Bhote Koshi Flash Flood Impact and Exposure Atlas

A second flood wave ran the corridor on 30 August, and Beijing told Nepal it was not a lake burst. A surge passed Dhunche at about 03:30 NPT; levels had receded by afternoon (The Rising Nepal / RSS). At about 11:30 NPT the Trishuli took the 130 m Ghyalchowk–Charaudi suspension bridge in Gorkha — the third suspension bridge lost, and itself the replacement route for the Dhawadighat bridge lost on 26 August (RSS). Nepal asked Beijing and was told the increased upstream flow came from an avalanche, not a breach (MoFA via The Rising Nepal); NDRRMA chief executive Dharam Raj Uprety confirms Sunday’s phone alerts followed a Chinese notification of increased flow (AP via NPR). Earlier the same day the Prithvi Highway subsided at Krishnabhir in Dhading, blocking the Muglin–Kathmandu section (The Kathmandu Post). Both failures lie beyond the 62.4 km corridor mapped here and outside all four EMS AOIs, and neither is added to the damage register: no source gives a coordinate, and this page does not geolocate from news text.

The lower lake has fully drained — but the dam is described as less stable, not more. China’s Ministry of Water Resources said on Sunday 30 August that the impounded water has been almost completely discharged and is flowing normally through natural channels, its hydrological emergency monitoring team cited by China Daily (The Kathmandu Post, 30 August). Its Saturday-evening remote-sensing analysis had already found the lake reduced, with risk manageable; G216 repair at Gyirong Port resumed on that basis (CGTN, 30 August). Its measured area fell from 120,000 m² on 27 August to 99,000 m² on 29 August (MWR via CCTV and Reuters); NDRRMA’s independent 0.11 km² for 27 August, from Planet Labs and Landsat via the Stimson Center, agrees to within 9 percent. The dam itself came down carrying ice and snow, leaving it high in water content and soft underfoot, and China Anneng now runs slope radar and drones on it around the clock so downstream teams can be pulled if it moves.

The upper feature is not a lake — MWR calls it an impact crater, and it now has an area, a volume and an altitude. China’s Ministry of Water Resources places it about 5.6 km upstream of the barrier lake at roughly 3,700 m, some 800 m higher than the lake, and at 11:23 on Sunday 30 August measured its water surface at 108,000 m² — down 19,000 m² from 07:00 — holding about 1,400,000 m³, with water flowing out continuously about 180 m below its dam (Xinhua, 30 August 20:28 Beijing time). Three further water-accumulation areas within 2.5 km downstream total under 70,000 m². The same MWR assessment reaches this page twice with two different names and two different days. China Daily, relayed by KathmanduPati on 31 August, calls the feature an erosion pit rather than an impact crater, places it on the Chhuchen Khola above its confluence with the Purepu Tsangpo, and dates the 108,000 m² reading to 11:23 on Saturday 29 August, not Sunday 30 August. Every figure matches — 1.4 Mm³, 5.6 km, 800 m, the 19,000 m² fall from 07:00, the three areas under 70,000 m² — so this is one assessment, not two, and the day it was taken is unresolved. The Sunday reading is retained here because Xinhua carried it on a 30 August dateline; the conflict is logged rather than settled. MWR says the crater and those three areas still pose risk hazards and require continued monitoring, so the hazard window does not close with the lower lake. This supersedes Chen Hongqi of the Ministry of Natural Resources’ Geological Disaster Technical Guidance Center, who on 29 August put the water at the foot of the collapse at over 120,000 m², depth not established and no visible outlet (Reuters), and it is the first agency reading against which Nitesh Khadka of the Chinese Academy of Sciences can be set: his 30 August account of more than half the upper water having run out (The Kathmandu Post) is neither confirmed nor contradicted by a 15 percent area loss over four hours. A Chinese geohazards expert warns over 100 glacial lakes in the area could pose a threat. MWR describes the lower lake as straddling the China–Nepal border, which makes the Nepal-versus-China placement dispute a false dichotomy rather than a settled question.

A third barrier lake formed on 29 August on the Nepali side, and has already breached. MWR reports it about 2.8 km downstream of the previous one, formed by a landslide on the Nepali side, its dam since breached by upstream flow and the channel back in a flowing state on drone footage and panoramic imagery (Xinhua, 30 August). That sits awkwardly against what Nepal was told — that Sunday’s increased flow came from an avalanche, not a breach — and no source reconciles the two accounts or timestamps the breach. It is not mapped here: no coordinate, area or volume has been published for it, and this atlas does not place a feature from a distance-along-river.

Nothing observed shows any of these features. None is mapped in this package: the marker is a single inferred point for the lower feature (medium confidence) and the two pool polygons are terrain-modelled screening surfaces, not observed water. Both overstate the lower lake on every published date — 19.1 ha and 29.5 ha against 9.9 ha and 12.0 ha — and have not been redrawn, because MWR published an area, not a geometry, and this atlas does not synthesise geometry from a scalar. No Sentinel-1 or Sentinel-2 acquisition covers the marker after the 28 August radar non-detection (see Observed Radar, below) and the 29 August optical pass that was 100.0 percent cloud within 300 m. Copernicus EMS EMSR927 has mapped nothing at any of them — all lie outside all four AOIs — and has delivered nothing new: AOI01 v1, AOI02 v2 and AOI03 v1 are unchanged, the first monitoring product for AOI03 Bidur is now in production on PeruSAT-1 imagery of 27 August and Resurs-P imagery of 28 August with delivery expected 31 August 02:50 UTC, and AOI04 Bharatpur’s acquisition remains slipped to 31 August. The full day-by-day record of how this hazard was reported from 27 August onward is kept in the errata log below.

The exposure envelope on this page is derived from terrain, not from the 26 August flood, so it applies to a second outburst down the same channel as readily as to the first. It is a screening surface and no substitute for official warnings — follow NDRRMA and local authority instructions.

Sentinel-1D acquired scene S1D_IW_GRDH_1SDV_20260828T122141_20260828T122206_004326_007FA4 at 28 August 12:21:54 UTC (18:07 NPT), the first acquisition of any kind over the inferred barrier-lake site since the fully clouded 27 August optical pass, and the first that cloud cannot block. It postdates the reported breach. This atlas has terrain-corrected it and searched a 39 × 55 km envelope spanning the whole mapped upstream drainage. The result is a non-detection.

The surface at the marker reads −8.7 dB in VV and is radiometrically identical to 16 August. Against a 2.23 dB noise floor measured from four pre-event 12-day pairs on the same orbit, a water surface as rough as −12 dB would have been detected at z = 14–24, and calm water at −22 dB at z = 74. Everything the search did flag totals 7.4 ha across 11 clusters, ten of them at 5,280–6,678 m and consistent with wet snow at the snowline; all eleven are published in the package rather than discarded.

The controls matter more than the result. Geocoding reproduces the product's own geolocation grid to 1.89 m over 210 points. The reference scene is the same relative orbit (85, ascending, 12-day repeat), so layover and shadow are pixel-identical and cancel. Run against the Copernicus EMS EMSR927 observed damage at Syapru Besi and Timure, the same detector finds significantly elevated change inside the mapped extent (Mann-Whitney p = 5.1×10^-21; >6 dB changes enriched 2.63×) — it sees real ground change where ground change is known to have happened.

This constrains where the lake is not. It does not prove no lake exists. C-band radar cannot separate a rough, turbulent or debris-covered water surface from land; 20.1% of the envelope is unusable for layover and shadow; the marker is an inferred location of medium confidence; and any impoundment beyond 28.60°N / 85.80°E was outside the search. The marker has not been moved and the modelled pool polygons stand — as modelled scenarios, now uncorroborated at this location.

A second pass has now looked from the other side, and found the same nothing. Sentinel-1D acquired the site again at 2026-08-31 00:10:50 UTC, this time descending on relative orbit 121 — a different track from the 28 August ascending pass, so layover and shadow fall on the opposite valley walls and a lake hidden by geometry on one pass should be visible on the other. Processed through the same Range-Doppler chain against its own 12-day reference of 19 August: geocoding reproduces the product geolocation grid to 1.85 m over 210 points, 87.7 percent of the envelope is usable (7.4 percent layover, 12.3 percent shadow), and the change noise floor is 2.50 dB from four pre-event 12-day pairs on this same track. At the marker the surface reads +0.29 dB VV over 335 valid pixels — 0.12 sigma, 95 percent CI +0.02 to +0.56 dB, which excludes any drop — and the two modelled pool footprints read +0.03 and +0.12 dB. Calm water at −22 dB would sit 163 standard errors below what was measured. The search flags 15 clusters totalling 67.24 ha, and every single one lies at 5,360–5,970 m: wet snow at the snowline. Not one candidate lies below 4,500 m, where an impoundment would have to be. The larger snow signal than 28 August is the acquisition hour — 00:10 UTC is 05:55 Nepal time, pre-dawn, against 18:07 on 28 August. The 1.50 ha low-elevation cluster flagged at 3,778 m on 28 August did not recur: the same site reads −2.38 dB here, short of the −4 dB threshold.

One control was lost on this pass, and the data says so rather than leaving it to be inferred. None of the Copernicus EMS damage polygons falls inside the 31 August slice’s data footprint — their pixels return sigma0 of exactly zero — so the Mann-Whitney and enrichment test that validated the 28 August chain could not be repeated. Sensitivity here rests on the four same-track pre-event pairs instead, and the sar_search_envelope_20260831 layer carries that limitation in its own positive_control field. Two independent geometries now agree, which is the strongest statement this atlas can make about a feature no camera has ever seen — and it is still a statement about where the lake is not.

• corrected 2026-10-18
  
  The processing code behind the two Sentinel-1 radar non-detections (28 and 31 August) was not retained when it was written, and has now been re-implemented from the published process steps. more ▾

  The Range-Doppler terrain correction, geolocation validation, layover and shadow masking, 12-day change and open-water search were written and run inside the processing environment on 28 and 31 August; the rasters, tables and FGDC process steps reached the package, the script itself did not. On 18 October the chain was rebuilt from those process steps and re-run on the 28 August pair against the same scenes (S1D_IW_GRDH_1SDV_20260828T122141 and its 16 August reference) and the same Copernicus GLO-30 terrain. Geolocation against the product grid: maximum error 1.89 m over 210 points (published 1.89 m). Change at the inferred barrier-lake marker: −0.17 to −0.45 dB (published −0.17 to −0.43), 0.08 to 0.21 standard deviations of the noise floor; a rough water surface would sit 15 to 31 standard errors below the measured value. Positive control against the EMS damage at Syapru Besi and Timure: changes above 6 dB enriched 3.2-fold (published 2.63-fold), Fisher p = 7e−24. Every flagged candidate larger than half a hectare sits at 5,280 m or above, at the snowline; none at impoundment elevations. The published layers stand unchanged. The re-implementation differs from the original in the layover/shadow formulation (it uses the 31 August integer-line test throughout, giving 75.5 percent usable against the 79.9 percent of the coarse-binned 28 August version) and in taking the noise floor from the pair itself rather than from four pre-event pairs; neither changes any conclusion. The program rules now require analysis code written during a run to be saved before the run ends. Nothing spatial changes and no Zenodo version is minted.
• corrected 2026-09-03
  
  Nepal death toll revised again: 1,127 → 1,252, and the missing count falls 4,858 → 4,216 — the headline authority returns to NDRRMA, which is now the higher of the two on deaths and the lower on missing. more ▾

  NDRRMA as of noon on 3 September gives 1,252 bodies recovered from eight districts, a table summing exactly to it — Chitwan 355, Nawalparasi East 218, Nawalparasi West 208, Nuwakot 177, Rasuwa 127, Gorkha 69, Dhading 60, Tanahun 38 — with 4,216 unaccounted for, of whom only 95 of the 1,252 recovered have been identified and returned to families (The Kathmandu Post, 3 September). Published components of the missing: about 900 hydropower workers, 1,558 from Nuwakot, 865 from Rasuwa, 583 foreign nationals, 45 Nepali Army, 25 police, 13 Armed Police Force, plus customs, immigration and local-government staff (KathmanduPati, 3 September); the categories overlap and are not additive against the headline. The fall in the missing count is a change of register, not a recovery of 642 people — the superseded 4,858 is Nepal Police at 05:00 NPT on 2 September and the 4,216 is NDRRMA, the two registries having stood 942 apart yesterday; no source reconciles them. NDRRMA's table carries no India entry, so the 9 bodies recovered across the border that entered the police tally on 2 September are outside this one. Rescued is given as 11,913 by the Post and 11,993 by KathmanduPati against the same update, and security personnel deployed as more than 21,000 and 21,402 — both unexplained. China's toll moves for the first time since 30 August: 16 → 21 dead, 546 → 541 missing, the Xizang regional government as of noon on 2 September via Xinhua. No spatial change; no Zenodo version.
• open 2026-09-03
  
  The corridor is under a fresh flash-flood warning, and MWR has given the crater a river name — the Cuojian — that no Nepali source uses. more ▾

  Nepal's Flood Forecasting Division puts the Bhotekoshi and Trishuli at high risk of flash flooding on 3 September with rain continuing in the upper reaches and water levels expected to rise significantly through Rasuwa and Nuwakot (The Kathmandu Post, 3 September). China's Ministry of Water Resources says it is now giving Nepal daily updates on the barrier lake above Gyirong Port, on cascading risks from the Cuojian River impact crater in Nepal, on glacial-lake risk in the Gyirong Zangbo basin and on new barrier-lake formation, and warns the geological setting remains highly unstable with further landslides and rockfalls possible (Xinhua, 2 September). Cuojian is a fourth label for the upper feature, after impact crater, erosion pit and the Chhuchen Khola placement; it is logged, not resolved, and it still has no published coordinate, so nothing is mapped from it. Hydrological stations destroyed on 26 August are still being rebuilt. No spatial change; no Zenodo version.
• open 2026-09-03
  
  A fourth post-event optical pass, and a fourth total cloud failure — 100.0 percent cloud within 300 m of the marker. more ▾

  Sentinel-2B acquired the site at 2026-09-03T05:00:50Z; the SCL band returns cloud for 716 of 716 pixels within 300 m, 7,847 of 7,860 (99.8 percent) within 1 km and 46,410 of 49,080 (94.6 percent) within 2.5 km. Scene-level cloud cover reads 69.5 percent and is again unrepresentative of the site. No Sentinel-1 acquisition since the 31 August descending pass, so there is still no observed water extent for the barrier lake or the crater. EMS EMSR927 re-queried: AOI01 v1, AOI02 v2, AOI03 MONIT01 v1 and AOI05 v1 unchanged; AOI05 MONIT01 and AOI06 Kyundi still W with no acquisition; AOI04 Bharatpur slips a sixth time — the 3 September 02:25 UTC image is gone from the tasking and a 4 September 05:09 UTC Legion acquisition replaces it, delivery expected 18:09 UTC. No spatial change; no Zenodo version.
• corrected 2026-09-02
  
  NDRRMA’s registry moves 1,050 → 1,114 dead; the police headline of 1,127 stands, and for the first time the gap between the two registries is fully accounted for. more ▾

