HeimdallResearch/western-terminal-lakes

Western Terminal Lakes

Not for water-rights, legal or engineering decisions.

Not in this download, free through the API: full-size summer satellite images of all 18 lakes and reservoirs, 2019 to 2026. Get a free key at https://heimdallresearch.com/go/signup

Contains modified Copernicus Sentinel data 2019-2026.

Not for water-rights, legal or engineering decisions. Provided as is, without warranty of any kind.

Monthly open-water area for 19 series on 18 western lakes and reservoirs from Sentinel-2 imagery, January 2019 to September 2026, with a low and high estimate and a comparison with the gauge-derived area where a published area table exists. The first seven series are the terminal lakes (Great Salt Lake north and south arms, Salton Sea, Mono, Pyramid, Walker and Abert), counted inside a hand-drawn outline that is larger than each lake. The other 12 are Lake Mead, Lake Powell, Elephant Butte, Falcon, Flaming Gorge, Amistad, Blue Mesa, Upper Klamath Lake, Choke Canyon, Lahontan, American Falls and Lake Corpus Christi, counted inside outlines built from public hydrography.

Each month, every clear Sentinel-2 view of each pixel is combined, so partly cloudy days still count. Two water indexes are computed for each view, their per-pixel monthly medians are taken, and water is classified once per month. Area is counted on a 20 m grid in an equal-area projection (EPSG:5070) inside a fixed reference outline. The outline is a hand-drawn polygon larger than the lake, so water_area_km2 is all open water inside it, including bays and ponds next to the lake. area_low_km2 and area_high_km2 are the lowest and highest of four water estimates. The share of the outline seen at least once is published with every value. A month with no usable view stays blank.

Four pixel categories are kept separate so frozen or cloudy areas never count as dry lakebed. (1) Open water: clear pixels classified as water and reported in water_area_km2. (2) Ice and snow: Sentinel-2 scene class 11 is dropped from November to April and is never counted as open water or as dry lakebed; December to March months with a high share of that class are flagged. (3) Not observed: cloud, cloud shadow, cirrus and other unusable classes are left out of the composite; their effect shows in seen_fraction, and a month with no usable view stays blank (quality_flag no_view). (4) Exposed lakebed: not classified in this version. Pixels that are not open water are not labelled as lakebed.

The 12 lakes and reservoirs added in this version use the same composite. Their outlines come from the USGS National Hydrography Dataset and the JRC Global Surface Water maximum extent, cut at the upstream end of each reservoir (data/lake_outlines.geojson and the outline sources in NOTICE.md). For Lake Mead, Lake Powell, Amistad, Flaming Gorge, Blue Mesa and Lake Corpus Christi, water_area_km2 is an AWEI-style estimate with its own Otsu threshold, because the MNDWI estimate left out water in the canyons. Lahontan also uses it, chosen differently (see Method). The AWEI-style index is not the published AWEI; Method gives the formula. The primary_estimate column says which estimate each row uses. The algae, mud, ice and canyon rules are in Method. Bear Lake was evaluated and is not in the table.

Bias is the mean difference between the satellite area and the gauge-derived area as a percent of the mean gauge-derived area, over unflagged months.

Key figures

• Great Salt Lake south: satellite -277.7 km2 (-13.4%), gauge-derived -154.3 km2 (-9.1%), 2019 to 2026, March, May to September average
• Great Salt Lake north: satellite -226.4 km2 (-16.2%), gauge-derived -129.3 km2 (-12.2%), 2019 to 2026, May to September average
• Great Salt Lake north, same months, by water estimate: MNDWI with the Otsu threshold (the figure above) -226.4 km2, MNDWI above 0 -120.7 km2, MNDWI above 0.1 -191.9 km2, AWEI with its own Otsu threshold -184.9 km2. The north-arm change therefore depends on the threshold choice. The south-arm change depends on it much less: its four estimates give -277.7, -283.0, -293.0 and -289.2 km2. This release does not say which estimate is right.
• Salton Sea: -75.9 km2 (-8.8%), 2019 to 2026, January to September average
• Walker Lake: -3.3 km2 (-2.8%), 2019 to 2026, March to September average
• Pyramid Lake: -3.6 km2 (-0.9%), 2019 to 2026, April to September average
• Mono Lake: +0.1 km2 (+0.0%), 2019 to 2026, April to September average
• Lake Abert: -6.6 km2 (-5.0%), 2019 to 2026, April to September average
• Lake Mead: -44.4 km2 (-12.6%), 2019 to 2026, January to September average
• Lake Powell: -115.4 km2 (-31.0%), 2019 to 2026, January to September average
• Elephant Butte Reservoir: -12.5 km2 (-33.2%), 2019 to 2026, January to June, August to September average
• Falcon Reservoir: -49.8 km2 (-32.1%), 2019 to 2026, January to February, July to September average
• Flaming Gorge Reservoir: -20.4 km2 (-13.0%), 2019 to 2026, April to September average
• Amistad Reservoir: -85.7 km2 (-50.9%), 2019 to 2026, January to September average
• Blue Mesa Reservoir: -10.4 km2 (-32.9%), 2019 to 2026, May to September average
• Upper Klamath Lake: -1.6 km2 (-0.7%), 2019 to 2026, January, April to May average
• Choke Canyon Reservoir: -35.7 km2 (-61.6%), 2019 to 2026, January to April, July average
• Lahontan Reservoir: -4.2 km2 (-12.5%), 2019 to 2026, January, March to September average
• American Falls Reservoir: -33.3 km2 (-19.7%), 2019 to 2026, April to September average
• Lake Corpus Christi: -32.9 km2 (-47.5%), 2019 to 2026, January to September average
• Each change above uses the same calendar months in both years: the months from January to September that are not flagged low_confidence in either year.
• Great Salt Lake: on the same months the satellite change is larger than the gauge-derived change on both arms (south 1.8 times, north 1.8 times). The two measure different things: water_area_km2 is open water inside a hand-drawn outline, gauge_area_km2 comes from the lake level and the USGS elevation-area table. The satellite minus gauge gap was +370.9 km2 (south) and +333.2 km2 (north) in 2019 and +247.5 and +236.1 km2 in 2026, so it is not a fixed offset. The cause is not established, and this release does not say which of the two series is closer to the true change.
• 19 series (Great Salt Lake counts as two). 1751 of 1767 lake-months have a value. 296 are flagged low_confidence.
• Great Salt Lake area here is open water inside an outline larger than the lake. It is above the area from the USGS tables by +18.1% (south arm, mean +291.7 km2, SD 65.6, n=79) and +27.9% (north arm, mean +268.7 km2, SD 49.3, n=60).
• Rank correlation with gauge-derived area (Spearman): Great Salt Lake south 0.84, Walker 0.87, Great Salt Lake north 0.67, Lake Abert 0.36

What the data shows

Great Salt Lake's south arm: July open water was 12% lower in 2026 than in July 2019.

