Forecast scoreboard Beta
Beta. New forecast, tested on past years but with only a short live record. Use it as background, not for decisions.
Question: will the river at a gauge rise 0.5 m or more above today's level within 7 days? Lower Brier score is better; skill above 0 means better than the baseline. Baselines: climatology (how often this happened at this time of year, 2007–2020) and persistence ("same as the last week").
Held-out test (2021–2026, fixed in advance)
| Gauge | Shown | Days | Rise rate | Brier model | Brier climatology | Brier persistence | Skill vs climatology (90% range) | Skill vs persistence (90% range) | Precision / recall | F1 model vs clim / pers |
|---|---|---|---|---|---|---|---|---|---|---|
| Chiang Saen | Public (beta) | 2123 | 28% | 0.170 | 0.188 | 0.203 | +0.10 (+0.06 to +0.13) | +0.16 (+0.14 to +0.18) | 0.48 / 0.60 | 0.53 vs 0.48 / 0.44 |
| Vientiane | Public (beta) | 2123 | 28% | 0.137 | 0.174 | 0.197 | +0.21 (+0.18 to +0.25) | +0.31 (+0.27 to +0.35) | 0.61 / 0.68 | 0.64 vs 0.57 / 0.43 |
| Khong Chiam | Public (beta) | 2123 | 20% | 0.084 | 0.120 | 0.147 | +0.30 (+0.26 to +0.34) | +0.43 (+0.39 to +0.48) | 0.64 / 0.83 | 0.72 vs 0.58 / 0.45 |
| Pakse | Public (beta) | 2108 | 18% | 0.082 | 0.110 | 0.134 | +0.25 (+0.18 to +0.31) | +0.39 (+0.34 to +0.43) | 0.60 / 0.79 | 0.68 vs 0.56 / 0.45 |
| Stung Treng | Public (beta) | 2120 | 17% | 0.076 | 0.104 | 0.124 | +0.27 (+0.15 to +0.38) | +0.39 (+0.29 to +0.47) | 0.62 / 0.77 | 0.69 vs 0.55 / 0.44 |
| Kratie | Public (beta) | 2123 | 25% | 0.090 | 0.134 | 0.161 | +0.33 (+0.24 to +0.41) | +0.44 (+0.37 to +0.51) | 0.71 / 0.84 | 0.77 vs 0.65 / 0.52 |
| Tan Chau | Public (beta) – tide-adjusted target | 2062 | 9% | 0.033 | 0.063 | 0.071 | +0.48 (+0.38 to +0.58) | +0.54 (+0.43 to +0.63) | 0.76 / 0.73 | 0.74 vs 0.44 / 0.43 |
| Chau Doc (Bassac) | Public (beta) – tide-adjusted target | 2062 | 6% | 0.029 | 0.050 | 0.054 | +0.42 (+0.28 to +0.53) | +0.47 (+0.34 to +0.56) | 0.72 / 0.65 | 0.68 vs 0.34 / 0.37 |
Live record (since release) separate from the backtest
Live = predictions as issued, scored only after their window closed. Separate from the held-out backtest. Nothing back-filled. Updated nightly.
7-day rise outlook (public gauges)
0 outlooks scored yet (8 waiting for their 7-day window to close)
Too few cases yet (0 scored; we need at least 30 before these numbers mean much). Treat them as anecdotes, not a track record.
Rain + rise heads-up (beta)
0 heads-up checks scored yet.
Too few cases yet (0 scored; we need at least 30 before these numbers mean much). Treat them as anecdotes, not a track record.
Live forecasts
Logged every night from release; each is scored once its 7 days have passed. Nothing is back-filled.
| Gauge | Issued | Scored | Pending | Brier model | Brier climatology | Brier persistence |
|---|---|---|---|---|---|---|
| Chiang Saen | 1 | 0 | 1 | – | – | – |
| Vientiane | 1 | 0 | 1 | – | – | – |
| Khong Chiam | 1 | 0 | 1 | – | – | – |
| Pakse | 1 | 0 | 1 | – | – | – |
| Stung Treng | 1 | 0 | 1 | – | – | – |
| Kratie | 1 | 0 | 1 | – | – | – |
| Tan Chau | 1 | 0 | 1 | – | – | – |
| Chau Doc (Bassac) | 1 | 0 | 1 | – | – | – |
Latest public forecasts
| Gauge | Data date | Model | Climatology | Outcome |
|---|---|---|---|---|
| Chiang Saen | 2026-10-31 | 14% | 19% | pending |
| Vientiane | 2026-10-31 | 8% | 16% | pending |
| Khong Chiam | 2026-10-31 | 5% | 9% | pending |
| Pakse | 2026-10-31 | 4% | 6% | pending |
| Stung Treng | 2026-10-31 | 4% | 10% | pending |
| Kratie | 2026-10-31 | 10% | 17% | pending |
| Tan Chau | 2026-10-31 | 12% | 2% | pending |
| Chau Doc (Bassac) | 2026-10-31 | 33% | 3% | pending |
Caveats
- Tan Chau and Chau Doc are tidal, so their rows use a different, tide-adjusted question (v7.7, separate locked protocol): will the 15-day average level rise 0.25 m or more within 7 days? Eight of those 15 days are already known when the forecast is issued, which makes this question easier than the raw one; persistence on the filtered level is the baseline that accounts for that.
- The model is retrained at most once a year, under a new protocol locked before fitting, and replaced only if it beats the current model and both baselines on the newest held-out year.
- Persistence is a weak baseline for a 7-day rise question; climatology is the fairer test.
- A rise can come from rain, tides or dam releases; the model does not attribute cause. Not a flood warning.
Protocol: docs/v73_rise_protocol.md, SHA-256 2c3a374efae300b3…, locked 2026-11-01T12:00:11+1100 before fitting.
Data: Mekong River Commission (MRC) PMFM public dashboard (daily series not redistributed); NASA IMERG.