Skip to main content
πŸ“‘ Offline β€” showing last-known data

🌊 Potomac Pulse

This app tracks water moving through the Potomac basin β€” every gauge, with travel times to Little Falls. And since Great Falls has no gauge, it estimates that too.
Built by Gordon Shumway.
⚠️ Unable to fetch data. Showing cached data.
🎯
Little Falls
--
cfs
--
ft
Point of Rocks to Little Falls: -- Great Falls to Little Falls: --
🌊 Great Falls
🏞️ Creeks
πŸ“Š All Gauges
πŸ“– How It Works
πŸ”’ Learning
Loading creek data...
⚠ Thresholds are approximate β€” always scout unfamiliar runs.
ESTIMATED CONDITIONS AT
Great Falls
ESTIMATED NOW
--
cfs
--
ft
on the Little Falls gauge scale
--
Confidence: --
24H HISTORY + 48H FORECAST*
*Experimental β€” collecting data to validate accuracy
ft
cfs
β–²β–Ό vs now β€’ tap card or graph for details
πŸ“₯ Estimation Inputs β–Έ
Point of Rocks (83.5%) -- cfs
+ Monocacy (7.1%) + -- cfs
+ Goose Creek (3.0%) + -- cfs
+ Broad Run (0.7%) + -- cfs
+ Seneca (0.9%) + -- cfs
Travel time, Point of Rocks to Great Falls -- hrs
Travel time, Great Falls to Little Falls -- hrs
Flow band --
Adjustment learned from past errors Β± -- cfs
90% CI --
Display cap (120% of Little Falls) --
Edwards Ferry check --
Estimate from the Edwards Ferry gauge alone

There is no gauge at Great Falls. Potomac Pulse estimates what is happening there right now and what is coming, so you can plan your time on the water.

1. The Nowcast

What is happening at Great Falls right now

Water takes roughly a day to flow from Point of Rocks to Great Falls. To estimate conditions at Great Falls right now, we look at what Point of Rocks was reading roughly 5 to 50 hours ago (depending on how fast the river is moving β€” about 19 hours at typical flows, much longer in low water), then adjust for everything that happens along the way.

Point of Rocks
~34 mi upstream · flow + level
↓
~5–50 hrs travel time
+ tributary inflows along the way
Monocacy 7.1% · Goose Creek 3.0% · Broad Run 0.7% · Seneca 0.9%
↓
Blended with Edwards Ferry
~16 mi upstream · level only · weight 0–40% by flow
↓
GREAT FALLS
Estimated CFS & stage
↓
~6 hrs later
Little Falls
Checks the prediction
Time-shifting
The primary input is the Point of Rocks gauge, ~34 river miles upstream. Travel time depends on flow: roughly 5 hours at very high water and about 19 hours at typical flows, lengthening to ~50 hours in low water. We look back by that amount to find the reading that corresponds to Great Falls right now. If the river is currently rising or falling at Point of Rocks, the estimate adjusts for how much of that change has reached Great Falls.
Tributaries and blending
Four tributaries enter the Potomac between Point of Rocks and Great Falls. Their current flows are added to the time-shifted estimate. The result is then blended with an independent estimate from the Edwards Ferry gauge, ~16 river miles above the falls. Edwards Ferry measures water level only, so a calibrated power-law model converts its stage reading to flow. At low flows the blend gives Edwards Ferry almost no weight; at high flows it contributes up to 40%.
Self-correcting
The server checks one earlier prediction at a time against what Little Falls actually recorded when that water arrived. A new prediction is stored only after the previous one has been checked, so most hourly predictions are never scored. The error is stored by flow level and flow direction (rising, falling, or steady), and a correction factor is updated so the next prediction in similar conditions is closer. The model learns from most checks, even when no one is looking at the site, but it can lag a fast-changing river. The estimate is also capped at 120% of the latest Little Falls reading, which can be over an hour old. Rain that falls between Point of Rocks and Great Falls can push the real flow above the capped number. When the cap holds, the Great Falls card says so and shows the model's own value. Comparisons where the gauge reading itself looks suspect (for example ice or an inconsistent stage reading) are quarantined: they don't update the correction factors or the accuracy average, but they still appear on the accuracy chart as hollow rings, so unusual events stay visible. An error far outside the model's normal range — typical during sudden local storms — is no longer quarantined (v39.0): it is learned through a clamp that limits how far one such miss can move the correction, and it counts in the accuracy average.
READING YOUR SCREEN
CFS — cubic feet per second (volume of water flowing)
Stage — water height in feet on the Little Falls gauge scale
Confidence — High / Medium / Low is the label shown with the estimate. It mostly reflects how complete the inputs are (tributary gauges reporting, upstream timing, Edwards Ferry agreeing). It also drops when the latest check against Little Falls was far off. It is not a measured error rate. In one event, the October 2026 flood, the typical (median) miss was about 9% at every level, and a few High readings were off by more than 25%.