Great Salt Lake south arm open-water area fell from 2017.2 km2 in July 2019 to 1770.6 km2 in July 2026 (loss 246.5 km2, 12.2 percent).

South-arm open water bears on dust, habitat, and lake-level policy around the south arm. The USGS Utah Water Science Center notes dust from the dry lakebed as a possible health risk for northern Utah communities (USGS Great Salt Lake, page opened 2026-10-06). Satellite open-water area is not a water-rights number.

import pandas as pd
from decimal import Decimal as D
m = pd.read_csv("data/lake_monthly.csv", dtype=str)
a, b = D(m.query("lake_id=='great-salt-lake-south' and month=='2019-07'").water_area_km2.iloc[0]), D(m.query("lake_id=='great-salt-lake-south' and month=='2026-07'").water_area_km2.iloc[0])
print(round(a,1), round(b,1), round(a-b,1), round((a-b)/a*100,1))

Great Salt Lake's north arm: July open water was 16% lower in 2026 than in July 2019.

Great Salt Lake north arm open-water area fell from 1369.9 km2 in July 2019 to 1149.3 km2 in July 2026 (loss 220.6 km2, 16.1 percent).

The causeway splits the lake, so north and south can move differently. On these Julys both arms shrank. Read this next to the south-arm finding when you need both sides of the release record.

import pandas as pd
from decimal import Decimal as D
m = pd.read_csv("data/lake_monthly.csv", dtype=str)
a, b = D(m.query("lake_id=='great-salt-lake-north' and month=='2019-07'").water_area_km2.iloc[0]), D(m.query("lake_id=='great-salt-lake-north' and month=='2026-07'").water_area_km2.iloc[0])
print(round(a,1), round(b,1), round(a-b,1), round((a-b)/a*100,1))

Salton Sea: July open water was 9% lower in 2026 than in July 2019.

Salton Sea open-water area fell from 850.4 km2 in July 2019 to 772.0 km2 in July 2026 (loss 78.3 km2, 9.2 percent).

Open-water change at Salton Sea sits in air quality and shoreline exposure debates in the Imperial and Coachella valleys. This figure is satellite open-water area for those two Julys only.

import pandas as pd
from decimal import Decimal as D
m = pd.read_csv("data/lake_monthly.csv", dtype=str)
a, b = D(m.query("lake_id=='salton-sea' and month=='2019-07'").water_area_km2.iloc[0]), D(m.query("lake_id=='salton-sea' and month=='2026-07'").water_area_km2.iloc[0])
print(round(a,1), round(b,1), round(a-b,1), round((a-b)/a*100,1))

Lake Abert: July open water was 7% lower in 2026 than in July 2019.

Lake Abert open-water area fell from 128.0 km2 in July 2019 to 119.5 km2 in July 2026 (loss 8.5 km2, 6.7 percent).

If you are comparing western terminals in the release package, Lake Abert lost open water on this July-to-July window. Do not reuse the Lakes 1.0 gain headline on 1.1.

import pandas as pd
from decimal import Decimal as D
m = pd.read_csv("data/lake_monthly.csv", dtype=str)
a, b = D(m.query("lake_id=='lake-abert' and month=='2019-07'").water_area_km2.iloc[0]), D(m.query("lake_id=='lake-abert' and month=='2026-07'").water_area_km2.iloc[0])
print(round(a,1), round(b,1), round(a-b,1), round((a-b)/a*100,1))

Coverage

Publisher	Heimdall Research
Version	1.1
Released	2026-10-10
DOI	10.5281/zenodo.23045007
License	CC BY 4.0
Web page	https://heimdallresearch.com/data/terminal-lakes/
Period	January 2019 to September 2026
Resolution	Lake by month, from 10 to 20 m imagery
Update cadence	The monthly table in this release runs through 2026-09. It is not refreshed on a schedule.

Versions

Version	Where published
1.0	Kaggle
1.1	Zenodo, Kaggle, Hugging Face (first version on Zenodo and Hugging Face)

Version 1.0 was published on Kaggle only and has no Zenodo DOI. Version 1.1 is the first version on Zenodo and Hugging Face.

Spatial units:

lake_id	name	period	outline_km2	water_estimate	reference_quality
great-salt-lake-north	Great Salt Lake north arm	January 2019 to September 2026	3,177	MNDWI Otsu	not graded
great-salt-lake-south	Great Salt Lake south arm	January 2019 to September 2026	4,055	MNDWI Otsu	not graded
salton-sea	Salton Sea	January 2019 to September 2026	2,461	MNDWI Otsu	not graded
mono-lake	Mono Lake	January 2019 to September 2026	360.6	MNDWI Otsu	not graded
pyramid-lake	Pyramid Lake	January 2019 to September 2026	912.3	MNDWI Otsu	not graded
walker-lake	Walker Lake	January 2019 to September 2026	513.7	MNDWI Otsu	not graded
lake-abert	Lake Abert	January 2019 to September 2026	421.7	MNDWI Otsu	not graded
lake-mead	Lake Mead	January 2019 to September 2026	628.5	AWEI-style Otsu	good
lake-powell	Lake Powell	January 2019 to September 2026	661.9	AWEI-style Otsu	good
elephant-butte-reservoir	Elephant Butte Reservoir	January 2019 to September 2026	149.1	MNDWI Otsu	fair
falcon-reservoir	Falcon Reservoir	January 2019 to September 2026	343.1	MNDWI Otsu	good
flaming-gorge-reservoir	Flaming Gorge Reservoir	January 2019 to September 2026	175.0	AWEI-style Otsu	good
amistad-reservoir	Amistad Reservoir	January 2019 to September 2026	267.1	AWEI-style Otsu	good
blue-mesa-reservoir	Blue Mesa Reservoir	January 2019 to September 2026	36.4	AWEI-style Otsu	good
upper-klamath-lake	Upper Klamath Lake	January 2019 to September 2026	257.0	MNDWI Otsu	weak
choke-canyon-reservoir	Choke Canyon Reservoir	January 2019 to September 2026	99.1	MNDWI Otsu	good
lahontan-reservoir	Lahontan Reservoir	January 2019 to September 2026	48.6	AWEI-style Otsu	weak
american-falls-reservoir	American Falls Reservoir	January 2019 to September 2026	231.7	MNDWI Otsu	weak
lake-corpus-christi	Lake Corpus Christi	January 2019 to September 2026	80.4	AWEI-style Otsu	good

Research use

Questions the tables can answer, each with the table and columns it uses, and what the tables leave open: docs/RESEARCH_USE.md.

