Roaring Fork River Basin

Source Our computation, no independent reference · computed here from the screened DWR record Source Our computation, no independent reference · no independent reference; see /methods

This year, in context

In the weeks around 2026-09-29 (the newest reading on file), flows at the 19 ranked gages are running at a median 74% of their own long-term norm for this window — 15 below a typical year, 4 at or above. Driest relative to its record: THOMPSON CREEK FEEDER DITCH NEAR HAYSTACK, CO, ranking 2nd lowest of 9 years.

Gage flow trend & changepoints

Of the 73 gages with enough record, 8 are trending down and 1 up, and 11 carry a detected shift. The clearest is North Fork Fryingpan River near Norrie, its annual mean flow changing -3.37 cfs per decade (-8.3% of its median annual flow) over 81 years (p<0.0001). The other 61 show no clear signal either way.

⚠️ 1 gage flagged: their records may have quietly shifted against their neighbours, so treat their trends with caution.

61 other gage(s) show no significant trend or changepoint (not shown).

How the measurement network itself changed

This basin was not always measured the way it is today — the API-era record is dense and recent, the early record sparse. 23 gage(s) and 263 structure(s) report in 2026, but structure reporting only reached half its modern level around ~1975. Every historical claim below is checked against this: long-term shifts that co-move with the growth of the network carry an explicit measurement-evolution caution, and abandonment screens whose idleness begins before the modern reporting era are flagged †.

Mass-balance (network closure) trend

Structure divrec-history backfill coverage as of the last check: 7% of 5741 schematic-relevant structures. 0 year(s) clear the 30% coverage gate so far -- the trend activates automatically once qualifying years span a full decade. No action needed; the nightly backfill keeps extending the record.

Reach gain/loss shifts (1950–2026)

Mass-balance kinks computed for every summer month back to 1950 (2628 reach-snapshots) — which reaches gain or lose water, and how that has shifted across decades. Each shift is cross-checked against the observation record above: as more diversions became metered, "unaccounted" water shrinks mechanically, which can masquerade as a hydrologic shift.

• CRYSTAL RIVER AT PLACITA, CO.: reach shifted from +123 cfs to +180 cfs summer mean (more gaining, p=0.011, 12 yr) [alluvial valley: real alluvial aquifer -- a decades-long shift toward losing is the classic signature of alluvial pumping and urbanization]
• ROARING FORK RIVER AT GLENWOOD SPRINGS, CO.: reach shifted from +2386 cfs to +1075 cfs summer mean (more losing, p=0.000, 77 yr) [alluvial valley: real alluvial aquifer -- a decades-long shift toward losing is the classic signature of alluvial pumping and urbanization]
• CRYSTAL RIVER ABV AVALANCHE CRK, NEAR REDSTONE, CO: reach shifted from +295 cfs to -136 cfs summer mean (more losing, p=0.001, 63 yr) [sedimentary uplands: moderate groundwater connectivity -- gain/loss shifts can be real aquifer response; corroborate with nearby well records]
• HUNTER CREEK AT ASPEN, CO: reach shifted from -54 cfs to -104 cfs summer mean (more losing, p=0.004, 17 yr) [sedimentary uplands: moderate groundwater connectivity -- gain/loss shifts can be real aquifer response; corroborate with nearby well records]
• MAROON CREEK ABOVE ASPEN, CO.: reach shifted from +17 cfs to +59 cfs summer mean (more gaining, p=0.011, 26 yr) [sedimentary uplands: moderate groundwater connectivity -- gain/loss shifts can be real aquifer response; corroborate with nearby well records]
• ROARING FORK RIVER ABV LOST MAN CR NEAR ASPEN, CO: reach shifted from +5 cfs to -25 cfs summer mean (more losing, p=0.004, 26 yr) [mountain crystalline canyon: minimal groundwater storage -- large or shifting unaccounted losses here are more consistent with data/measurement issues or unmodeled plumbing than with aquifer exchange]

Groundwater now has its own tab: 46 monitored well(s), water-table trends, stream corridors, and the advisory depletion screen.

Reservoirs now has its own tab: 165 reservoir(s), end-of-year storage back to 1953, click-to-chart histories.

Cross-gage pattern fingerprint

Over the 63-year window with full coverage (1964–2026), the dominant shared pattern across 4 gage(s) explains 55% of year-to-year variance (almost certainly a basin-wide climate/snowpack driver).

All gages in this window move with the shared pattern -- no localized outlier.

