Flood Frequency, an Outlier, and a High-Hazard Dam Spillway
A 60-ft-high earth embankment dam is classified high hazard (about 40 homes in the inundation zone) and was built in 1968 with an uncontrolled service spillway that passes 5,200 cfs at the dam crest. The gage a mile downstream has 45 years of record. The consultant's Bulletin 17C fit gives a 100-year peak of 8,400 cfs; the 1985 event of 22,000 cfs is the largest in the record and 2.8 times the second-largest value. Dropping 1985 from the fit lowers the 100-year estimate to 3,900 cfs. The state's rules require high-hazard dams to pass one-half of the probable maximum flood without overtopping; the PMF peak is estimated at 65,000 cfs. The owner proposes adopting the 3,900-cfs 100-year event as the design basis to avoid a spillway upgrade.
1. Is dropping the 1985 peak from the frequency analysis defensible? Explain the statistical issue and what the analysis should actually do with that value.
Read the worked answer
No, it is not defensible as presented.
- 1985 is an observed systematic record value, not an error, a censored value, or a measurement artifact. Removing a real peak because it is inconvenient biases the fit low by construction - the 100-year estimate halves, which is the tell.
- Bulletin 17C's outlier treatment is aimed at low outliers (multiple Grubbs-Beck test), because low values are more often the suspect ones. High values are retained; a high value may be flagged or given an assigned probability, but it is not simply discarded.
- The real issue a 2.8× gap raises is record length and sampling error, not the validity of 1985. With 45 years of record, the 100-year estimate carries a wide confidence interval - commonly a factor of two or more on the upper bound. The difference between 3,900 and 8,400 cfs is well inside that uncertainty, which means neither number is a reliable basis for a dam safety decision on its own.
- The correct treatment is to keep 1985, use EMA with any available historical or paleoflood information to extend the effective record, state the confidence limits explicitly, and do a sensitivity run to show the decision-maker how much the answer moves.
Adopting the lower number without disclosing that it comes from deleting the largest flood would be a serious documentation failure.
2. What is the governing design basis for this dam, and outline the analysis program you would require before any spillway decision.
Read the worked answer
Governing basis: the state rule - pass one-half PMF (about 32,500 cfs) without overtopping the embankment. Flood frequency estimates inform the design of appurtenances and the risk assessment; they do not override the regulatory inflow design flood for a high-hazard structure.
Scale check that should end the discussion about using 3,900 cfs: the existing service spillway passes 5,200 cfs at the crest. The rule requires roughly 32,500 cfs. That is a factor of six, not a rounding difference. Even the consultant's own 100-year value of 8,400 cfs exceeds the existing spillway capacity - meaning the dam does not meet even a much weaker standard.
Required program:
- Hydrometeorological review of the PMF estimate (PMP depth, temporal and spatial distribution, snowmelt or antecedent conditions if relevant) - this is the number that governs, and it should be independently checked, not taken from a single consultant run.
- Rainfall-runoff routing of the PMF through the 45-year-calibrated model; confirm the watershed model reproduces the 1985 event reasonably, which is the best available calibration point.
- Full reservoir routing through both the service spillway and the auxiliary/emergency spillway, with the auxiliary crest geometry and erodibility stated.
- Geotechnical evaluation of overtopping tolerance and the auxiliary spillway erosion resistance - the freeboard conclusion depends on this.
- Risk assessment / consequence evaluation, and interim risk reduction measures (reservoir restriction, warning system, outlet inspection) while the upgrade is designed.
Only after that does the owner have a defensible basis to size a spillway.
3. The basin has urbanized substantially since 1968 and the gage record mostly predates that development. How does that change the analysis, and how would you present the uncertainty to the owner?
Read the worked answer
It changes the analysis in the direction that matters: the frequency curve is probably not stationary, and the historical record understates today's peaks.
- Urbanization raises runoff coefficients and shortens the time of concentration, which increases peak discharge for the same rainfall - most strongly at the more frequent return periods and less at extreme events where saturation dominates, but the direction is unambiguous.
- A stationary Bulletin 17C fit to a 45-year record that spans a land-use transition is therefore a description of the past, not of the current hazard. The 1985 peak should not be read as an upper bound simply because the watershed was less developed then.
- Practical handling: trend testing on the annual peaks and on the rainfall record, a calibrated model run with current imperviousness versus 1968 conditions to quantify the shift, and explicit disclosure of nonstationarity rather than silently fitting a single stationary curve. Where the trend is significant, the design should use the current-condition model with the upper end of the frequency estimate.
Presenting uncertainty to the owner: give a table, not a single number - 100-year peak with 90% confidence limits under both stationary and current-land-use assumptions, alongside the PMF and the regulatory fraction. State plainly: (a) the regulatory design inflow is set by rule and is about six times the existing spillway capacity; (b) the frequency estimate is uncertain by roughly a factor of two, so no defensible reading of it produces a number the existing spillway passes; (c) interim risk reduction measures are available immediately. Frame the decision as how to upgrade and on what schedule, not whether.