Experimental River Flow Forecasts for Pagosa Springs

The Black Hat Hydrology Lab is Pagosa Outside’s public-facing river-forecasting experiment. It combines historical San Juan River flows, headwater snowpack, current conditions and weather information to make longer-range flow estimates and seasonal predictions. It is part river science, part local experience and part ongoing attempt to make better educated guesses than we used to make over coffee.

The name is a wink, not a confession. In our version, “black hat” means taking public river, snowpack, weather and climate information traditionally scattered across separate sources, hacking it together in unconventional ways and testing whether it can produce a better-informed home-field advantage.

We are not breaking into government computers or hiding the methods: every source is public, every official guess is locked and every miss stays on the scoreboard. The only thing we are trying to outsmart is Mother Nature—and she still has the administrator password.

The River Desk shows current measurements, published government forecasts, snowpack, weather and today’s river options. The Black Hat Lab goes beyond those published forecasts and experiments with what might happen 30 to 90 days from now or later in the river season.

The Lab’s predictions are independent Pagosa Outside experiments—not official government forecasts. They should be treated as entertainment rather than instructions for river use, travel or business decisions. Use the River Desk for current conditions and published short-term forecasts.

The Lab estimates San Juan River flow 30, 60 and 90 days ahead and displays an experimental 90-day hydrograph. It calculates the probability of reaching Pagosa’s general tubing and rafting-flow ranges, operates a Live Seasonal Watch and publishes eight fixed predictions about important events during the water year.

Those seasonal predictions cover headwater snowpack, the beginning and strength of spring runoff, peak flow, rafting-flow days, Tubing Season Opening Day and tubing-range days.

The honest answer is: promising, but still experimental. In historical testing, the seasonal prediction methods scored 73 hits in 96 development tests. A later five-season test—using methods selected without seeing those results—scored 30 hits in 40 predictions. That works out to roughly three correct calls out of four.

The 30-day flow model reduced its typical error by approximately 28% compared with using historical averages alone, while the 90-day model improved by approximately 11%. The experimental ranges contained the actual result about four times out of five during historical testing.

Accuracy still changes considerably by season, forecast distance and type of prediction. One unusual storm, heat wave or stubborn snowpack can embarrass the entire contraption. That is why the scoreboard displays every public hit, miss and pending result instead of showing only the guesses that make us look clever.

The Lab uses river flow from the U.S. Geological Survey gauge at Pagosa Springs; snowpack from the Upper San Juan and Wolf Creek Summit Snow Telemetry stations; temperature and precipitation forecasts from the National Weather Service; selected climate outlooks from the National Oceanic and Atmospheric Administration; and historical guidance from the Colorado Basin River Forecast Center.

Middle Creek and Lily Pond may be used as regional storm-pattern checks on predictions where historical testing showed an improvement. Other stations—including Beartown and Grayback—were tested but did not improve the Pagosa predictions enough to earn a regular seat at the workbench.

The Lab does not ask a chatbot to stare into the river and invent a number. Its predictions come from repeatable formulas, historical comparisons and backtested models. The calculations run automatically, but the targets, inputs, scoring rules and operating thresholds were developed around the actual questions Pagosa river managers and guides try to answer.

The Lab compares current river flow, recent movement, seasonal timing, headwater snowpack, temperature and precipitation with similar conditions from previous water years. It produces a center estimate and an experimental range for each future date. The historical average is displayed as a baseline, not as a second prediction.

The center estimate is the Lab’s single best guess. The experimental range shows a broader group of reasonably plausible outcomes and is calibrated to contain the actual result about four times out of five during historical testing. It is not a best-case and worst-case limit, and the river is under no contractual obligation to remain inside it.

Each horizon is tested separately. A 60-day range can occasionally be wider than a 90-day range when the 60-day date lands during a volatile runoff or seasonal transition and the 90-day date lands during historically steadier flow.