  NDRRMA’s 09:00 NPT bulletin of 2 September gives 1,114 bodies recovered — Chitwan 348, Nawalparasi East 217, Nawalparasi West 190, Nuwakot 155, Gorkha 66, Dhading 59, Tanahun 38 and Rasuwa 41, summing exactly — with 3,916 missing (unchanged), 292 injured and 11,993 rescued (OnlineKhabar, 2 September). Seven of the eight districts are identical to the police table of 05:00 NPT; the 13-body difference is exactly Rasuwa (45 against 41) plus the 9 recovered in India, which NDRRMA does not carry. The missing counts remain 3,916 (NDRRMA) against 4,858 (police), 942 apart. Wikipedia now carries 1,114 / 3,916, the NDRRMA figures. Headline unchanged at 1,127 / 4,858. No spatial change; no Zenodo version.
• open 2026-09-02
  
  EMS AOI04 Bharatpur slips a fifth time — the 2 September Legion image is gone from the tasking and a 3 September 02:25 UTC acquisition replaces it, delivery expected 15:25 UTC. more ▾

  Re-queried 13:00 UTC (EMSR927): AOI01 v1, AOI02 v2, AOI03 MONIT01 v1 and AOI05 v1 unchanged; AOI05 MONIT01 and AOI06 Kyundi still status W with no acquisition. No Sentinel-1 or Sentinel-2 pass over the barrier lake since the 1 September optical pass. No spatial change; no Zenodo version.
• corrected 2026-09-02
  
  A shapefile bundle is now offered alongside the GeoPackage package — 23 layers, 1.7 MB, a format export and nothing more. more ▾

  Every vector layer in the v13 package is exported to ESRI Shapefile with a field-name crosswalk (10-character limit), booleans as integers (1/0, −1 untested), text over 254 characters truncated and flagged, and mixed-geometry layers split by type; 10 EMS road records with null geometry are dropped from the shapefile and remain in the GeoPackage. The hot_extent files stay ODbL and the ems_* files carry Copernicus EMS attribution terms, as in the package. No geometry, attribute or figure changed and the GeoPackages remain the citable data, so no Zenodo version; the bundle will be folded into the archived package at the next version that has a spatial reason to exist. Built with scripts/nepal_flood_build_shp.py, kept in the repository for reuse.
• corrected 2026-09-02
  
  The page map now draws the Copernicus EMS observed extent for all four AOIs — the first observed geometry on this map, seven days in. more ▾

  Until now the map showed only Keystone-modelled surfaces (envelope bands, pool polygons) and points; the EMS polygons lived in the package alone. A toggleable layer is added from a display copy of ems_observed_event simplified at 4 m (178 KB against 3.7 MB; areas within 0.2 percent of the packaged layer, which is unchanged and remains the citable geometry). Corridor AOIs render amber; Phosretar renders purple to mark that it is not tested against the envelope. Presentation only: no package file, figure or Zenodo version changed.
• corrected 2026-09-02
  
  Nepal death toll revised again: 1,003 → 1,050 → 1,127, and the missing count jumps 3,916 → 4,858 — two registries, and the headline authority changes from NDRRMA to Nepal Police. more ▾

  Nepal Police at 05:00 NPT on 2 September give 1,127 bodies recovered with a district table summing exactly to it — Chitwan 348, Nawalparasi East 217, Nawalparasi West 190, Nuwakot 155, Gorkha 66, Dhading 59, Rasuwa 45, Tanahun 38 and India 9, the first time bodies recovered across the border enter a Nepali tally — with 4,858 missing, 6,541 rescued, 1,113 post-mortems completed and 911 unidentified bodies posted online (The Kathmandu Post, 2 September). Between the two sits NDRRMA’s first situation report of 1 September: 1,050 dead with a table summing exactly to it — Chitwan 321, Nawalparasi East 216, Nawalparasi West 170, Nuwakot 142, Gorkha 65, Dhading 57, Rasuwa 41, Tanahun 38 — and 3,916 missing, 11,814 rescued, 292 injured, 1.6 million affected (The Kathmandu Post, 1 September). Against the 1,003 of 13:00 NPT the +47 to 1,050 is entirely Nuwakot, 95 → 142, the district this page noted had not moved for three days. The two agencies’ missing counts — 3,916 and 4,858 — are 942 apart and are not reconciled by any source; neither is averaged, both are published, and the police figure is adopted as the later one. Wikipedia’s lead now carries 1,050 on The Guardian and its body 1,010; both are superseded. China unchanged at 16 dead and 546 missing. Casualty figures are not stored in the spatial data; the Zenodo version minted today is for the EMS layers below, not for this.
• corrected 2026-09-02
  
  Copernicus EMS delivered a fourth area of interest, AOI05 Phosretar — 422.56 ha of observed extent downstream of the corridor — and it is added to the package without touching the validation figures, because the envelope is not defined there. more ▾

  EMSR927 delivered AOI05 GRA_PRODUCT v1 at 2026-09-02T03:21:15 UTC, photo-interpreted from Pleiades (31 August 05:27 UTC) and Pleiades Neo (05:01 UTC), covering 23.9 km² of the Trishuli between Devighat and the Prithvi Highway (EMSR927). It grades 563 buildings (264 Destroyed, 69 Damaged, 230 Possibly damaged), 21 bridges (11 Destroyed, including the Keurintar Hanging Bridge) and 395 road segments with 10.44 km Destroyed, 1.1 km of it Prithvi Highway. All are appended to ems_observed_event, ems_damage_buildings, ems_damage_roads and ems_damage_bridges as aoi_number 5. Why the 97.0 percent does not move: the AOI begins 0.99 km below the channel terminus at Devighat, where the height-above-channel raster has no channel reference — its 424 residual pixels there read a median of −3.5 m. A naive overlay would have reported 19.8 of 35.8 ha (55.2 percent) inside the 80 m band; that is a raster-edge artefact and is withheld. Every AOI05 record carries env_tested = False and null envelope fields, the corridor tables are relabelled ‘AOIs 1–3’, and 942.00 ha / 97.0 percent stand unchanged. Schema additions: env_tested; env_tested_ha, env_test_note; geom_source, span_km (EMS delivered these bridges as lines, carried as centroids). Also at this query: AOI04 Bharatpur has its acquisition at last (Legion, 2 September 05:24 UTC, delivery expected 18:24 UTC); AOI05 monitoring 01 and a new AOI06 Kyundi in Chitwan are declared, status W. No new Sentinel pass over the barrier lake since 1 September. Spatial data changed, so a new Zenodo version was minted — v13 — reached through the concept DOI 10.5281/zenodo.22118798.
• corrected 2026-09-01
  
  Nepal death toll passes 1,000: 939 → 987 → 1,003 in one day, and the two 1 September figures are the same registry four hours apart, not a conflict. more ▾

  NDRRMA at 09:00 NPT gives 987 with a district table summing exactly to it — Chitwan 321, Nawalparasi East 216, Nawalparasi West 170, Nuwakot 95, Gorkha 65, Dhading 55, Tanahun 38, Rasuwa 27 (The Kathmandu Post, updated 11:38 NPT). NDRRMA at 13:00 NPT gives 1,003 with the same table and two cells moved: Rasuwa 27 → 41 and Dhading 55 → 57. Those two increments are the whole +16 (ANI, 17:10 IST, fetched first-hand). Against the 903 table of 31 August the full +100 also decomposes: Chitwan +49, Nawalparasi West +19, Rasuwa +17, Nawalparasi East +9, Gorkha +3, Dhading +2, Tanahun +1 — and Nuwakot +0, unmoved at 95 for three days while carrying 1,558 missing. Missing falls 3,925 → 3,916, entirely in the foreign-national component, 592 → 583; the Nepali component holds at 3,333 and hydropower workers at 639. The Nuwakot registry conflict logged 31 August has closed: NDRRMA publishes 1,558, superseding AFP's 1,158 and ANI's 1,548. Rasuwa missing stays 865. Security-force missing is unchanged but disaggregated for the first time — 45 Nepal Army, 25 Nepal Police and 13 Armed Police Force (the 38 'police'), 15 Customs and 3 Immigration (the 18) — plus 3 Langtang National Park and 3 Quarantine Office newly itemised. Rescued 10,451 → 11,884. A fourth figure is in circulation and is NOT adopted here: The Guardian's 1 September live blog is cited for 1,010 dead in Nepal and carried into the English Wikipedia lead and infobox; theguardian.com is not reachable from this page, so the 1,010 is recorded but not published and its timestamp is unknown. A second Wikipedia figure is stale rather than disputed: its infobox carries 1,473 injured on a 28 August BBC citation, against NDRRMA's 279 injured (111 in treatment at 09:00, 168 discharged by 13:00) — a five-fold gap that no source reconciles. China unchanged at 16 dead and 546 missing, as of 18:00 on 30 August. Casualty figures are not stored in the spatial data; no Zenodo version.
• open 2026-09-01
  
  The cloud over the barrier lake has broken for the first time in six days — and there is no water there. more ▾

  Sentinel-2C acquired the marker at 2026-09-01T05:10:46 UTC. Scene-level cloud cover reads 87.9 percent, but at pixel level the SCL band returns cloud or cloud shadow (classes 3, 8, 9, 10) on 564 of 716 pixels within 300 m of 28.31249 N, 85.55406 E — 78.8 percent, against 100.0 percent on all three earlier optical passes — and on 6,142 of 7,860 within 1 km. The same code reproduces the published 27, 29 and 31 August figures exactly (100.0 / 100.0 / 100.0 percent within 300 m), so the break is in the sky, not in the method. That leaves 152 pixels within 300 m not classified as cloud — 96 vegetation, 32 bare, 24 SCL class 7 'unclassified' — and 1,718 within 1 km. On them: zero SCL water pixels, and MNDWI (green against SWIR-1, 20 m) above 0.3 on 0 of 152 within 300 m and 7 of 1,718 within 1 km, which is 0.28 ha in scattered pixels. Median MNDWI within 300 m is −0.348 and the maximum is 0.055, nowhere near open water. This is a fourth non-detection and the first optical one, after the 28 August ascending and 31 August descending radar passes. It is not an observed extent and does not replace the modelled pool polygons. Twenty-one percent clear cover in fragments cannot delineate anything, and MNDWI cannot separate water from snow: at 2.5 km, where 1,134 snow pixels sit inside the clear mask, the index runs to 0.93. No geometry and no attribute claim changed, so no Zenodo version was minted; this evidence line will be folded into the barrier-lake record at the next version that has a substantive reason to exist.
• corrected 2026-08-31
  
  Planet Labs imagery assessed for this event, and it closes none of the atlas's observation gaps — a negative result, published. more ▾

  Planet released 26 scenes under CC BY-NC 4.0 through its Crisis Response Program: 19 PlanetScope (3.4–3.9 m), 4 SkySat (0.64–0.81 m) and 3 Pelican (0.55 m), up to seven times sharper than anything else available here. All 26 footprints, acquisition parameters and measured usability figures now ship as planet_scene_coverage. Barrier lake: exactly one scene covers the marker (PlanetScope, 26 Aug 05:45 UTC) and its UDM2 reads 0.0 percent clear within 300 m and within 1 km — 52.8 percent cloud, 47.2 percent light haze. Bidur: the 31 August SkySat pair covers 71 percent of the 80 m envelope at 0.65 m and is opaque cloud throughout, confirmed by its mask and by inspecting the pixels. Downstream: the 28 August PlanetScope strip leaves only 7.7 percent of the envelope clear, in fragments. Every observation gap this atlas carries is a cloud gap rather than an effort gap, which is now documented rather than assumed. Spatial data changed, so a new Zenodo version was minted — v12 — through the concept DOI 10.5281/zenodo.22118798.
• corrected 2026-08-31
  
  The Copernicus EMS extent at Syapru Besi is independently corroborated by a different sensor 64 minutes later — AUC 0.946. more ▾

  Planet's Pelican frame of 27 August 06:09:59 UTC images AOI01 at 0.55 m; Copernicus EMS photo-interpreted its AOI01 polygon from a WorldView-3 frame of 05:05 UTC the same morning. Over the 75.6 ha of the 111.1 ha polygon that is co-visible in the Pelican frame, NDVI separates ground inside the polygon from ground outside it at AUC 0.946, with median NDVI −0.085 inside against +0.013 outside. Visible-band brightness alone reaches only 0.661, blue/NIR 0.934 and red-edge II/NIR 0.885. This does not re-measure the EMS extent; it confirms that the ground EMS outlined really is spectrally distinct, from an independent sensor at an independent hour. The statistic is published as planet_ems_corroboration.
• open 2026-08-31
  
  A Planet-derived flood extent was attempted and deliberately NOT published, and the reason is worth stating. more ▾

  Haze has compressed the NDVI dynamic range of the 27 August Pelican frame to a 95th percentile of 0.198, where healthy forest exceeds 0.6 — so the flood signal is real but the contrast is not. A global Otsu threshold classifies 65.7 percent of clear ground as bare; a local-background NDVI anomaly at 250, 500 and 1,000 m radii does no better, at 64–68 percent. Tuning the threshold directly to the EMS polygon — which would already forfeit independence — peaks at F1 0.470, precision 0.614 at recall 0.381. A surface 39 percent wrong, fitted to the polygon it purports to check, is not an independent observation of it. Only a Planet acquisition with clear sky would change this.
• corrected 2026-08-31
  
  Planet data is used here without any licence change to this package, and the reasoning is recorded rather than assumed. more ▾

  Planet Crisis Response imagery is CC BY-NC 4.0. This package is CC BY 4.0 and stays that way, because no Planet pixels and no geometry traced from Planet pixels are redistributed. What ships is scene footprints and acquisition parameters, which are catalog metadata, and usability measurements, which are facts — neither is Adapted Material, so no non-commercial term propagates into the dataset. Had a traced extent polygon been published it WOULD have been Adapted Material and would have had to carry CC BY-NC 4.0 in a fenced file, on the model already used for the ODbL hot_* layers. That question is moot because no such polygon was produced. Keeping the package CC BY matters for reuse: OpenStreetMap and HOT cannot ingest non-commercial data, so an NC term would have blocked the crisis-mapping community this atlas is partly written for.
• corrected 2026-08-31
  
  Nepal death toll revised again: 903 → 939 — and the missing count FALLS for the first time, 4,247 → 3,925. more ▾

  NDRRMA's Monday-evening update, carried by AFP at 18:43 GST / 20:28 NPT and fetched first-hand. The fall is not a correction to the total but a recount of one component: missing hydropower-project workers drop 933 → 639, a figure static since 29 August. The headline still decomposes exactly: 3,333 Nepali citizens + 592 foreign nationals = 3,925. No district breakdown has been published for the 939; the 903 it supersedes had one that summed exactly. A registry conflict logged this morning has closed: the morning AFP copy gave Rasuwa 562 against DEOC's 865, and the evening copy adopts 865, with Nuwakot 1,158 in both; ANI's 1,548 for Nuwakot is not repeated and stays unresolved. Security-force missing now 45 Nepali Army, 38 police, 18 customs and immigration — the customs-and-immigration figure falls 19 → 18. China unchanged at 16 dead and 546 missing, so both countries together are 955 dead and 4,471 missing, superseding 919 and 4,793. The 17 bodies recovered by Uttar Pradesh authorities from rivers flowing out of Nepal remain outside both tolls. Casualty figures are not stored in the spatial data; no Zenodo version for this entry.
• corrected 2026-08-31
  
  A second Sentinel-1 pass looked at the barrier lake from the opposite geometry and found the same nothing. more ▾