Method

• Scenes: for each month, all Sentinel-2 Level-2A scenes over the lake from the first to the last day of the month are read from Microsoft Planetary Computer. Digital numbers are converted to surface reflectance scene by scene. Scenes with processing baseline 04.00 or later store reflectance with a +1000 offset, which is removed before scaling; earlier baselines have no offset. Where a scene exists as an original and a reprocessed copy, the reprocessed copy is used.
• Grid and masking: pixels are resampled to 20 m in EPSG:5070 (Albers equal area). Pixels the scene classification layer marks as no data, saturated, dark area, cloud shadow, cloud or thin cirrus are dropped (not observed). Snow and ice (class 11) pixels are also dropped from November to April and are never classified as open water or as dry lakebed.
• Composite: two water indexes are computed for every clear view, MNDWI = (B03 - B11) / (B03 + B11), with B03 green and B11 SWIR1, and an AWEI-style index = 4 x (B03 - B11) - (0.25 x B08 + 2.75 x B11), with B08 near-infrared. The published no-shadow AWEI (Feyisa and others, 2014) uses SWIR2 (B12) in place of the second B11, so this index is not the published one; its validation here is ours (see Limits). The monthly composite is the per-pixel median of each index over the clear views in the month.
• Water area: water_area_km2 counts composite pixels inside the reference outline with MNDWI above an Otsu threshold. One threshold is chosen per lake and month from that month's composite. Area is the pixel count times 400 m2.
• Added lakes, estimate: for Lake Mead, Lake Powell, Amistad, Flaming Gorge, Blue Mesa and Lake Corpus Christi, water_area_km2 is the AWEI-style estimate with its own Otsu threshold, because the MNDWI estimate left out water in the canyons. On 12 months each of Lake Powell and Lake Mead the AWEI-style estimate was within 10 percent of the gauge-derived area and the month-to-month changes moved together; the MNDWI estimate was not. The other canyon lakes use the same estimate. area_low_km2 and area_high_km2 stay the lowest and highest of the four estimates. The column primary_estimate says which estimate each row uses (aweish_otsu is the stored name of the AWEI-style estimate).
• Lahontan: water_area_km2 is also the AWEI-style estimate. Lahontan has almost no narrow arms (canyon_fraction below 0.005). The 12 months used for Lahontan had already been seen across all 93 months, so that check is not blind. On those 12 months both estimates were within 10 percent of the gauge-derived area. The bias of both estimates over all unflagged months, and the 2019 to 2026 change with each, are in Limits.
• Canyon and flag rules: canyon_fraction is the share of the outline in arms narrower than 3 pixels (60 m), found by opening the outline with a 30 m disc, and canyon_limited is true from 0.005. Ice is flagged from November to May on lakes that can freeze when more than 5% of observed pixels are snow or ice class. Algae or turbid water is flagged on Upper Klamath Lake, Blue Mesa and Elephant Butte when 10% or more of the water pixels have NDVI above 0; the low end of the range is then lowered by the flagged area and set to 0 if that would go below 0. Wet mud on Elephant Butte is counted as an uncertain edge within 0.10 of the threshold, which widens the range. A month with seen_fraction below 0.2 is blank (no_view). A month whose images could not all be read, or that ran out of time, is blank (processing_failed).
• Range: area_low_km2 and area_high_km2 are the lowest and highest of four estimates from the same composite: MNDWI with the Otsu threshold, MNDWI above 0, MNDWI above 0.1, and the AWEI-style index with its own Otsu threshold.
• Coverage: seen_fraction is the share of outline pixels with at least one clear view in the month. n_views is the number of days with a scene in the composite. A month with no usable scene stays blank (quality_flag no_view; processing_failed when the month could not be computed).
• Flags: low_confidence is true (and the value is kept) when (1) seen_fraction is below 0.8, (2) the area is more than 10% from the 3-month rolling median (5% for Pyramid and Mono), (3) the month is December to March and more than 5% of the outline is snow or ice class, or (4) the low-estimate rule (area_low_km2 under 80% of the value) applies to the MNDWI series only, and AWEI-style series are governed by the 125% high-estimate rule (area_high_km2 above 125% of the value), which applies to every series, including the first seven. quality_flag is ice for rule 3, low_coverage for rules 1 and 2, methods_disagree for rule 4 alone, bright_surface when more than 15% of the outline is snow or ice class, no_view for a blank month, and ok otherwise.
• Gauge comparison: gauge_area_km2 converts the USGS monthly median water level (storage for Walker) to area where a published elevation-area table exists. Great Salt Lake arms use Root (2023) DOI 10.5066/P9DGG75W. Walker and Lake Abert use Root and Hynek (2026) DOI 10.5066/P147WRTT. Salton Sea has gauge elevation but no published elevation-area table. Pyramid has a table but USGS 10336500 has no usable daily elevation in this period. Mono has a table but no USGS daily gauge series was used.

Intended use: Tracking long-term change in the size of western terminal lakes, comparing lakes on one definition, and checking satellite area against gauge-derived area.

Not intended for: Water rights, engineering or regulatory decisions, volume or storage estimates, dust emission estimates, air-quality forecasts, or health-risk claims.

