Forecasts, Methods and Data

Review the held-out forecast queue, inspect the observations behind Revir's dated citations and compare all 18 benchmark arms.

2024 Forecast Replay

In the recent-backwater cohort, 51 of the first 109 forecasts in the kept queue preceded a measured inland tilt, compared with 23 in the matched filtered queue. Some forecasts share a verifying pass. The readout counts both forecast hits and distinct passes.

Forecasts Followed by a Measured Tilt
CohortReviewsMatched modelKept hitsDeleted hits
Recent backwater109Grounded Qwen2.5-7B5123
Recent coastal30Grounded Qwen2.5-7B205
Recent backwater109Hydraulic trees · strong tilt181
All unseen reaches1691Hydraulic trees · strong tilt10359

The interactive queue uses the same held-out forecasts as this table.

Dated Observations and Prompt Audit

Forecast row 7170, unnamed European reach 24294000061, issued 3 May 2024. The archived explanation cites the 24 April 2024 and 14 April 2024 readings.

Recorded downstream fall
Datem per 10 km
24 April 20247.360
14 April 20246.327
Reveal verifying pass

5 May 2024 · 6.843 m downstream fall per 10 km.

The frozen grounded audit reproduced 98/98 expected date-and-value fields for the latest and recent minimum observations, versus 0/98 for the latent-only version. Given both dates in its two-line prompt anchor, the grounded model reproduced 149/188 wider expected facts and made 31/180 unsupported numeric mentions.

Benchmarks and Scope

Brier error scores the probability forecast. Queue hits count how many reviewed forecasts preceded a measured event. The two measures can favour different models.

Full 18-arm benchmark and scope limits

Recent-backwater confirmation in 2024, with 545 orbital forecasts at 26 reaches. Each issue counts once. Ties are ordered by frozen membership row. Random expectation is an additional comparison.

Table 1 · Any Inland Tilt
ArmBrierBrier skillAUROCHitsDistinct passesCut-off tie range
Both inputs, tilts kept0.20173+11.7%0.648511950–62
Both inputs, tilts deleted0.21437+6.2%0.577231120–25
Dated text summary only0.22787+0.3%0.436191113–29
Time-series latents only0.22465+1.7%0.674441834–47
Off-the-shelf LLM (raw table)0.22926−0.3%0.521291828–29
Encoder + Perceiver, no LLM0.20030+12.3%0.637411640–41
Hydraulic trees · kept0.18832+17.6%0.592351235–35
Hydraulic trees · deleted0.19889+13.0%0.601391239–39
Persistence · kept0.29632−29.7%0.451141014–14
Persistence · deleted0.19406+15.1%0.655541650–54
Basic OpenTSLM tokenizer (CNN, no LLM)0.22941−0.4%0.603301229–31
PhysicalTimeRoPE0.23029−0.8%0.674491543–60
RankRoPE0.22135+3.1%0.627411741–42
CD Terminal 1870 · kept0.19698+13.8%0.636451845–45
CD Terminal 1870 · deleted0.18412+19.4%0.668601860–60
CD Candidate 80 · 1h interim0.18427+19.4%0.656471547–47
Four-visit noise · kept0.17973+21.3%0.617271427–27
Four-visit noise · deleted0.18686+18.2%0.655491449–49
Random review, expectation··0.50026.0019.67·
Table 2 · Strong Inland Tilt Beyond 1σ
ArmBrierBrier skillAUROCHitsDistinct passesCut-off tie range
Both inputs, tilts kept0.10077 †+0.1%0.300110–1
Both inputs, tilts deleted0.10037 †+0.5%0.339424–4
Dated text summary only0.09815 †+2.7%0.56415109–20
Time-series latents only0.09808 †+2.7%0.415220–3
Off-the-shelf LLM (raw table)0.09899 †+1.8%0.48511510–11
Encoder + Perceiver, no LLM0.10616 †−5.3%0.354222–2
Hydraulic trees · kept0.09456+6.2%0.60318718–18
Hydraulic trees · deleted0.10102−0.2%0.289111–1
Persistence · kept0.11446−13.5%0.417141014–14
Persistence · deleted0.10152−0.7%0.341535–5
Basic OpenTSLM tokenizer (CNN, no LLM)0.10137−0.5%0.38610510–10
PhysicalTimeRoPE0.10110−0.3%0.352320–6
RankRoPE0.09901+1.8%0.347442–6
CD Terminal 1870 · kept0.11217 †−11.2%0.316222–2
CD Terminal 1870 · deleted0.11579 †−14.8%0.298323–3
CD Candidate 80 · 1h interim0.11506 †−14.1%0.329646–6
Four-visit noise · kept0.10235 †−1.5%0.539191019–19
Four-visit noise · deleted0.10991 †−9.0%0.316111–1
Random review, expectation··0.50011.209.09·

† Any-tilt model ranking evaluated against the outcome beyond 1σ. Probability error is a cross-target diagnostic.