  The 2026-08-31T00:10:50Z acquisition is descending relative orbit 121, a different track from the 28 August ascending relative-orbit-85 pass, so layover and shadow fall on the opposite valley walls and an impoundment hidden by geometry on the first pass should have been visible on this one. Processed through the same Range-Doppler chain against its own 12-day reference of 19 August. Geocoding reproduces the product geolocation grid to 1.85 m over 210 points, with incidence agreeing to 0.0000 degrees and slant-range time to under a millimetre; 7.4 percent layover, 12.3 percent shadow, 87.7 percent usable; change noise floor 2.50 dB from four pre-event 12-day pairs on this same track (2.66 / 2.52 / 2.48 / 2.31 dB, pooled mean +0.03 dB). Result: non-detection. 15 clusters of 1.5 ha or more, 67.24 ha total, every one at 5,360–5,970 m — wet snow at the snowline — and not one below 4,500 m. At the marker: +0.29 dB VV over 335 valid pixels, 0.12 sigma, 95 percent CI +0.02 to +0.56 dB, excluding any drop at all; modelled pools +0.03 and +0.12 dB. Calm water at −22 dB would sit 163 standard errors below the measurement. The larger snow signal than 28 August tracks the hour: 00:10 UTC is 05:55 NPT, pre-dawn, against 18:07 NPT. The 1.50 ha cluster flagged at 3,778 m on 28 August did not recur (−2.38 dB against a −4 dB threshold). Spatial data changed, so a new Zenodo version was minted — v11 — reached through the concept DOI 10.5281/zenodo.22118798.
• open 2026-08-31
  
  The positive control that validated the 28 August radar chain could not be repeated on the 31 August pass, and the data now says so in its own field. more ▾

  None of the Copernicus EMS EMSR927 damage polygons falls inside the 31 August slice's data footprint — their pixels return sigma0 of exactly zero, and the STAC footprint confirms the scene does not reach Timure or Syapru Besi — so the Mann-Whitney and >6 dB enrichment test that showed the detector sees real ground change where ground change is known cannot be run here. Sensitivity on this pass rests instead on the four same-track pre-event 12-day pairs, which bound the noise floor but do not demonstrate detection of a known positive. sar_search_envelope_20260831 carries this in a positive_control field reading UNAVAILABLE rather than leaving the omission to be inferred. The non-detection is reported with that limitation attached.
• corrected 2026-08-31
  
  The barrier-lake GeoPackage had been wrong since v1: 13 of 25 fields, with the values it did carry shifted into the wrong columns. more ▾

  nepal_flood_20260826_barrier_lake.gpkg put the location description in reporting_authority, the OSM-node reasoning in described_location, the string 'medium' in location_source, 17.6 in location_confidence, and the string 'lower' in evidence_url_2 — a one-column shift caused by a dropped lake_id field, and it dropped twelve fields entirely including the whole measured-area series. The CSV and GeoJSON siblings were correct throughout, and every figure this page has published came from them, so no published number was ever wrong because of this. The GeoPackage is rebuilt from them in v11 with all 25 fields. Anyone who read attributes out of the .gpkg before v11 should re-read. A second, smaller fault was found in the same sweep: upstream_rivers.geojson shipped without the length_km field its own GeoPackage carries; both siblings are regenerated.
• corrected 2026-08-31
  
  Every GeoPackage in the package was audited against its CSV and GeoJSON siblings after two file-level faults in two consecutive versions. more ▾

  The zero-byte observed-event GeoPackage repaired in v10 and the field-shifted barrier-lake GeoPackage repaired in v11 are the same class of fault — a spatial file silently disagreeing with the tabular file beside it — so all 18 GeoPackages were compared for column set and row count against their siblings. 2 faults found, 16 clean. Both faults are recorded above and both are fixed. The audit method and its result are stored in the manifest under layer_audit_20260831 so the check can be repeated and the claim can be falsified. This kind of fault is invisible from the page, which is why it went nine versions undetected; the audit is now part of the publishing routine rather than a one-off.
• corrected 2026-08-31
  
  Lake attribute records refreshed to match what the page says. more ▾

  The barrier-lake point, both modelled pool polygons and the upper-feature record all carried an observed-extent note whose most recent evidence was the 28 August radar pass and the 29 August cloudy optical pass. All three now carry the second radar non-detection, the third consecutive fully clouded optical pass (Sentinel-2A, 31 August: high-probability cloud on 100.0 percent of pixels within 300 m, 1 km and 2.5 km by the SCL band), and MWR's 108,000 m² surface and 1,400,000 m³ volume for the upper feature at roughly 3,700 m, with the impact-crater versus erosion-pit naming conflict and its disputed day logged in the record rather than resolved. No geometry moved: the marker coordinates and both pool outlines are byte-identical to v10.
• open 2026-08-31
  
  Copernicus EMS EMSR927 is unchanged since this morning — AOI04 Bharatpur is still undelivered on day six. more ▾

  Re-queried at 22:45 UTC: AOI01 v1, AOI02 v2 and AOI03 GRA_MONIT01 v1 all unchanged, no re-deliveries in products[].version.number, and AOI04 still statusCode W with a null deliveryTime against an expectedDelivery of 2026-08-29 17:01 UTC. Its Legion acquisition of 31 August 07:16 UTC is now sixteen hours old with no filename assigned. The activation remains open. No spatial change from this entry.
• corrected 2026-08-31
  
  Nepal death toll revised again: 788 → 903, and the missing count nearly doubles, 2,502 → 4,247. more ▾

  NDRRMA's Monday-morning update of 31 August, carried by ANI at 05:23 UTC / 11:08 NPT and by AFP/RSS via myRepublica, both fetched first-hand. Both headlines decompose exactly: 903 = Chitwan 272 (including partial remains), Nawalparasi East 207, Nawalparasi West 151, Nuwakot 95, Gorkha 62, Dhading 55, Tanahun 37, Rasuwa 24; 4,247 = 3,655 Nepali citizens + 592 foreign nationals. This is the first headline since 734 that is fully broken down, and it closes the ‘no district breakdown has been published’ note carried against the 788. 903 supersedes 788, 781, 768, 752, 734, 675, 579, 567, 547, 538, 489, 475 and 469; 4,247 supersedes 2,502, 2,498, 2,426/2,418 and 1,924. Rescued rises to 10,451 (8,730 by ground forces, 3,081 airlifted on 363 helicopter flights including 236 foreigners); 411 flights flown, 7 of them on Monday. Foreign missing 589/591 → 592. Security-force missing now 25 Nepal Police, 45 Nepal Army, 13 Armed Police Force, 15 Customs and 3 Immigration — the police component falls from 27 to 25 and AFP's ‘19 customs and immigration officers’ is one more than the 15+3 ANI lists; neither discrepancy is reconciled. China's Gyirong County figures are unchanged at 16 dead and 546 missing, so the two countries together are 919 dead and 4,793 missing. Casualty figures are not stored in the spatial data; no Zenodo version.
• open 2026-08-31
  
  The missing count nearly doubled on one district, and the two registries reporting it disagree by 1,073 people. more ▾

  Nuwakot's missing goes from 115 on Sunday to 1,158 in NDRRMA's Monday figures as reported by AFP, with Rasuwa unchanged at 562 — that single district is essentially the whole 1,745-person rise. But District Emergency Operation Centre figures published by ANI in the same hour, against the same 4,247 headline, give Nuwakot 1,548 and Rasuwa 865. Both sets sit under one national total, so at least one is not what the national total is built from: AFP's 3,655 Nepali citizens + 592 foreigners reconciles to 4,247 exactly, while the DEOC components (865 + 1,548 + 933 hydropower + 65 Makwanpur + 592 foreign + 101 security, customs and immigration) reach 4,104 and fall 143 short. Neither registry is adopted over the other and the gap is not explained by any source. Separately, Indian authorities in Uttar Pradesh have recovered 17 bodies from rivers flowing out of Nepal; they are not in either country's toll. No spatial change; no Zenodo version.
• corrected 2026-08-31
  
  Copernicus EMS superseded AOI03 Bidur with a monitoring product and the observed extent grows again: 829.52 → 942.00 ha. more ▾

  EMSR927 delivered AOI03 GRA_MONIT01 v1 at 2026-08-31T01:07:28 UTC (statusCode F), photo-interpreted from PeruSAT-1 of 27 August 04:47 UTC and Resurs-P of 28 August 08:29 UTC — the acquisitions this page flagged as tasked on 30 August. Every AOI03 feature is replaced by it and every derived figure recomputed; AOI01 v1 and AOI02 v2 are untouched. Bidur observed extent 589.079 → 701.559 ha. Validation against the unchanged 80 m height-above-channel band: observed extent inside 801.77 ha (96.7%) → 913.34 ha (97.0%); buildings graded 3,207 → 3,922, of which Destroyed 2,497/2,521 (99.0%) → 3,150/3,174 (99.2%), Damaged 285 → 311 (100%), Possibly damaged 397/401 → 433/437 (99.1%); destroyed bridges 24/24 → 30/30; destroyed road 46.64/47.25 km (98.7%) → 56.14/56.76 km (98.9%); graded facility area 34.09 → 34.40 ha, still 100% inside. The control moves with it and still constrains the headline: undamaged road inside the band 74.4% → 74.7%, a 24-point spread against the destroyed road rather than a near-total one. The 80 m band discriminates well in the upstream gorge and poorly in the broad Bidur valley, and 97.0 percent must not be quoted without that. Spatial data changed, so a new Zenodo version was minted — v10 — reached through the concept DOI 10.5281/zenodo.22118798.
• corrected 2026-08-31
  
  The paragraph under the EMS validation table had been contradicting the table above it since 29 August, and three of its four figures were stale. more ▾

  It read that the band holds ‘96.3 percent of the destroyed road and only 11.6 percent of the road EMS could see no damage on’ — both two-AOI figures, superseded when Bidur was added, with the table on the same screen already showing 98.7 and 74.4 percent. It also gave the tightened-band coverage as 73.6 / 50.2 / 35.9 percent at 40 / 20 / 10 m against the validation table's own 89.9 / 68.5 / 41.4, and put the area outside the band at 7.2 percent when 100 − 96.7 is 3.3. All are now recomputed on the v10 product: 98.9 and 74.7 percent, 90.6 / 69.6 / 42.3, and 3.0 percent outside. The 11.6 percent figure is not deleted — it is the genuine Syapru-Besi-and-Timure-only value and is what makes the gorge-versus-valley caveat legible — but it is no longer presented as the current control. The published CSVs were always right; this was a prose error.
• corrected 2026-08-31
  
  The EMS observed-event GeoPackage shipped empty in every version from v1 to v9 — a zero-byte file. more ▾

  nepal_flood_20260826_ems_observed_event.gpkg was 0 bytes in the distributed ZIP; anyone who opened it got nothing. The GeoJSON and CSV of the same layer were always complete and correct, so no published figure on this page was ever derived from the empty file. It is written properly in v10: 561 KB, layer observed_event, 3 features, with FGDC metadata embedded. Re-download if you took the .gpkg before v10.
• open 2026-08-31
  
  EMS AOI04 Bharatpur is still undelivered on day six, but has its first booked acquisition — Legion, 31 August 07:16 UTC. more ▾

  The product remains statusCode W with a null deliveryTime against an expectedDelivery of 2026-08-29 17:01 UTC that passed two days ago. What has changed is that the image slot is no longer empty: the activation API now lists a new Legion optical acquisition for 07:16 UTC today, filename not yet assigned. Bharatpur is the furthest downstream AOI and lies well beyond the 62.4 km corridor this atlas maps, so its delivery would not extend the envelope validation — it would measure how far past the modelled corridor the observed damage runs. No spatial change from this entry.
• open 2026-08-31
  
  A third post-event optical pass, and a third total cloud failure — 100.0 percent within 300 m, 1 km and even 2.5 km of the marker. more ▾

  Sentinel-2A acquired the barrier-lake site at 2026-08-31T05:01:08 UTC. Scene-level cloud cover reads 88.8 percent, but read at the pixel level the SCL band returns class 9, high-probability cloud, for 716 of 716 pixels within 300 m of 28.31249 N 85.55406 E, 7,860 of 7,860 within 1 km and 49,080 of 49,080 within 2.5 km. There is no usable optical observation of the lake, the crater or the third breached lake on any date. A Sentinel-1D radar pass does exist for 2026-08-31T00:10:50 UTC, unprocessed as of this update; the 28 August radar was processed and did not find the lake. The pool polygons therefore remain terrain-modelled screening surfaces and the marker remains an inferred point of medium confidence.
• open 2026-08-31
  
  The upper feature's own measurement reaches this page twice, under two names and on two different days. more ▾

  China Daily via KathmanduPati, 31 August calls it an erosion pit rather than an impact crater, places it on the Chhuchen Khola above the Purepu Tsangpo confluence, and dates the 108,000 m² surface to 11:23 on Saturday 29 August; Xinhua, on a 30 August 20:28 Beijing dateline, gave the same reading as Sunday 30 August. Every number is identical — 1.4 Mm³, 5.6 km upstream, 800 m higher, the 19,000 m² fall from 07:00, three further areas under 70,000 m² — so this is one MWR assessment relayed twice, and only its date is in dispute. The Sunday reading is retained. MWR's hazard language is unchanged: significant risks and hidden hazards, continued monitoring required. No spatial change; no Zenodo version.
• corrected 2026-08-31
  
  The upper feature has its first agency measurement, and it is not a lake — MWR calls it an impact crater, 108,000 m² holding 1,400,000 m³. more ▾

  China’s Ministry of Water Resources places it about 5.6 km upstream of the barrier lake at roughly 3,700 m, some 800 m higher than the lake. At 11:23 on Sunday 30 August its water surface was about 108,000 m², down 19,000 m² from 07:00, with a storage volume of about 1,400,000 m³ and water flowing out continuously about 180 m below its dam; three further accumulation areas within 2.5 km downstream total under 70,000 m². MWR says the crater and those three areas still pose risk hazards and require continued monitoring (Xinhua, 30 August 20:28 Beijing time). This is the first area series and the first volume for the feature carried here since 29 August as “over 120,000 m², depth not established” (Chen Hongqi, Ministry of Natural Resources, via Reuters), and the first agency reading against which Nitesh Khadka’s 30 August account of more than half the upper water having run out can be set: 127,000 → 108,000 m² in four hours is a 15 percent area loss, which neither confirms nor contradicts him. Xinhua gives clock times without a zone on a Beijing dateline; read as Beijing time, 11:23 is 09:08 NPT. Not mapped: MWR published scalars, not geometry. No spatial change; no Zenodo version.
• open 2026-08-31
  
  A third barrier lake formed on 29 August, on the Nepali side, and has already breached — which sits badly against the ‘avalanche, not a breach’ account Nepal was given. more ▾

  The same MWR statement reports a new barrier lake formed on Saturday about 2.8 km downstream of the previous one after a landslide on the Nepali side, its dam since breached by upstream flow and the channel back in a flowing state on drone footage and panoramic imagery (Xinhua, 30 August). This page publishes Nepal’s Ministry of Foreign Affairs being told by Beijing that Sunday’s increased upstream flow came from an avalanche, not a breach. Both can hold — a dam formed on Saturday breaching is not the 2,000,000 m³ lake bursting — but a breach did occur upstream in the window before the 03:30 NPT surge past Dhunche, and no source reconciles the two accounts or timestamps it. Left open. The feature is not mapped: no coordinate, area or volume has been published for it, and this atlas does not place a feature from a distance-along-river. No spatial change; no Zenodo version.
• corrected 2026-08-31
  
