Timing Projection Methodology

Every 5 minutes, Steelhead Almanac projects what each Steelhead Alley river is about to do over the next 18 hours, and surfaces the result as a Timing badge next to the current Rating on every river page. This page explains how that projection is computed, what each state means, and where the inputs come from.

By looking ahead at rising and falling water levels, our modeling helps anglers target the optimal fishing windows. Timing projections should always be used alongside the current rating to determine when conditions are best for fishing.

How to use a timing projection

Start with the current gauge reading, then look at the forecast target time and projected flow.
Follow the projected rise or recession against the river’s reference range to plan the next water window.
Recheck as rain arrives and the gauge changes. We retain issued forecasts and compare them with later readings at the target time.

Forecast confidence: Confidence describes the strength of the forecast inputs and model fit. The recorded checks below show how the flow forecasts actually performed.

Forecast check: Grand River, October 6, 2026

Dated example, not current conditions. Captured Oct 8, 2026 22:54:07 UTC.

Reading or estimate	Value	Recorded time / basis
Flow when the forecast was prepared Gauge reading	31.3 CFS	Oct 4, 2026 at 8:00 PM EDT
Forecast flow Recorded before the target time	37.8 CFS	Issued Oct 4, 2026 at 8:45:48 PM EDT; target Oct 6 at 8:00 PM EDT
Actual flow at the target time Same river, same gauge, exact target time	24.1 CFS	Oct 6, 2026 at 8:00 PM EDT
Forecast error Forecast compared with actual flow	13.7 CFS high	37.8 − 24.1 CFS
Unchanged-flow comparison Starting flow held unchanged	7.2 CFS high	31.3 − 24.1 CFS

The forecast called for a rise from 31.3 to 37.8 CFS. Instead, the Grand fell to 24.1 CFS. Holding the starting flow unchanged would have been closer on this one. That is why we keep both the forecast and the gauge result: the river gets the last word, and we keep the score.

Recorded check 68 · own-gauge-flow-v1 · model c5c4a91f. Latest eligible scored case at capture, selected by target time rather than by how well the forecast performed.

Grand River at Painesville · USGS gauge 04212100

Sources and further explanation

Read the Grand River briefing →

Published by Steelhead Almanac / Buell EA LLC. Agency observations, Almanac estimates and fishing reports serve different purposes. Suggest a correction with its source.

Live Example: Black River (Watertown)

Timing Badge:

Low water — projection flat

Projected State: low

Next Event: none

Confidence: 45%

Active Sources: NWS AHPS

The 7 Timing States

State	Technical Definition	Angler Meaning	Badge Display
opening	Flow is moving into the optimal window.	Flow may enter the reference range; check the current rating and access separately.	Renders (Green)
closing	Flow is currently inside the optimal window, but projected to leave it.	Flow is in range now but may leave it; this does not establish safe access or fish presence.	Renders (Yellow)
blown	Flow is above the optimal window and peak is ahead or no recovery within 18h.	Flow is above the reference range. Check local conditions; clarity is not directly measured by this state.	Renders (Red)
recovering	Flow is dropping back toward fishable, but clarity typically lags flow.	Water dropping; watch clarity before driving out.	Renders (Cyan)
low	Flow is below the optimal window and projected to remain below it.	Flow is below the reference range. This state alone does not establish clarity or fish behavior.	Renders (Gray)
stable	Flow is inside the optimal window and projected to remain inside it.	Projected flow stays in range; other inputs can still change the rating.	Silent
unknown	Insufficient signals or confidence is below floor (0.2).	No usable timing projection. Check fresh observations and local conditions before planning.	Silent

Next-Event Indicators

When a state transition is imminent, the projector identifies the next inflection point within the 18-hour window:

Event Indicator	Definition
rises_into_range	Flow is projected to rise into the optimal window from below.
drops_into_range	Flow is projected to fall into the optimal window from above (the drop).
rises_above_range	Flow is projected to rise above the optimal range, blowing the river out.
drops_below_range	Flow is projected to drop below the optimal fishable range.
peaks	Flow is projected to hit its maximum storm peak and begin receding.

How Confidence is Computed

Confidence is a 0.0 – 1.0 score that blends four factors: how trustworthy the strongest input source is, how far ahead in time we're projecting, how well a flow trend can be fit to the recent 24-hour data, and how active the regional storm lifecycle is.

High (0.8+)

Trust the projection. Strongly backed by NWS AHPS forecast points or high-correlation regressions.

Moderate (0.5–0.8)

Projection is directional. Combine with current ratings before making a final trip choice.

