How FriendOrFlow works
Methods, data sources, and limits.
Part 1. The short version
What this site tells you
FriendOrFlow answers one question for popular river trips: if I launch on this date, how likely is it that things work out? For each river and each launch date, it shows how often the river had a runnable flow, how often the weather was pleasant, and how crowded the permit lottery tends to be.
Every number comes from historical data. We look at what happened on the same calendar dates in past years and count how often each outcome occurred. That makes the results historical odds, not a forecast. They say what is typical and how much it varies, not what next summer will do.
Where the numbers come from
River flow comes from U.S. Geological Survey stream gauges, using records back to 1976. Some gauges are incomplete and these are noted in our summaries.
Flow levels (what counts as too low, too high, and just right) come from American Whitewater.
Weather comes from weather stations near each river: daily highs, lows, rain, and snow. These data were obtained from the National Centers for Environmental Information. For some rivers we created a blend of stations that we think better reflects the weather on the river. On the Middle Fork Salmon for example, instead of using only the weather from the Middle Fork Lodge, we created a blend of weather from Stanley, the Lodge, and Salmon that more closely tracks where a typical boater would be on each day of the trip.
Permit lottery demand comes from public recreation.gov historical reservation records (the RIDB data set) for the permit lotteries on each river.
El Niño and La Niña years are identified with NOAA's Oceanic Niño Index.
We did not collect any of this data ourselves. What we did was summarize it by launch date and trip length, and the summaries, charts, and text are original to this site.
How we handle a trip
A river trip lasts several days, so a launch date is not one day of weather or flow but a stretch of days. Each river uses a typical trip length and for each launch date we look at the whole trip--in every year on record--and ask what happened over those days. A trip counts as having flow in range only if every day stayed in range, and as having fair weather only if no day brought trouble.
What the charts mean
Chance is the share of past years in which something happened on a trip starting that date. A 70% chance of flow in range means about 7 of 10 past trips stayed in range.
Shaded bands show how sure we can be. The band is wider where few years are on record or the thing we’re summarizing is variable. Bands on flow and temperature charts show the middle 90% of years, so a typical year falls inside and an unusual one falls near the edge. With 50 years of data that means only the 5 most-extreme years are not shown.
Fair weather means a trip with none of four problems: rain, a hot day, a freezing night, or a miserable stretch. ‘Miserable’ means at least one cold and wet day (more than 0.05” rain and a high less than 55 or a low below 40), or rain on half or more of the trip days.
Flow and weather together multiplies the two chances to find the best launch dates. It treats flow and weather as independent, which is a good estimate except during a flooding event.
Permit demand compares the popularity of first-choice lottery requests on each date. For most lotteries, the actual odds are low for any day, but choosing a more-popular launch will make your odds that much lower versus a less-popular launch.
El Niño and La Niña years are shown so you can see how these large-scale events might change flows or weather as you plan your trip. Because there are only a handful of these on record, they should be treated carefully.
Things to keep in mind
Past conditions do not guarantee future ones. Climate, snowpack, and river management change.
Flow on some rivers is affected by dams and releases, and this site does not know about future year's operations.
A weather station can be some distance from the river, and conditions on the water can differ from the station.
Permit rules, quotas, and lottery dates change. Always confirm current rules with the managing agency (the USDA Forest Service or the Bureau of Land Management) before planning.
Part 2. The detailed version
2.1 Calendar handling
Every day of every year is mapped to a month-and-day key (for example 0714 for July 14) and to a placeholder date in a leap year (2000) so that February 29 is carried correctly and all years line up on one axis. A launch date's statistics are computed across all years that have a complete record for that date's trip window. The number of years behind each point is stored with it and shown in the tooltips.
2.2 Flow metrics
Trip window. For launch date d and trip length L days, the window is days d through d + L − 1.
Trip-average flow is the mean of daily flows over the window. The median line is the 50th percentile across years. The band is the 5th to 95th percentile.
Flow levels. Each river has up to four flow dividers that split flows into five named levels. The flow levels come from AmericanWhitewater.org and are typically described as ‘below recommended; ‘low runnable’; ‘medium runnable’; ‘high runnable’; and ‘above recommended’. We use the lowest divider as the minimum flow for the river and the second-highest for the maximum flow. Some rivers only have two dividers and three named levels: ‘below recommended’; ‘runnable’; ‘above recommended’. In this case we use the two dividers as our minimum and maximum values. We make no claim that the flows from AmericanWhitewater are suitable for your specific purpose.
Any day below minimum is the share of years in which any day in the window fell below the minimum flow. Any day above maximum is the same for the maximum. Flow in range is the share of years with neither. These are counted directly from the record.
Level chances (P_LEVEL_0 and so on) is used to populate the flow histogram and gives the share of years whose trip-average flow fell in each level.
2.3 Weather metrics
For each launch date and each year, the trip window is checked against the following definitions. Each chance is the share of years in which the condition occurred at least once in the window (or, for rain, in total).
| Measure | Definition |
|---|---|
| Wet | Total precipitation over the trip is more than 0.1 inch. |
| Hot | At least one day with a high above 100°F. |
| Freezing night | At least one night with a low at or below 32°F. |
| Rainy day | A day with 0.05 inch or more of precipitation, or any snow. |
| Miserable | At least one cold, wet day (0.05 inch or more of rain, or any snow, with a high below 55°F or a low below 40°F), or rain on half or more of the trip's days (rounded down, at least 1 day). |
| Fair weather | None of wet, hot, freezing night, or miserable. |
| Snow | Any snowfall during the trip. |
| Trip high / trip low | The hottest daytime high and the coldest overnight low during the trip. Medians and the 5th to 95th percentile band are shown across years. |
Shaded bands on chance charts are 90 percent confidence intervals on the proportion of years, computed from the number of years with data using the Wilson score. They widen where fewer years contribute.