  Both outstanding casualty-source discrepancies closed on their own: Wikipedia’s 794 is withdrawn and AP’s ‘at least 760’ now reads ‘at least 800’. No adopted figure moves. more ▾

  The English Wikipedia article 2026 Nepal floods gave 794 dead in Nepal at revision 2026-08-30T15:23:04Z, cited to the BBC live page; at revision 2026-08-30T22:54:02Z it gives 788 and 2,502 for Nepal and 804 and 3,048 for both countries, sourced to the same AP story of 30 August fetched first-hand here, and 794 no longer appears — withdrawn rather than superseded, since bbc.com was never reachable from this job. The AP copy, updated 2026-08-30T20:30:42Z, now says in its second paragraph that the floods killed ‘at least 800 people’ in Nepal and Tibet, consistent with 788 + 16; the ‘at least 760’ flagged here belonged to its earlier 752 filing, which still stands at NPR. One new discrepancy replaces them and it is Wikipedia-internal: its Casualties section reads 2,504 missing against 2,502 in its own lead and infobox. Nepal remains 788 and 2,502; China 16 and 546. No spatial change; no Zenodo version.
• corrected 2026-08-30
  
  Both barrier lakes are now draining — the lower one fully, and more than half the upper lake has run out. more ▾

  China’s Ministry of Water Resources said on Sunday 30 August that the roughly 2,000,000 m³ impounded at the Chhochen Khola / Purepu Tsangpo confluence has been almost completely discharged and water is flowing normally through natural channels — its hydrological emergency monitoring team, cited by China Daily and reported by The Kathmandu Post, 30 August, updated 12:39 NPT. That supersedes largely drained, which came from MWR’s Saturday-evening remote-sensing analysis via CGTN, and it ends the measured area series at 120,000 m² (27 August) and 99,000 m² (29 August). The upper lake is the larger correction: this page carried it, on Reuters of 29 August, as unaffected, the bigger of the two, and growing with no visible outlet. Nitesh Khadka, a Nepali scientist at the Chinese Academy of Sciences in Chengdu, says more than half its water has flowed out, with photographs sent by colleagues. That is one named scientist relaying imagery, not an agency measurement, and no area or volume accompanies it. NDRRMA’s condition — risk remains until both lakes have fully drained — is therefore still not met, and the dam body is still the soft, ice-rich material China Anneng watches by slope radar. No spatial change; no Zenodo version.
• corrected 2026-08-30
  
  A district breakdown now exists for 768, and NDRRMA’s own 18:00 NPT figure was 781 — the 788 headline stands, but the chain between 734 and 788 was missing two rungs. more ▾

  Nepal Police put bodies recovered at 768 as of 14:00 NPT on 30 August with a full district breakdown — Chitwan 264, Nawalparasi East 194, Nawalparasi West 100, Gorkha 58, Nuwakot 52, Dhading 50, Tanahun 37, Rasuwa 13, summing exactly to 768 (RSS via The Rising Nepal). It replaces the 09:00 NPT breakdown for 734 that this page was still showing as the most recent one. Separately, Xinhua reports NDRRMA directly at 781 bodies recovered and 2,502 missing as of 18:00 NPT (Xinhua, Kathmandu, 30 August) — the first agency-attributed figure between 734 and 788, and an independent corroboration of 2,502. 788 (AP via CBS, 20:02 NPT) remains the latest and is unchanged. Two components move: the Nepal Tourism Board gives 591 foreign nationals out of contact at 18:00 NPT, with 289 foreigners rescued and five re-contacted — three Spanish, one American, one Belgian — against the 589 foreign citizens in NDRRMA’s 09:00 NPT components; and Nepal Police give the missing security personnel as 44 Nepal Army, 27 Nepal Police and 13 Armed Police Force, 84 rather than the 85 published here, the Army figure falling 45 → 44. One count is not reconciled: Nepal Police put total rescued at 2,715 against NDRRMA’s 8,186. Both are shown; neither is adopted over the other. No spatial change; no Zenodo version.
• corrected 2026-08-30
  
  The EMS Bidur monitoring product is in production on 27–28 August imagery, not the 30 August acquisition reported here this midday. more ▾

  AOI03 GRA_MONIT01 (id 2868) has moved from status W to I and now names its two source frames: PeruSAT-1 at 27 August 04:47 UTC and Resurs-P at 28 August 08:29 UTC, both listed as ortho COGs. The 30 August 18:00 UTC acquisition this page reported was a placeholder and did not happen. Expected delivery moves 31 August 07:00 → 02:50 UTC. The Resurs-P frame is still a genuine second date at Bidur — the delivered v1 grading used 27 August imagery — so a differenced extent remains possible, a day earlier than predicted. No vector layers and no map sheets are published yet, so there is still no new observed extent. AOI04 Bharatpur is unchanged at status W, Legion at 31 August 07:16 UTC, expectedDelivery still the long-passed 29 August 17:01 UTC. AOI01 v1, AOI02 v2 and AOI03 v1 are unchanged. Method note: the EMS backend endpoint this job queries has moved. /backend/api/activations/{code}/ now returns 404; the current path is /backend/dashboard-api/public-activations/?code={code}. Re-queried 30 August 20:05 UTC. No spatial change; no Zenodo version.
• corrected 2026-08-30
  
  A full-corridor observed flood extent exists and is now in the package — 31.71 km², 3.8× the 8.30 km² Copernicus EMS mapped — and it independently replicates the 80 m result at 93.7 percent. more ▾

  This page has said since 28 August that outside three EMS AOIs every surface here is terrain-modelled. That is superseded. The Humanitarian OpenStreetMap Team has published an observed extent for 27 August covering the whole Bhote Koshi and Trishuli, photo-interpreted from drone, Landsat, PlanetScope and Sentinel imagery (HDX, dataset hot_flood_npl), reached via International Charter activation 1052. Restricted to the reach this atlas models — within 3 km of the 62.42 km channel, 16.54 km² of observed extent — the share inside each height-above-channel band is 10 m 42.5%, 20 m 64.2%, 40 m 84.5%, 80 m 93.7%, against the 96.7 percent measured on the 8.30 km² EMS extent. Twice the area, different sensors, different interpreter, same threshold. The 80 m band is only 39.4 percent covered by observed water, which is expected and is the point: it screens for what the terrain permits, not for what was wetted. The 96.7 percent caveat carries over unchanged — the band discriminates in the upstream gorge and poorly in the broad Bidur valley. LICENCE, AND IT IS NOT THE PAGE DEFAULT: the layer ships as ODbL 1.0, not CC-BY, because a Derivative Database of an ODbL source stays ODbL — converting it to a shapefile does not reset that. It is a separate file, joined to nothing, so the package is a Collective Database and share-alike reaches the hot_* files alone; every Keystone layer is unchanged and keeps its own terms. The validation table is CC-BY, because aggregate statistics are a Produced Work under ODbL s.4.5. HOT publishes the vector under ODbL although it was interpreted partly from PlanetScope; this package relies on that grant and still redistributes no Planet imagery, which remains CC-BY-NC and excluded. Package republished as v9.
• open 2026-08-30
  
  The scope limitation is now measured, and it is worse than stated: 15.16 km² — 47.8 percent of all observed flooding — lies outside the mapped corridor. more ▾

  This page has said the 62.4 km corridor covers ‘roughly three-quarters’ of the affected channel, on the strength of reporting that the flow travelled up to 80 km. Measured against the HOT observed extent, it covers about half. Observed water reaches 54.8 km past the downstream endpoint at Devighat, out to 84.5564 E, across Nuwakot, Dhading, Gorkha and Chitawan; HOT’s own corridor of interest runs to the Narayani confluence at Devghat in Chitwan, roughly 70 km beyond this atlas’s Devighat in Nuwakot — two different places with near-identical names, and EMS AOI04 ‘Bharatpur’ is the Chitwan one. HOT’s ground-reported bridge register puts seven washed-out bridges beyond the corridor, up to 27.4 km off-channel. The corridor was not extended. Doing so means regenerating the channel, the HAND raster, all four envelope bands and every exposure and population figure in the package, which is not a change to make inside a recurring update; it is recorded here as a known limit instead. Every exposure figure on this page is a lower bound on extent as well as on population. No geometry changed for this entry.
• open 2026-08-30
  
  Cross-checking against HOT and OSM raises two questions about this package’s own positions, and finds one defect in theirs. more ▾

  Four hydropower sites in the damage register sit some distance from their OpenStreetMap counterparts: Trishuli (24 MW) 3.39 km, Upper Trishuli 3A 2.66 km, Rasuwagadhi 1.58 km, Devighat 1.51 km. Two of those — Trishuli and Devighat — are carried here at location_confidence=high. On a page that reduces everything to a height above the channel, a 3 km shift along a gorge changes the answer, so this is flagged rather than silently reconciled: neither source is authoritative, and no coordinate has been altered pending a proper check. Separately, Upper Trishuli 3A carries hand_m = −5.6, below the channel surface — either a real intake-elevation artefact of a DSM or a georeferencing error, and it should not have gone unremarked. In the other direction, HOT’s bridge damage register contains a defective record: Thulo bharkhu has longitude equal to latitude (28.1390111, 28.1390111), placing it in China. Only the flood extent was ingested here, so nothing in this package is affected, but the register would need coordinate screening before any future use.
• corrected 2026-08-30
  
  Nepal death toll revised again: 752 → 788; the missing count does not move, 2,502. more ▾

  The same AP wire story this page cited at 15:37 NPT has been refreshed and now gives Nepal’s disaster agency at 788 dead with 2,502 missing (CBS News carrying the AP wire, 30 August 10:17 ET / 20:02 NPT, fetched first-hand); the NPR copy of the same story still reads 752 and has not been updated, so the 752 published here this afternoon was correct to its source at the time. No district breakdown accompanies the 788. China unchanged, so the two-country total moves 768 → 804 dead. A higher figure is in circulation and is not adopted: the English Wikipedia article 2026 Nepal floods at revision 2026-08-30T15:23:04Z carries 794 for Nepal, cited to the BBC live page of 30 August. bbc.com is not reachable from this job, so 794 could not be checked against its own source and is recorded here rather than published; a later run should adopt it once a reachable source carries it. No spatial change; no Zenodo version.
• corrected 2026-08-30
  
  Active Hazard block condensed from nine paragraphs to four — presentation only, no figure touched. more ▾

  The block had accreted to about 17,700 characters and had become a day-by-day log rather than a statement of current hazard, repeating material already held in this errata log and in the Observed Radar section below it. What was removed: the 28 August breach report and the slow-drainage account that followed it, the Chinese Embassy’s contradicting high-water statement of the same evening, the 2,000,000 and 2,500,000 m³ volume series, the extension of the inflow window to peak flow around 1 September, the Khadka two-lake attribution, the China Anneng deployment detail, the archived status = "open - high breach risk" marker attribute, and the narrative description of the Sentinel-1D pass. Every one of those remains published in full in its own errata entry, and the radar work keeps its own section with the controls intact. What stays on the block is current state only: the 30 August surge, the lower lake, the upper lake, and what this page does and does not observe. No spatial change; no Zenodo version.
• corrected 2026-08-30
  
  Nepal death toll revised again: 734 → 752, and the missing count moves up, 2,498 → 2,502. more ▾

  Nepal’s disaster agency gave 752 dead and 2,502 missing on Sunday 30 August, reported by the Associated Press via NPR, 30 August 05:52 ET (15:37 NPT), fetched first-hand; the same pair is carried by the English Wikipedia article 2026 Nepal floods at revision 2026-08-30T11:54:34Z, sourced to the BBC live page of 30 August. 752 supersedes the 734 of the NDRRMA 09:00 NPT update carried here this morning. No district breakdown has been published for the 752, so the 734 breakdown is retained on the page and marked to its own date. China is unchanged at 16 dead and 546 missing, so the two-country total moves 750 → 768 dead and 3,044 → 3,048 missing. One caveat on the AP copy: its own second paragraph says the floods ‘killed at least 760 people’ across Nepal and Tibet, which is neither 752 nor 752+16; the figure adopted here is the one AP attributes directly to the disaster agency later in the same piece. No spatial change; no Zenodo version.
• open 2026-08-30
  
  A second flood wave ran the corridor on 30 August — 03:30 NPT past Dhunche, a 130 m suspension bridge gone at 11:30 NPT — and it is not a lake burst. more ▾

  Residents of Dhunche fled to the hospital ground and the playground when the surge came through at about 03:30 NPT; levels had receded by the afternoon and a Nepali Army helicopter based at Dhunche was sent upstream to assess damage (The Rising Nepal / RSS, 30 August). At about 11:30 NPT, after a flood lasting roughly three hours, the Trishuli swept away the Ghyalchowk–Charaudi suspension bridge — 130 m, built six years ago, linking Gandaki Rural Municipality-8 in Gorkha to the Prithvi Highway and carrying the dairy and vegetable traffic of about 400 households to the Charaudi cooperative. It was itself the alternative to the Dhawadighat bridge lost on 26 August; it is the third suspension bridge the Trishuli has taken (The Rising Nepal / RSS, 30 August). Nepal asked Beijing what was happening and was told the increased flow upstream came from an avalanche, not a breach — the impoundment has not burst (MoFA via The Rising Nepal); NDRRMA chief executive Dharam Raj Uprety confirms the Sunday phone alerts followed a Chinese notification of increased upstream flow (AP). Like the Krishnabhir highway failure, this is fresh damage outside the mapped 62.4 km corridor and outside all four EMS AOIs, and it is not added to the damage register: no source gives a coordinate and this page does not geolocate from news text. No spatial change; no Zenodo version.
• open 2026-08-30
  
  EMS has tasked its first monitoring product of the activation — Bidur again — while Bharatpur’s acquisition slips to 31 August, two days past its own stated delivery. more ▾

  Re-queried 30 August 12:0x UTC. A new AOI03 Bidur product (id 2868, monitoring: true, monitoringNumber: 1) sits at status W on VHR1 optical imagery to be acquired 30 August 18:00 UTC, sensor not yet named, expected delivery 31 August 07:00 UTC. That would be the first repeat observation anywhere on this event — a second-date extent at Bidur, against which the 29 August grading could be differenced. The AOI04 Bharatpur tasking has changed underneath its unmoved metadata: the image is now Legion at 31 August 07:16 UTC, not the 29 August 04:01 UTC acquisition reported here this morning, while expectedDelivery still reads the long-passed 29 August 17:01 UTC. Devghat, where Chitwan’s unidentified bodies are being buried, therefore has no observed extent and no credible delivery date. The three delivered products are unchanged at AOI01 v1, AOI02 v2 and AOI03 v1. The activation now publishes its own headline stats — 830 ha maximum extent, 3,207 identified buildings, 5,300 population, 46 km of roads, 11 ha built-up — consistent with the 829.52 ha this page computed from the delivered vectors. No spatial change; no Zenodo version.
• corrected 2026-08-30
  
  Nepal death toll revised again: 675 → 734, and the missing count moves up, 2,418 → 2,498. more ▾