Low (Below 0.4)

Do not commit to the timing. Default purely to the river's current Rating.

Data Signal Cascade

The projector gathers four primary input signals to construct the sub-daily hydrologic projection:

1. NWS AHPS (Advanced Hydrologic Prediction Service)
Where available, we fetch official NOAA/NWS river forecast graphs. These are highly accurate, hydrologist-reviewed data points. Not all Steelhead Alley streams have AHPS monitoring stations.
2. Storm Pulses
We maintain a database of storm events. Rain-driven flow projections require quantitative NOAA rainfall forecasts, converted to inches over the weather periods. Weather phrases and rain probabilities alone do not supply a rainfall amount. When quantitative coverage is missing, this rain-driven model is skipped; that does not mean no rain is expected.
3. Recent Flow Regression
Local flow models require a dated gauge measurement within the ingestion freshness limit (normally two hours). Saving an old reading again does not make it current. Trend fitting uses distinct observation times and excludes readings unavailable at issuance. A confirmed peak followed by three consecutive hourly observations and a decline of at least 5% replaces an older rising trend with a recession fit; that short fit is extrapolated only across the number of hours it covers. Otherwise, rising trends are limited to 24 hours ahead and recession extrapolation to 48 hours. Dam-controlled rivers use official guidance where available, otherwise unchanged fresh flow. Missing or stale local flow remains unavailable. These rules do not predict dam releases or establish forecast accuracy.
4. Storm Lifecycle Context
Our system monitors regional weather conditions to determine the storm phase (quiet, watching, active, recovering). This context modulates the confidence score rather than driving the flow projection curve directly.

UI Integration

River Show Pages: Renders as an inline badge (e.g. Low water, Closing ~4h) adjacent to the main rating. Hovering or clicking reveals confidence and sources.
Smart Gauge: Displays a dashed 18-hour projection curve extending from the "NOW" indicator.
Day Analysis Chart: The flow timeline is stretched to 90 hours, showing a dashed future projection with crossing markers when events occur.
Explorer Map: Hovering over a river pin shows the timing badge directly inside the popup.
Public API: Accessible via the timing JSON payload on our public API endpoints.

Prospective flow forecast checks

We are recording future flow predictions from the twice-daily 7-day Index forecast before their target time, then comparing them with that river's own primary USGS gauge. This checks flow in cubic feet per second, not catches, fishing safety, clarity, the whole Index, or the separate 18-hour timing display.

Each prediction keeps its declared target time: the selected official river-forecast point, or the exact hour used by our flow model. We take the closest observed gauge reading within 30 minutes of that target, allowing 24 hours for arrival. Missing readings and excluded proxy rivers stay in the counts. The comparison forecast is the last fresh measured flow available when the prediction was prepared, held unchanged.

Aggregate last checked: 2026-11-04T10:05:37+00:00. First locked prediction: 2026-10-05 00:45:49. 12,462 recorded cases.

Excluded not prospective: 3
Excluded proxy: 1,836
Missing actual: 13
Missing prediction: 36
Pending: 1,667
Scored: 8,907

Model version c5c4a91f9199

1860 cases. Versions are reported separately. CFS errors across different rivers are not directly comparable; use the river breakdown too.

Elapsed forecast lead

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
0-24h	170 / 145	21.0	21.1 / 18.1 (n=144)	87 / 144
unknown	3 / 0	Pending	No paired outcomes	0 / 0
24-72h	614 / 514	42.6	42.7 / 22.9 (n=512)	349 / 512
>72h	1073 / 898	63.8	64.0 / 42.3 (n=895)	615 / 895