Validation

Check	Result	Source
Change in open-water area, 2019 to 2026 (km2 and %). For each lake both averages cover the same calendar months: the months from January to September with a value that is not flagged low_confidence in either year.	GSL south 2074.0 to 1796.3 (-277.7, -13.4%; March, May to September); GSL north 1393.6 to 1167.2 (-226.4, -16.2%; May to September); Salton 860.5 to 784.6 (-75.9, -8.8%; January to September); Walker 118.0 to 114.7 (-3.3, -2.8%; March to September); Pyramid 420.3 to 416.7 (-3.6, -0.9%; April to September); Mono 179.2 to 179.3 (+0.1, +0.0%; April to September); Abert 130.7 to 124.1 (-6.6, -5.0%; April to September). Lake-months with a value: 1751 of 1767. Flagged low_confidence: 296.	lake_monthly.csv. Differences and percentages are taken before rounding.
Months flagged low_confidence, per series and by quality_flag	GSL south 14; GSL north 33; Salton 0; Walker 9; Pyramid 21; Mono 26; Abert 42. By quality_flag among the flagged months: ice 178, low_coverage 29, methods_disagree 37, bright_surface 1, algae_turbid 32, jump 19, mud 0. These parts add up to 296, the number of flagged months (296).	lake_monthly.csv
Range checks: area_low_km2 <= water_area_km2 <= area_high_km2 on every row; area_low_km2 at least 80% of water_area_km2 in unflagged months of the series that use the MNDWI estimate; area_high_km2 at most 125% of water_area_km2 in every unflagged month	The order low <= value <= high holds on all 1751 rows with a value. The 80% rule applies to the series whose main value is the MNDWI estimate (the first seven and every added lake except the seven that use the AWEI-style estimate): 0 of 883 unflagged months are below it. It does not apply to the seven series that use the AWEI-style estimate, because their low end is the MNDWI family, which leaves out canyon water (296 of 572 unflagged months are below 80%); they are governed by the 125% rule. The 125% rule applies to every series: 0 of 1455 unflagged months are above it. Among the first seven series, area_low_km2 is at least 0.899 of the value in every unflagged month. The release tool stops the build on any row that breaks the rule it applies.	lake_monthly.csv
Rows where area_low_km2 equals water_area_km2 (the lowest of the four estimates is the reported one)	Great Salt Lake south 93 of 93, Great Salt Lake north 93 of 93, Mono Lake 93 of 93. On those series the range extends upward only: water_area_km2 is its lower end.	lake_monthly.csv
Satellite area against gauge-derived area, unflagged months with both values: mean of water_area_km2 minus gauge_area_km2, bias (defined in the description), standard deviation (SD), months with satellite above gauge, Spearman rank correlation (rho)	GSL south +291.7 km2 (+18.1%; SD 65.6; above gauge in 79 of 79 months; rho 0.84). GSL north +268.7 km2 (+27.9%; SD 49.3; 60 of 60; rho 0.67). Walker +0.8 km2 (+0.7%; SD 1.6; 59 of 84; rho 0.87). Lake Abert +20.1 km2 (+18.3%; SD 16.3; 25 of 25; rho 0.36; the gauge series starts partway through the record). No paired months for Pyramid, Salton Sea or Mono (reasons are in Method).	lake_monthly.csv; USGS Water Data for the Nation; DOI 10.5066/P9DGG75W; DOI 10.5066/P147WRTT
Great Salt Lake and Lake Abert: why satellite area is above gauge-derived area	water_area_km2 is all open water inside the reference outline, and the outlines are hand-drawn envelopes larger than the lake: 4,055 km2 for the south arm (12 points), 3,177 km2 for the north arm (8 points) and 421.7 km2 for Lake Abert. The south-arm outline takes in Farmington Bay and evaporation ponds. The north-arm outline takes in the west side of Bear River Bay and evaporation ponds. The area of water in those bays and ponds is not quantified. The low estimate is also above the gauge area (area_low_km2 minus gauge_area_km2: south mean +291.7 km2, north +268.7 km2, Lake Abert +20.1 km2), so the choice among the four water estimates does not account for the difference.	lake_monthly.csv; data/lake_outlines.geojson
Imagery: every GeoTIFF read in full, blank share measured in the GeoTIFF and in the preview	0 images are marked partial (more than 5% of the image has no data). The share for each image is in imagery/index.csv. It counts pixels with no data, not cloud. Four summers still show cloud or sun glint over part of the lake, and imagery/IMAGERY.md names them.	imagery/index.csv
Added lakes: change in open-water area, 2019 to 2026 (km2 and %), same calendar months in both years, January to September, unflagged	Lake Mead 351.8 to 307.5 (-44.4, -12.6%; January to September); Lake Powell 372.8 to 257.4 (-115.4, -31.0%; January to September); Elephant Butte Reservoir 37.7 to 25.2 (-12.5, -33.2%; January to June, August to September); Falcon Reservoir 155.0 to 105.2 (-49.8, -32.1%; January to February, July to September); Flaming Gorge Reservoir 156.5 to 136.1 (-20.4, -13.0%; April to September); Amistad Reservoir 168.4 to 82.7 (-85.7, -50.9%; January to September); Blue Mesa Reservoir 31.5 to 21.2 (-10.4, -32.9%; May to September); Upper Klamath Lake 240.5 to 238.9 (-1.6, -0.7%; January, April to May); Choke Canyon Reservoir 58.0 to 22.3 (-35.7, -61.6%; January to April, July); Lahontan Reservoir 33.7 to 29.4 (-4.2, -12.5%; January, March to September); American Falls Reservoir 169.0 to 135.6 (-33.3, -19.7%; April to September); Lake Corpus Christi 69.3 to 36.4 (-32.9, -47.5%; January to September)	lake_monthly.csv
Added lakes: months flagged low_confidence	Lake Mead 1; Lake Powell 0; Elephant Butte Reservoir 9; Falcon Reservoir 17; Flaming Gorge Reservoir 25; Amistad Reservoir 1; Blue Mesa Reservoir 29; Upper Klamath Lake 28; Choke Canyon Reservoir 9; Lahontan Reservoir 9; American Falls Reservoir 21; Lake Corpus Christi 2	lake_monthly.csv
Added lakes: satellite area against gauge-derived area (published area table at the gauge level), unflagged months with both values: mean of water_area_km2 minus gauge_area_km2, bias, SD, months above, Spearman rank correlation of levels (rho), Pearson correlation of month-to-month changes (r), and the reference quality	Beyond the 10% bias rule: none. Per lake: Lake Mead +11.3 km2 (+3.6%; SD 5.8; above in 92 of 92; rho 0.95; r 0.65; good reference); Lake Powell +23.8 km2 (+8.0%; SD 9.1; above in 91 of 91; rho 0.98; r 0.90; good reference); Elephant Butte Reservoir -0.8 km2 (-2.2%; SD 2.1; above in 35 of 84; rho 0.99; r 0.95; fair reference); Falcon Reservoir -0.5 km2 (-0.5%; SD 6.2; above in 42 of 76; rho 0.95; r 0.71; good reference); Flaming Gorge Reservoir +3.5 km2 (+2.5%; SD 2.8; above in 56 of 66; rho 0.99; r 0.83; good reference); Amistad Reservoir +6.9 km2 (+6.4%; SD 4.9; above in 83 of 91; rho 0.99; r 0.88; good reference); Blue Mesa Reservoir +0.1 km2 (+0.4%; SD 1.1; above in 31 of 57; rho 0.99; r 0.98; good reference); Choke Canyon Reservoir -3.8 km2 (-7.6%; SD 4.0; above in 1 of 84; rho 0.96; r 0.48; good reference); Lahontan Reservoir +1.3 km2 (+4.9%; SD 4.7; above in 65 of 84; rho 0.98; r 0.87; weak reference); American Falls Reservoir -7.1 km2 (-4.7%; SD 14.8; above in 29 of 63; rho 0.98; r 0.93; weak reference); Lake Corpus Christi -1.9 km2 (-3.4%; SD 3.5; above in 30 of 86; rho 0.95; r 0.94; good reference). Upper Klamath Lake has no gauge-derived area.	lake_monthly.csv; Reclamation RISE and area-capacity tables; Texas Water Development Board and IBWC tables; USGS gauges
Added lakes in Texas: satellite area against the Texas Water Development Board published surface area, unflagged months: mean difference, SD	Falcon Reservoir -2.2 km2 (-1.9%; SD 5.9; n 76); Amistad Reservoir +5.4 km2 (+4.9%; SD 6.0; n 91); Choke Canyon Reservoir -3.8 km2 (-7.6%; SD 4.0; n 84); Lake Corpus Christi -2.1 km2 (-3.7%; SD 3.8; n 88)	lake_monthly.csv; waterdatafortexas.org
Added lakes: satellite area against the JRC Global Surface Water monthly history, 2019 to 2021 only, months where JRC saw at least 80 percent of the outline	Beyond the 10% bias rule used for the canyon test: Elephant Butte Reservoir -12.0%, Amistad Reservoir +10.3%. Per lake: Lake Mead +9.1 km2 (+2.7%; SD 9.2; n 24); Lake Powell +29.6 km2 (+9.0%; SD 11.4; n 25); Elephant Butte Reservoir -5.0 km2 (-12.0%; SD 5.9; n 28); Falcon Reservoir +1.2 km2 (+0.9%; SD 9.3; n 28); Flaming Gorge Reservoir +4.8 km2 (+3.3%; SD 3.7; n 21); Amistad Reservoir +12.9 km2 (+10.3%; SD 6.1; n 28); Blue Mesa Reservoir +0.7 km2 (+2.7%; SD 1.2; n 18); Upper Klamath Lake -1.8 km2 (-0.8%; SD 4.8; n 14); Choke Canyon Reservoir +5.0 km2 (+9.6%; SD 6.6; n 23); Lahontan Reservoir -0.2 km2 (-0.6%; SD 1.6; n 24); American Falls Reservoir -6.2 km2 (-3.8%; SD 7.4; n 21); Lake Corpus Christi +1.9 km2 (+3.0%; SD 2.9; n 28)	lake_monthly.csv; JRC Global Surface Water v1.4, doi:10.1038/nature20584
Tests of the AWEI-style estimate before it was adopted, 12 months each (lakes/results/new13/diag and lakes/results/new13/noncanyon/awei_extend_test.py)	Lake Powell: AWEI-style bias +6.5%, correlation of changes 0.95, rank correlation 0.97 (MNDWI: -35.6%, 0.77, 0.59). Lake Mead: +3.0%, 0.82, 0.94 (MNDWI: -12.1%, -0.17, 0.48). Lahontan, not blind: +5.9%, 0.97, 0.97 (MNDWI: -8.8%, 0.85, 0.98). Bear Lake, against JRC on 22 months: -6.4%, 0.37, 0.34 (rank correlation below 0.8, so it failed and is withheld).	run files in the repository; rules sections 12 and 13
Alternative definition, for comparison with other sources: the mean of the monthly percent differences between satellite and gauge-derived area (not used elsewhere)	Lake Mead +4%; Lake Powell +9%; Flaming Gorge +3%; Amistad +8%; Blue Mesa +1%; Lake Corpus Christi -2%; Lahontan +18%. Small lakes at low water give large monthly percentages, which is why this differs from the bias.	lake_monthly.csv