2. The 48-Hour Forecast

What is coming to Great Falls

The forecast uses a different method from the nowcast. The National Weather Service predicts what will arrive at Little Falls, the downstream gauge. Since water passes Great Falls before it reaches Little Falls, we shift the NWS prediction earlier in time by the travel time between the two gauges (roughly 6 hours at typical flows, but under 2 hours in a flood and up to 17 hours in very low water). If NWS says Little Falls will rise at 6 PM, Great Falls should rise around noon.

Be honest about what this is. Because the forecast is a time-shift of the NWS Little Falls prediction, it is not an independent forecast β€” it is the NWS forecast re-expressed for Great Falls, nudged by the gap between what NWS currently predicts for Little Falls and what Little Falls actually reads. So we no longer show a "vs NWS" score: on the same clock, that comparison is the forecast against itself. We compare it instead against persistence β€” simply assuming the river holds its current flow. Beating persistence is the real test of whether the forecast is telling you anything.

Each forecast is scored when the water it describes actually reaches Little Falls β€” one travel time after the hour it was predicting for. Before v37.16 we scored it at the predicted hour itself, which compared the forecast against the river before that water had arrived. Errors under steady flow were small, but the mistake bit hardest on rising and falling rivers, which is exactly when a forecast matters.

NWS forecasts can drift from actual gauge readings, so we apply an additive bias correction: the difference between what NWS currently predicts for Little Falls and what the gauge currently reads. This offset is recalculated every 15 minutes and automatically shrinks as NWS updates its models.

FORECAST RELIABILITY
6 hr
Good
12 hr
Fair
24 hr
Rough
48 hr
Directional
This is an estimate, not a measurement
Do not use it for life-safety decisions. Localized storms between gauges will not show until that water arrives, and fast-rising floods may arrive sooner than predicted. Always check NWS warnings independently.
Travel times by flow level
FlowSpeedPoR → Little Falls
Very low (~1.2k cfs)~0.7 mph~60 hrs
Low (~2k cfs)~0.9 mph~45 hrs
Typical (~5k cfs)~1.6 mph~26 hrs
High (~15k cfs)~3 mph~14 hrs
Flood (~50k+ cfs)~6 mph~7 hrs
Based on USGS Circular 438 (Searcy, 1961), updated with modern 15-minute gauge data.
Data: USGS Water Services (15-min updates) · NWS river forecasts · USGS water temperature at Point of Rocks

Full methodology, equations, validation data, and references

πŸ”’ Protected Section

The adaptive learning system is restricted to the model administrator.
Enter PIN to access learning data and corrections.



Incorrect PIN

ADMIN DASHBOARD
LF ACTUAL
--
-- ft
GF ESTIMATE
--
-- ft
PoR NOW
--
--
EF Stage
--
Monocacy
--
Goose Ck
--
Travel
--
GF Validations
--
Validation Err*
--
Flagged
--
Sched Runs
--
*Validation error is approximate β€” compares GF predictions from ~7h ago against LF readings taken 2-4h late. Feeds EMA corrections but overstates true model error. See forecast accuracy for clean metrics.
❄️ ICE-AFFECTED GAUGES
--
Last fetch: --

πŸ”„ Validation Status

Next Validation --
Last Validation No validations yet

Scheduled Function Health

Last Run --
Consecutive Runs --
Missed Runs (total) --
Pending Predictions --

πŸ”§ System Diagnostics

Throughput --
Stage Error --
Bin Writes --

Stage Skips --
Last Flag --

EF Fit --

🧾 Recent Admin Actions

Loading...

GF Flow-Bin Corrections (System 2)

Loading bin statistics...

Clears flow-bin corrections.

Prediction Accuracy (7d)

Collecting validation data...
Predicted:
Actual:
Error:
● Predicted LF   ● Actual LF   β—‹ Hard-flagged (excluded from avg)

πŸ‡ Shadow Model Horse Race

Three alternative GF estimation approaches running alongside production. Display-only β€” never affects the main estimate.

🟒 PRODUCTION
--
-- ft
πŸ”Ά LF FEEDBACK
--
-- ft
--
πŸ”· ONLINE REGRESSION
--
-- ft
--
🟣 KALMAN FILTER
--
-- ft
--
LEADERBOARD (awaiting validated predictions)
πŸ“Š Leaderboard will populate after Phase 2 (server-side validation)
πŸ”§ Model Diagnostics
LF Feedback
Correction: 0.0% | Pending prediction: none
Online Regression
Weights: not initialized | Train count: 0
Kalman Filter
State: not initialized | P: --

Clears all shadow model learned state. Models restart from scratch.

Exports current GF correction bins + learning metadata + forecast accuracy as a JSON file. Read-only β€” changes nothing.

GS Β© 2026 Gordon Shumway β€’ Licensed under CC BY-NC 4.0 β€’ Open Source β€’ v39.13