River breakdown

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
ashtabula-river	60 / 0	Pending	No paired outcomes	0 / 0
rocky-river	60 / 60	16.0	16.0 / 4.9 (n=60)	48 / 60
conneaut-creek	60 / 60	4.2	4.2 / 3.9 (n=60)	32 / 60
chagrin-river	60 / 60	9.5	9.5 / 7.4 (n=60)	33 / 60
vermilion-river	60 / 60	5.9	6.4 / 5.0 (n=54)	35 / 54
grand-river	60 / 59	4.2	4.2 / 3.3 (n=59)	36 / 59
cuyahoga-river	60 / 59	17.2	17.2 / 27.2 (n=59)	16 / 59
brandy-run	60 / 60	0.2	0.2 / 0.2 (n=60)	52 / 60
elk-creek	60 / 0	Pending	No paired outcomes	0 / 0
walnut-creek	60 / 60	1.3	1.3 / 0.8 (n=60)	37 / 60
twenty-mile-creek	60 / 0	Pending	No paired outcomes	0 / 0
crooked-creek	60 / 0	Pending	No paired outcomes	0 / 0
cattaraugus-creek	60 / 59	10.4	10.4 / 10.9 (n=59)	23 / 59
chautauqua-creek	60 / 60	1.1	1.1 / 0.6 (n=60)	38 / 60
eighteen-mile-creek	60 / 60	1.2	1.2 / 0.8 (n=60)	41 / 60
huron-river	60 / 60	14.5	14.5 / 10.7 (n=60)	37 / 60
black-river	60 / 60	17.8	17.8 / 17.1 (n=60)	42 / 60
euclid-creek	60 / 60	2.4	2.4 / 1.3 (n=60)	47 / 60
canadaway-creek	60 / 60	1.0	1.0 / 0.8 (n=60)	20 / 60
silver-creek	60 / 60	0.4	0.4 / 0.2 (n=60)	51 / 60
buffalo-creek	60 / 60	8.0	8.0 / 3.8 (n=60)	47 / 60
cazenovia-creek	60 / 60	15.0	15.0 / 4.2 (n=60)	40 / 60
salmon-river	60 / 60	7.1	7.1 / 3.6 (n=60)	46 / 60
oak-orchard-creek	60 / 60	571.0	571.0 / 282.8 (n=60)	45 / 60
genesee-river	60 / 60	200.4	200.4 / 160.2 (n=60)	46 / 60
sandy-creek-hamlin	60 / 60	27.6	27.6 / 10.9 (n=60)	46 / 60
irondequoit-creek	60 / 60	33.6	33.6 / 12.2 (n=60)	51 / 60
johnson-creek	60 / 0	Pending	No paired outcomes	0 / 0
eighteenmile-creek-niagara	60 / 60	36.7	36.7 / 14.8 (n=60)	48 / 60
black-river-watertown	60 / 60	279.1	279.1 / 226.4 (n=60)	45 / 60
sandy-creek-adams	60 / 60	84.6	84.6 / 57.3 (n=60)	49 / 60

Modeled storm state when issued (not observed storm truth)

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
quiet	1704 / 1431	51.1	51.3 / 32.8 (n=1425)	945 / 1425
recovering	72 / 48	66.0	66.0 / 32.3 (n=48)	36 / 48
watching	72 / 66	68.7	68.7 / 54.2 (n=66)	59 / 66
active	12 / 12	111.0	111.0 / 23.9 (n=12)	11 / 12

Model version 395a5868624c

1860 cases. Versions are reported separately. CFS errors across different rivers are not directly comparable; use the river breakdown too.

Elapsed forecast lead

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
0-24h	214 / 184	42.6	42.8 / 29.8 (n=183)	139 / 183
24-72h	604 / 504	86.3	86.7 / 61.1 (n=502)	374 / 502
>72h	1042 / 872	187.3	187.9 / 176.1 (n=869)	497 / 869

River breakdown

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
ashtabula-river	60 / 0	Pending	No paired outcomes	0 / 0
rocky-river	60 / 60	47.7	47.7 / 39.2 (n=60)	33 / 60
conneaut-creek	60 / 60	3.8	3.8 / 2.8 (n=60)	45 / 60
chagrin-river	60 / 60	72.8	72.8 / 71.7 (n=60)	24 / 60
vermilion-river	60 / 60	45.8	45.8 / 34.1 (n=60)	31 / 60
grand-river	60 / 60	14.5	14.5 / 5.4 (n=60)	52 / 60
cuyahoga-river	60 / 60	217.3	217.3 / 105.1 (n=60)	39 / 60
brandy-run	60 / 60	0.5	0.5 / 0.5 (n=60)	36 / 60
elk-creek	60 / 0	Pending	No paired outcomes	0 / 0
walnut-creek	60 / 60	4.1	4.1 / 3.1 (n=60)	42 / 60
twenty-mile-creek	60 / 0	Pending	No paired outcomes	0 / 0
crooked-creek	60 / 0	Pending	No paired outcomes	0 / 0
cattaraugus-creek	60 / 60	63.4	63.4 / 37.2 (n=60)	50 / 60
chautauqua-creek	60 / 60	2.9	2.9 / 1.9 (n=60)	35 / 60
eighteen-mile-creek	60 / 60	7.4	7.4 / 6.7 (n=60)	33 / 60
huron-river	60 / 60	93.5	93.5 / 27.4 (n=60)	53 / 60
black-river	60 / 60	80.1	80.1 / 57.7 (n=60)	30 / 60
euclid-creek	60 / 60	25.4	25.4 / 24.2 (n=60)	46 / 60
canadaway-creek	60 / 60	3.7	3.7 / 2.5 (n=60)	39 / 60
silver-creek	60 / 60	1.0	1.0 / 0.9 (n=60)	28 / 60
buffalo-creek	60 / 60	25.9	25.9 / 20.8 (n=60)	42 / 60
cazenovia-creek	60 / 60	29.6	31.8 / 28.1 (n=54)	41 / 54
salmon-river	60 / 60	13.8	13.8 / 10.1 (n=60)	41 / 60
oak-orchard-creek	60 / 60	347.6	347.6 / 265.0 (n=60)	35 / 60
genesee-river	60 / 60	517.2	517.2 / 295.1 (n=60)	45 / 60
sandy-creek-hamlin	60 / 60	15.5	15.5 / 9.7 (n=60)	41 / 60
irondequoit-creek	60 / 60	24.2	24.2 / 18.1 (n=60)	44 / 60
johnson-creek	60 / 0	Pending	No paired outcomes	0 / 0
eighteenmile-creek-niagara	60 / 60	32.9	32.9 / 19.1 (n=60)	36 / 60
black-river-watertown	60 / 60	1,585.1	1,585.1 / 1,809.1 (n=60)	30 / 60
sandy-creek-adams	60 / 60	302.1	302.1 / 259.3 (n=60)	39 / 60