Limits and appropriate use

• Salt crust and shallow margins confuse spectral water indexes; flagged months are marked.
• Winter ice and snow limit Mono and the northern lakes in some months. Ice, snow, and cloud are never counted as dry lakebed.
• Open water, ice and snow, and not observed are separate categories. Do not treat a blank or ice-flagged month as drying.
• The reference outline is a hand-drawn polygon. It was not taken from a shoreline survey, and it is larger than the lake. The outlines are in data/lake_outlines.geojson.
• Great Salt Lake and Lake Abert values are open water inside the outline, which takes in more than the lake body. They run above the area from the USGS elevation-area tables (sizes are in the Validation table). On the Great Salt Lake arms the satellite change from 2019 to 2026 is also larger than the gauge-derived change, and the gap between the two is not constant. The cause is not established (see Key figures).
• The Great Salt Lake north change from 2019 to 2026 depends on the water estimate. On the same months it is -226.4 km2 with MNDWI and the Otsu threshold (the reported value), -120.7 km2 with MNDWI above 0, -191.9 km2 with MNDWI above 0.1 and -184.9 km2 with AWEI and its own Otsu threshold. The south-arm values are -277.7, -283.0, -293.0 and -289.2 km2. This release does not say which estimate is right. The four estimates are in data/water_estimates.csv.
• Salton Sea, January 2022: one catalog tile (T11SPS, scenes of 2022-01-25) has no files in the archive, which return HTTP 404. That day is built from the other three tiles.
• The latest month in the table is marked provisional. It can change when late or reprocessed scenes arrive.
• The 12 added lakes use outlines built from public hydrography and cut by eye at the upstream end of each reservoir. They are not surveyed pool limits, and arms beyond the cuts are not counted. The published full-pool area, its source and its URL are in the properties of data/lake_outlines.geojson (published_full_pool_km2, published_source, published_url; Upper Klamath Lake has none for the main lake alone).
• Lake Mead, Lake Powell, Amistad, Flaming Gorge, Blue Mesa and Lake Corpus Christi use the AWEI-style estimate. It does not match the gauge-derived reference exactly. Bias against the reference over unflagged months: Lake Mead +3.6% (months above 92 of 92); Lake Powell +8.0% (months above 91 of 91); Flaming Gorge +2.5% (months above 56 of 66); Amistad +6.4% (months above 83 of 91); Blue Mesa +0.4% (months above 31 of 57); Lake Corpus Christi -3.4% (months above 30 of 86).
• Lahontan uses the AWEI-style estimate too, for a different reason (see Method). Lahontan: AWEI-style bias +4.9% (months above 65 of 84); MNDWI bias -11.0%. The matched-month 2019 to 2026 change is -12.5% with the AWEI-style estimate and -12.6% with MNDWI.
• The AWEI-style index is not the published AWEI: it uses SWIR1 (B11) in both terms where the published no-shadow form uses SWIR2 (B12) in the second. Its validation here is ours: the 12-month tests and the full-run comparison with the gauge-derived area in Validation, not the original paper's.
• Lake Powell, Amistad, Flaming Gorge, Blue Mesa and Lake Corpus Christi have arms narrower than three pixels (canyon_limited is true). Pixels in those arms mix water and shore. Lake Mead is just under the line (canyon_fraction 0.0046).
• Reference quality is graded good, fair or weak for each added lake (reference_quality column). Graded below good: Upper Klamath Lake (weak), Lahontan (weak), American Falls (weak) and Elephant Butte (fair). Agreement is measured against a reference of that quality, so a weak reference limits what a match shows.
• Upper Klamath Lake has no gauge-derived area, because the published table also counts Agency Lake and flooded wetlands. Elephant Butte, Blue Mesa and Upper Klamath Lake months with algae scum, floating plants or wet mud are flagged and carry a wider range.
• In 2 Blue Mesa months (frozen Januaries) the widened low end would be below zero and is set to 0. Those months are flagged ice.
• 2 months are blank with quality_flag processing_failed: Lake Powell 2025-09, where one image day could not be read, and Lake Powell 2025-11, which did not finish within two hours. 14 months are blank with no_view because fewer than 20% of the outline was seen.
• Bear Lake was evaluated and withheld. Its AWEI-style estimate did not follow the JRC monthly history closely enough to pass the checks set before the test (correlation of changes 0.37, rank correlation of levels 0.34; lakes/results/new13/noncanyon/awei_extend_test.py), so no values are published and nothing is claimed for it.
• A month where area_high_km2 is more than 1.25 times water_area_km2 is flagged methods_disagree in every series. In the first seven series, the months flagged methods_disagree with a low estimate at or above 80% of the value, so flagged by this rule alone, are: Great Salt Lake north 3, Lake Abert 3, Great Salt Lake south 0, Salton Sea 0, Walker 0, Pyramid 0, Mono 0.
• The JRC monthly history ends in 2021, so the comparison with it covers 2019 to 2021 only. The Texas Water Development Board area is a table lookup at the gauge level, not a measurement.