Shaded bands on temperature, precipitation, and flow lines show the a range which covers the middle 90 percent of years, which is a different thing: they describe spread in outcomes, not uncertainty in an estimate.
2.4 Blended weather station
Where a river passes several stations, the Blended option estimates conditions along the trip. For each trip day, it weights the stations by how near each is to where the boat is likely to be, so early days lean on the upstream station and later days on the downstream one. Only years in which every station has data are used, so the blend is internally consistent. The share each station contributes on each trip day is shown on the site in the ‘i' information bubble (the BLEND_MIX field). We chose the weights based on a review of historical weather at the contributing sites and boater experience.
2.5 Combined score
The best-date score is the chance of flow in range times the chance of fair weather: SCORE = P(flow in range) × P(fair weather). Multiplying assumes the two are independent. Where flow and weather are related (for example, a wet spring that raises flows and brings rain), the true joint chance differs somewhat from the product, so the combined figure is labeled an estimate.
2.6 Strong El Niño and La Niña years
Years are classified with NOAA's Oceanic Niño Index (ONI) averaged over December to February. A year is labeled El Niño when that value is at or above +1.0 and La Niña when it is at or below −1.0 (the EL_MIN and LA_MAX thresholds in the year file); all other years are "Other". The tab compares trips starting in El Niño or La Niña years with all other years. Because only a few years qualify, the tab shows medians for weather and individual hydrographs for flow rather than chances.
2.7 Permit lottery demand
Source. First-choice lottery applications by date from recreation.gov reservation records. Ranked second to fourth choices do not exist in the data from recreation.gov, so demand reflects what people wanted most.
Interest index. For each date in each year, the date's requests are divided by the season's average requests per offered date. A value of 1 is a typical date. Dates with zero requests are floored at 1 request, and the number of such cases is reported (N_ZERO).
Fold change. The index is shown as a signed fold change: +3 means about three times busier than typical, −3 about three times quieter, and values between −1 and +1 are shown as "about typical". On the charts each step up or down is a doubling.
Across years. The summary files store the minimum, average, and maximum over the years used, so you can see both the usual level and the spread.
Offered or not. The Selway has dates that were offered to private parties in some years and not others, so each date is marked as lottery, mixed, or not offered, based on recent seasons. The other rivers offer every date in the season.
Day-of-week effect. Weekends and holiday-adjacent dates draw more requests. For each river we estimate a multiplier for each weekday, with a minimum and maximum across years, and report how much of the date-to-date variation is explained by time of season alone, by weekday alone, and by both (FIT_DATE, FIT_WEEKDAY, FIT_BOTH). Where the fit is weak, as on the Main Salmon, the weekday figures should be read loosely.
2027 forecast. The calendar view's lottery row starts from each date's historical demand by time of season, then rescales it by the day-of-week multiplier for the weekday that date falls on in 2027. This removes the luck of which weekday a date happened to land on in past years. It is an estimate for 2027 only, built from past seasons, not a prediction of the real counts.
What it is not. Demand is not your chance of winning. The reservation data show requests, not the number of permits or applicants per group, so the site shows crowding and popularity, not odds.
2.8 Assumptions and limitations
Stationarity. Treating all years as equally likely assumes the past represents the future. Warming trends, changing snowpack, and fire all push against this.
Sample size. Flow and weather records span decades, but permit data span only a few seasons, so lottery demand is much less certain. Similarly, single-station weather data typically has a longer record than the blended stations.
Independence. The combined score treats flow and weather as independent, which is an approximation.
Measurement location. Gauges and weather stations are points, and a long river changes along its length.
Management. Dam operations, closures, wildfire, and rule changes can override history.
Missing data. Gaps in gauge or station records reduce the number of years behind a date. The count is always stored with the value.
Part 3. Credits, license, and how to cite
Data credits
Streamflow data: U.S. Geological Survey. Weather data: NOAA National Centers for Environmental Information. Oceanic Niño Index: NOAA Climate Prediction Center. Permit reservation data: recreation.gov RIDB. Permit program information: USDA Forest Service and Bureau of Land Management. Flow levels: American Whitewater.
FriendOrFlow is an independent project and is not affiliated with or endorsed by these agencies.
Software
Charts are drawn with D3 (ISC license). Data processing is done in Python with the author's own tools.
License
Unless noted, the original content of this site (analysis, summaries, charts, and text) is © 2026 Alex Steely and licensed under the Creative Commons Attribution 4.0 International license (CC BY 4.0). You may share and adapt it, including commercially, if you give appropriate credit, link to the license, and indicate if you made changes. The license: https://creativecommons.org/licenses/by/4.0/
The license covers our original work only. It does not extend to the underlying government data, which remain subject to their sources' terms, or to third-party software such as D3.
Suggested citation
Steely, A. (2026). FriendOrFlow: historical flow, weather, and permit-lottery odds for river launch dates. https://friendorflow.com