Free calculator · Frequency & data

Flood frequency analysis software, in the browser

Paste an annual maximum series or import one from a DWS, USGS or Bureau of Meteorology gauge, and Studio fits 14 distributions by method of moments, L-moments, maximum likelihood and probability weighted moments, ranks them on goodness of fit and returns quantiles with confidence intervals. Log-Pearson Type III runs the Bulletin 17C procedure: the iterative Multiple Grubbs-Beck test, the Expected Moments Algorithm for historical floods and weighted regional skew.

Flood frequency calculator (LP3, GEV, Gumbel)

Annual maximum series

20 values read. One flow per line, or year, flow pairs; commas, spaces and tabs all work.

Distributions
Fitting
Parameter estimation

Bulletin 17C fits Log-Pearson III by the moments of log₁₀ Q. L-moments are steadier on short records.

Plotting position

Weibull, m/(n+1), is unbiased in probability; Gringorten suits Gumbel and GEV.

1:100 · LP3
272.6m³/s
1:100 · GEV
259.3m³/s
1:100 · Gumbel
229.7m³/s
With 20 years of record, return periods beyond about 1:40 are extrapolations of the fitted curve, and the spread between distributions widens with T.

Flood quantiles

Annual maximum flood by return period for each fitted distribution
Return periodLP3GEVGumbel
1:250 % AEP86.686.689.9
1:520 % AEP125.4123.5127.3
1:1010 % AEP155.2151.3152.1
1:205 % AEP187180.8175.9
1:502 % AEP233.5223.5206.6
1:1001 % AEP272.6259.3229.7
1:2000.5 % AEP315.8298.5252.7

Annual maximum flows in m³/s, fitted by L-moments. AEP is the annual exceedance probability, 1/T.

Frequency curves

050100150200250300350400125102050100200500Return period (years, log scale)Annual maximum flow (m³/s)
LP3 (L-moments)GEV (L-moments)Gumbel (L-moments)Observed (Weibull positions)

Fitted parameters

Fitted distribution parameters
Log-Pearson Type IIIlog₁₀ Q, Q in m³/sȳ 1.952 · s 0.1809 · G 0.479
Generalised extreme valueξ and α in m³/sξ 75.9 · α 28.6 · k -0.137
Gumbel (EV1)ξ and α in m³/sξ 77.8 · α 33

Sample statistics

Sample statistics of the annual maximum series
Record length, n20years
Mean96.84m³/s
Standard deviation41.33m³/s
Skew coefficient, G1.143
Coefficient of variation0.427
Smallest and largest45.2 / 201.5m³/s
Mean of log₁₀ Q, ȳ1.952
Standard deviation of log₁₀ Q, s0.1738
Skew of log₁₀ Q, G0.333
Low outliers (Grubbs-Beck)None below 34.5m³/s
Trend (Mann-Kendall)None (Z = 0.68, p = 0.5)

Keep this calculation

Studio accounts open soon. Join the waiting list and we'll e-mail you when you can save it in the full Flood Frequency Analysis.

  • 14 distributions ranked on fit
  • Confidence intervals
  • Historical floods by EMA (Bulletin 17C)
  • DWS, USGS and BoM gauge import
  • Word and Excel report

Runs the same calculation engine as Delta Hydro Studio. Results are computed in your browser; nothing you type leaves it until you choose to save.

Using the calculator above

The calculator opens on the 20-year record from the worked example further down this page, fitted by L-moments, so its 1:100 floods of 272.6 m³/s (Log-Pearson III), 259.3 m³/s (GEV) and 229.7 m³/s (Gumbel) are the ones in the table there. Switch to method of moments and Log-Pearson III gives the hand calculation’s 250.3 m³/s.

  • Annual maximum series. One peak per year, pasted one per line or as year, flow pairs; commas, spaces and tabs all work. It needs at least 10 positive values and warns when the record is shorter than 20 years.
  • Distributions. Log-Pearson Type III, GEV and Gumbel, with Log-normal as a fourth. Each gets a column in the quantile table and a curve on the chart.
  • Parameter estimation. Method of moments (of log₁₀ Q for Log-Pearson III, as in Bulletin 17C), L-moments or maximum likelihood.
  • Plotting position. Weibull or Gringorten, for the observed floods drawn on the frequency curve.

It also runs the Grubbs-Beck low-outlier test and the Mann-Kendall trend test, and lists the sample statistics, the statistics of log₁₀ Q and the fitted parameters.

What flood frequency analysis calculates

Flood frequency analysis fits a probability distribution to a gauge’s annual maximum series, one peak per water year, and reads the design flood off the fitted curve. A 1:100-year flood has a 1 % annual exceedance probability (AEP). Three distributions cover most national guidance.