  The NDRRMA Search, Rescue and Relief Update issued at 09:00 NPT on 30 August gives 734 bodies and human remains recovered and 2,498 people missing (The Kathmandu Post, 30 August, fetched first-hand). District breakdown of the 734: Chitwan 259, Nawalparasi East 184, Nawalparasi West 82, Gorkha 58, Nuwakot 52, Dhading 50, Tanahun 36, Rasuwa 13 — summing exactly to the total. Supporting counts also move: 242 injured treated or discharged (from 219), 8,186 rescued (from 7,514) and 19,895 security personnel mobilised (from 18,708) — that last exceeding its own published components, 8,399 Nepal Army, 7,213 police and 4,203 Armed Police Force, which sum to 19,815, by 80. Unidentified bodies are held at 21 hospitals for identification. No spatial change; no Zenodo version.
• corrected 2026-08-30
  
  The 80-person gap in NDRRMA’s missing count closes by adoption, not by explanation — and a 22-person gap opens the other way. more ▾

  On 29 August this page carried NDRRMA’s headline of 2,418 while recording that the authority’s own published components summed to 2,498, and declined to adopt the larger figure. NDRRMA has now published 2,498 as its headline. Its components have moved the opposite way: 933 hydropower-project workers, 589 foreign citizens, 127 Nepalis travelling alongside them and 85 security personnel (45 Nepal Army, 27 Nepal Police, 13 Armed Police Force), plus district reports of Rasuwa 562, Nuwakot 115 and Makawanpur 65, now total 2,476 — 22 short of the headline, and short by exactly the categories dropped since 29 August: 15 Customs, 4 Immigration and the 3 in Langtang National Park. No source explains either the adoption or the omission. Separately, AFP describes NDRRMA’s Saturday-evening figure as 2,426 (AFP via BSS, 30 August), not the 2,418 this page took from ANI; that discrepancy is unresolved. No spatial change; no Zenodo version.
• corrected 2026-08-30
  
  China’s Gyirong County figures adopted at 16 dead and 546 missing — the numbers this page rejected a day ago. more ▾

  On 29 August 16 and 546 appeared only in the English Wikipedia infobox, contradicting that article’s own lead and every source then reachable here, and were not adopted; this page held 7 dead and 554 missing from Reuters citing Xinhua. Both figures are now carried independently by AFP, 30 August, which states that Chinese state media had confirmed 16 dead in Tibet and 546 missing, and by CNN’s live coverage reporting the Tibet regional government. 16 supersedes 7, 5 and 3; 546 supersedes 554 and 558 — the second time the Chinese missing count has moved down. AFP puts the two countries together at 750 dead and 3,044 missing; whether the Nepali and Chinese counts overlap is still not established. No spatial change; no Zenodo version.
• open 2026-08-30
  
  EMS AOI04 Bharatpur has missed its own delivery deadline — 29 August 17:01 UTC has passed with the product still at status W. more ▾

  Re-queried 30 August 08:0x UTC: the tasking is unchanged (Legion, acquired 29 August 04:01 UTC), no delivery time is set and no layers exist. This is the AOI covering Devghat in Bharatpur-1, where Chitwan’s unidentified bodies are being buried, so it is the one place where an observed extent would bear directly on the recovery. AOI01 v1, AOI02 v2 and AOI03 v1 are unchanged in products[].version.number. Earth Search returns no Sentinel-1 or Sentinel-2 acquisition over the barrier-lake marker since the fully clouded 29 August 05:00:50 UTC pass. No spatial change; no Zenodo version.
• corrected 2026-08-30
  
  The lower barrier lake is now reported largely drained, not merely shrinking — but the dam holding it is described as getting less stable, not more. more ▾

  China’s Ministry of Water Resources, in a remote-sensing analysis on the evening of Saturday 29 August, finds the lake reduced in size with water flowing out naturally and the barrier body largely drained; its risk was assessed as manageable that day, and ground rescue and repair of the G216 at Gyirong Port resumed on that basis (CGTN, 30 August). That extends the measured area series recorded below rather than contradicting it. Two things keep this open. The barrier dam is less stable than a rock-and-earth dam would be: the material came down carrying large amounts of ice and snow, leaving it high in water content and soft underfoot, per the engineer Tian quoted in the same report, and China Anneng now runs slope radar and drones on it around the clock so downstream teams can be withdrawn if it moves. And none of this touches the upper lake, for which no new figure has been published since Chen Hongqi’s >120,000 m² on 29 August; NDRRMA’s position that risk remains until both lakes have drained is unaffected. No spatial change and no Zenodo version: status already reads “draining — area shrinking; burst risk assessed as decreased”, which this read extends rather than falsifies, and status_as_of is already 2026-08-29, the date of the read itself.
• open 2026-08-30
  
  211 of the 669 bodies Nepal Police had recovered by Saturday evening are partial or severed remains — the published toll counts bodies, not people. more ▾

  Police spokesperson DIG Abi Narayan Kafle, as of 16:00 NPT on 29 August: 669 recovered, comprising 268 men, 148 women, 22 girls, 20 boys and 211 partial, mutilated or severed remains; 2,362 missing, of whom 1,863 Nepali (1,591 men, 264 women) and 499 foreign (223 men, 300 women). Both sets sum exactly to their headline (The Kathmandu Post, 29 August). Twenty had been identified by Saturday evening; 229 of the 233 bodies in Chitwan are unidentified and are being buried at Devghat in Bharatpur-1, inside EMS AOI04, still undelivered (The Kathmandu Post, 30 August). This does not supersede the 675 and 2,418 above: NDRRMA’s cut-off is 18:30 NPT, two and a half hours later, and this page attributes rather than reconciles. It does bear on the open entry below — Police components sum and NDRRMA’s do not, which locates that arithmetic gap in NDRRMA’s publication rather than in the underlying count. The stat chip above is relabelled bodies recovered accordingly. No spatial change; no Zenodo version.
• open 2026-08-30
  
  The Trishuli severed the Prithvi Highway at Krishnabhir on 30 August — fresh damage four days on, downstream of the corridor mapped here. more ▾

  A section of the highway subsided in Benighat Rorang Rural Municipality, Dhading, after the river eroded its base, completely blocking the Muglin–Kathmandu section and cutting a main route into the capital; the water level had been rising since early Sunday (Dhading CDO Laxmi Pandey Gautam and Bagmati Province Traffic Police chief Dilli Narayan Pandey, via The Kathmandu Post, 30 August). This is new damage dated 30 August, not a late report of 26 August damage, so the event is still generating impacts. It lies beyond the downstream end of the 62.4 km corridor at Devighat and outside all four EMS AOIs, which is further support for the standing open entry that 62.4 km is a lower bound on extent. It is not added to the damage register, which is closed at the 16 source-cited records of the 26 August flood itself. No spatial change; no Zenodo version.
• corrected 2026-08-30
  
  The missing at Gyirong Port now have a nationality breakdown: 261 foreign nationals, 104 of them Nepali. more ▾

  CGTN, citing a press conference on the morning of Sunday 30 August, gives 261 foreign nationals unaccounted for — 104 Nepalis, 49 Indians, 33 Latvians, 18 Americans and 15 British among them — with the relevant embassies and consulates notified and no foreign tourists left stranded in Gyirong County (via The Kathmandu Post, 30 August). This is a composition, not a new total. The Xizang emergency management department’s count as of 01:00 on 29 August remains 7 dead and 554 missing (CGTN, 30 August), which is what this page carries. No source states whether the 261 sits inside the 554 or is counted on a different basis, so it is recorded alongside and not subtracted from it, and it is not additive with the 499 foreign nationals Nepal Police report missing on the Nepali side. No spatial change; no Zenodo version.
• corrected 2026-08-30
  
  The two lake layers carried stale attributes and the map popups had drifted from the page — corrected in an attribute-only republication, Zenodo v8. No geometry changed. more ▾

  The marker still served volume_m3 = 2,000,000 and expected_inflow_m3_to_30aug = 3,000,000, and both pool polygons still carried a caveat asserting that no post-event Sentinel-1 acquisition was available — false since 28 August, when one was acquired and processed here to a non-detection. A reader clicking the marker was being given figures the prose above it had already retired. What changed in the data. barrier_lake gains the measured surface-area series — 120,000 m² (27 Aug) and 99,000 m² (29 Aug) from the PRC Ministry of Water Resources, 110,000 m² (27 Aug) from NDRRMA via the Stimson Center — plus lake_id, status_as_of, area_source, observed_extent, expected_peak_inflow_date (about 1 September, superseding the 30 August window) and volume_m3_reported_20260828_min. volume_m3 is renamed volume_m3_reported_20260827 and expected_inflow_m3_to_30aug is dropped as a dead concept. modelled_pool now quantifies its own error: the 19.1 and 29.5 ha polygons overstate the measured lake by 1.59× and 2.46× against 27 August, 1.93× and 2.98× against 29 August. The polygons were deliberately not refitted. Every published area is a scalar; no authority has released a geometry. A pool re-fitted to 99,000 m² would be a DEM flood-fill tuned to a number and would look more like evidence than it is. The polygons stay as modelled scenarios with the error stated in the attributes. New file nepal_flood_20260826_upper_lake_record.csv: the second, larger lake at the foot of the glacier collapse, carried as an attribute-only record with no geometry for the same reason — it is described only by words, and this atlas already has one inferred point its own radar failed to corroborate. Geometry was verified byte-identical to v7 against a pre-edit copy before release. Attributes in a GeoPackage changed, so a Zenodo version was minted: v8 (10.5281/zenodo.22168268), reached through the concept DOI 10.5281/zenodo.22118798.
• corrected 2026-08-29
  
  Nepal death toll revised again: 579 → 675, and the missing count moves back up, 1,924 → 2,418. more ▾

  The NDRRMA Search, Rescue and Relief Update timestamped 18:30 NPT on 29 August gives 675 bodies recovered and 2,418 people missing or out of contact (ANI, 29 August 21:35 IST, fetched first-hand). District breakdown of the 675: Chitwan 246, Nawalparasi East 165, Nawalparasi West 63, Gorkha 57, Nuwakot 51, Dhading 45, Tanahun 35, Rasuwa 13 — summing exactly to the total. Every district rose. 675 supersedes 579 and the chain below it; 2,418 supersedes 1,924, reversing the unexplained fall from 2,381 that this page recorded on 28 August. Two other Nepal totals are in circulation for the same day and neither is adopted: the English Wikipedia article 2026 Nepal floods gives 2,426 citing NBC News (revision 2026-08-30T02:18:34Z), and ABC News gives 2,498. Supporting counts also move: 219 injured treated or discharged, against the 101 remaining in hospital carried here from 28 August — different metrics, both shown; rescued 4,451 → 7,514 across 261 helicopter sorties, including 227 foreign nationals airlifted and 279 workers brought out of flooded hydropower tunnels; security personnel mobilised 15,431 → 18,708 (8,135 Nepal Police, 6,370 Nepal Army, 4,203 Armed Police Force); 145 of 198 damaged telephone towers restored and an automated flood-monitoring station established at Galchi. China is unchanged at 7 dead and 554 missing (Reuters citing Xinhua, 29 August), so the 16+ Chinese deaths and 546 Chinese missing carried in the Wikipedia infobox at that same revision contradict both the article's own lead and every source reachable here; not adopted. No spatial change; no Zenodo version.
• open 2026-08-29
  
  NDRRMA's published missing-person components no longer sum to its own headline — 2,498 against 2,418, an 80-person gap. more ▾

  On 28 August the eleven components it published summed exactly to the headline 1,924, which is why this page reproduced them as an arithmetic check. On 29 August they do not. The categories are 933 hydropower-project workers, 589 foreign citizens and 127 Nepalis travelling alongside foreign tourists, plus 104 security and government personnel (45 Nepal Army, 27 Nepal Police, 15 Customs, 13 Armed Police Force, 4 Immigration); the district reports are Rasuwa 562, Nuwakot 115, Makwanpur 65 and Langtang National Park 3. Read additively those total 2,498 — which is precisely the figure ABC News published — against the 2,418 NDRRMA gives as its own total. Neither NDRRMA nor ANI states whether the district figures are a separate geographic cut of the same population or a further additive category, so the identity with the ABC total may be coincidence or may be the origin of it. 2,418 is shown here as the authority's own total and 2,498 is not adopted. Two components fell while the total rose: Nepal Police 28 → 27 and Immigration 15 → 4. Foreign nationals rose 517 → 589, the Rasuwa district figure 161 → 562 and Nuwakot 1 → 115.
• corrected 2026-08-29
  
  The barrier lake now has a measured area on two dates and it is shrinking — 120,000 → 99,000 m² between 27 and 29 August — and both modelled pool polygons overstate it on every published date. more ▾

  China's Ministry of Water Resources, via CCTV and Reuters, 29 August, puts the lake at the Purepu Tsangpo confluence at 99,000 m² (9.9 ha) early Saturday morning, down 21,000 m² from Thursday's 120,000 m² (12.0 ha), with water flowing slowly through and the burst risk assessed as decreased. Two consequences. First, independent corroboration of the area. NDRRMA gave 0.11 km² (11.0 ha) for 27 August from Planet Labs and Landsat imagery via the Stimson Center; MWR gives 12.0 ha for the same day from its own satellite analysis. Two agencies, two imagery chains, one date, agreeing to within 9 percent — the first time any figure on this page has been corroborated across independent sources. Second, the modelled pools are now wrong by a measured margin, not merely unverified. The 2 Mm³ and 5 Mm³ polygons cover 19.1 and 29.5 ha against 9.9 ha observed — 1.93× and 2.98× — and against 12.0 ha on 27 August, 1.59× and 2.46×. They have not been redrawn: MWR published an area, not a geometry, and this page does not synthesise geometry from a scalar. MWR also describes the lake as straddling the China–Nepal border at the confluence of two rivers. That makes the Nepal-versus-China placement dispute carried here since 28 August a false dichotomy rather than a settled question, and it is recorded as such. The shrinking area is not reconciled with the >2,500,000 m³ volume reported for Friday 28 August; no source publishes both an area and a volume for the same instant. No spatial change; no Zenodo version.
• open 2026-08-29
  
  The second lake now has a number from a government technical agency, and it is the larger of the two: over 120,000 m², depth unknown, no visible outlet. more ▾

  Chen Hongqi, senior engineer at the Ministry of Natural Resources' Geological Disaster Technical Guidance Center, gives the body of water at the foot of the glacier collapse as over 120,000 m² with depth not yet established (Reuters, 29 August); Reuters separately reports satellite imagery showing this lake growing with no visible outlet, so pressure may still be building. This page has carried the two-lake account since 28 August on the attribution of one academic source, Nitesh Khadka via Al Jazeera; it is now stated by a Chinese government agency with a figure attached, and the upper lake is larger than the lower one this atlas has been tracking. An NDRRMA official says flood risk remains until both lakes have fully drained, so the de-escalation on the lower lake does not close the hazard window; a Chinese geohazards expert separately warns the lake may be unstable and that over 100 glacial lakes in the area could pose a threat. Neither lake is mapped in this package. The marker at 28.31249 N, 85.55406 E is a single inferred point for the lower feature and no authority has published a geometry for either. active_hazard.status moves from breached-disputed to two-lakes-draining-disputed.
• open 2026-08-29
  
  A second post-event optical pass exists and is unusable — 100.0 percent cloud within 300 m of the site. more ▾

