Frequency & data · South Africa

The index flood method for ungauged catchments

The index flood method estimates the T-year flood at a site as a local index flood, usually the mean annual flood, multiplied by a dimensionless growth factor shared by a homogeneous region of gauged catchments. It turns many short gauge records into one well-defined frequency curve and transfers it to catchments with no gauge at all. Studio implements the South African calibration of Calitz and Smithers (2020).

How the index flood method works

The method splits a flood quantile into a local size and a regional shape:

QT=μ  qTQ_T = \mu \; q_T
SymbolMeaningUnit
QTQ_TFlood with return period TT at the sitem³/s
μ\muIndex flood, usually the mean annual floodm³/s
qTq_TRegional growth factor for return period TT–

Dalrymple (1960) set it out for the USGS: scale each gauge’s floods by its own mean annual flood, combine the scaled frequency curves of a region that floods alike, and adopt one regional growth curve. Ten gauges with 30 years each give the curve’s shape 300 station-years of information instead of 30, which matters most at the 1:50 to 1:200-year return periods where a single short record is extrapolating.

Hosking and Wallis (1997) put the method on L-moments in four steps:

  1. Screen the data with the discordancy measure DiD_i, which flags sites whose L-moment ratios sit far from the group’s.
  2. Form homogeneous regions and test them with the heterogeneity measure HH: below 1 acceptable, 1 to 2 possibly heterogeneous, 2 or more definitely heterogeneous.
  3. Choose the regional distribution with an L-moment goodness-of-fit test.
  4. Fit the growth curve to the record-length-weighted regional L-moments.

At a gauged site, μ\mu is the mean of its own annual maxima. At an ungauged site it comes from a regression on catchment descriptors. That regression is where most of the uncertainty lives: an error in μ\mu scales every return period by the same factor. The UK Flood Estimation Handbook uses the same structure with the median annual flood as its index and pooling groups in place of fixed regions.

The South African calibration (Calitz and Smithers, 2020)

The Studio tool runs for South African sites only. It implements the regional index flood model of Calitz and Smithers (2020), WRC Report 2748/1/20:

  • Data: 383 records of at least 20 years, made up of 296 DWS river gauges and 87 synthetic dam inflow series from the DWS flood studies group.
  • Regions: 42 relatively homogeneous clusters, formed by clustering on outlet latitude, longitude and distance from the coast, then adjusted by hand. Studio’s cluster map has 41 clusters, numbered 1 to 42 with no cluster 7.
  • Distribution: the Generalised Pareto, chosen as the most suitable parent distribution for South Africa.
  • Index flood: the mean annual flood scaling factor, from a log-linear regression with its own coefficients in each cluster.
  • Assessment: jack-knife testing found the cluster-based index flood model performed best on Nash-Sutcliffe efficiency and regression slope, and the authors recommend it for South Africa.
MAF=exp⁡(aln⁡A+bln⁡MAP+cln⁡Dc+d),QT=MAF×GFTMAF = \exp\left(a \ln A + b \ln MAP + c \ln D_c + d\right), \qquad Q_T = MAF \times GF_T
SymbolMeaningUnit
MAFMAFMean annual flood scaling factor (the index flood)m³/s
AACatchment areakm²
MAPMAPMean annual precipitation over the catchmentmm
DcD_cStraight-line distance from the catchment centroid to the nearest coastline pointdecimal degrees
a,b,c,da, b, c, dCluster regression coefficients–
GFTGF_TCluster growth factor for return period TT (1:2 to 1:200)–

DcD_c is in decimal degrees, as in the calibration, not kilometres. A centroid on the coastline gives Dc=0D_c = 0 and ln⁡0\ln 0 is undefined, so the tool stops with an error rather than clamping the value.

Inputs and where they come from

  • Catchment centroid: click it, type it, or let Studio take it from an uploaded polygon or a saved watershed delineation. It selects the cluster.
  • Catchment area: typed, or measured from the same polygon.
  • MAP: filled from the South African rainfall grid, as the catchment average when a polygon is loaded or the grid cell at the centroid when only a point is given. You can overwrite it.
  • Distance to coast: measured automatically to the nearest point on the southern African coastline. You can place the coast point by hand where the automatic choice is wrong.

Worked example: a 120 km² catchment in KwaZulu-Natal

The inputs are the example in the Studio documentation: A=120A = 120 km², MAP=900MAP = 900 mm, centroid at 29.5°S, 30.2°E. Running the Studio engine places the centroid in cluster 18 and finds the nearest coast at 29.81°S, 31.04°E, a distance Dc=0.894D_c = 0.894°. Cluster 18’s coefficients are a=0.82a = 0.82, b=1.37b = 1.37, c=−0.32c = -0.32 and d=−9.27d = -9.27:

ln⁡MAF=0.82(4.7875)+1.37(6.8024)−0.32(−0.1125)−9.27=4.0111\ln MAF = 0.82(4.7875) + 1.37(6.8024) - 0.32(-0.1125) - 9.27 = 4.0111

so MAF=e4.0111=55.2MAF = e^{4.0111} = 55.2 m³/s.