Log-Pearson Type III, the Bulletin 17C distribution, fitted to base-10 logarithms:

log⁡10QT=yˉ+KT(G) sy\log_{10} Q_T = \bar{y} + K_T(G)\, s_y

Generalised Extreme Value (GEV), in the Hosking parameterisation used by L-moments:

QT=ξ+αk[1−(−ln⁡(1−1T))k]Q_T = \xi + \frac{\alpha}{k}\left[1 - \left(-\ln\left(1 - \tfrac{1}{T}\right)\right)^{k}\right]

Gumbel, the GEV with k=0k = 0:

QT=ξ−αln⁡[−ln⁡(1−1T)]Q_T = \xi - \alpha \ln\left[-\ln\left(1 - \tfrac{1}{T}\right)\right]
SymbolMeaningUnit
QTQ_TFlood with return period TTm³/s (or ft³/s)
TTReturn period, the reciprocal of AEPyears
yˉ\bar{y}, sys_y, GGMean, standard deviation and skew of log⁡10Q\log_{10} Q–
KT(G)K_T(G)Pearson Type III frequency factor for TT and skew GG–
ξ\xi, α\alpha, kkLocation, scale and shapem³/s, m³/s, –

In the GEV, k<0k < 0 gives a heavy upper tail and k>0k > 0 a bounded one.

Bulletin 17C, as implemented

Bulletin 17C (England et al., 2019) is the US federal procedure for Log-Pearson Type III. Studio’s Log-Pearson Type III fit includes:

  • Multiple Grubbs-Beck test (MGBT). Low outliers, called potentially influential low floods, are flagged in log space against a threshold recomputed after each round of removal, until a round flags nothing.
  • Conditional probability adjustment. Low outliers and zero-flow years are removed from the fit but kept in the year count, and exceedance probabilities are scaled to match, so Log-Pearson Type III and Log-Normal still fit records from ephemeral rivers.
  • Expected Moments Algorithm (EMA) (Cohn et al., 1997). You enter historical floods and perception-threshold periods, spans of years in which any flood above a known level would have been recorded. EMA treats the quiet years as observations below the threshold instead of missing data.
  • Weighted regional skew, combining station skew GsG_s and regional skew GrG_r by their mean square errors:
Gw=MSEr Gs+MSEs GrMSEr+MSEsG_w = \frac{MSE_r\, G_s + MSE_s\, G_r}{MSE_r + MSE_s}

Confidence intervals default to a parametric bootstrap of 500 samples, which works for every distribution. Switch to analytic intervals and Pearson Type III under method of moments uses the Bulletin 17B (Appendix 9) non-central-t approximation, while maximum likelihood fits use delta-method intervals from the Hessian; other pairs stay on the bootstrap. The report states which method produced each interval. Bulletin 17C itself computes intervals from the variance of the EMA estimator (Cohn et al., 2001), which Studio does not implement, so for a Bulletin 17C submission take the interval from PeakFQ or HEC-SSP.

Getting the annual maximum series in

  • From a gauge. The Stream Gauge Finder sends DWS, USGS NWIS or Bureau of Meteorology annual peaks with their gauge heights, qualification codes and catchment area. USGS peaks carrying code 7, historic peak, move into the historical-data panel with an inferred threshold for you to check.
  • From a file. Upload CSV, TSV or text; the tool guesses the year and flow columns from the headers and lets you correct them.
  • By pasting, comma, space, tab or one value per line, with or without a year column.

The record must be annual maxima from a stationary series. The Mann-Kendall test with Sen’s slope runs by default on any record of ten or more years and warns on a significant trend.

Worked example: a 20-year record

The series is the example from the Studio flood frequency documentation, annual maxima in m³/s:

45.2, 67.8, 123.4, 89.1, 56.3, 201.5, 78.9, 95.6, 110.3, 54.7, 142.8, 63.4, 87.2, 175.3, 92.1, 58.9, 134.6, 71.5, 105.8, 82.4

Running the Studio engine gives a mean of 96.84 m³/s, standard deviation 41.33 m³/s and skew 1.143, close to the Gumbel’s fixed 1.14. MGBT found no low outliers below its threshold of 34.5 m³/s, and Mann-Kendall found no trend (Z = 0.68, p = 0.50).

Log-Pearson Type III by hand. The logs have yˉ=1.952\bar{y} = 1.952, sy=0.1738s_y = 0.1738 and G=0.333G = 0.333. At T=100T = 100 the frequency factor is K=2.569K = 2.569, so log⁡10Q100=1.952+2.569×0.1738=2.398\log_{10} Q_{100} = 1.952 + 2.569 \times 0.1738 = 2.398 and Q100=250.3Q_{100} = 250.3 m³/s.