  Sentinel-2C scene S2C_45RUM_20260829_0_L2A, acquired 29 August 05:00:50 UTC, is the first optical acquisition over the barrier-lake marker since the clouded 27 August pass. Reading the SCL band over a 5 km window rather than trusting the scene header: cloud and cloud shadow (SCL classes 3, 8, 9, 10) cover 100.0 percent of pixels within 300 m of 28.31249 N, 85.55406 E — 46.8 percent of them high-probability cloud — 99.2 percent within 1 km and 99.1 percent within 2.5 km. Scene-level cloud cover is 82.6 percent and understates the site, as it did on 27 August. No new Sentinel-1 acquisition since the 28 August 12:21:54 UTC scene already processed here. Copernicus EMS EMSR927 re-queried 30 August 02:3x UTC is unchanged in products[].version.number: AOI01 Syapru Besi grading v1, AOI02 Timure v2, AOI03 Bidur v1, AOI04 Bharatpur still status W. There is still no observed water extent for either lake. No spatial change; no Zenodo version.
• corrected 2026-08-29
  
  Errata log condensed and collapsed — presentation only, no figures touched. more ▾

  This log had reached 41 entries and about 64,000 characters, longer than the atlas it annotates. Two changes. First, every entry was rewritten shorter. Nothing was deleted, no date or status changed, and every published figure, supersession chain and source attribution was carried across intact; what came out is repetition — the boilerplate about casualty counts not being stored in the spatial package and the concept DOI always resolving to the newest version, which recurred verbatim in fifteen entries, and each entry’s restatement of AOI statuses and casualty totals already carried by its neighbours. Second, entries now collapse to their headline behind a disclosure control, so the log stays scannable as it grows. That is a display change only: the full text of every entry is in the page, one click away, and expands when printed. Nothing has been summarised away, and no entry will be removed from this log as the event develops. Later entries are written to the same tighter standard.
• corrected 2026-08-29
  
  Copernicus EMS delivered a third AOI — Bidur — and the observed extent more than triples, 240.44 → 829.52 ha. more ▾

  This page stated that EMSR927 had released Syapru Besi and Timure only (240.44 ha) and that AOI03 Bidur was at status I with no vector layers; both are superseded. On re-query at 29 August 03:0x UTC, AOI03 Bidur was delivered as a grading product at version 1, deliveryTime 2026-08-29T02:57:09 UTC — about four minutes before retrieval — photo-interpreted from BlackSky imagery of 27 August 12:09 UTC and Satellogic imagery of 27 August 04:22 UTC (EMSR927). At 589.079 ha it is the first observed extent anywhere below Syapru Besi, 50–60 river km downstream around Bidur and Devighat in Nuwakot, and larger than the two upstream AOIs together. It brings 2,343 graded buildings (1,826 Destroyed), 613 graded road segments (117.57 km), 20 graded bridge points (18 Destroyed) and 14 graded facility features. Every EMS count here is superseded: buildings 864 → 3,207, roads 44 → 650 segments, bridges 6 → 26. A new layer, ems_damage_facilities, spans all three AOIs (16 features, 14 Destroyed) and carries three named destroyed hydropower assets — the Langtang Khola Hydroelectric Power House 20MW, the Rasuwagadhi Hydropower Dam and the Trishuli Power House — each wholly inside the 80 m band; those features were in the AOI01 and AOI02 deliveries from v4 onward and had not been carried in the package, a gap stated here rather than quietly filled. Validation recomputed across all three AOIs against the unchanged 80 m height-above-channel band: observed extent inside the band 240.44 ha (92.1%) → 801.77 ha (96.7%); buildings graded Destroyed 673/695 (96.8%) → 2,497/2,521 (99.0%); destroyed bridges 6/6 → 24/24; destroyed road 15.09 km (96.3%) → 46.64 km (98.7%). AOI03 alone is 98.5% inside, the strongest of the three. The control weakens sharply, and matters more than the headline. Road graded ‘No visible damage’ falling inside the band goes from 11.6% (0.45 of 3.87 km) to 74.4% (61.25 of 82.38 km), and 77.4% at Bidur alone: in the confined upstream gorge the 80 m band separates damaged from undamaged ground, but in the broad Bidur valley most ground of any kind lies below 80 m above the river, so there it behaves closer to a floodplain mask than a damage predictor. The 96.7% should not be quoted without that. Spatial data changed, so a new Zenodo version was minted — v7 (10.5281/zenodo.22152427), reached through the concept DOI 10.5281/zenodo.22118798. Unchanged: AOI04 Bharatpur at status W with no layers, EMS has still mapped nothing at the barrier lake, and the pool polygons remain terrain-modelled.
• corrected 2026-08-29
  
  China’s Gyirong County figures revised again — 5 → 7 dead, and the missing count moves DOWN, 558 → 554 — but the two accounts are concurrent, not cleanly sequential. more ▾

  Xinhua reports 7 dead and 554 missing (CNN live coverage, 28 August; the same pair is carried by the BBC News live page of 28 August and by the English Wikipedia article 2026 Nepal floods, infobox at revision 2026-08-29T00:59:57Z). Against that, Reuters copy timestamped 28 August and still running on the ABC News live blog at 29 August 03:05 AEST gives 5 dead and 558 missing, attributing the missing figure to CCTV on Thursday and noting it includes 260 foreign nationals. The later Xinhua pair is shown here; the Reuters/CCTV pair is retained alongside it rather than deleted, because no source reachable here states that one supersedes the other, and the four-person reduction is unexplained. Nepal is unchanged at 579 dead and 1,924 missing (NDRRMA, 19:00 NPT 28 August). The two countries’ counts are still not summed and their overlap is unestablished. No spatial change; no Zenodo version.
• open 2026-08-29
  
  The barrier-lake hazard window does not close on 30 August — peak inflow is now reported for around 1 September, with rain forecast on the impoundment through the coming week. more ▾

  This page states that China’s Ministry of Water Resources expected a further 3,000,000 m³ of inflow through 30 August. Reuters reports that the same Thursday warning put peak flow at around 1 September, and that light to moderate rainfall around the Gyirong border port has loosened and destabilised soil and rock upstream, with rain forecast to continue over the coming week (Reuters via AsiaOne, 28 August). Nothing spatial changes — the pool polygons and the exposure envelope remain terrain-modelled screening surfaces — but the period in which they are the relevant screening surface for a second outburst runs at least two days longer than this page states.
• corrected 2026-08-29
  
  The first published AREA of the barrier lake has surfaced — 0.11 km² — and both modelled pool polygons are larger than it. more ▾

  NDRRMA states that a 0.11 km² lake formed about 18 km upstream of the Rasuwagadhi border point after the Lhende River was blocked again, and warns the natural dam could burst at any time. The figure comes from preliminary analysis of Planet Labs and Landsat imagery received 11:44 NPT on 27 August, using imagery provided by the Stimson Center (myRepublica, 27 August, fetched first-hand). This page has carried the 18 km placement since 28 August but never the area. 0.11 km² is 11.0 ha; the two modelled pools cover 19.1 ha (the 2 Mm³ scenario) and 29.5 ha (the 5 Mm³ scenario) — 1.74× and 2.68× the only published area. The active-hazard note says the polygons ‘bracket the reported size’: true on volume, false on area, where both overstate the lake and the smaller is 74 percent too large. That is the expected direction of error for a flood-fill on a pre-event Copernicus DEM containing neither the debris dam nor the post-event scour, but it should be stated rather than implied. The polygons have not been redrawn. NDRRMA published an area, not a geometry — no coordinates, no outline, no vector file — and 0.11 km² to two significant figures spans 10.5 to 11.5 ha. The area also sharpens the radar non-detection recorded here: the terrain-corrected Sentinel-1D search found no new open water larger than 1.5 ha in a 39 × 55 km envelope, and 11 ha is 7.3× that floor. The readings reconcile only if the lake drained or breached between the 27 August Planet/Landsat analysis and the 28 August 12:21 UTC radar pass, or its surface was too rough or debris-covered to read as water, or it lies outside the searched envelope; nothing published resolves which. Licensing: the underlying Planet Labs imagery is CC-BY-NC and is not used in this openly licensed dataset; the government figure derived from it is cited as such. No spatial change; no Zenodo version.
• open 2026-08-29
  
  Nepal’s own technical agency has placed the avalanche source on Nepali territory — the first Nepali government attribution of where the collapse began. more ▾

  The Department of Hydrology and Meteorology, working from satellite imagery shared by the Chinese side, concludes the trigger was a combination of ice avalanche, permafrost movement and rockslide. Binod Parajuli, senior divisional hydrologist in the Flood Forecasting Division, told The Kathmandu Post (28 August, updated 21:04 NPT / 15:19 UTC) that ‘the avalanche originated on Nepali territory, but after it came down it crossed the Nepal-China border, causing extensive damage on both sides’. This page describes the source as a glacier collapse north of Langtang Lirung and Tsangbu Ri and has deliberately not asserted which side of the border it sits on; that description is unchanged and the source point is not moved, but a Nepali government agency now states an answer. Two qualifications travel with it: Parajuli says a full technical assessment is still pending, the department’s field team not having reached the site; and the placement of the avalanche source is a separate question from the placement of the barrier lake, which remains disputed — Nepal Police, in the statement behind Friday’s downstream alert, describe a dam that ‘has burst again on the Tibetan side’ (India TV citing the Nepal Police statement, 28 August), against Al Jazeera placing it on the Nepali Lhende system and the Chinese Embassy placing the avalanche in a tributary on the Nepali side. This page adopts none of the barrier-lake placements. No spatial change; no Zenodo version.
• open 2026-08-29
  
  A first magnitude for the source collapse has been published — 100 to 200 million cubic metres — with a mechanism and a fall height that do not match this page. more ▾

  This page describes ‘an unnamed glacier roughly 610 m wide that fell about 1,200 m, most of it melting from the friction of the impact’, and carries no volume at all. Dr Mohd. Farooq Azam, cryosphere specialist at ICIMOD, told The Kathmandu Post (28 August, updated 21:04 NPT) that the failure began in the bedrock — ‘when the bedrock gave way, the glacier on top also got swept’ — and that ‘the glacier collapse was 100 to 200 million cubic meters in size and hit the valley 1000 meters below’. Three differences: the ordering (this page has the ice failing, ICIMOD the rock failing first); the fall height (1,000 m against 1,200 m); and the volume, published as a factor-of-two range and given verbally to a newspaper rather than measured in a released dataset. Neither figure has been adopted — substituting one uncited number for another would not improve the record, so both are on it with attribution and the discrepancy is stated rather than resolved. ICIMOD’s Sunwi Maskey adds the destructiveness mechanism: frictional melting of the ice into water during the descent, plus entrained soil and rock, turned an ice-rock avalanche of a kind ‘frequent in Nepal’ into a valley-filling flood. No spatial change; no Zenodo version.
• corrected 2026-08-29
  
  Bodies have been recovered outside Nepal, in India, and the recovery-district list is superseded again. more ▾

  This page states that bodies were recovered in Dhading, Gorkha, Chitwan, Tanahun and East Nawalparasi; that list is incomplete in both directions. Nepal Police via The Kathmandu Post (28 August) now list West Nawalparasi as well. More consequentially, seven bodies believed to be flood victims have been recovered across the international border in Uttar Pradesh, India: four in Maharajganj district on 27 August — ‘we think they have been swept into India from Nepal following flash floods there’, district magistrate Gaurav Singh Sogarwal told AFP (India TV / AFP, 28 August) — and three in Kushinagar, roughly 70 km downstream of the frontier. Wikipedia (revision 2026-08-28T19:57:00Z) carries the same two districts, citing BBC live coverage. No number on this page changes: the seven are not stated by any source to be additional to Nepal’s 579, they are unidentified, and NDRRMA has not counted them. But they sharpen the standing limitation that the 62.4 km corridor understates the flood — transported remains several hundred kilometres beyond the modelled corridor, through the Trishuli and Narayani into the Indian plain, well outside every layer in this package. The envelope, damage register and height-above-channel surface all stop at Devighat and are a lower bound on reach. No geometry was extended and the corridor is not redrawn on the strength of body-recovery reports.
• corrected 2026-08-28
  
  Copernicus EMS re-delivered the Timure observed extent and it grew, so the envelope validation is recomputed — and the corroboration weakened. more ▾

  This page states that AOI01 Syapru Besi and AOI02 Timure are unchanged at delivered v1. On re-query of the EMSR927 activation API at 19:00 UTC, AOI02 Timure is at grading version 2 (deliveryTime 2026-08-28T17:10:43 UTC, statusCode F); the download path silently repointed from the v1 to the v2 archive with no status field changing, so the re-delivery is visible only in products[].version.number. The observed-event polygon is revised upward from 125.875446 to 129.360664 ha: 3.588 ha added across 71 disjoint additions along the margins of the mapped scour, 0.112 ha removed, bounding box unchanged to seven decimal places — a refinement of interpretation, not a re-mapping. AOI01 remains at v1, untouched. The graded damage layers did not change: the v2 building (431 points), road (15 segments, 9.6991 km) and bridge (1 point) layers were compared feature by feature against v1 and are identical in geometry, damage grade and length. The validation did. Against the unchanged 80 m band, Timure falls from 115.95 ha (92.1%) to 117.46 ha (90.8%) inside, and the two-AOI headline from 236.95 ha at 92.8% to 240.44 ha at 92.1%: more of the newly mapped ground sits above the band than below it, so the independent test of this atlas’s terrain surface got weaker rather than stronger, and is reported in the direction it fell. The threshold still holds — 92.1% of observed ground inside a band drawn from terrain alone, against 11.6% of road graded ‘No visible damage’. AOI01 band figures are carried forward verbatim rather than recomputed, so no spurious sub-hectare drift enters the record. Spatial data changed, so a new Zenodo version was minted — v6 (10.5281/zenodo.22149681), reached through the concept DOI 10.5281/zenodo.22118798. Unchanged: AOI03 Bidur at status I with no vector layers (expected 28 August 19:30 UTC), AOI04 Bharatpur at status W (Legion planned 29 August 04:01 UTC), nothing mapped at the barrier lake, pool polygons still modelled, and casualties at 579 dead and 1,924 missing.
• corrected 2026-08-28
  
  Nepal death toll revised again: 567 → 579 — and this figure comes straight from NDRRMA, first-hand, replacing a Sky-News-via-Wikipedia citation chain. more ▾

  The NDRRMA Search, Rescue and Relief Update timestamped 19:00 NPT on 28 August reports 579 bodies recovered and 101 injured still in hospital (ANI via The Tribune, 28 August 20:28 IST, fetched first-hand; the same update is carried by UPI, cited in turn by Wikipedia at revision 2026-08-28T16:50:11Z). District breakdown: Chitwan 233, Nawalparasi East 154, Gorkha 46, Dhading 40, Nuwakot 37, Tanahun 30, Nawalparasi West 27, Rasuwa 12 — summing exactly to 579, with NDRRMA noting the Gorkha figure includes separated human remains. Three districts moved down against the Nepal Police 15:00 NPT list carried here (Nuwakot 38→37, Nawalparasi West 28→27, Rasuwa 13→12), consistent with de-duplication as bodies are identified, though no source states that. The 567 figure is retired: it entered only through Sky News as cited by Wikipedia, its underlying agency was never established, and NDRRMA’s own first-hand count supersedes it. No spatial change; no Zenodo version.
• corrected 2026-08-28
  
