Statistics
Gumbel Return-Period Quantiles and Multi-Year Exceedance Risk
Calculate an illustrative Gumbel annual-maximum quantile from a sample mean and standard deviation, and distinguish return period from the chance of an exceedance during several years.
Use annual maxima, not all daily observations
The example assumes a mean annual peak flow of 450 m3/s and sample standard deviation of 120 m3/s. These summarize one maximum per year, not annual mean flow or every daily measurement. For wind, use consistently defined annual maximum gusts and matching units; do not mix flood and wind records.
Gumbel is one candidate model for maxima, not an automatic law for every flood or hurricane record. Its right tail is exponential, not a power-law heavy tail. The fact that maxima are skewed does not establish that a normal model always underestimates them.
Moment-based Gumbel maximum model
The maximum-case CDF is F(x)=EXP(-EXP(-(x-mu)/beta)). Moment estimates are beta=s*SQRT(6)/PI() and mu=mean-gamma*beta, where gamma=0.577215664901533. For annual exceedance probability 1/T, use y=-LN(-LN(1-1/T)) and x=mu+beta*y. Equivalently, x=mean+K*s with K=SQRT(6)/PI()*(y-gamma).
This uses the moment-based factor, without sample-size corrections. Keep raw outputs E3:E9 and round only for display. E7 is the estimate, not a validated design specification or a prediction of when an event occurs.
| T (years) | Annual exceedance | Estimated magnitude (m3/s) |
|---|---|---|
| 10 | 10% | 606.55 |
| 100 | 1% | 826.40 |
=IF(E2="OK",B2+E6*B3,E2)Inputs and the working output cells
B2:B5 must be numeric. This flood/wind example requires a nonnegative mean, positive SD, T greater than 1 and no greater than 1,000,000, and a nonnegative whole number of exposure years N. The upper T limit is a numerical scope for this worksheet, not evidence that such extrapolation is credible. It avoids taking logarithms after 1-1/T rounds to 1 at extreme inputs.
E2 reports Invalid input when a check fails. E3 and E4 show scale and location; E5 the reduced variate; E6 the factor; E7 the magnitude; E8 the annual probability; E9 the exposure probability. Zero SD is a degenerate sample, not a positive-scale Gumbel distribution. A low quantile may be negative: the model has unbounded support, so inspect its suitability instead of silently clamping to zero.
A 100-year event can happen in successive years
Under independent years with an unchanged annual exceedance probability p=1/T, the probability of at least one exceedance in N years is 1-(1-p)^N. At T=100 and N=30 this is about 26.03%. It does not mean an event occurs on a 100-year schedule.
Set N to 0 and the exposure probability is zero; set N to 1 and it equals 1/T. If annual risks vary, the independent-year expression becomes 1-PRODUCT(1-p_year). Dependence between years requires a different analysis.
=IF(E2="OK",1-(1-1/B4)^B5,E2)Try changes, then assess model uncertainty
Change B4 to 10: E7 is about 606.55 m3/s and E9 about 95.76% for 30 years. Restore 100, then double the mean and SD: E7 doubles while E8 and E9 stay unchanged. Clear an input or set SD to zero to test the validation.
Before practical flood or wind design, assess data quality, sample length, independence, stationarity, fit and parameter uncertainty, and use the applicable design standard. A point estimate extrapolated beyond the observations is uncertain; this worksheet supplies neither confidence bounds nor a site-specific model assessment.
Try the example
Use the original values or change them to test your understanding.