Modeled storm state when issued (not observed storm truth)

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
quiet	1200 / 1002	110.7	110.7 / 84.0 (n=1002)	654 / 1002
recovering	102 / 90	345.0	345.0 / 372.4 (n=90)	59 / 90
watching	528 / 438	146.3	148.2 / 152.3 (n=432)	276 / 432
active	30 / 30	286.0	286.0 / 187.2 (n=30)	21 / 30

Model version c2a5feaf2ad6

186 cases. Versions are reported separately. CFS errors across different rivers are not directly comparable; use the river breakdown too.

Elapsed forecast lead

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
0-24h	31 / 26	262.0	262.0 / 169.3 (n=26)	19 / 26
24-72h	62 / 52	233.2	233.2 / 225.2 (n=52)	30 / 52
>72h	93 / 78	355.4	355.4 / 249.7 (n=78)	41 / 78

River breakdown

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
ashtabula-river	6 / 0	Pending	No paired outcomes	0 / 0
rocky-river	6 / 6	79.3	79.3 / 55.6 (n=6)	2 / 6
conneaut-creek	6 / 6	8.0	8.0 / 7.3 (n=6)	4 / 6
chagrin-river	6 / 6	95.8	95.8 / 108.7 (n=6)	2 / 6
vermilion-river	6 / 6	52.6	52.6 / 49.0 (n=6)	6 / 6
grand-river	6 / 6	11.9	11.9 / 10.1 (n=6)	4 / 6
cuyahoga-river	6 / 6	224.4	224.4 / 195.7 (n=6)	4 / 6
brandy-run	6 / 6	0.7	0.7 / 0.6 (n=6)	5 / 6
elk-creek	6 / 0	Pending	No paired outcomes	0 / 0
walnut-creek	6 / 6	4.6	4.6 / 4.1 (n=6)	3 / 6
twenty-mile-creek	6 / 0	Pending	No paired outcomes	0 / 0
crooked-creek	6 / 0	Pending	No paired outcomes	0 / 0
cattaraugus-creek	6 / 6	72.4	72.4 / 60.8 (n=6)	3 / 6
chautauqua-creek	6 / 6	3.6	3.6 / 3.2 (n=6)	1 / 6
eighteen-mile-creek	6 / 6	6.4	6.4 / 5.9 (n=6)	3 / 6
huron-river	6 / 6	60.5	60.5 / 49.3 (n=6)	6 / 6
black-river	6 / 6	42.3	42.3 / 74.7 (n=6)	1 / 6
euclid-creek	6 / 6	27.6	27.6 / 26.3 (n=6)	4 / 6
canadaway-creek	6 / 6	4.2	4.2 / 3.4 (n=6)	5 / 6
silver-creek	6 / 6	1.4	1.4 / 1.3 (n=6)	4 / 6
buffalo-creek	6 / 6	9.6	9.6 / 14.1 (n=6)	3 / 6
cazenovia-creek	6 / 6	9.7	9.7 / 12.6 (n=6)	4 / 6
salmon-river	6 / 6	33.1	33.1 / 10.7 (n=6)	4 / 6
oak-orchard-creek	6 / 6	150.3	150.3 / 150.3 (n=6)	0 / 6
genesee-river	6 / 6	752.5	752.5 / 448.8 (n=6)	6 / 6
sandy-creek-hamlin	6 / 6	15.1	15.1 / 10.8 (n=6)	5 / 6
irondequoit-creek	6 / 6	30.6	30.6 / 28.7 (n=6)	4 / 6
johnson-creek	6 / 0	Pending	No paired outcomes	0 / 0
eighteenmile-creek-niagara	6 / 6	12.6	12.6 / 25.8 (n=6)	1 / 6
black-river-watertown	6 / 6	747.4	747.4 / 2,886.7 (n=6)	0 / 6
sandy-creek-adams	6 / 6	5,320.3	5,320.3 / 1,687.5 (n=6)	6 / 6