  The missing count is no longer a live disagreement between two agencies — NDRRMA has adopted 1,924 as its own figure, and its earlier 977 is superseded rather than concurrent. more ▾

  Every previous entry framed 977 and the 1,924 reaching this page via Sky News as concurrent counts from different agencies, neither adopted. That no longer holds. The NDRRMA 19:00 NPT update gives 1,924 and, for the first time, its composition: 933 hydropower-project workers, 517 foreign nationals, 127 Nepali citizens resident abroad, 161 through the Rasuwa District Emergency Operations Centre, 66 from Makwanpur and 1 from Nuwakot, plus 45 Nepal Army, 28 Nepal Police, 13 Armed Police Force, 15 Customs and 15 Immigration personnel and 3 from Langtang National Park (ANI via The Tribune, 28 August). The 517 foreign nationals NDRRMA previously reported as its whole missing category are one component of the larger total, which resolves how 977 and 1,924 coexisted: they counted different populations, not the same one differently. Security-personnel counts also move against situation report 6 (Nepal Army 44→45, Immigration 4→15). The 2,381 from Nepal Police at 11:32 NPT remains superseded and the downward move to 1,924 remains unexplained by any source reachable here. The published figure does not change; its attribution and standing do.
• open 2026-08-28
  
  The barrier lake has grown to more than 2,500,000 m³ — larger than the volume the modelled pool polygon was built to match. more ▾

  Every figure carried here has been China’s Ministry of Water Resources estimate of about 2,000,000 m³ (1.5–2 million as a range) from the morning of 27 August. Al Jazeera and Reuters, 28 August 09:30 UTC report the volume swelled past 2.5 million by Friday, measured by the eight-member Chinese engineering team flying aerial surveys and 3D modelling. The current volume sits between the 2 Mm³ and 5 Mm³ scenarios already drawn, so both remain valid bracketing screening surfaces and no geometry changed — but the smaller polygon is now the lower bracket, not ‘the reported size’. The lake growing while simultaneously being reported as breached and draining is not reconciled by any source here, and is consistent with the Chinese Embassy’s 28 August statement that it remains high and overflowing with a breach risk. The same report notes the overflow has blocked delivery of relief supplies (IFRC), and puts Nepal’s deployment at 30,000 security personnel against NDRRMA’s own 15,431 — not reconciled here. No spatial change; no Zenodo version.
• open 2026-08-28
  
  Copernicus EMS EMSR927 has moved on the two unreleased AOIs, but not in the direction this page predicted, and still yields no new observed extent. more ▾

  This page states that AOI03 Bidur and AOI04 Bharatpur are both at status W, tasked on Legion imagery acquired 28 August 04:01 UTC with delivery expected 28 August 17:01 UTC. On re-query: AOI03 Bidur has moved to status I (in production), mapped from BlackSky imagery of 27 August 12:09 UTC and Satellogic imagery of 27 August 04:22 UTC — not Legion — with both orthorectified COGs published and delivery moved to 28 August 19:30 UTC; mapsCount is 0 and no observed event, building, road or bridge layer exists. AOI04 Bharatpur has slipped a full day: still status W, Legion acquisition now scheduled 29 August 04:01 UTC with delivery expected 29 August 17:01 UTC, flagged as newly planned. AOI01 and AOI02 are unchanged at delivered v1. The consequence is unchanged: the observed extent still covers only two settlements (236.95 ha), everything else remains terrain-modelled, and EMS has mapped nothing at the barrier lake, 17.6 km upstream of the border and outside all four AOIs. The activation is now cross-linked to International Disasters Charter activation 1052, and EMS publishes activation-level totals of 864 identified buildings, 16 km of roads, 11 ha of built-up area and 900 estimated population across the mapped AOIs.
• open 2026-08-28
  
  Still no second satellite pass — there is as yet no observed water extent for the barrier lake, and none is guaranteed. more ▾

  An Earth Search query over 28.31249 N, 85.55406 E across 26–30 August returns exactly two acquisitions, both already reported here: the Sentinel-2B L2A scene of 27 August 05:10:45 UTC (78.5 percent scene cloud, fully clouded over the site) and the Sentinel-1D IW GRDH scene of 28 August 12:21:54 UTC, terrain-corrected on the previous run and finding no new open water anywhere in a 39 × 55 km envelope. Nothing has been acquired since. The pool polygons therefore remain terrain-modelled screening surfaces, the barrier-lake marker remains an inferred medium-confidence location, and the ‘no observed extent’ caveat on the exposure envelope stands everywhere outside the two EMS-graded settlements.
• corrected 2026-08-28
  
  The first post-event radar has been processed, and it does not find the barrier lake. more ▾

  The Sentinel-1D pass flagged in the previous entry has been terrain-corrected and searched. Across a 39 × 55 km envelope covering the entire mapped upstream drainage — the Lhende Khola, Chhochen Khola, Purepu Tsangpo and Chusumdo Tsangpo — there is no new open-water body larger than 1.5 ha. At the inferred marker the surface reads −8.7 dB in VV, changed by −0.33 dB since 16 August; across the modelled 2 Mm³ pool footprint, −8.2 dB and −0.43 dB. Those are 0.15 and 0.19 standard deviations against a 2.23 dB noise floor from four pre-event 12-day pairs — not a small signal, no signal; the surface on 28 August is radiometrically the same as 12 days earlier. Water as rough as −12 dB would have shown at z = 14–24, calm water at −22 dB at z = 74. This establishes that the impoundment is not at the location this page marks and that nothing observable happened there — no inundation, no scour, no debris. It does not establish that no lake exists: C-band radar cannot distinguish a sufficiently rough, turbulent or debris-covered water surface from land, 20.1% of the envelope is unusable for layover and shadow, the marker was always location_confidence=medium, and a lake beyond 28.60°N / 85.80°E fell outside the search. The marker has not been moved and the pool polygons have not been withdrawn — they are now explicitly uncorroborated at the inferred location rather than merely unverified. Method and controls. A Range-Doppler terrain correction written for this package: trajectory from the product orbit state vectors, zero-Doppler solved per DEM cell, slant-to-ground range via the product srgr polynomial, σ0 from the product calibration LUT, layover/shadow from exact radar geometry; the geocoding reproduces the product’s own geolocation grid to a maximum error of 1.89 m over 210 points. The reference is the 16 August pass on the same relative orbit (85, ascending, Sentinel-1D, 12-day repeat), so layover and shadow cancel. As a positive control the same detector was run against the EMSR927 observed extent at Syapru Besi and Timure: absolute VV change is significantly elevated inside the photo-interpreted damage (Mann-Whitney p = 5.1×10^-21; changes above 6 dB enriched 2.63-fold, Fisher p = 6.7×10^-18), confirming the chain detects real ground change. Package republished as v5.
• open 2026-08-28
  
  China now says the barrier lake is still high and overflowing and that the breach risk remains — which does not sit with the ‘draining, risk smaller than feared’ account this page has carried. more ▾

  The Chinese Embassy in Kathmandu, reported by The Kathmandu Post, says the landslide-dammed lake is at a high water level and has begun to overflow, that a risk of breach remains, and that the situation is ‘at a critical stage for rescue and emergency response in both Nepal and China’, urging downstream communities to monitor river levels. This is not presented as superseding the breach report — it is a later statement from a different authority describing the impoundment as still impounding, against the Rasuwa Chief District Officer’s Army-sourced breach report and CCTV’s reported ~10 m drop. Both are recorded and neither adopted; active_hazard moves from breached-draining to breached-disputed so the page stops asserting a settled outcome it cannot support. On the placement dispute, the embassy puts the ice-rock avalanche ‘in a tributary on the Nepali side of the upper reaches of the China-Nepal boundary river’ — a third source, leaning toward Nepal, against Al Jazeera (Nepal) and The Kathmandu Post (Chinese side) — but it describes the avalanche site rather than naming the impoundment, so the marker is not moved and location_confidence stays medium. The embassy gave no size and no volume, so the last published figures stand (about 1.5–2 million m³, China’s Ministry of Water Resources, 27 August). No spatial change; no Zenodo version.
• open 2026-08-28
  
  The first post-event radar acquisition over the barrier lake now exists. more ▾

  Every earlier entry states that an Earth Search query over 28.31249 N, 85.55406 E returns no Sentinel-1 or Sentinel-2 acquisition after the fully clouded 27 August 05:10 UTC Sentinel-2 pass. That is no longer true. Scene S1D_IW_GRDH_1SDV_20260828T122141_20260828T122206_004326_007FA4 was acquired 28 August 12:21:54 UTC (18:07 NPT) — ascending, relative orbit 85, dual-pol VV/VH, 10 m pixel spacing — and processed to Fast-24h at 15:05 UTC, about an hour before this check. Its footprint contains the marker, radar is unaffected by the cloud that has blocked every optical pass since the event, and it is post-breach, postdating both the morning breach report and the reported ~10 m fall in level. Nothing here is derived from it yet and the pool polygons remain terrain-modelled. The reason is stated rather than glossed: the AWS product is a Level-1 GRD in radar geometry with orbit-and-GCP metadata, not a map-projected analysis-ready raster, and a defensible water extent requires radiometric terrain correction against a DEM. In a gorge of this relief at 28°N on an ascending pass, layover and shadow mimic the low-backscatter signature of standing water, so an uncorrected threshold would produce a plausible-looking polygon that is largely artefact — worse than publishing nothing. Filed open: the imagery exists, the processing does not, and the modelled polygons stand until it does. No spatial change; no Zenodo version.
• corrected 2026-08-28
  
  Nepal casualty figures revised again: 547 → 567 dead, and the missing count moves DOWN, 2,381 → 1,924. more ▾

  Wikipedia’s 2026 Nepal floods (revision 2026-08-28T15:00:56Z, retrieved 15:05 UTC / 20:50 NPT) states 567 dead and 1,924 missing, citing Sky News (28 August live page). The Sky News page was not reachable from this run — the domain is blocked to this environment — so the figure enters only through that chain, stated in full rather than presented as first-hand. The underlying Nepali agency behind 567 and 1,924 is not established by anything reachable here: Nepal Police’s 15:00 NPT release gave 547 across eight districts, NDRRMA via The Kathmandu Post gave 538 at 15:52 NPT, and neither has been restated. The missing figure moving downward is the more consequential change. The published 2,381 came from Nepal Police via The Himalayan Times at 11:32 NPT; 1,924 is later and lower, consistent with people being accounted for as district lists reconcile, but no source reachable here explains the 457-person reduction. It is adopted as the latest sourced Nepal-wide total and the earlier figure recorded rather than erased. NDRRMA’s concurrent 977 is unchanged and still not superseded by either number. One caution: the same revision carries 1,473 in its infobox as injuries, but the Nepal Police release it comes from describes 1,473 as people injured and rescued (Rasuwa 572, Nuwakot 898, Dhading 3, with 94 treated or discharged in Kathmandu Valley hospitals) — a composite, not an injury count. This page publishes no injury figure. China unchanged at 5 dead and 558 missing. No spatial change; no Zenodo version.
• open 2026-08-28
  
  Two new impact counts are published that this page’s exposure envelope cannot yet be checked against. more ▾

  IFRC head of delegation in Nepal David Fisher told a Friday-morning press conference that roughly 93,000 people have been affected on an initial estimate, and appealed for CHF 25 million, naming a second flood wave from the newly formed lakes as the principal concern (Ratopati, 28 August). Separately the independent power producers’ association reports 934 hydropower workers and staff missing across 11 of 13 damaged projects — 730 in Rasuwa (Upper Trishuli-1 alone 576, Rasuwagadhi 93, Langtang Khola 42, Chilime 8, Mailung Khola 7, Rasuwa-Bhotekoshi 4) and 204 in Nuwakot (Upper Trishuli-3B 158, Upper Trishuli-3A 42, Middle Trishuli Ganga 2, Trishuli 1, Devighat 1), with 254 rescued from Upper Trishuli-1 (Ratopati, 28 August). Neither reconciles with anything here. The 93,000 is an affected-population estimate on an unstated footprint and is not comparable to the 17,167 residents this atlas counts inside the 80 m envelope, which is a lower bound on a narrow ribbon. The 934 exceeds the 60 hydropower workers the government reported missing on 26 August by more than an order of magnitude and is not separable from the 1,924 Nepal-wide total. This page maps six damaged Nepal Electricity Authority facilities, five inside the envelope; the 13 damaged projects named here include several not in that register. The damage register has not been rewritten, because these are press-reported counts without published geometry. Status open pending an official facility-level damage assessment.
• open 2026-08-28
  
  The barrier lake is now described as one of two new lakes, and a second source places the confluence impoundment on the Nepali side of the Lhende system. more ▾

  Nitesh Khadka, a Nepali physical geographer specialising in glacial floods, told Al Jazeera (28 August, updated ~12:50 UTC) that satellite images and drone video show two lakes: an upper lake in China near the glacier, where meltwater is trapped, and a lower lake — the larger and more dangerous — on the Lhende, where a landslide blocked the valley and dammed the river. Pawan Bhattarai of Tribhuvan University described the lower feature to the same outlet as an avalanche-induced barrier lake formed where the ice-rock avalanche deposited rock, ice and debris. Al Jazeera places the Chhochen Khola / Purepu Tsangpo confluence lake ‘on the Lhende river system in Nepal’, whereas The Kathmandu Post has placed it on the Chinese side; Al Jazeera is itself internally inconsistent, elsewhere calling it a lake ‘on the Nepal-China border’. This page adopts neither placement. No authority has published coordinates, the marker stays at its inferred position (28.31249 N, 85.55406 E, location_confidence=medium), and the ‘same feature?’ question remains open. Two further details change no published figure: China’s Ministry of Water Resources is quoted giving the impoundment as about 1.5 to 2 million m³ (a range, against the single 2,000,000 m³ carried here from The Kathmandu Post), and CCTV reported on 28 August that an eight-member Chinese engineering team has been mobilised to fly aerial surveys and build a 3D model, and that risks from the overflowing lake were ‘manageable’. That survey, if published, would be the first authoritative geometry for this feature. No spatial change; no Zenodo version.
• corrected 2026-08-28
  
  Nepal death toll revised again: 538 → 547, and the authority behind the headline figure changes from NDRRMA to Nepal Police. more ▾

  Nepal Police, in an update timestamped 15:00 NPT on 28 August, put bodies recovered at 547 across eight districts: Chitwan 221, Nawalparasi East 134, Gorkha 44, Dhading 40, Nuwakot 38, Tanahun 29, Nawalparasi West 28, Rasuwa 13 (Ratopati, 28 August, an AI-translated republication of the Nepali original, corroborated independently by PTI via The Tribune, updated 28 August 18:11 IST / 12:41 UTC, carrying the same breakdown). The districts sum exactly to 547. This supersedes the 538 from NDRRMA earlier the same afternoon; the two are the same kind of count from different agencies, and the later, higher, more granular police figure is adopted. Note the shift in what the headline is sourced to: 547 is bodies recovered, as counted by police, not a separately adjudicated confirmed-dead register. The missing count is not changed by this update: the police release says only that thousands remain missing, so the 2,381 from Nepal Police at 11:32 NPT is held, as is NDRRMA’s concurrent 977 — which PTI restates at the later cut-off as 977 including 86 security personnel, 517 foreign tourists and 127 Nepali travellers. The two remain concurrent and unreconciled. Chitwan recoveries, given here as 186, are now 221; foreign tourists rescued, given as 117, are now 121 (Nepal Tourism Board via PTI: 85 Indian, 16 Chinese, 9 South Korean). No spatial change; no Zenodo version.
• corrected 2026-08-28
  