Return periodGrowth factor GFTGF_TQTQ_T (m³/s)
1:20.4424.3
1:51.4177.8
1:102.40132.5
1:203.69203.7
1:506.01331.8
1:1008.38462.6
1:20011.46632.6

The 1:100-year flood is 19 times the 1:2-year flood. That spread is the signature of a strongly skewed flood series, and it is why the median annual flood here is less than half the mean.

Sensitivity. Cluster 18 raises MAP to the power 1.37, so a MAP 15 % higher gives floods 21 % higher, and 15 % lower gives floods 20 % lower. Use the catchment-average MAP, not a station value from outside the catchment.

Cluster coefficients compared

The coefficients differ between clusters: the MAP exponent is 1.37 in cluster 18 and 1.87 in cluster 22, so the same 10 % error in MAP moves the flood by 14 % in one and 20 % in the other. Two clusters from the Studio cluster map:

Clusteraa (area)bb (MAP)cc (DcD_c)ddGF2GF_2GF10GF_{10}GF50GF_{50}GF100GF_{100}GF200GF_{200}
18 (KwaZulu-Natal midlands, worked example)0.821.37−0.32−9.270.442.406.018.3811.46
22 (central Free State, near Bloemfontein)1.021.870.39−13.040.522.295.357.309.76

Limitations and when to use something else

  • Calibrated on South African gauges. The clusters, coefficients and coastline term do not transfer to other countries.
  • Calibration range. Treat catchments far outside the calibration data’s range of size and MAP with caution.
  • Local anomalies. Urbanisation, unusual geology and large upstream dams are not represented by the regional parameters.
  • Cluster boundaries. A centroid near a cluster boundary can switch cluster and change every quantile. Check the neighbouring cluster’s answer; Design Flood Estimation has a cluster override for this.
  • One method of several. The SANRAL Drainage Manual expects design floods to be compared across methods. Run the probabilistic rational method and the QRT calculator on the same catchment, check against the regional maximum flood as an upper bound, and where a nearby gauge has a long record, fit it directly in flood frequency analysis.

What the full Studio tool adds

The calculator shows the cluster map and the nearest coast point, reports the cluster’s coefficients and all seven growth factors so the estimate can be reproduced by hand, and also reports the C10 scaling factor used by the probabilistic rational method for comparison. Calculations save to projects and export as Word, Excel or PDF reports. The same method can be ticked on a South African design flood estimation request, which derives the area, catchment MAP and centroid from its own delineation and puts the result beside the other methods.

FAQ

Common questions

What is the index flood method?
It is a regional flood frequency method in which the flood for return period T equals an index flood times a growth factor. The index flood sets the size of floods at the site; the growth factor, shared by every catchment in a homogeneous region, sets how fast floods grow with return period. Dalrymple described it for the USGS in 1960, and Hosking and Wallis rebuilt it on L-moments in 1997.
How is the index flood estimated at an ungauged site?
At a gauged site the index flood is the mean of the annual maximum series. At an ungauged site it comes from a regression on catchment descriptors, calibrated on the gauged sites in the region. The South African version of Calitz and Smithers (2020) uses catchment area, mean annual precipitation and distance from the catchment centroid to the coast, with separate coefficients for each cluster.
What is a homogeneous region in regional flood frequency analysis?
A group of sites whose flood series differ only by a scale factor, so that their annual maxima divided by the index flood follow one distribution. Hosking and Wallis test this with the heterogeneity measure H, which compares the spread of L-CV between sites with the spread expected by chance. H below 1 is acceptably homogeneous, 1 to 2 possibly heterogeneous, and 2 or more definitely heterogeneous.
Why is the 1:2-year flood smaller than the mean annual flood?
Because flood series are skewed. A few very large floods pull the mean above the median, and the 1:2-year flood is the median annual maximum. In the South African cluster used in the worked example on this page, the 1:2-year growth factor is 0.44, so the median annual flood is less than half the mean.
Can I use the South African index flood method outside South Africa?
No. The clusters, regression coefficients and growth curves were fitted to South African gauge records and the distance-to-coast term is measured to the southern African coastline. The general method transfers well; the coefficients do not. Elsewhere, use the national procedure, such as the Flood Estimation Handbook pooling-group method in the UK, or calibrate your own regions from local gauges.
Full documentation: Index Flood Method
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_index_flood. The agent passes the inputs, the engine does the calculation, and the record is saved to your project for you to review.

References

  1. Dalrymple, T. (1960) Flood-Frequency Analyses. Manual of Hydrology, Part 3: Flood-Flow Techniques. U.S. Geological Survey Water-Supply Paper 1543-A.
  2. Hosking, J.R.M. and Wallis, J.R. (1997) Regional Frequency Analysis: An Approach Based on L-Moments. Cambridge University Press, Cambridge.
  3. 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.
  4. Institute of Hydrology (1999) Flood Estimation Handbook, Volume 3: Statistical Procedures for Flood Frequency Estimation. Institute of Hydrology, Wallingford.
  5. SANRAL (2013) Drainage Manual, 6th ed. South African National Roads Agency SOC Ltd, Pretoria.

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.