Structures most implicated by reconciliation (as of 2026-09-21)

Which structures' data absorbed the most correction when the current mass-balance snapshot was reconciled -- see the Historical tab for the full picture on any date.

• BOWLES PL & DITCH AP (diversion): +15.3 cfs (50% of that reach's correction)
• BOWLES PIPELINE AND DITCH (diversion): +15.3 cfs (50% of that reach's correction)
• BRANDYWINE SPRING (return): +2.4 cfs (1% of that reach's correction)
• HOMESTEAD SPRING (return): +2.4 cfs (1% of that reach's correction)
• PASTURES PIPELINE (diversion): +2.4 cfs (1% of that reach's correction)
• PASTURES POND (diversion): +2.4 cfs (1% of that reach's correction)
• ODONNELL PUMP & PIPELINE (diversion): +2.4 cfs (1% of that reach's correction)
• KINDER POND NO 1 (diversion): +2.4 cfs (1% of that reach's correction)
• KINDER POND NO 2 (diversion): +2.4 cfs (1% of that reach's correction)
• KINDER PUMP & PL (diversion): +2.4 cfs (1% of that reach's correction)

Source DWR administrative record · verify at DWR groundwater wells ↗ Source DWR administrative record · verify at DWR well permits ↗

149 water-level measurements across 46 monitored well(s) — 19 alluvial, 0 bedrock — plus 28 administered well(s) reporting metered pumping.

Well map

3D aquifer x-ray

Every aquifer unit as its own head surface, animated through time.

Water-table trends

Of 0 well(s) with enough record for a trend test: 0 declining, 0 rising.

Stream corridors — the surface connection

Where a reach's gain or loss has shifted, the wells within two miles of its anchoring gage and what their water table did over the same era.

CRYSTAL RIVER AT PLACITA, CO.: 0 alluvial well(s) nearby

CRYSTAL RIVER ABV AVALANCHE CRK, NEAR REDSTONE, CO: 1 alluvial well(s) nearby. Glover screen: 0.1 cfs of nearby metered alluvial pumping → est. lagged depletion 0.0–0.1 cfs (uncalibrated T/S ranges; flagged for SPDSS calibration, AWAS cross-check, and expert review — DWR's AWAS is the authority).

HUNTER CREEK AT ASPEN, CO: 0 alluvial well(s) nearby. Glover screen: 0.0 cfs of nearby metered alluvial pumping → est. lagged depletion 0.0–0.0 cfs (uncalibrated T/S ranges; flagged for SPDSS calibration, AWAS cross-check, and expert review — DWR's AWAS is the authority).

Wells labeled by administrative world before anything else is said about them: tributary alluvium participates in the priority system; designated-basin and Denver Basin bedrock aquifers are governed differently and are excluded from stream-connection reasoning. Read bedrock declines differently too: deep bedrock aquifers recharge negligibly on human timescales, so a falling bedrock level is storage mining — effectively permanent — not a drought response that a wet year undoes. All screens advisory.

Source DWR administrative record · verify at DWR structure records ↗

165 reservoir(s) with 7,787 storage measurement(s), 1953–2025. Basin storage as of 2024-10-31 (the record's own end — reservoir records publish ~annually): 79,357 AF across 3 reservoir(s).

Reservoir map — how full is each vessel?

Basin storage, end of each water year

End-of-water-year totals; per-year reservoir counts shown because reporting coverage varies -- a total is only comparable to years with similar counts.

Largest reservoirs on record

Click a row to chart its full storage history, or open its dossier for rights, documents and dam-safety facts.

Source Independent agency measurement · fetched 2026-10-14 · verify at NRCS report generator ↗ Source Our computation, scored against a reference · NRCS forecast, graded here against the gage record

Snow-water equivalent, WY2027

NRCS SNOTEL, daily. Sites are assigned to the district whose boundary contains them — the snow that feeds this district can also sit in a neighboring headwater district.

Two ways a snow year goes wrong (1981–2026)

A bad snow year here comes in two flavors (both called snow droughtA winter that shorts the snowpack. It happens two ways: a dry one, where the snow never comes, and a warm one, where near-normal snow falls but melts off weeks early -- and the river feels them differently.), and the river feels them differently. In 6 of the last 46 winters the snow never came — the pack peaked below 80% of its median. In 0 more, the snow came but left early — a near-normal peak melted out two weeks or more ahead of its usual date, sending the water down before the season needed it.

The snow never came: 1981, 1990, 2002, 2012, 2015, 2026.