Modeled storm state when issued (not observed storm truth)

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
quiet	108 / 84	42.4	42.4 / 38.0 (n=84)	50 / 84
watching	66 / 60	182.1	182.1 / 363.8 (n=60)	33 / 60
active	12 / 12	2,681.3	2,681.3 / 881.1 (n=12)	7 / 12

Model version 467e92d71b50

372 cases. Versions are reported separately. CFS errors across different rivers are not directly comparable; use the river breakdown too.

Elapsed forecast lead

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
0-24h	31 / 26	262.0	262.0 / 169.3 (n=26)	19 / 26
24-72h	124 / 104	137.4	139.2 / 206.0 (n=102)	55 / 102
>72h	217 / 182	231.0	234.7 / 217.6 (n=178)	96 / 178

River breakdown

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
ashtabula-river	12 / 0	Pending	No paired outcomes	0 / 0
rocky-river	12 / 12	57.2	55.5 / 55.6 (n=6)	0 / 6
conneaut-creek	12 / 12	9.3	9.3 / 8.3 (n=12)	10 / 12
chagrin-river	12 / 12	109.7	109.7 / 110.3 (n=12)	2 / 12
vermilion-river	12 / 12	59.5	59.5 / 52.9 (n=12)	11 / 12
grand-river	12 / 12	12.3	12.3 / 10.5 (n=12)	7 / 12
cuyahoga-river	12 / 12	227.4	227.4 / 184.9 (n=12)	10 / 12
brandy-run	12 / 12	0.7	0.7 / 0.5 (n=12)	11 / 12
elk-creek	12 / 0	Pending	No paired outcomes	0 / 0
walnut-creek	12 / 12	4.5	4.5 / 4.1 (n=12)	6 / 12
twenty-mile-creek	12 / 0	Pending	No paired outcomes	0 / 0
crooked-creek	12 / 0	Pending	No paired outcomes	0 / 0
cattaraugus-creek	12 / 12	76.9	76.9 / 63.5 (n=12)	9 / 12
chautauqua-creek	12 / 12	4.5	4.5 / 3.6 (n=12)	5 / 12
eighteen-mile-creek	12 / 12	6.7	6.7 / 6.0 (n=12)	6 / 12
huron-river	12 / 12	58.3	58.3 / 47.9 (n=12)	10 / 12
black-river	12 / 12	82.4	82.4 / 77.9 (n=12)	2 / 12
euclid-creek	12 / 12	28.2	28.2 / 26.7 (n=12)	11 / 12
canadaway-creek	12 / 12	4.4	4.4 / 3.4 (n=12)	11 / 12
silver-creek	12 / 12	1.4	1.4 / 1.3 (n=12)	8 / 12
buffalo-creek	12 / 12	15.1	15.1 / 14.5 (n=12)	7 / 12
cazenovia-creek	12 / 12	15.8	15.8 / 13.2 (n=12)	8 / 12
salmon-river	12 / 12	15.8	15.8 / 15.0 (n=12)	2 / 12
oak-orchard-creek	12 / 12	121.7	121.7 / 121.7 (n=12)	0 / 12
genesee-river	12 / 12	658.9	658.9 / 348.7 (n=12)	12 / 12
sandy-creek-hamlin	12 / 12	19.8	19.8 / 12.3 (n=12)	11 / 12
irondequoit-creek	12 / 12	29.3	29.3 / 28.9 (n=12)	1 / 12
johnson-creek	12 / 0	Pending	No paired outcomes	0 / 0
eighteenmile-creek-niagara	12 / 12	32.9	32.9 / 30.0 (n=12)	7 / 12
black-river-watertown	12 / 12	1,705.8	1,705.8 / 2,737.5 (n=12)	0 / 12
sandy-creek-adams	12 / 12	1,903.6	1,903.6 / 1,394.0 (n=12)	3 / 12