Distribution / estimatorQ2Q_2Q10Q_{10}Q50Q_{50}Q100Q_{100}95 % interval on Q100Q_{100}
Log-Pearson III / moments87.6151.3218.4250.3186.9–422.6 (Bulletin 17B)
Log-Pearson III / L-moments86.6155.2233.5272.6154.4–630.5 (bootstrap)
GEV / L-moments (k=−0.137k = -0.137)86.6151.3223.5259.3145.1–487.3 (bootstrap)
Gumbel / L-moments89.9152.1206.6229.7168.3–309.8 (bootstrap)
Generalised Pareto / L-moments85.2157.9213.1232.5145.7–363.6 (bootstrap)

All values in m³/s. Under L-moments the composite ranking put Log-Pearson III first, then Pearson III and Log-Normal; the Generalised Pareto ranked last with an Anderson-Darling statistic of 1.63.

The point estimates of Q100Q_{100} span 230 to 273 m³/s, while one interval alone runs from 187 to 423 m³/s. Twenty years of record cannot pin down the 1:100 flood, and no choice of distribution changes that. Information from outside the record can:

  • An illustrative regional skew of 0 with mean square error 0.302 pulls the weighted skew to 0.167 and Q100Q_{100} to 238.6 m³/s.
  • A hypothetical historical flood of 380 m³/s in 1987, with a 250 m³/s perception threshold for 1951 to 2000 and the gauged record placed at 2001 to 2020, gives EMA an effective record of 70 years. Q100Q_{100} rises to 301.2 m³/s and its Bulletin 17B interval, computed on the 70-year effective record, narrows to 247.7–392.5 m³/s.

The 14 distributions

DistributionParametersWhere it is used
Log-Pearson Type III3Bulletin 17C, United States
Generalised Extreme Value3Widely used for annual maxima; unifies the three extreme value types
Generalised Logistic3UK Flood Estimation Handbook
Generalised Pareto3Peaks over threshold; selected for South Africa by Calitz and Smithers (2020)
Pearson Type III3Untransformed counterpart of LP3
Log-Logistic3Heavier tail than Log-Normal; tail sensitivity check
Gumbel (EV1)2Fixed skew of 1.14; long records with moderate skew
Log-Normal2When log-space skew is near zero
Gamma2Pearson Type III with zero location; stable on short records
Exponential2Fixed skew of 2; peaks over threshold
Weibull (minimum)2Bounded-tail comparison
Normal2Baseline only; flood series are rarely symmetric
TCEV4Two flood populations, such as frontal rain and tropical cyclones; maximum likelihood only
IPZA3South African quantile model (van der Spuy and du Plessis, 2024) from the mean, the standard deviation and the standard deviation without the largest flood, so one extreme peak moves it less

Each can be fitted by method of moments, L-moments, maximum likelihood and probability weighted moments, except Log-Logistic (no probability weighted moments), TCEV (maximum likelihood only) and IPZA (method of moments only, with published frequency factors for AEP 99.5 % to 0.01 %).

Limitations

The method assumes the annual maxima are independent, identically distributed and stationary. Upstream dams, urbanisation and climate shifts break that, and a Mann-Kendall warning is a reason to shorten the record or model the trend, not to carry on. The confidence interval covers sampling error only: it ignores the choice of distribution and the error in the rating curve, which is often largest for the biggest floods because they sit above the highest gauged flow. A 1:1 000-year estimate from a 50-year record is speculation, whatever its interval says.

Ungauged sites need a regional method. In South Africa use the index flood method, the probabilistic rational method or the quantile regression technique, with the regional maximum flood as an upper-bound check. Design flood estimation runs them together.

What a free Studio account adds

A free Delta Hydro Studio account includes the full Flood Frequency Analysis tool, which runs the same engine with the rest of it exposed:

  • All 14 distributions, fitted by method of moments, L-moments, maximum likelihood and probability weighted moments in one run.
  • Confidence intervals on every quantile, by parametric bootstrap or by the analytic methods described above.
  • Bulletin 17C historical floods and perception thresholds through the Expected Moments Algorithm, and weighted regional skew.
  • Annual peaks imported from DWS, USGS NWIS and Bureau of Meteorology gauges through the Stream Gauge Finder, or from a CSV, TSV or text file.
  • Goodness of fit by Anderson-Darling, Kolmogorov-Smirnov and chi-square, with parametric-bootstrap p-values for the top five candidates, plus AIC and BIC, combined into one ranking.
  • Eight plotting positions, including Cunnane, Gringorten, the DWS Z-set formula and Hirsch-Stedinger for censored records.
  • P-P and Q-Q plots, an L-moment ratio diagram and an exceedance-risk calculator for a given design life.
  • Regional consistency for three or more gauges: Hosking-Wallis discordancy and heterogeneity.
  • Saved analyses in projects; CSV quantile tables, PNG charts, and Word and Excel reports that state which method produced every interval.
FAQ