The probability table estimates the chance that flow on each forecast date will fall within Pagosa’s general operating ranges:

• 30–399 cubic feet per second: tubing range

• 300 cubic feet per second or higher: possible Splash & Dash rafting flow

• 600 cubic feet per second or higher: possible Mesa Canyon rafting flow

• 800 cubic feet per second or higher: possible West Fork rafting flow

These are flow probabilities—not promises that a particular trip will operate. Water temperature, weather, debris, access and current river conditions still influence Pagosa Outside’s daily lineup.

The Watch changes jobs as the water year progresses. During winter, Spring Runoff Watch follows headwater snowpack development and early runoff potential. As spring runoff begins rising, Predict the Peak Runoff Watch continuously updates the Lab's best guess for the peak date, Mountain Time and flow. After a sustained decline indicates that the spring crest has likely passed, it becomes Tubing Opening Day Watch and looks for the first complete 24 hours with every available Pagosa gauge reading from 30 through 399 cubic feet per second. After the tubing transition, Monsoon Flow Watch follows possible rain-driven tubing boosts, low-water recoveries and returns to rafting flow. After Labor Day, it becomes Bonus Flow Watch.

Unlike the eight official predictions, the Live Seasonal Watch updates whenever the Lab runs and new information becomes available. Its changing estimates never rewrite the fixed guesses on the scoreboard.

The eight official predictions are:

1. First persistent headwater snowpack

2. Peak headwater snowpack

3. First 400-plus cubic-feet-per-second day

4. May–June river strength

5. Spring peak date, time and flow

6. Rafting-flow days from May 1 through Labor Day

7. Tubing Season Opening Day

8. Tubing-flow days from May 1 through Labor Day

Each question uses its own combination of snowpack, river flow, weather, climate outlooks, seasonal timing and historical comparisons. Predicting peak snowpack is not the same problem as predicting spring peak flow or tubing days, so one magic formula would not be very magical.

Each official call also has a fixed issue date selected to balance useful lead time with historical accuracy. Once published, the guess is locked and cannot be quietly changed when better information arrives.

Every prediction has a measurable finish line and scoring rule established before the outcome is known. A hit earns one point for the Black Hat Lab. A miss earns one point for Mother Nature. Unfinished predictions remain pending.

The scoreboard follows the October 1 through September 30 water year and divides the season into four quarters. A new game begins each October, while the previous season’s record remains visible. The Lab does not get to quietly drag its misses behind the shed.

A retrospective season recreates earlier predictions using only information that would have been available on each stated issue date. The model is trained on previous years and is not allowed to peek at what happened later. Retrospective results are clearly labeled because they were historical tests—not predictions publicly posted at the time.

Sometimes. Archived National Oceanic and Atmospheric Administration one-month and three-month outlooks improved selected predictions, including the first 400-plus cubic-feet-per-second day, May–June river strength and spring peak flow.

Those outlooks did not improve every part of the Lab. They are excluded from the 30-, 60- and 90-day flow estimates, hydrograph, operating-threshold probabilities and Monsoon Flow Watch because historical testing found that the other methods performed as well or better. The Live Seasonal Watch still uses the National Weather Service’s seven-day headwater forecast for short-range weather and storm information.

More data does not automatically produce a better prediction, so every new ingredient has to earn its way into the formula.

Absolutely. This is a continuing public experiment, and the build number shows which version is running. New water years create new tests, misses expose weak formulas and better data sources may improve individual predictions.

A method changes only when backtesting shows a real improvement without peeking at the answer. Past guesses and results remain visible so a new formula cannot rewrite history or polish an old miss into a hit.

Because once we put the whole river story onto one desktop, somebody asked the dangerous question: What happens after the official forecast ends—and can we guess it better than usual?

The Lab combines 25 years of river and snowpack history with current conditions, weather forecasts, tested formulas and the PO River Crew’s accumulated hunches to make longer-range and seasonal predictions. It is where actual science, river-guide instinct and old-timer theories are forced to share one questionable workbench while Mother Nature controls the scoreboard.

We built it to find out whether our educated guesses can beat ordinary historical averages—and because arguing about peak runoff at the coffee shop was not producing enough charts.