Hydrology Validation And P114 Closeout
Purpose
P114.9 closes the FEMIC forest-hydrology producer phase with a bounded public gauge pilot, a published-evidence review, explicit data gaps, a model card, and the WS4/FHOPS handoff recommendation.
This page records what was validated, what failed, and what remains
unvalidated. The closeout records use femic.hydrology.validation.v1,
femic.hydrology.pilot.v1, and femic.hydrology.model_card.v1. This is
not a regulatory flood map and does not claim design-flood suitability or
operational safety certification.
What Was Validated
The validation layer checks:
public-gauge annual-maximum frequency backcasts with GEV and LP3;
split-sample early/late tail stability;
tail plausibility against the observed record;
directional expectations from published paired-watershed work;
full-pilot readiness and missing inputs; and
the model-card and consumer-handoff conditions.
The published benchmark cases are directional and order-of-magnitude checks. They are not point estimates to be copied into another watershed.
Public Gauge Pilot
Three public Water Survey of Canada gauges were used:
Carnation Creek at the Mouth (08HB048), coastal rain-dominated regime;
Coldwater River near Brookmere (08LG048), interior snow-transient regime;
Squamish River near Brackendale (08GA022), coastal mountain rain-on-snow regime.
The bounded fixture retains only annual maximum daily discharge, date, and source symbol under the Open Government Licence - Canada. It does not include daily series or private/Indigenous data.
100-year return levels use 200 bootstrap samples. Values are in m3/s.
Gauge |
Regime |
Years |
GEV 100-year |
LP3 100-year |
Early/late 100-year ratio |
Usable fits |
|---|---|---|---|---|---|---|
08HB048 Carnation Creek |
Rain |
51 (1973-2024) |
27.05 |
27.07 |
0.94 |
GEV, LP3 |
08LG048 Coldwater River |
Snow-transient |
58 (1966-2024) |
112.55 |
114.03 |
0.89 |
GEV, LP3 |
08GA022 Squamish River |
Rain-on-snow |
71 (1923-2025) |
Rejected |
2,928.24 |
Unusable |
LP3 |
Carnation Creek has a missing 2018 annual maximum. Coldwater River has a missing 2021 value. Squamish River has gaps in 1926-1954, 1996, and 2019-2020. Several recent values carry an Estimated source symbol. These are recorded as quality caveats.
Tail Plausibility Failure
The unconstrained GEV fit for Squamish River produced a shape of about -5.68 and a non-physical 100-year tail. The validation layer rejected that fit: the shape is outside the screening range [-0.5, 0.5] and the fitted 100-year quantile is many orders of magnitude above the observed maximum. The early and late split-sample GEV fits failed the same check, so the split-sample ratios are not usable.
This is a P114.5 fit-stability finding, not a data-entry error. The LP3 fit for Squamish River remains plausible. A response bundle must not use a fit that fails the tail check.
P118.2 remediation: the stationary GEV now falls back to a bounded shape
[-0.5, 0.5] and records fit_strategy="constrained". The Squamish GEV
fit now returns a finite constrained tail and is usable under the same tail
screen. The original finding remains the reason for the guard; the constrained
strategy and warning must still be reviewed before operational use.
Published Benchmark Review
The published benchmark catalogue records five directional expectations:
snow-transient high-return-period increase;
rain high-return-period increase;
frequency dependence, where a long-return-period ratio exceeds a short-return-period ratio;
snowmelt high-return-period increase; and
paired control-watershed no-change.
Sources include Alila et al. (2009, doi:10.1029/2008WR007207), Kuraś et al. (2012, doi:10.1029/2011WR010705), Green & Alila (2012, doi:10.1029/2012WR012449), Schnorbus & Alila (2013, doi:10.1029/2012WR011901), Pham et al. (2025, Journal of Hydrology 661:133704), and Rong, Bishop & Alila (2026, doi:10.1016/j.advwatres.2026.105432).
No public paired baseline/treatment frequency result was assembled in P114.9.
All five cases therefore remain not_evaluated for the public pilot. The
directional assessment harness is implemented and tested with synthetic paired
ratios; it is ready for a reviewed paired-watershed dataset in a follow-up
phase. A passed directional screen would not prove hydrologic correctness.
Pilot Readiness And Data Gaps
The full two-watershed process pilot was not completed. The bounded run records these gaps:
no public hydrology domain manifest with LiDAR, stream, road, and receptor layers was supplied;
no annual harvest/road disturbance ledger was supplied;
no executable process model was available for the pilot;
no downstream receptor capacity or damage curve was available; and
no hydrologist review record was supplied.
The gauge backcast is therefore a frequency-estimation check, not a treatment or process-model validation.
Uncertainty And Deviations
Bootstrap intervals are percentile intervals, not a Bayesian posterior.
Short records and nonstationarity limit tail inference.
Missing years and estimated values affect the fits.
The GEV fit for Squamish River is rejected by the tail screen.
No paired treatment/control response was available.
Gauge-frequency behaviour does not identify harvest or road causality.
Climate projection, regionalization, and damage-curve validation are outside P114.9.
No hydrologist review has been recorded.
Model Card
The closeout writes a versioned model card with:
intended use and out-of-scope uses;
regimes and validation summary;
data gaps and usable fits;
uncertainty and limitations;
a review checklist; and
the consumer handoff conditions.
The model card status is not_reviewed. It is provisional until a
hydrologist records approval, conditions, or rejection.
WS4/FHOPS Handoff
The FEMIC response contract is stable for screening and decision support, with conditions:
WS4 P49 may ingest a pinned response bundle for spatial scheduling screens.
FHOPS P14 may ingest the same bundle for operational road/harvest tradeoff reporting.
Both consumers must verify
bundle_id,bundle_version, and artifact SHA-256 values.Both must carry the screening-only caveat and refuse queries outside the recorded applicability domain.
A fit that fails the tail check must not be used for risk reporting.
Hydrologist review and a completed process-model pilot are required before operational use or regulatory interpretation.
The phase can hand off a bounded, versioned screening layer. It cannot hand off a regulatory or design-flood product.
Closeout Statement
P114.9 closes the FEMIC producer phase with a conditional handoff:
public gauge backcasts completed and recorded;
published evidence catalogued as directional benchmarks;
the GEV instability recorded and rejected by the tail screen;
data gaps, uncertainty, and review status recorded; and
WS4/FHOPS consumer conditions documented.
The next work is a reviewed paired-watershed/process-model validation and a hydrologist-approved model card, not a claim that the current layer is fully validated.
Reproduce The Bounded Pilot
python -m femic hydrology pilot \
--gauges tests/fixtures/hydrology/pilot/gauges.json \
--annual-maxima tests/fixtures/hydrology/pilot/annual_maxima.csv \
--output-dir runtime/hydrology/p114_9_closeout \
--bootstrap-samples 200 \
--seed 42
The command writes pilot_run.json, validation_report.json, and
model_card.json. Passing a public domain manifest, disturbance ledger, or
review record upgrades the readiness audit; those inputs were not available for
this bounded closeout.