Common questions

Which distribution is best for flood frequency analysis?
It depends on where you are and what your record looks like. Bulletin 17C prescribes Log-Pearson Type III in the United States. The UK Flood Estimation Handbook uses the Generalised Logistic. Calitz and Smithers (2020) selected the Generalised Pareto for South Africa. Fit several, compare goodness of fit and the plotted curves, and report the spread at your design return period rather than one number.
How do I calculate a Log-Pearson Type III flood?
Take base-10 logarithms of the annual maxima and compute their mean, standard deviation and skew. Look up the frequency factor K for your return period and that skew, then add K times the standard deviation to the mean and raise 10 to that power. For a 20-year example with log mean 1.952, standard deviation 0.1738 and skew 0.333, K for 1:100 is 2.569 and the flood is 250.3 cubic metres per second.
What changed from Bulletin 17B to Bulletin 17C?
Bulletin 17C (England et al., 2019) replaced the single Grubbs-Beck low-outlier test of Bulletin 17B with the Multiple Grubbs-Beck test, and replaced the 17B historical weighting with the Expected Moments Algorithm, with confidence intervals computed from the EMA estimator's variance. EMA can use floods from outside the gauged record and whole periods in which floods are known to have stayed below a threshold, so documentary evidence enters the fit as data instead of as an adjustment afterwards.
Is there an online alternative to HEC-SSP?
HEC-SSP, the US Army Corps of Engineers' free desktop package, and USGS PeakFQ are the established Bulletin 17C programs for US practice. Studio runs in a browser, fits Log-Pearson Type III with EMA, MGBT and weighted skew alongside twelve other distributions, imports DWS, USGS and Bureau of Meteorology peaks directly, and exports Word and Excel reports. For a US federal submission, check the result against one of those programs.
What is the difference between L-moments and method of moments?
Method of moments matches the sample mean, variance and skew, and the skew of a short record is unstable because it cubes deviations from the mean. L-moments are linear combinations of the ranked data, so one large flood moves them less and they are nearly unbiased in small samples. Bulletin 17C uses moments in log space; regional analysis following Hosking and Wallis uses L-moments.
Full documentation: Flood Frequency Analysis
Every input, option and equation in the Studio tool, with its limitations.
Run it from your AI agent
This engine is open to Claude, ChatGPT, Cursor and other MCP clients as run_flood_frequency_analysis. The agent passes the inputs, the engine does the calculation, and the record is saved to your project for you to review.

References

  1. England, J.F., Cohn, T.A., Faber, B.A., Stedinger, J.R., Thomas, W.O., Veilleux, A.G., Kiang, J.E. and Mason, R.R. (2019) Guidelines for Determining Flood Flow Frequency, Bulletin 17C. U.S. Geological Survey Techniques and Methods, Book 4, Chapter B5.
  2. Cohn, T.A., Lane, W.L. and Baier, W.G. (1997) An algorithm for computing moments-based flood quantile estimates when historical flood information is available. Water Resources Research, 33(9), 2089-2096.
  3. Cohn, T.A., Lane, W.L. and Stedinger, J.R. (2001) Confidence intervals for Expected Moments Algorithm flood quantile estimates. Water Resources Research, 37(6), 1695-1706.
  4. Hosking, J.R.M. and Wallis, J.R. (1997) Regional Frequency Analysis: An Approach Based on L-Moments. Cambridge University Press, Cambridge.
  5. Stedinger, J.R., Vogel, R.M. and Foufoula-Georgiou, E. (1993) Frequency analysis of extreme events. Chapter 18 in Maidment, D.R. (ed.) Handbook of Hydrology. McGraw-Hill, New York.
  6. Institute of Hydrology (1999) Flood Estimation Handbook, Volume 3: Statistical Procedures for Flood Frequency Estimation. Institute of Hydrology, Wallingford.
  7. Calitz, J.P. and Smithers, J.C. (2020) Development and Assessment of Regionalised Approaches to Design Flood Estimation in South Africa. WRC Report No. 2748/1/20, Water Research Commission, Pretoria.
  8. Interagency Advisory Committee on Water Data (1982) Guidelines for Determining Flood Flow Frequency, Bulletin 17B. U.S. Geological Survey, Office of Water Data Coordination, Reston, Virginia.

Last reviewed by the Delta Hydro engineering team.

Keep the whole study in one place

Delta Hydro Studio puts this tool alongside catchment, rainfall, runoff and hydraulics tools, with projects, team sharing and Word and Excel reports. Run them by hand, or let your AI agent run them for you to review.