Modeled storm state when issued (not observed storm truth)

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
quiet	210 / 162	42.1	41.5 / 36.6 (n=156)	102 / 156
watching	120 / 108	265.9	265.9 / 358.0 (n=108)	48 / 108
active	30 / 30	650.6	650.6 / 422.8 (n=30)	18 / 30
recovering	12 / 12	673.8	673.8 / 590.8 (n=12)	2 / 12

Model version ae671bd0df74

2790 cases. Versions are reported separately. CFS errors across different rivers are not directly comparable; use the river breakdown too.

Elapsed forecast lead

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
24-72h	930 / 774	87.1	87.1 / 87.3 (n=774)	294 / 774
>72h	1643 / 1366	96.4	96.4 / 113.0 (n=1366)	452 / 1366
0-24h	217 / 182	85.1	85.1 / 56.1 (n=182)	91 / 182

River breakdown

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
ashtabula-river	90 / 0	Pending	No paired outcomes	0 / 0
rocky-river	90 / 84	58.8	58.8 / 68.6 (n=84)	14 / 84
conneaut-creek	90 / 90	8.8	8.8 / 9.1 (n=90)	19 / 90
chagrin-river	90 / 90	97.6	97.6 / 118.8 (n=90)	18 / 90
vermilion-river	90 / 84	44.8	44.8 / 55.0 (n=84)	15 / 84
grand-river	90 / 90	11.0	11.0 / 12.4 (n=90)	33 / 90
cuyahoga-river	90 / 90	96.2	96.2 / 147.0 (n=90)	31 / 90
brandy-run	90 / 90	0.7	0.7 / 0.5 (n=90)	68 / 90
elk-creek	90 / 0	Pending	No paired outcomes	0 / 0
walnut-creek	90 / 90	5.9	5.9 / 7.0 (n=90)	62 / 90
twenty-mile-creek	90 / 0	Pending	No paired outcomes	0 / 0
crooked-creek	90 / 0	Pending	No paired outcomes	0 / 0
cattaraugus-creek	90 / 90	85.7	85.7 / 71.0 (n=90)	28 / 90
chautauqua-creek	90 / 90	4.8	4.8 / 3.0 (n=90)	58 / 90
eighteen-mile-creek	90 / 90	6.7	6.7 / 7.7 (n=90)	20 / 90
huron-river	90 / 90	30.7	30.7 / 36.0 (n=90)	31 / 90
black-river	90 / 90	66.3	66.3 / 75.4 (n=90)	23 / 90
euclid-creek	90 / 90	10.4	10.4 / 15.1 (n=90)	44 / 90
canadaway-creek	90 / 90	6.3	6.3 / 5.4 (n=90)	46 / 90
silver-creek	90 / 90	1.4	1.4 / 1.3 (n=90)	36 / 90
buffalo-creek	90 / 90	20.5	20.5 / 28.0 (n=90)	22 / 90
cazenovia-creek	90 / 90	17.2	17.2 / 19.7 (n=90)	26 / 90
salmon-river	90 / 90	31.3	31.3 / 21.0 (n=90)	52 / 90
oak-orchard-creek	90 / 84	131.9	131.9 / 131.9 (n=84)	0 / 84
genesee-river	90 / 90	357.2	357.2 / 161.8 (n=90)	54 / 90
sandy-creek-hamlin	90 / 90	13.6	13.6 / 12.1 (n=90)	45 / 90
irondequoit-creek	90 / 90	37.9	37.9 / 30.8 (n=90)	43 / 90
johnson-creek	90 / 0	Pending	No paired outcomes	0 / 0
eighteenmile-creek-niagara	90 / 90	14.2	14.2 / 15.5 (n=90)	17 / 90
black-river-watertown	90 / 90	1,045.2	1,045.2 / 1,312.4 (n=90)	17 / 90
sandy-creek-adams	90 / 90	194.6	194.6 / 229.6 (n=90)	15 / 90

Modeled storm state when issued (not observed storm truth)

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
quiet	1902 / 1560	69.7	69.7 / 86.6 (n=1560)	531 / 1560
active	132 / 108	134.5	134.5 / 96.7 (n=108)	26 / 108
watching	450 / 390	80.1	80.1 / 76.4 (n=390)	216 / 390
recovering	306 / 264	227.5	227.5 / 215.0 (n=264)	64 / 264

Model version 48521c910d45

2976 cases. Versions are reported separately. CFS errors across different rivers are not directly comparable; use the river breakdown too.

