Regional Flood Frequency Analysis (L-Moments)
Also known as: Regional Frequency Analysis, Index-Flood Method, L-Moments Regionalization, Pooled Flood Frequency Analysis
Regional flood frequency analysis estimates flood quantiles by pooling data across many hydrologically similar sites rather than relying on a single short record, which sharply reduces the uncertainty of rare-flood estimates and—crucially—allows estimation at ungauged sites. The dominant framework, codified by Hosking and Wallis in their 1997 book Regional Frequency Analysis: An Approach Based on L-Moments, rests on the index-flood assumption: within a homogeneous region, the flood frequency distributions at all sites are identical apart from a site-specific scale factor, the index flood. The method uses L-moments — linear combinations of order statistics that are far more robust than conventional moments for small samples and heavy tails (building on Hosking, Wallis, and Wood's earlier probability-weighted-moment work) — to test regional homogeneity, choose a common distribution, and fit a dimensionless regional growth curve that is then rescaled by each site's index flood. It is the standard approach for design-flood estimation where individual records are short or absent.
Key highlights
- Dramatically reduces the uncertainty of rare-flood quantiles by pooling information across many sites instead of relying on one short record.
- Enables flood estimation at ungauged sites by predicting the index flood from catchment and climate attributes.
- L-moments give robust, low-bias estimation of distribution shape and reliable homogeneity and goodness-of-fit testing for small samples and heavy tails.
- Estimates the distribution shape at the regional level, mitigating the tail-misspecification risk inherent in single-site analysis.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use regional flood frequency analysis when at-site records are too short for reliable estimation of rare floods, when you must estimate design floods at ungauged catchments, or when you want to reduce the large sampling uncertainty of single-site quantiles by pooling information across hydrologically similar sites. It is the natural choice for regional design standards, flood mapping over many catchments, and any setting where a defensible estimate is needed where little or no local data exist. The method is appropriate when a credible homogeneous region can be formed and when the index-flood assumption — identical frequency distributions apart from scale — is reasonable within that region. It is less suitable when sites are strongly heterogeneous and cannot be grouped into homogeneous regions, when the index flood cannot be reliably predicted at ungauged sites, or when a single long, high-quality record makes at-site analysis sufficient. Strong non-stationarity and regulation also challenge the pooling assumptions, as they do for at-site extreme-value methods.
Strengths & limitations
- Dramatically reduces the uncertainty of rare-flood quantiles by pooling information across many sites instead of relying on one short record.
- Enables flood estimation at ungauged sites by predicting the index flood from catchment and climate attributes.
- L-moments give robust, low-bias estimation of distribution shape and reliable homogeneity and goodness-of-fit testing for small samples and heavy tails.
- Estimates the distribution shape at the regional level, mitigating the tail-misspecification risk inherent in single-site analysis.
- Validity rests on the index-flood assumption that sites share a common scaled distribution, which real regions only approximately satisfy.
- Delineating genuinely homogeneous regions is difficult and subjective, and misgrouping biases the pooled growth curve.
- Quantile estimates at ungauged sites are only as good as the index-flood prediction model and its covariates.
- Like other extreme-value methods it assumes stationarity and is strained by climate change, land-use change, and regulation.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
What is the index-flood assumption?
It is the assumption that within a homogeneous region all sites have the same flood frequency distribution except for a single site-specific scale factor, the index flood. Equivalently, if you divide each site's floods by its index flood (typically the mean or median annual flood), the resulting dimensionless distributions are identical across sites and described by one regional growth curve. This assumption is what lets data from many short records be combined to estimate a common shape, and what makes estimation at ungauged sites possible: you only need to predict the index flood there and multiply by the regional curve.
Why are L-moments used instead of ordinary moments?
L-moments are linear combinations of ordered sample values, whereas conventional moments involve squaring and cubing deviations. That difference makes L-moments far less sensitive to outliers and far less biased in small samples, which matters enormously for floods, where records are short and tails are heavy and a single huge flood can dominate ordinary skewness. L-moment ratios (L-CV, L-skewness, L-kurtosis) also have bounded, well-behaved sampling distributions, which is what allows Hosking and Wallis's homogeneity and goodness-of-fit tests to be calibrated reliably. In short, L-moments give more stable shape estimates exactly in the regime where flood analysis operates.
How is a homogeneous region identified and tested?
Candidate regions are formed by grouping sites with similar catchment and climate attributes, using clustering or a region-of-influence approach that selects neighbors for each target site. Homogeneity is then tested with Hosking and Wallis's heterogeneity measure H, which compares the observed scatter of sites' L-moment ratios to the scatter expected under homogeneity, simulated from a fitted regional distribution: H below 1 indicates acceptable homogeneity, between 1 and 2 possible heterogeneity, and 2 or above clear heterogeneity that calls for redefining the region. A companion discordancy measure flags individual sites whose L-moments are inconsistent with the group, which may indicate data errors or sites that do not belong.
Sources
- 1.Hosking, J. R. M., & Wallis, J. R. (1997). Regional Frequency Analysis: An Approach Based on L-Moments. Cambridge University Press, Cambridge.ISBN 9780521430456
- 2.Hosking, J. R. M., Wallis, J. R., & Wood, E. F. (1985). Estimation of the Generalized Extreme-Value Distribution by the Method of Probability-Weighted Moments. Technometrics, 27(3), 251-261.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 23). Regional Flood Frequency Analysis. ScholarGate. https://scholargate.app/disaster-studies/regional-flood-frequency-analysis