26 August 2026, minute by minute and day by day
Bottom Line
- The parts existed; the joins didn’t. On 26 August, warning and response ran on phone calls and loudspeakers, not on design.
- The hazard outran the instruments. It covered about 20 km in 7 minutes. Gauges reporting every 10 minutes never saw it: the last upstream reading, at 08:40, was normal.
- 38 minutes from collapse to the first SMS. 23 of them were spent waiting for a phone call.
- The one official message said “remain on high alert”. No action, no place, no time. At about 200 SMS a second, it may have taken an hour to send.
- After impact, no common picture. The confirmed death toll more than doubled after the first 72 hours.
This is the first part of a three-part blog series examining Nepal’s Bhotekoshi disaster.
Part 1 of 3: what happened. Part 2, “A Decade-Long Cycle: Nepal’s Emergency Communication Resilience Report Card, 2016–2026”, asks why. Part 3, “Degrade by Design, or Collapse by Default: A Governance Choice”, sets out how to design for the next crisis.
On 26 August 2026, every part of Nepal’s emergency communication system existed: seismic monitoring, river gauges, a forecasting agency, a national SMS service, two mobile networks, volunteer mappers and a command post. What failed were the joins between them. Where a join should have been, an individual improvised: a district officer with a phone, a police officer with a loudspeaker, a head teacher with a direction.
A Hazard That Outran the Warning System
At 08:37 on 26 August, a slab of bedrock and the glacier ice on top of it broke away at about 5,200 m on the north side of Langtang Lirung and fell into the Lhende Khola, a border stream that joins the Bhote Koshi at Rasuwagadhi (USGS; Kathmandu Post). The avalanche began on Nepali territory, then crossed the border four times in its first 14 km (Gujree, preprint). Its impact registered as the equivalent of a magnitude 5.2 earthquake (USGS; GFZ estimated Mw 5.7), and was at first taken for one (The Landslide Blog). No significant rain had fallen: this was an ice-rock avalanche that became a debris flow, possibly after damming the Lhende for a few minutes (HiRISK). It reached the Gyirong–Rasuwagadhi border crossing, about 20 km downstream, roughly seven minutes later (CCTV). A second, smaller collapse followed at 11:45.
One rapid analysis argues that debris had dammed the river for 18 to 19 hours before 08:37 (IRDR). The seismic record, satellite imagery and gauge readings published so far do not support that, and ICIMOD lists the question as open (ICIMOD). This post follows the weight of evidence: a single collapse at 08:37.
The mechanism matters for policy. A rain flood builds over hours and gives forecasters a rising signal to track. This hazard announced itself through the rock at the source, then outran every instrument downstream.
The human cost is still being counted. Nepal Police put the toll at 1,457 dead and 4,720 missing as of 10 October (ePardafas).
Reading the Record: One Clock, Two Timelines
Disaster reviews usually split early warning from response: different agencies, budgets and reports. This post keeps both on one clock, with T = 08:37, the moment of collapse.
Figure 1, The Clock (early warning) counts in minutes (T+3, T+23, T+38) and in distance down the river from ground zero (D). It asks whether a usable message reached people before the water did. Each box is labelled with its time
Figure 2, The Golden 72 Hours (incident management) counts in hours and days. It asks whether access came back and whether responders shared one picture. Each box is labelled with its date.
In both figures, black boxes are the system and green boxes are people on the ground. Red bands on Figure 1 mark where people were killed. References in brackets, such as (T+23) or (29 Aug), point to the matching box.
Early Warning: The First 77 Minutes
Figure 1 tells the first 77 minutes on one line, from the collapse at ground zero to the last deaths about 70 km downstream. Appendix A lists the evidence behind each box.
[Figure 1: The Clock (T = 08:37, D = Ground Zero).]
What the Numbers Reveal
The reported times and readings support a few simple calculations. None of them appears in the official record. Each one points to a policy gap.
| Measure | Calculation from the record | What it means for policy |
| Speed of the hazard (T+0 to T+7) | ~20 km in ~7 min ≈ 170 km/h | Faster than a gauge that reports every 10 minutes |
| Last upstream reading (T+3, T+13) | Normal 1.62 m at 08:40, against a 6 m warning; surge at the border ~08:44; the 08:50 reading never came | The upstream gauges never saw the flood. Downstream, Betrawati died at 3.55 of 4.1 m (87%, T+43) |
| First threshold crossed (T+111 onwards) | Malekhu, T+163, ~100 km down | Threshold alerting fired nearly three hours in, long after the deaths upstream |
| The window (T+23 to T+43) | First phone call T+23 to water at Betrawati T+43 = 20 min | The only stretch with both time and a person holding a phone; the head teacher used ten minutes of it |
| Collapse to first SMS (T+0 to T+38) | T+0 to T+38 = 38 min: 23 to detect by phone, 5 to confirm, 10 to send | Most of the delay came before any message existed |
| SMS queue (T+38 to T+77) | 679,295 ÷ ~200 SMS/s ≈ 3,400 s ≈ 57 min | Sending from 09:15 would run to about 10:12 (T+95), after the last deaths |
Sent is not delivered. Nepali Times reported that all 679,295 messages went out “within one minute” (Nepali Times). But DHM’s own stated sending rate is about 200 SMS per second, and a 2025 study found alert delays of up to 1 hour 43 minutes; Nepal Telecom says delivery rates are not measured (NIMJN). If that rate held across both networks, dispatch alone took close to an hour. This is our estimate, not a reported figure.