  China’s death toll in Gyirong County revised: 3 → 5. more ▾

  Chinese authorities now report 5 killed in Gyirong County, Tibet Autonomous Region (PTI via The Tribune, 28 August); the same figure is carried by CCTV as cited by Wikipedia’s 2026 Nepal floods (retrieved 28 August 13:01 UTC). The Chinese missing count is unchanged at 558, including 260 foreign nationals. It remains unestablished whether the Chinese and Nepali counts overlap, and they are not summed anywhere on this page.
• corrected 2026-08-28
  
  Nepal death toll revised again: 489 → 538 — and the missing count is a live disagreement between two agencies, not a supersession. more ▾

  NDRRMA put the confirmed toll at 538 by Friday afternoon, per The Kathmandu Post (day-three explainer, updated 28 August 15:52 NPT / 10:07 UTC, fetched first-hand), superseding the 489 shown here, which reached this page only through a citation chain. The same cut-off gives 977 still missing, including 517 foreign nationals — the figure NDRRMA published in situation report 6 on the night of 27 August. This corrects a framing error carried here: the 2,381 missing reported by Nepal Police via The Himalayan Times at 11:32 NPT was described as superseding NDRRMA’s 977. It does not — the two are concurrent, from different agencies counting different things, and NDRRMA has not moved off 977. The published missing figure is held at 2,381 (the higher and broader police register) and both are now recorded side by side. Nepal Police separately still list 28 of their own personnel unaccounted for. Two narrower counts moved: Chitwan recoveries 178 → 186, and the Nepal Tourism Board now reports 117 foreign tourists rescued from Rasuwa (85 Indian, 16 Chinese, 9 South Korean, 2 American, 2 Italian, 3 nationalities unestablished), superseding the 31 rescued reported on 27 August. China unchanged at 3 dead and 558 missing. Figures are attributed rather than reconciled, averaged or aggregated. No spatial change; no Zenodo version.
• open 2026-08-28
  
  Nepal’s government has not confirmed that the barrier lake collapsed, and the accounts are not reconciled. more ▾

  Government spokesperson and Education Minister Sasmit Pokharel said Nepal had information that glaciers had broken, causing water mixed with debris to overflow, that there was no information a dam itself had collapsed, and that China had indicated no immediate major threat (The Kathmandu Post, 28 August 15:52 NPT, which states explicitly that this and the breach report ‘have not yet been fully reconciled’). That stands against the Rasuwa Chief District Officer’s Army-sourced breach report earlier the same day and CCTV’s reported ~10 m fall in level. active_hazard remains breached-draining, but it is now recorded that no Nepali government confirmation of a collapse exists and that Nepal has not established how much water, if any, was released. Separately, NDRRMA issued a fresh downstream alert at 14:15 NPT: the Bhote Koshi level in Rasuwa has risen again, carrying mud and debris, and spokesperson Shanti Mahat urged crews along the Bhote Koshi and Trishuli to exercise caution (The Himalayan Times, 28 August 14:15 NPT). Two standing checks could not be completed on this run and are not claimed as verified: the EMSR927 activation page renders its product list client-side and returned no product table, and the Earth Search STAC query over 28.31249 N, 85.55406 E could not be issued. The last verified status is carried forward and is now stale — AOI01 and AOI02 delivered at v1, AOI03 and AOI04 at status W with delivery expected 28 August 17:01 UTC, and no acquisition after the fully clouded 27 August 05:10 UTC pass. No observed extent of any second flood exists, and the pool polygons remain terrain-modelled.
• corrected 2026-08-28
  
  The breached barrier lake is now draining, and the second-flood risk has been downgraded. more ▾

  Following the breach reported this morning, officials on both the Nepali and Chinese sides said the risk was smaller than feared because the lake is draining slowly rather than failing catastrophically, and CCTV reported the water level had fallen by about 10 m. Rescue work at Gyirong Port, halted when the lake burst its banks, has resumed, as has rescue work in Nepal (CNN live coverage, 28 August, as cited by Wikipedia’s 2026 Nepal floods at revision 2026-08-28T09:00:21Z, retrieved 09:00 UTC; the primary outlet was not reachable from this run, so the chain is stated in full). This supersedes the statement carried here that no assessment of the breach had been published. Two things remain unestablished: no confirmed release volume, and no observed extent of any second flood. Nothing spatial changed. Earth Search over 28.31249 N, 85.55406 E still returns no Sentinel-1 or Sentinel-2 acquisition after the 27 August 05:10 UTC pass (78.5 percent scene cloud, fully clouded over the site), the only Sentinel-1 pass in the window being pre-event on 24 August. EMSR927 AOI03 and AOI04 remain at status W with no layers, delivery expected 28 August 17:01 UTC; AOI01 and AOI02 unchanged at v1. The pool polygons and the archived barrier_lake status attribute are unchanged and no Zenodo version was minted — a narrative and casualty update only.
• corrected 2026-08-28
  
  Nepal death toll revised again: 475 → 489. more ▾

  Nepal News reports the confirmed toll at 489, with over 2,300 missing, superseding the 475 shown here (The Kathmandu Post citing police, 28 August 11:22 NPT). Retrieved as cited by Wikipedia’s 2026 Nepal floods at revision 2026-08-28T09:00:21Z (retrieved 09:00 UTC); the primary outlet was not reachable from this run, so the chain is stated in full rather than the figure presented as first-hand. The missing count is left at 2,381 (Nepal Police via The Himalayan Times, 11:32 NPT): the same revision carries 2,381 in its infobox but 2,389 in body text, and that eight-person discrepancy is unresolved by any source reachable here, so the lower published figure is retained rather than silently moved. China unchanged at 3 dead and 558 missing, including 260 foreign nationals, overlap with Nepal still unclear. Figures are attributed rather than reconciled or aggregated.
• open 2026-08-28
  
  The barrier lake has breached. more ▾

  Rasuwa Chief District Officer Narendra Pariyar said on 28 August that the Nepali Army had just informed him the impoundment near the Chhochen Khola / Purepu Tsangpo confluence had breached, per The Kathmandu Post (updated 28 August 12:49 NPT / 07:04 UTC). NDRRMA says the Bhote Koshi level has surged, and Nepal is preparing aerial reconnaissance over the Bhote Koshi and surrounding Rasuwa using helicopters already at Dhunche. This supersedes the statement carried here that no breach had been reported as of 28 August 06:00 UTC. Two things are not established: no detailed assessment and no confirmed release volume have been published, and the report is single-source pending that reconnaissance. The last published volume figures remain China’s Ministry of Water Resources estimate of about 2,000,000 m³ on Thursday morning with a further 3,000,000 m³ of inflow expected over three days. For this atlas: the two modelled pools (2 Mm³ and 5 Mm³) were screening surfaces for a potential outburst that is now reported to have occurred; they remain terrain-modelled and are not an observed extent, and none exists — Earth Search returns no acquisition after the fully clouded 27 August 05:10 UTC pass, and EMSR927 AOI03 and AOI04 remain at status W, delivery expected 28 August 17:01 UTC. The archived barrier_lake layer still carries status="open - high breach risk" against its 27 August reporting date; that attribute has deliberately not been rewritten and no Zenodo version minted, because the breach is a single-source report awaiting official assessment. Casualties unchanged at 475 dead and 2,381 missing. One narrower count has appeared and does not reconcile: the US State Department is tracking 90 Americans unaccounted for, with three rescued and no reported US fatalities, against the 65 Americans on the Nepal Police missing list. Both are recorded and neither adopted.
• corrected 2026-08-28
  
  Nepal casualty figures revised again: 469 → 475 dead; a far larger missing count now published. more ▾

  The Kathmandu Post, citing police, put the confirmed toll at 475 at 28 August 11:22 NPT (05:37 UTC), superseding the 469 shown here. Separately The Himalayan Times, citing Nepal Police at 11:32 NPT, reports 2,381 unaccounted for — including 644 foreign nationals (178 Indian, 65 American, 53 Ukrainian, 51 Malaysian, 35 Australian, 33 British, 25 Canadian) and 127 Nepalis resident abroad — and gives the police death toll as 469 across eight districts, largest recoveries Chitwan 178 and Nawalparasi East 110. The 2,381 supersedes the 977 from the NDRRMA sixth situation report of Thursday night. The jump is a change in the register, not the event: it reflects consolidation of district-level and tourism-sector lists, and the two agencies were counting different things. Both are recorded, not reconciled or averaged. Two further narrower counts also disagree: the Ministry of Foreign Affairs tabulated 596 foreign nationals still missing out of 627 (31 found) in its 27 August update, and the Nepal Tourism Board counts 667 tourists and travellers missing (127 Nepali). The Post and the Himalayan Times disagree on the death toll by six; the higher, later-timestamped figure is shown. China unchanged at 3 dead and 558 missing. No spatial change; no Zenodo version.
• open 2026-08-28
  
  NDRRMA independently locates the blockage, close to but not identical with this page’s inferred marker. more ▾

  NDRRMA states the obstruction sits on the Lhende Khola about 18 km upstream of the Rasuwagadhi border, confirmed on a satellite image acquired 27 August, and repeats its warning that a breach could send a second major flood down the Lhende, Bhote Koshi and Trishuli. This atlas’s marker at 28.31249 N, 85.55406 E lies 17.6 km straight-line from Rasuwagadhi, on the Purepu Tsangpo. Straight-line distance is a lower bound on along-channel distance, so if NDRRMA’s 18 km is measured along the river the true blockage likely sits somewhat downstream of the marker. It is also not established that NDRRMA’s Lhende blockage and the impoundment China’s Ministry of Water Resources placed at the Chhochen Khola / Purepu Tsangpo confluence are the same feature — different agencies, one obstruction or two. The marker has not been moved and location_confidence remains medium. No breach has been reported as of 28 August 06:00 UTC. Separately, EMSR927 AOI03 Bidur and AOI04 Bharatpur are now formally tasked on Legion imagery acquired 28 August 04:01 UTC, delivery expected 28 August 17:01 UTC, but both remain at status W with no layers. Earth Search still returns no acquisition after the 27 August 05:10 UTC pass, 78% cloud at scene level and fully clouded over the site. The pool polygons therefore remain terrain-modelled.
• corrected 2026-08-28
  
  Nepal casualty figures revised again: 389 → 469 dead, 910 → 977 missing. more ▾

  The Kathmandu Post, citing Nepal Police, put the confirmed toll at 469 on the morning of 28 August (09:05 NPT / 03:20 UTC). The missing total is from the NDRRMA sixth situation report of 27 August night: 977 unaccounted for, including 517 foreign nationals and 127 Nepalis resident abroad. The two come from different agencies and cut-offs and are reported as such, not reconciled. The Nepal Tourism Board separately counted 667 tourists and travellers still missing on Thursday night (127 Nepali), superseding the 668 previously shown. China unchanged at 3 dead and 558 missing. Some wire coverage quotes a combined ‘more than 1,300 missing’ across both sides; that aggregate is not adopted here. No spatial change; no Zenodo version.
• corrected 2026-08-28
  
  Data-package card corrected: it described v3, the file served is v4. more ▾

  After the 28 August republication the download card still read ‘ZIP v3 — 9 GeoPackages’. The archive at that link has been the v4 package since upload: 13 GeoPackages, 13 FGDC sidecars, 15 CSVs and 4 GeoJSONs, including the Copernicus EMS EMSR927 observed event extent, the graded buildings, roads and bridges, and the envelope validation tables. A label correction only — no figure, geometry or layer changed, and no Zenodo version minted.
• corrected 2026-08-28
  
  First observed extent published, and the envelope validated against it. more ▾

  Copernicus EMS Rapid Mapping activation EMSR927 released grading products for AOI01 Syapru Besi and AOI02 Timure, photo-interpreted from WorldView-3 and Legion imagery acquired 27 August 05:05 UTC. Four layers enter the package: an observed event extent (111.078 ha and 125.875 ha), 864 graded buildings, 44 graded road segments and 6 graded bridges. EMS types the mapped surface as 6-Mass Movement / Landslide — the scoured and debris-covered ground the flow left, not a standing water extent — and the activation carries no delineation product. Tested against this atlas’s terrain envelope, 92.8% of the 236.95 ha observed area, 96.8% of the 695 buildings graded Destroyed, all 6 destroyed bridges and 96.3% of the 15.67 km of destroyed road fall inside the 80 m band, against only 11.6% of the 3.87 km graded ‘No visible damage’. Package republished as v4.
• open 2026-08-28
  
  The observed extent covers two settlements, not the corridor. more ▾

  EMSR927's released AOIs are Syapru Besi (river km 13.8) and Timure (river km 2.3) — together 236.95 ha against a 39.4 km² envelope. AOI03 Bidur and AOI04 Bharatpur were tasked on Legion for 28 August 04:01 UTC but had no layers published at retrieval. Everywhere else on this page the surfaces remain terrain-modelled screening products. The barrier lake pool polygons are unaffected and remain modelled: EMS did not map the impoundment, which sits 17.6 km upstream of the border, and an Earth Search query over the site returns no Sentinel-1 or Sentinel-2 acquisition after the fully clouded 27 August 05:10 UTC pass.
• open 2026-08-28
  Barrier lake pool polygons added, MODELLED from the Copernicus DEM to match the reported volumes - not an observed extent. The only post-event optical pass is 100 percent cloud within 300 m of the site and no post-event SAR was available. Package republished as v3 (10.5281/zenodo.22135493).
• corrected 2026-08-28
  Data package republished as v2 (10.5281/zenodo.22135085) adding the barrier lake and upstream drainage layers and raising the glacier-collapse source record to high confidence. This entry now cites the concept DOI 10.5281/zenodo.22118798, which always resolves to the newest version.
• open 2026-08-28
  Barrier lake added to the map at 28.31249 N, 85.55406 E - an inferred location. No authority has published coordinates; the marker sits at the only OSM-verified confluence in the described area. location_confidence=medium. Postdates the Zenodo deposit and is not in the archived package.
• corrected 2026-08-28
  Casualty figures revised 157 -> 177 -> 389 dead with 910 missing in Nepal; China reports 3 dead and 558 missing in Gyirong County. Spatial data and Zenodo deposit unchanged.
• corrected 2026-08-28
  Source location resolved in favour of USGS - the seismic origin sits 1.5 km due north of Langtang Lirung and 64.4 km from Kathmandu, matching both published descriptions of the collapsed glacier.
• corrected 2026-08-28
  Official damage counts updated to 41 bridges and 42 km of road, Rs 200 billion national estimate, and 431 MW removed from the grid. Prefer official figures for damage totals.
• corrected 2026-08-28
  The 80 m envelope threshold is corroborated by reporting that floodwaters reached settlements about 80 m above river level.
• open 2026-08-28
  The 62.4 km corridor understates the flood, which satellite analysis puts at up to 80 km downstream. Exposure figures are a lower bound on extent as well as on population.
• corrected 2026-08-27
  Downstream scope limitation superseded: bodies recovered in Dhading, Gorkha, Chitwan, Tanahun and East Nawalparasi.

Keystone GIS documents known data issues here rather than silently revising published datasets. Entries are collapsed to their headline; open one for the full text, evidence and attribution. Nothing is removed from this log.