Skip to main content
Research · operator view · build 2026.09.15

Model health &
deployment traceability.

The operator view tracks production model status, source freshness, and current public coverage. Shorelife is a daily forecast product and does not detect intra-day spills.

Registry winner
xgb-undersample-ensemble-curated-v0
promotion-gate winner; a serve-time router picks per beach by sample age
Test AUCPR
0.8094
Test Brier
0.0570
Coverage
252 / 341
74% of California beaches
Public release
Eligible
Spatial backtests

Performance on unobserved counties

spatial beach persistence
AUCPR
0.824
Brier
0.158
15 folds · 12242 samples
spatial county persistence
AUCPR
0.501
Brier
0.138
6 folds · 69566 samples
spatial beach xgb undersample ensemble
AUCPR
0.961
Brier
0.094
15 folds · 12242 samples
spatial county xgb undersample ensemble
AUCPR
0.748
Brier
0.099
6 folds · 69566 samples
spatial beach hist gbm
AUCPR
0.960
Brier
0.097
15 folds · 12242 samples
spatial county hist gbm
AUCPR
0.667
Brier
0.123
6 folds · 69566 samples
spatial beach hist gbm positive persistence guard
AUCPR
0.853
Brier
0.135
15 folds · 12242 samples
spatial county hist gbm positive persistence guard
AUCPR
0.603
Brier
0.127
6 folds · 69566 samples
spatial beach hist gbm persistence blend
AUCPR
0.958
Brier
0.107
15 folds · 12242 samples
spatial county hist gbm persistence blend
AUCPR
0.705
Brier
0.100
6 folds · 69566 samples
spatial beach xgb undersample offset
AUCPR
0.929
Brier
0.194
15 folds · 12242 samples
spatial county xgb undersample offset
AUCPR
0.619
Brier
0.122
6 folds · 69566 samples
Deep dives

Transparency documentation

Labels · Thresholds

Risk Labels

How exceedance probability maps to the four public bands, legal basis, and what the model can and cannot predict.

READ →
Provenance · Cadence

Data Sources

The six primary datasets feeding the pipeline — BeachWatch, NDBC, CDIP, Open-Meteo, USGS NWIS, and CEDEN — with freshness status.

READ →
AUCPR · Brier · Spatial CV

Calibration

Live production metrics, spatial backtest results, and candidate model registry from the latest CI run.

READ →
Scientific artifacts

Open data & methodology

Technical Methodology

Model design, coverage limits, risk-band calibration, spatial holdout protocol, and known limitations.

Shorelife Team
READ →

Source Code & Data

Full pipeline, training scripts, and curated datasets available on GitHub under an open-source license.

Automated Pipeline
GITHUB →