Keeping inland tilts lowers Brier error in 3 of the 5 paired model families. The kept grounded model scored +11.7% Brier skill. Arms with higher skill on this cohort were Encoder + Perceiver, no LLM, Hydraulic trees · kept, Hydraulic trees · deleted, Persistence · deleted, CD Terminal 1870 · kept, CD Terminal 1870 · deleted, CD Candidate 80 · 1h interim, Four-visit noise · kept and Four-visit noise · deleted.

At 109 reviews, filtered persistence found 54 hits on 16 distinct passes. The kept grounded model found 51 on 19 distinct passes, 33 coastal and 18 inland. The signed four-visit noise model had a Brier error of 0.17973.

The recent-backwater cohort had 56 strong-tilt forecast issues across 23 distinct passes. The matched trees found 18 vs 1 hits at 109 reviews. In the top fifth (1,691 reviews) across all 80 reaches, the trees found 103 vs 59.

The replay ranks archived forecasts across a fixed set of issues. Forecasts share verifying passes, and measured tilts smaller than the reported uncertainty count for the any-tilt endpoint. Test membership is disjoint from project training by reach, level-six SWORD basin and network group. Foundation-model pretraining is outside this audit. Operational lead time, salinity at intakes, drainage decisions and flood losses require separate observations and prospective evaluation.

Forecast Method and Ablations

We keep the sign of quality-screened SWOT slopes acquired before issue, as reprocessed in Version D. Across irregular satellite intervals, the model encodes elapsed time, orbit and tide information and receives a short dated summary.

PhysicalTimeRoPE encodes the target stretch's signed tilt, water-height anomaly and measurement uncertainty with 32 rotary periods. The OpenTSLM masked Perceiver supplies 64 latent prefix tokens. A two-line dated summary provides the latest observation, the four-visit nonnegative baseline, the recent minimum and the lunar phase contrast when available. Qwen2.5-7B uses LoRA adaptation. The replay uses the calibrated probability from its Yes/No token scores.

Signed SWOT passesElapsed time + dated summaryNext-pass risk p
The target is the first UTC day with a valid pass 2 to 21 days after issue (using the lowest-uncertainty pass when a reach is observed twice in one day). A negative downstream fall means the downstream end stands higher.

Matched Input Ablations

Matched Qwen2.5-7B Brier skill against the 2023 event rate, on 545 forecast issues at 26 recent-backwater reaches in 2024.

Brier skill, relative to the training event rate
Off-the-shelf LLM (raw table)−0.3%
Dated text summary only+0.3%
Time-series latents only+1.7%
Both inputs, tilts deleted+6.2%
Both inputs, tilts kept+11.7%

Paired reach-bootstrap intervals for the kept model's gains over filtered, text-only and latent-only inputs excluded zero. The kept–filtered difference is +5.53 percentage points (+0.53 to +7.67).

Elapsed Time and Coastal Terms

The basic OpenTSLM arm uses the upstream CNNTokenizer in a standalone CNN encoder and forecast head with 1.10M trainable parameters, scored on the same held-out forecast issues using signed SWOT visits.

On recent-backwater stretches, that CNN arm scored −0.4% Brier skill (0.603 AUROC), catching 30 hits in 109 reviews and 8 versus 20 for the full model in 30 coastal reviews (2.5 times as many coastal hits). PhysicalTimeRoPE also lowered synthetic irregular-pulse Brier error by 12.7% compared with the CNN.

In the separate clock diagnostic, the PhysicalTimeRoPE encoder ranked inland tilts at AUROC 0.674 versus 0.627 for RankRoPE at recent-backwater reaches, though its Brier error was higher in that cohort (13.7% lower error than RankRoPE on synthetic pulses).

On 6 recent coastal reaches, symbolic equation search (terminal CD Candidate 1870) found 13 hits with tilts kept and 21 with tilts deleted in 30 forecast reviews. Brier errors were 0.24181 and 0.20120, respectively. One reach accounted for 41 of the 50 positive forecast issues.

Calibrated with equal issue weights, the signed four-visit uncertainty baseline found 10 hits, whereas disclosed 1-hour Candidate 80 found 24 with 17.7% lower coastal Brier error. That Brier advantage disappeared after the 24-hour 2023 search. Terminal selection lowered Brier error on T2 steepening with tilts kept (0.15840 versus 0.16203 across 80 reaches).