Imagery

One Sentinel-2 true-color composite for each of the 19 series for each summer, 2019 to 2026 (20 m, EPSG:5070). The month of each composite is the month in the date column of imagery/index.csv. July is the default. A summer uses August or September where July had cloud or sun glint over the lake. In the first seven series a summer uses August where more than 5% of the July image was blank. Four summers still show cloud or sun glint over part of the lake: Choke Canyon Reservoir in 2019 and 2020 and Lake Corpus Christi in 2019 and 2020. This download has one WebP preview per image. The previews are in imagery/previews.zip. Unzip it inside imagery/ to get imagery/previews/, the folder that preview_path names. The full-size composites are not in the download. imagery/index.csv gives the share of blank pixels in each image; images marked partial (more than 5% of the image has no data): 0. See imagery/IMAGERY.md for the month of every image that is not from July.

Attribution: WebP previews burn credit into a bottom-right semi-transparent band. GeoTIFF/COG files never stamp text onto pixels; the same credit is in TIFF copyright/description tags and the index CSV. See imagery/IMAGERY.md.

Files and fields

File	Description	Bytes	SHA-256
data/lake_monthly.csv (CSV)	One row per lake (or arm) per month	282537	bd8464b04ce68064...
data/lake_monthly.parquet (Parquet)	Same rows, columnar format	110414
data/water_estimates.csv (CSV)	The four water estimates behind water_area_km2, area_low_km2 and area_high_km2, one row per lake and month	118032	02591a54000d7d82...
data/water_estimates.parquet (Parquet)	Same rows, columnar format	62677
data/reference_sources.csv (CSV)	One row per series, naming the gauge series, the area curve and who produced the curve	9384	ef60897cc8ea4217...
data/reference_sources.parquet (Parquet)	Same rows, columnar format	8037
data/lake_outlines.geojson	Reference outline for each series (GeoJSON, EPSG:4326). Open water is counted inside these polygons; outline_area_km2 is the polygon area.	7575414	acb27049ebde7a22...
imagery/IMAGERY.md	Imagery method and attribution notes	3119	43549ec5c307e6ec...
imagery/index.csv	Index of packaged imagery (paths, credits, status). preview_path names a file in previews.zip, which unzips inside imagery/ into previews/	64387	9a1e3ee5a79e8fd8...
imagery/previews.zip	Every WebP preview in one ZIP. Unzip it inside imagery/ to get imagery/previews/	46218867	9e52ef9297fd749e...
docs/RESEARCH_USE.md	Research-use notes built from the tables	38862	4efc20d71477d67f...
docs/questions/great-salt-lake-south-arm-july-open-water-change.md	Research-use notes built from the tables	1854	1ec660bfc81b4ab6...
docs/questions/great-salt-lake-north-arm-july-open-water-change.md	Research-use notes built from the tables	1596	691774d0611c548d...
docs/questions/salton-sea-july-open-water-change.md	Research-use notes built from the tables	1527	c67b479297a5dc4f...
docs/questions/lake-abert-july-open-water-change.md	Research-use notes built from the tables	1598	8d691f79a0b72074...

On Zenodo downloads, nested paths are flattened: data/foo.csv appears as data__foo.csv. On Kaggle the CSV files sit at the top level: foo.csv.