Computed by this site from 8 SNOTEL sites’ full daily records (NRCS AWDB, back to 1981). Each year is measured against each site’s own multi-decade median — not the 1991–2020 normal the chart above uses — so the 1980s grade against a fair baseline. Peak and melt-out need 85% of the season’s days on the record; thin years are dropped, not guessed. Derived from snow measurements alone — it cannot tell dry-from-warm causes apart, only what the snowpack did.

Runoff outlook — Ruedi Reservoir Inflow

NRCS's final issue (2026-06-01) calls it even odds that at least 15 kAF arrives over 06-01 to 07-31 — 18% of a normal season; almost certainly at least 8, probably not more than 29. New forecasts begin in January. How to read these numbers →

We test a record-based adjustment of the April benchmark at every graded point. Here it did not beat NRCS over 32 replayed seasons (14% typical miss unadjusted vs 16% adjusted), so none is shown — NRCS’s own forecast, unadjusted, is the best April benchmark available for this river. Why we test →

Runoff outlook — Roaring Fk at Glenwood Springs

NRCS's final issue (2026-06-01) calls it even odds that at least 88 kAF arrives over 06-01 to 07-31 — 20% of a normal season; almost certainly at least 40, probably not more than 150. Our own record at this gage shows 99 kAF actually delivered. New forecasts begin in January. How to read these numbers →

The record-adjusted April benchmark

NRCS’s April issue (2026-04-01) called even odds at 250 kAF over 04-01 to 07-31. Re-anchoring that number on this river’s own history — how its April issues have verified, the winter’s baseflow, and the district’s April 1 snowpack — moves it to 184 kAF. The river delivered 185 kAF.

Computed by this site — not an NRCS number, and shown only because it earned it: replayed over 69 past seasons here (1962–2025), never seeing the year it predicts, the adjustment cut the typical April miss from 24% to 14%. At points where it does not beat NRCS, no adjusted number appears. Coefficients are frozen each August; the model never changes mid-season. Method and honest limits →

How good are these forecasts here? (1950–2026)

Grading every April issue against what the river then delivered, measured at this gage’s own record — 76 seasons: the median miss of the 50% forecast is 24%, with a high bias of 22%, and the observed volume landed inside the stated 10–90% band 52% of the time (a calibrated forecast would say 80%).

Computed by this site: NRCS's archived April issues (their period as issued each year) against the observed volume in DWR’s published record; years with under 90% daily coverage are dropped, not summed. A grade of the forecast, never of the forecasters — these are hard rivers. How this point ranks among every graded Colorado forecast point →

What April snow has meant for this river (1981–2026)

Across 46 measured years, the season’s water has closely followed the April snowpack: a year opening April with 80% of median snow has typically gone on to deliver about 62% of a normal season (a snow-to-runoff elasticityHow strongly one number moves with another. A snow-to-runoff elasticity of 1.5 means a 10% shortfall in April snowpack has typically become a 15% shortfall in the season's water. of 1.3).

Computed by this site: the district’s April 1 SWE (% of each SNOTEL site’s own-record median) against the observed April–July volume at this gage (% of its median season) — a fixed window every year, unlike the forecast grades above, which use each issue’s own period. The slope is fit so single wild years can’t steer it (Theil–Sen, log space); years missing 10% of their days are dropped.

Source Our computation, no independent reference · computed here from the screened DWR record Source Independent agency measurement · fetched 2026-10-06 · verify at U.S. Drought Monitor ↗

Right now

This is the 13th-driest week 39 in 120 years of record at ROARING FORK RIVER AT GLENWOOD SPRINGS, CO. (as of 2026-09-21; this week’s flow sits at the 11th percentileRank against the record: a river at its 5th percentile is lower than it was in 95 of the past 100 years on this same date. for the date). On this week’s US Drought Monitor, Gunnison County is 22% in D4 exceptional drought (2026-10-06). The law agrees: the most senior active call leaves 100% of this district’s decreed flow rights junior to the calling right this week.

Every year since 1906, week by week

Each thin row is a year at ROARING FORK RIVER AT GLENWOOD SPRINGS, CO.; each cell is one week’s flow ranked against the same calendar week of every other year — warm means dry for the date, blue wet, so seasonality cancels and droughts appear as warm bands. Gaps are missing record, never a dry river. Click a year to reconstruct it on the Historical tab.

This year against its analogs

Over the 39 weeks recorded so far, 2026’s weekly trace most resembles 2002 (mean gap 11.3 percentile points; next closest 2021 (12.3), 1977 (13.8)). The analog’s remainder is drawn as context — its ending is what happened to it, not a forecast for this year.