Elapsed forecast lead

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
0-24h	248 / 216	26.4	26.4 / 26.4 (n=216)	126 / 216
24-72h	992 / 858	25.4	25.4 / 28.2 (n=858)	390 / 858
>72h	1736 / 1393	34.7	34.7 / 40.3 (n=1393)	497 / 1393

River breakdown

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
ashtabula-river	96 / 0	Pending	No paired outcomes	0 / 0
rocky-river	96 / 92	3.3	3.3 / 2.7 (n=92)	43 / 92
conneaut-creek	96 / 92	2.1	2.1 / 1.7 (n=92)	49 / 92
chagrin-river	96 / 92	4.4	4.4 / 4.0 (n=92)	38 / 92
vermilion-river	96 / 92	1.8	1.8 / 6.9 (n=92)	5 / 92
grand-river	96 / 92	12.4	12.4 / 14.9 (n=92)	29 / 92
cuyahoga-river	96 / 92	56.7	56.7 / 36.7 (n=92)	66 / 92
brandy-run	96 / 92	0.1	0.1 / 0.0 (n=92)	48 / 92
elk-creek	96 / 0	Pending	No paired outcomes	0 / 0
walnut-creek	96 / 81	1.9	1.9 / 0.7 (n=81)	57 / 81
twenty-mile-creek	96 / 0	Pending	No paired outcomes	0 / 0
crooked-creek	96 / 0	Pending	No paired outcomes	0 / 0
cattaraugus-creek	96 / 92	6.4	6.4 / 7.3 (n=92)	28 / 92
chautauqua-creek	96 / 92	1.2	1.2 / 0.8 (n=92)	80 / 92
eighteen-mile-creek	96 / 92	1.6	1.6 / 1.1 (n=92)	67 / 92
huron-river	96 / 92	3.0	3.0 / 4.2 (n=92)	27 / 92
black-river	96 / 92	2.4	2.4 / 8.1 (n=92)	16 / 92
euclid-creek	96 / 86	2.1	2.1 / 1.3 (n=86)	52 / 86
canadaway-creek	96 / 92	0.4	0.4 / 0.3 (n=92)	40 / 92
silver-creek	96 / 92	0.6	0.6 / 0.6 (n=92)	39 / 92
buffalo-creek	96 / 92	3.2	3.2 / 1.9 (n=92)	63 / 92
cazenovia-creek	96 / 92	2.6	2.6 / 1.3 (n=92)	63 / 92
salmon-river	96 / 92	10.2	10.2 / 9.3 (n=92)	29 / 92
oak-orchard-creek	96 / 92	160.6	160.6 / 160.6 (n=92)	0 / 92
genesee-river	96 / 92	245.8	245.8 / 200.2 (n=92)	37 / 92
sandy-creek-hamlin	96 / 92	8.9	8.9 / 8.3 (n=92)	34 / 92
irondequoit-creek	96 / 92	2.7	2.7 / 2.3 (n=92)	43 / 92
johnson-creek	96 / 92	19.7	19.7 / 17.3 (n=92)	29 / 92
eighteenmile-creek-niagara	96 / 92	9.5	9.5 / 13.2 (n=92)	0 / 92
black-river-watertown	96 / 92	255.7	255.7 / 410.3 (n=92)	18 / 92
sandy-creek-adams	96 / 92	5.6	5.6 / 18.8 (n=92)	13 / 92

Modeled storm state when issued (not observed storm truth)

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
quiet	2976 / 2467	30.7	30.7 / 34.9 (n=2467)	1013 / 2467

Model version b3f961a97961

2418 cases. Versions are reported separately. CFS errors across different rivers are not directly comparable; use the river breakdown too.

Elapsed forecast lead

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
24-72h	806 / 319	26.9	26.9 / 25.1 (n=319)	153 / 319
>72h	1426 / 108	21.7	21.7 / 21.7 (n=108)	44 / 108
0-24h	186 / 106	24.7	24.7 / 26.7 (n=106)	66 / 106