Fields

lake_monthly

One row per lake (or arm) per month

Field	Type	Unit	Description
lake_id	string		Stable lake identifier
month	string		Calendar month, YYYY-MM
water_area_km2	number	km2	Open-water area inside the reference outline (km2), from the monthly composite with the MNDWI Otsu threshold (the AWEI-style Otsu estimate where primary_estimate is aweish_otsu); ice/snow and cloud (not observed) are excluded
area_low_km2	number	km2	Lowest of the four water estimates for the month (km2)
area_high_km2	number	km2	Highest of the four water estimates for the month (km2)
seen_fraction	number		Share of outline pixels with at least one clear view in the month (0-1)
n_views	integer		Number of days with a Sentinel-2 scene in the composite
gauge_area_km2	number	km2	Area from the gauge level and the published elevation-area table (km2); blank where there is no paired gauge series
quality_flag	string		ok, low_coverage, bright_surface, ice, methods_disagree, no_view, processing_failed, algae_turbid, mud or jump (rules in Method; no_view is a month with fewer than 20% of the outline seen, processing_failed is a month that could not be computed; the first flag that applies, in the order processing_failed, no_view, ice, mud, algae_turbid, methods_disagree, jump, low_coverage, bright_surface); ice and no_view are not dry lakebed
provisional	boolean		True for the latest month, which can change when late or reprocessed scenes arrive
low_confidence	boolean		True when a plausibility rule flagged the month (value kept)
primary_estimate	string		Which estimate water_area_km2 uses: mndwi_otsu or aweish_otsu (the AWEI-style estimate described in Method; the name is historical). Blank for the first seven series, which use mndwi_otsu
method_low_km2	number	km2	Lowest of the four estimates before any widening (added lakes only)
method_high_km2	number	km2	Highest of the four estimates before any widening (added lakes only)
interval_widened	boolean		True when area_low_km2 or area_high_km2 is wider than the four-estimate range because of algae or wet mud (added lakes only)
uncertain_edge_km2	number	km2	Area of observed pixels whose MNDWI is within 0.10 of the month's threshold: wet mud, shallow edge or mixed pixels, neither confirmed water nor confirmed lakebed (added lakes only)
canyon_fraction	number		Share of the outline in arms narrower than 3 pixels (60 m), from the outline shape (added lakes only)
canyon_limited	boolean		True when canyon_fraction is 0.005 or more: pixels in narrow arms mix water and shore (added lakes only)
flag_ice	boolean		Snow or ice class above 5% of observed pixels, November to May, on lakes that can freeze (added lakes only)
flag_mud	boolean		Elephant Butte only: uncertain edge at least 25% of the area, or an Otsu threshold outside -0.10 to 0.30
flag_algae_turbid	boolean		10% or more of the water pixels have NDVI above 0 (scum, floating plants or very turbid water); Upper Klamath Lake, Blue Mesa and Elephant Butte only
flag_methods_disagree	boolean		the low-estimate rule (area_low_km2 under 80% of the value) applies to the MNDWI series only, and AWEI-style series are governed by the 125% high-estimate rule (area_high_km2 above 125% of the value), which applies to every series, including the first seven.
flag_jump	boolean		More than 10% from the median of the month and its two neighbours, and the reference area does not show half of that change (added lakes only)
flag_low_coverage	boolean		seen_fraction below 0.8 (added lakes only)
flag_bright_surface	boolean		June to October with more than 15% of observed pixels in snow or ice class (added lakes only)
reference_quality	string		good, fair or weak: how much weight the gauge-derived reference can carry (survey age, datum, how much the reference area changes). Added lakes only
jrc_area_km2	number	km2	Water area inside the outline from the JRC monthly history, 2019 to 2021, months where JRC saw at least 80% of the outline (added lakes only)
twdb_area_km2	number	km2	Surface area published by the Texas Water Development Board at the gauge level (Falcon, Amistad, Choke Canyon, Corpus Christi); a table lookup, not a measurement

water_estimates

The four water estimates behind water_area_km2, area_low_km2 and area_high_km2, one row per lake and month

Field	Type	Unit	Description
lake_id	string		Stable lake identifier
month	string		Calendar month, YYYY-MM
area_mndwi_otsu_km2	number	km2	Open-water area with MNDWI above its Otsu threshold (km2); equals water_area_km2 except where primary_estimate in lake_monthly is aweish_otsu
area_mndwi_0_0_km2	number	km2	Open-water area with MNDWI above 0 (km2)
area_mndwi_0_1_km2	number	km2	Open-water area with MNDWI above 0.1 (km2)
area_awei_otsu_km2	number	km2	Open-water area with the AWEI-style index above its own Otsu threshold (km2)
low_confidence	boolean		Same flag as in lake_monthly

reference_sources

One row per series, naming the gauge series, the area curve and who produced the curve

Field	Type	Unit	Description
series	string		Series identifier, the same value as lake_id in lake_monthly
agency	string		Agency that supplies the gauge series for this row
series_id_or_url	string		Gauge series identifier or the page the series was read from
curve	string		Area curve used for gauge-derived area, or not applicable when no curve is used
curve_date	string		Date of the curve, or not applicable when no curve is used
curve_producer	string		Who produced the curve, or not applicable when no curve is used
notes	string		Public facts about the series and the curve

Example use

import pandas as pd

monthly = pd.read_csv("data/lake_monthly.csv")
# Walker Lake trend in open-water area since 2019.
walker = monthly[monthly.lake_id == "walker-lake"].dropna(subset=["water_area_km2"])
print(walker[["month", "water_area_km2", "area_low_km2", "area_high_km2"]].tail(12))

API:

