When are flood-frequency estimates reportable? A selective-coverage audit of short hydrological records

This study is a selective-coverage audit, not a claim of a universally reliable reporting rule. It asks whether a three-state reporting policy (PASS, HOLD, WITHHOLD) for stationary generalized extreme-value (GEV) return levels remains reliable after selection, and whether pooled coverage conceals tail-specific failure. The policy combines profile-likelihood intervals, split-conformal expansion and observable stability diagnostics. In an independent paired synthetic test (1,000 in-domain cases), a tail-protected policy accepted 201 cases (20.1%); 191/201 intervals contained the true Q100 (95.02%; Wilson 91.1–97.3%) and none exceeded 50% error, against 313/338 (92.60%) and 17 large errors for the baseline policy. The apparent pooled gain stems mainly from withholding difficult cases rather than improved interval construction: on the 201 cases accepted by both policies the baseline intervals covered 194 and the protected intervals 191 (median widths 76% and 50%), while the 137 cases rejected only by the protected policy were covered 119 times (86.9%). A post hoc ablation shows that a fitted-shape cap, ξ̂ ≤ 0, alone reproduces the result (221/231, 95.7%). Acceptance was concentrated at true ξ = −0.1 (185 of 201); coverage was 10/16 at ξ = 0.2 and no case was accepted at ξ = 0.5 or 0.8. Without recalibration, the frozen policy covered 232/245 accepted cases (94.7%) under a finer GEV shape grid, Pearson type III and serially dependent maxima, but only 51/56 (91.1%) for GEV with ξ ≥ 0. In five Australian gauges (282 annual maxima of daily-mean flow, 83.7% graded C or E at the peak date) every training-fit ξ̂ exceeded 0.2, so no site could be accepted, and GEV did not outperform a lognormal benchmark at four sites. The evidence delineates a boundary of reportability for light-tailed regimes; it does not validate design-flood estimates, heavy-tail coverage, or superiority over established practice such as Bulletin 17C. Keywords: flood-frequency analysis; generalized extreme value; selective coverage; abstention; split-conformal calibration; heavy-tail identifiability; daily-flow maxima; Australian gauges.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24412517
Primary Topic
Flood Risk Assessment and Management
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

When are flood-frequency estimates reportable? A selective-coverage audit of short hydrological records

Ayad Kadhim Hussein
Zenodo (CERN European Organization for Nuclear Research)
Flood Risk Assessment and Management
article

When are flood-frequency estimates reportable? A selective-coverage audit of short hydrological records

Ayad Kadhim Hussein
article en

Abstract

This study is a selective-coverage audit, not a claim of a universally reliable reporting rule. It asks whether a three-state reporting policy (PASS, HOLD, WITHHOLD) for stationary generalized extreme-value (GEV) return levels remains reliable after selection, and whether pooled coverage conceals tail-specific failure. The policy combines profile-likelihood intervals, split-conformal expansion and observable stability diagnostics. In an independent paired synthetic test (1,000 in-domain cases), a tail-protected policy accepted 201 cases (20.1%); 191/201 intervals contained the true Q100 (95.02%; Wilson 91.1–97.3%) and none exceeded 50% error, against 313/338 (92.60%) and 17 large errors for the baseline policy. The apparent pooled gain stems mainly from withholding difficult cases rather than improved interval construction: on the 201 cases accepted by both policies the baseline intervals covered 194 and the protected intervals 191 (median widths 76% and 50%), while the 137 cases rejected only by the protected policy were covered 119 times (86.9%). A post hoc ablation shows that a fitted-shape cap, ξ̂ ≤ 0, alone reproduces the result (221/231, 95.7%). Acceptance was concentrated at true ξ = −0.1 (185 of 201); coverage was 10/16 at ξ = 0.2 and no case was accepted at ξ = 0.5 or 0.8. Without recalibration, the frozen policy covered 232/245 accepted cases (94.7%) under a finer GEV shape grid, Pearson type III and serially dependent maxima, but only 51/56 (91.1%) for GEV with ξ ≥ 0. In five Australian gauges (282 annual maxima of daily-mean flow, 83.7% graded C or E at the peak date) every training-fit ξ̂ exceeded 0.2, so no site could be accepted, and GEV did not outperform a lognormal benchmark at four sites. The evidence delineates a boundary of reportability for light-tailed regimes; it does not validate design-flood estimates, heavy-tail coverage, or superiority over established practice such as Bulletin 17C. Keywords: flood-frequency analysis; generalized extreme value; selective coverage; abstention; split-conformal calibration; heavy-tail identifiability; daily-flow maxima; Australian gauges.

Zenodo (CERN European Organization for Nuclear Research)
University of Kufa (IQ)
Flood Risk Assessment and Management
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon