an accompanying publication documenting the updates to the skew weighting algorithm.
As of June 28, 2022:
Potential computational issues in the PeakFQ software continue to be investigated in collaboration with the U.S. Army Corps of Engineers (USACE). Refined user guidance and/or updated methodology documentation will be developed and made available, if needed.
Specifically, the issues are related to:
- (NOTE: this issue has been resolved in the new release PeakFQv7.4) Incorrect period-of-record determination: Currently, missing periods of record indicated by the perception threshold range (inf, inf) are incorrectly counted. This error will be fixed in the next release.
- Weighted skew calculation: Weighted skew values are computed iteratively within the Expected Moments Algorithm (EMA). In some cases, the regional skew value may be given too much weight. The outcome can be especially noticeable in cases where regional skew and at-site skew have a large difference; however, these errors are present whenever observed data is censored and replaced by flow intervals, or when historical periods of unobserved flows are represented by flow intervals. Alternative solutions are being evaluated that can be implemented in PeakFQ, but their implementation may be delayed if their development requires updates to B17C.
- Computation of skew mean square error (MSE): Computational procedures for skew MSE are different when potentially influential low flood (PILF) thresholds are determined by the Multiple-Grubbs Beck Test (MGBT) versus a fixed user-specified threshold. A solution is being evaluated that can be implemented in PeakFQ where the skew MSE is computed identically for both the MGBT and user-entered thresholds.
Monte Carlo evaluations are being performed to test the use and limitations of current computational methods on the MSE skew computations and weighting calculation. Results of these evaluations will be used primarily to inform which weighting method performs better and may be used to further refine B17C guidelines.
Based on current understanding (March 17, 2022), the impacts of the identified issues on most at-site flood frequency estimates, and to an even lesser extent USGS regional regression equations, appear to be limited. At this time, there is no recommendation to stop the use of USGS published regional skew relations, USGS published regional regression equations, or estimates of peak-flow from StreamStats which use USGS published regional regression equations. The current recommendation for use of PeakFQ for at-site analysis is to proceed with caution and closely review outputs. Specifically, users should use more caution when analyses are near the boundaries defined in B17C which include warnings for excessive skew values and recommendations on when to weight skew with regional values. Future user guidelines and possibly edits to B17C will provide more specific details.
Update as of 6/28/22, PeakFQv7.4:
With the resolution of the ‘incorrect period-of-record determination’ issue (see above for more details), new USGS flood frequency analyses will proceed with the use of PeakFQv7.4. With two issues (see above) still outstanding, analysts are encouraged to pay special attention to the outputs of PeakFQ analyses as they relate to weighted skew values (especially those where the value of the at-site skew and regional skew have large differences) and the mean squared error of the skew (especially at gages with highly negative at-site skews).
Further information will be posted on PeakFQ, StreamStats, and other USGS websites as the issues are investigated and resolved.
If users encounter issues or problems with PeakFQ, please consider submitting an issue request or bug report to h2osoft@usgs.gov