GET /v1/lakes

GET /v1/lakes/{lake_id}/monthly

Sources and attribution

Input	Role	Terms
Copernicus Sentinel-2 Level-2A surface reflectance	Measured	Open under the Copernicus Sentinel data terms
Microsoft Planetary Computer (Sentinel-2 STAC catalog and signed asset URLs)	Access path for Sentinel-2 L2A	Planetary Computer terms of use; upstream Copernicus Sentinel terms apply to the imagery
US Geological Survey. Great Salt Lake south USGS-10010000; Great Salt Lake north USGS-10010100; Salton Sea USGS-10254005; Pyramid Lake USGS-10336500; Walker Lake USGS-10288500; Lake Abert USGS-10387150; Lahontan Reservoir USGS-10312100; Upper Klamath Lake USGS-11507000; Choke Canyon Reservoir USGS-08206900; Lake Corpus Christi USGS-08210500. https://api.waterdata.usgs.gov/ogcapi/v0/collections/daily/items	Gauge level or storage series	Public domain
USGS Great Salt Lake elevation-area-volume tables, DOI 10.5066/P9DGG75W	Converting Great Salt Lake north and south arm gauge level to area	CC0
USGS elevation-area-volume tables for selected Great Basin lakes, DOI 10.5066/P147WRTT	Converting Walker and Lake Abert gauge readings to area	CC0
US Bureau of Reclamation. Lake Mead https://data.usbr.gov/catalog/4370/item/6123; Lake Powell https://data.usbr.gov/catalog/2362/item/508; Flaming Gorge Reservoir https://data.usbr.gov/rise/api/result?itemId=341; Blue Mesa Reservoir https://data.usbr.gov/rise/api/result?itemId=78; Elephant Butte Reservoir https://data.usbr.gov/rise/api/result?itemId=332; Lahontan Reservoir area-capacity table https://data.usbr.gov/rise/content-rise-public/rise/catalog-item/binary/Lahontan%20Reservoir%20Bathymetric%20Survey%20Report%202004_508.pdf; American Falls Reservoir https://www.usbr.gov/pn-bin/daily.pl?list=amf%20fb,amf%20af	Gauge level and area-capacity tables	Federal data marked provisional; reuse terms to be confirmed with Reclamation. Provisional real-time data; reuse terms to be confirmed with Reclamation.
US Section, International Boundary and Water Commission. Amistad Reservoir https://waterdatafortexas.org/reservoirs/individual/amistad; Falcon Reservoir https://waterdatafortexas.org/reservoirs/individual/falcon	Level records for Amistad and Falcon	Level records from the US Section, hosted by the Texas Water Development Board; IBWC terms to be confirmed
Texas Water Development Board. Choke Canyon Reservoir https://waterdatafortexas.org/reservoirs/individual/choke-canyon; Lake Corpus Christi https://waterdatafortexas.org/reservoirs/individual/corpus-christi	Reservoir figures for Choke Canyon and Lake Corpus Christi	Copying and distribution freely permitted; acknowledgment requested. https://waterdatafortexas.org/policies. Water Data for Texas figures, with US Geological Survey levels.
USGS National Hydrography Dataset, https://hydro.nationalmap.gov/arcgis/rest/services/nhd/MapServer/12	Lake polygons for the outlines of the added lakes	Public domain
Global Surface Water, Monthly History, European Commission Joint Research Centre. Pekel, J.-F., Cottam, A., Gorelick, N. and Belward, A. S. High-resolution mapping of global surface water and its long-term changes. Nature 540, 418-422 (2016), doi:10.1038/nature20584.	Comparison statistics	JRC data is used only for comparison statistics.

How to cite

Heimdall Research (2026). Western Terminal Lakes (Version 1.1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.23045007 Concept DOI https://doi.org/10.5281/zenodo.23045006.

@misc{western_terminal_lakes_1_1,
  author    = {Heimdall Research},
  title     = {Western Terminal Lakes},
  year      = {2026},
  version   = {1.1},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.23045007},
  url       = {https://doi.org/10.5281/zenodo.23045007}
}

Also cite the upstream sources listed above.

Changes

• Adds 12 lakes and reservoirs, one series each, to the 7 series of version 1.0. The table has 19 series on 18 lakes and reservoirs (Great Salt Lake has two arms). Corrects the 1.0 areas: version 1.0 divided Sentinel-2 values by 10000 and did not remove the +1000 offset that processing baseline 04.00 and later scenes carry (scenes from 2022-01-25). Lake Mead, Lake Powell, Amistad, Flaming Gorge, Blue Mesa and Lake Corpus Christi use an AWEI-style index because the MNDWI estimate left out water in the canyons. Lahontan uses that index too. The 12 months used for Lahontan had already been seen, so that check is not blind. Flag rules: on an MNDWI series a month is flagged when area_low_km2 is under 80% of the value, and on every series a month is flagged when area_high_km2 is above 125% of the value. 1.0 numbers should not be used.
• Version 1.0 is the table public on Kaggle on 2026-09-30 (sha256 98b1d505). Every lake-month of the first seven series was rebuilt, 651 rows.
• water_area_km2, 2022 to 2026 (399 rows): changed by a median of 13.54 km2, by up to 856.8 km2; 379 rows moved by more than 1 km2 and 127 by more than 5%. In 1.0 the lowest estimate was as low as 0.056 of the reported area from 2022 on; it is now at least 0.899 in every month that is not flagged.
• water_area_km2, 2019 to 2021 (252 rows): median change 0.05 km2; 30 rows moved by more than 1 km2, the largest 254.4 km2 (Great Salt Lake north 2019-04, flagged). 194 of the 651 rows in 1.0 had included a scene from the first day of the next month; the window now ends on the last day of the month. Salton Sea 2019-03 went from 810.2 to 860.1 km2 for that reason (1.0 included the 2019-04-01 scene, which moved the water threshold from 0.258 to 0.209); the new value is in line with the months before and after and is not flagged.
• 2019 to 2026 change, 1.0 against this release: Great Salt Lake south -205.2 km2 (-9.9%) now -277.7 (-13.4%); Great Salt Lake north -102.4 (-7.3%) now -226.4 (-16.2%); Salton Sea -38.3 (-4.5%) now -75.9 (-8.8%); Walker Lake +1.2 (+1.1%) now -3.3 (-2.8%); Pyramid Lake +5.5 (+1.3%) now -3.6 (-0.9%); Mono Lake +6.2 (+3.4%) now +0.1 (+0.0%); Lake Abert +8.6 (+6.6%) now -6.6 (-5.0%). Version 1.0 averaged different months in the two years; this release uses the same months.
• low_confidence changed on 19 rows of the first seven series (16 newly flagged, 3 no longer flagged). A fourth rule, quality_flag methods_disagree, flags months where area_low_km2 is below 80% of water_area_km2 or area_high_km2 is above 125% of it. Gauge-derived area is unchanged except Walker Lake 2026-09 (provisional USGS storage, 0.02 km2).
• Added: data/lake_outlines.geojson, data/water_estimates.csv (the four water estimates behind the value and its range), the gauge-derived change beside the satellite change for Great Salt Lake, documentation for n_views, provisional and low_confidence, and a blank-share column in imagery/index.csv. The summer images of the first seven series were rebuilt: three of the 1.0 images were 45% to 86% blank and marked ok. The 12 added series have summer images too, so every series has one image for each summer from 2019 to 2026.
• Added: 12 series (Lake Mead, Lake Powell, Elephant Butte, Falcon, Flaming Gorge, Amistad, Blue Mesa, Upper Klamath, Choke Canyon, Lahontan, American Falls, Lake Corpus Christi), 1116 lake-months, with outlines from public hydrography, a gauge-derived reference where a published area table exists, and new columns in lake_monthly (primary_estimate, canyon and ice, algae and mud flags, reference quality, JRC and Texas Water Development Board areas). Seven of them use the AWEI-style estimate as the main value. Bear Lake was evaluated and withheld. The table now has 1767 rows.
• Added data/reference_sources.csv: one row per series, with the gauge series, the area curve and who produced the curve.

Access and contact

Access: download the files here, or use the same data through the Heimdall Research API and MCP server (heimdallresearch.com).

Corrections and questions: https://heimdallresearch.com/contact/collaborate/.