River breakdown

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
ashtabula-river	78 / 0	Pending	No paired outcomes	0 / 0
rocky-river	78 / 20	1.0	1.0 / 0.7 (n=20)	5 / 20
conneaut-creek	78 / 20	1.6	1.6 / 1.0 (n=20)	8 / 20
chagrin-river	78 / 20	3.8	3.8 / 2.3 (n=20)	9 / 20
vermilion-river	78 / 20	1.5	1.5 / 1.6 (n=20)	5 / 20
grand-river	78 / 20	6.0	6.0 / 4.0 (n=20)	9 / 20
cuyahoga-river	78 / 20	64.9	64.9 / 63.7 (n=20)	7 / 20
brandy-run	78 / 20	0.1	0.1 / 0.1 (n=20)	9 / 20
elk-creek	78 / 0	Pending	No paired outcomes	0 / 0
walnut-creek	78 / 13	1.4	1.4 / 1.0 (n=13)	7 / 13
twenty-mile-creek	78 / 0	Pending	No paired outcomes	0 / 0
crooked-creek	78 / 0	Pending	No paired outcomes	0 / 0
cattaraugus-creek	78 / 20	12.4	12.4 / 5.6 (n=20)	18 / 20
chautauqua-creek	78 / 20	0.7	0.7 / 0.4 (n=20)	16 / 20
eighteen-mile-creek	78 / 20	1.2	1.2 / 1.2 (n=20)	11 / 20
huron-river	78 / 20	2.7	2.7 / 1.1 (n=20)	16 / 20
black-river	78 / 20	1.7	1.7 / 1.1 (n=20)	10 / 20
euclid-creek	78 / 20	3.6	3.6 / 2.9 (n=20)	15 / 20
canadaway-creek	78 / 20	0.2	0.2 / 0.3 (n=20)	8 / 20
silver-creek	78 / 20	0.4	0.4 / 0.3 (n=20)	18 / 20
buffalo-creek	78 / 20	4.9	4.9 / 2.3 (n=20)	13 / 20
cazenovia-creek	78 / 20	4.0	4.0 / 1.8 (n=20)	16 / 20
salmon-river	78 / 20	3.3	3.3 / 2.8 (n=20)	5 / 20
oak-orchard-creek	78 / 20	129.3	129.3 / 129.3 (n=20)	0 / 20
genesee-river	78 / 20	238.4	238.4 / 210.5 (n=20)	7 / 20
sandy-creek-hamlin	78 / 20	13.1	13.1 / 12.3 (n=20)	7 / 20
irondequoit-creek	78 / 20	5.8	5.8 / 5.1 (n=20)	12 / 20
johnson-creek	78 / 20	16.0	16.0 / 9.5 (n=20)	11 / 20
eighteenmile-creek-niagara	78 / 20	7.2	7.2 / 8.9 (n=20)	6 / 20
black-river-watertown	78 / 20	142.9	142.9 / 180.0 (n=20)	8 / 20
sandy-creek-adams	78 / 20	9.7	9.7 / 8.8 (n=20)	7 / 20

Modeled storm state when issued (not observed storm truth)

Group	Cases / scored	Mean absolute error, CFS	Paired model / unchanged-flow error, CFS	Worse than unchanged flow
quiet	2274 / 525	25.8	25.8 / 25.0 (n=525)	262 / 525
recovering	12 / 6	0.9	0.9 / 1.3 (n=6)	0 / 6
active	6 / 2	7.2	7.2 / 3.3 (n=2)	1 / 2
watching	126 / 0	Pending	No paired outcomes	0 / 0

Gauge values are provisional as stored when evaluated; corrections made before evaluation can be included. Newly inserted readings after the arrival deadline are excluded. Once evaluated, results and missing outcomes are frozen. Neighboring rivers and repeated forecasts share weather and are not independent trials. Small samples do not establish forecast skill, and no score here automatically enables stronger fishing claims.

Frequently Asked Questions

How often does the projection update?

Projections are updated every 5 minutes server-side through cache invalidation, utilizing fresh feeds from NWS AHPS forecast points, rain gauges, and NWS QPF predictions.

What's the difference between stable and low?

stable means flow is currently inside the optimal window and projected to stay there for 18 hours. low means flow is currently below the optimal window and projected to stay below for 18 hours. Both describe steady projections; the difference is whether the steady level is inside or below the optimal flow window. For trip advice, combine timing.state with the river's current rating and is_fishable — Timing answers what flow will do, not whether you should fish.

What's the difference between Timing and the current Rating?

Current Rating tells you what the river is doing right now (e.g. Excellent, Good, Fair, Poor). The Timing projection looks ahead over the next 18 hours to tell you how that window is changing.

What if confidence is low?

If the timing confidence is low (below 0.4), treat the timing trend cautiously and check the observation time, current conditions and local access before planning.

Why doesn't the projection go beyond 18 hours?

Hydrologic modeling confidence degrades rapidly past the 18-hour mark for small, steep tributaries. For multi-day planning, use our 7-day Index Forecast instead.