The Drought Monitor’s view since 2000

Mean share of area in each class across Eagle, Garfield, Gunnison, Pitkin counties — the counties this district’s gages sit in, which overlap the district imperfectly. The Drought Monitor weighs soils, vegetation and local reports; it is CONTEXT beside the river’s own record above, not the water-rights picture.

The law’s own drought gauge

When water runs short, administration answers: calls reach further back in time, and more decreed rights go junior to the calling right. This is that response, weekly since 2002 — the share of this district’s decreed flow rights junior to the most senior active call. Watch it spike in the named droughts below, and linger after the river recovers.

Computed from DWR’s call record (407 calls touching this district) against the district’s rights index. Scoping is deliberately coarse — a call’s own district, or districts named in its comments — and this is administration’s RESPONSE, not a per-structure verdict (each dossier’s curtailment card does that, with stream miles). A deepening call can also reflect growing demand, not only shrinking supply.

This basin’s droughts, by its own record

Detected mechanically: the 26-week smoothed percentile below the 30th for at least 20 weeks; recovered on regaining the 50th. The water-table clock compares each monitoring well against its own pre-drought median (the 3 years before onset) and dates the year the median well got back — the aquifer remembers longest, and “still below” is a finding, not a gap. The knobs are stated because the episode list depends on them — a different reasonable choice merges or splits episodes at the edges. Names are attached where an episode overlaps a drought Colorado already talks about.

When this basin claimed its water

Every bar is a year’s new appropriations (5,598 rights across 121 years of shared record), colored by that year’s wetness from the barcode above. Here the correlation is +0.06 — essentially no wet-dry pattern; here, law and settlement set the pace, not the weather. Across the state both behaviors exist: ephemeral southwest basins file after wet demonstrations, transbasin-source basins under drought — and the biggest statewide filing waves sat in the 1950s drought and after 2002, not in the wet 1980s–90s.

Appropriation dates are the CLAIMED initiation of a right, not the courthouse filing date; the record is survivor-biased (abandoned claims are absent, so failed dry-era scrambles are undercounted); and the big 1960s and 2000s spikes sit on legal events — the pre-1969-Act adjudication era and 2000s legislation — not on weather. A correlation here is a description, never a cause.

New to how a dry winter becomes a call on the river? Anatomy of a drought walks the chain in four minutes — snow to rivers to reservoirs to calls to the water table.

The water itself, in this district

Four registers, each on its own schedule: what USGS sensors read right now, where samples have been taken, which stretches the state lists as impaired, and how the tap water of the systems here stands with EPA. None is a DWR record; each links its own page.

Right now

Source Provisional telemetry · fetched 2026-10-15 · verify at USGS current conditions ↗

4 USGS sensors report temperature, conductance, dissolved oxygen or turbidity from inside this district; provisional readings, never merged into the verified record.

Where water has been sampled

Source Register of another agency · fetched 2026-10-08 · verify at Water Quality Portal ↗

The portal lists 748 sampling stations in this district's watershed (Roaring Fork 14010004), of which 132 reported a sample in the last 5 years on the state and tribal side (USGS activity is in the station count, not in that figure). Most active: Colorado River Watch (83), Colorado Dept. of Public Health & Environment-WQCD (40), USGS Colorado Water Science Center (5). Most-sampled groups: Inorganics, Major, Metals, Physical, Nutrient, Inorganics, Minor, Metals. Every station as a file ↗

Results themselves are not fetched here; USGS samples added after March 2024 may not be in the portal's legacy profiles yet.

Drinking water

Source Register of another agency · fetched 2026-10-07 · verify at EPA SDWIS ↗

51 community systems touch this district. None carries an open health-based violation on EPA's record. EPA sampled 5 of them for PFAS in 2023–25; none had PFOA or PFOS at or above the 2024 limit.

and 41 more; the full directory filters by district.

EPA's record carries violations and lead/copper percentiles; nitrate and disinfection by-product results are not in it. UCMR 5 samples are monitoring, not compliance, and the 2024 PFOA/PFOS limits equal its reporting limit, so any detection of those two is at or above them.

Impaired waters

Source Register of another agency · verify at EPA ATTAINS ↗

The state's 303(d) impaired-waters list has not been joined to this basin's own watershed boundary yet, so nothing is shown rather than a guess; the basins with a join carry the full list here. Why the join is geometry, not a name match.