Download the Summary Figure

PDFSVGPNG

Discarded Passes and Lunar Timing

The discharge rule would remove 586 of 20,083 quality-screened SWOT passes at 95 of 320 stretches. This includes 211 passes more than 1.96 reported measurement uncertainties below zero.

M2 Lunar Timing at 12.42 Hours
Reach GroupStretchesMean RNominal pHolm p
Coastal130.3670.0050.060
Flat inland280.3030.495·

The permutation draws event times from the actual passes at each stretch and averages the phase-vector lengths with equal reach weights. The coastal adjustment covers 12 tests. The pooled inland result leaves the cause of individual passes open.

All 95 Reaches With Discarded Passes
Quality-screened SWOT passes
ReachScreenedDeletedBeyond 1.96σShare
Eider15252034.2%
Unnamed Bahia coast reach, Brazil128361028.1%
24480900071152251616.4%
Unnamed Gulf coast reach, Mexico4325058.1%
Tombigbee River floodplain, US4323053.5%
Yata River8821223.9%
Trent, UK8318621.7%
Weser River, Germany88151517.0%
85104000031107141213.1%
146310001414012030.0%
Kuban4112729.3%
261704001611321097.6%
Nile River129907.0%
Tanaro River, Italy469819.6%
24440500271569216.1%
Rio Negro, Brazil130956.9%
Unnamed James Bay coast reach, Canada549216.7%
81156000011289032.1%
83192000271759712.0%
Tano River, Ghana92888.7%
14633000341388321.1%
Psel548614.8%
26190100141528015.4%
Selawik River, Arctic Alaska186824.3%
Tevere307523.3%
Courantyne467015.2%
Malani123755.7%
La Mana, French Guiana115716.1%
62269300551327121.9%
64140400031477014.9%
Pantanal Basin, South America134755.2%
Noatak River110766.4%
Wolf River74608.1%
73255000131346217.6%
24328900031335515.2%
Oulujoki73506.8%
2619040025180546.3%
2816070002158508.6%
Beni River100525.0%
Savannah River265019.2%
75350400081265019.2%
Meade River65537.7%
14443000121264115.4%
Mauwa294413.8%
25210800031213421.9%
Orinoco123443.3%
7123030004169405.8%
Alabama River284114.3%
75377000171284414.3%
81360000091111423.6%
Mouhoun River31329.7%
Bamingui River41337.3%
Itomampy38307.9%
Khopyor River59315.1%
Gauja54325.6%
2618400043197323.1%
2832410021131339.7%
Uruguay River41307.3%
Rio Cayapas303010.0%
Caloosahatchee River193015.8%
75371200171253112.0%
Kafue River39205.1%
Nyong River47214.3%
Ikopa42204.8%
2525400032173222.7%
2810010027141204.9%
Malani128211.6%
Isiboro River104201.9%
6310340023143204.7%
Riacho Monte Lindo Grande82222.4%
7112000025141204.9%
Savannah River29226.9%
St. Johns River, US122116.7%
8317050093183222.4%
Ngounie River25114.0%
14221400031135100.7%
Daka41112.4%
White Volta42102.4%
17211100441111100.9%
Manambavana23104.3%
Tevere41112.4%
Snov65101.5%
2256020010188101.1%
River Shannon86101.2%
Nemunas River47102.1%
2429400006155101.8%
Lakselva123110.8%
2526040014189101.1%
2614800030179111.3%
2710600014183111.2%
Malani83111.2%
L'Approuague37102.7%
Yata River104101.0%
6720500108136112.8%
7237000005140112.5%

Gauge Appendix for the River Trent at Keadby

This 2024 development example is separate from the unseen forecast test. At Keadby, the pumping station drains at high tide into a 10.46 km tidal Trent stretch towards the Humber, where a gauge reports pier height while SWOT measures tilt along the reach. Keadby Pumping Station

Satellite pass retained Inland10 kmTowards sea
12 March 2024 · tide coming in
Satellite pass retained Inland10 kmTowards sea
17 March 2024 · tide going out
Keadby gauge and satellite observations
DateSWOT measuredGauge either side of pass
12 March 20240.86 ± 0.20 m seaward rise per 10 km3.188 → 3.965 m
17 March 20240.84 ± 0.21 m seaward fall per 10 km1.972 → 1.870 m

A single pier height fits both an inland and a seaward tilt. The flow filter retained 65 of 83 quality-screened SWOT passes in 2023–24. Keeping both signs retained all 83.

On 12 March 2024, the Keadby gauge rose +0.78 m in 15 minutes around the pass. The Environment Agency marked the readings Unchecked. SWOT supplies a spatial gradient. Neither instrument measures salinity or flow speed directly.