Whose clock? Start times differ: 09:15 (DHM, via Nepali Times), 09:16 (Spotlight Nepal), an NDRRMA warning at 09:28 (OnlineKhabar opinion), and “by 9:35” (the Prime Minister). The one dated receipt, 09:54 in Nuwakot, came 19 to 39 minutes after dispatch, depending on whose clock is used.
What the message said. “Remain on high alert” told people that danger existed, not what to do about it, where, or by when. By 09:05 the flood had been observed upstream and confirmed (T+23, T+28). The message still carried no instruction, no place and no time.
Where the Warning Chain Broke: Sensing, Authority, Message
Figure 1 shows three breaks, and an individual stepped into each one.
Sensing. The first network to fail was not the phone network. It was the sensor network. Four hydrological stations were destroyed: Rasuwagadhi, Syabrubesi, Betrawati and Malekhu (National Herald). They were placed to measure a river’s height, not to survive or detect a hazard that covered 20 km in 7 minutes. The upstream gauges’ last readings were normal; their next reports never came (T+3, T+13, T+43). The one signal that did come from the source, the seismic trace, was read as an earthquake (T+0). Detection fell to a phone call (T+23).
Authority. The Flood Forecasting Division could confirm the flood within five minutes of the call (T+28). What it could not do was turn that confirmation into an order. The duty of officials to act on a DHM warning has never been written down (T+28). On the ground, the people who told others to move were police officers on loudspeakers (T+23 to T+43) and a head teacher acting on an urgent call from upstream (T+33).
Message. One channel, SMS, carried one message: “remain on high alert” (T+38). It reached at least one household with the water (T+58 to T+77). The head teacher acted on a phone call that gave him what the SMS did not: a reason to move now.
Credit belongs to the people who improvised. The policy finding is that the system depended on them.
Incident Management: The Golden 72 Hours
Once the water passed, the clock switched from minutes to days, and the network was dark for the whole first 72 hours. Figure 2 tells that story; Appendix B lists the evidence behind each box.
[Figure 2: The Golden 72 Hours (T = 08:37, 26 August).]
Where the Response Broke: Restoration and Coordination
Figure 2 shows two breaks, and again people filled them by hand.
Restoration. Access came back through effort, not design. Temporary microwave links, helicopters and a ropeway (27 Aug–3 Sep) are self-healing, but improvised. By day six, 92% of Nepal Telecom’s sites were back (110 of 120, 1 Sep). By day fourteen, only 71% of fibre-to-the-home lines were (1,093 of 1,544, 9 Sep). The fibre that failed had been strung in the air, across the terrain most exposed to the hazard (T+0 to +10 h).
Coordination. Every piece of situational awareness existed. China warned of a second flood (27 Aug), volunteers mapped the damage (27 Aug), and fifteen coordination teams stood up in Kathmandu (Kathmandu box). Our reading of the record is that nothing joined these into one common operating picture. When the barrier lake overflowed, rescue paused for everyone at once (28 Aug).
A toll that took weeks to know. Without a common picture, even the size of the disaster had to be assembled slowly, from different counters:
| Date | Dead | Missing | Source |
| 10 Oct | 1,457 | 4,720 | Nepal Police, via ePardafas |
| 6 Sep | 1,342 | 4,896 | NDRRMA SitRep #06, via RRN |
| 1 Sep | 987 | 3,916 | NDRRMA SitRep #01 |
| 29 Aug | 669 | 2,426 | Police (dead), NDRRMA (missing), via HimalPress |
The confirmed death toll more than doubled between day three and day 45. The record also mixes police and NDRRMA counts. For responders deciding where to send helicopters, the picture in the first 72 hours was a fraction of the real one.
The Handover Gap Between Warning and Response
The most important finding sits where the two figures meet. Figure 1 ends with the last deaths at about T+60 (last-deaths band) and an SMS queue that may have run past T+90. Figure 2 begins with the network already dark (T+0 to +10 h). Nothing carried information across that seam: no record shows the warning chain telling responders who had been warned, where, or with what result.
Read on one clock, both halves share one pattern:
- The parts existed. Seismic monitoring, gauges, a forecasting division, a national SMS service, two operators, volunteer mappers, a command post.
- The joins did not. Detection was not wired to authority, authority was not wired to action, and responders were not wired to one another.
- Individuals filled the gaps. A district officer and a gauge staffer on the phone, police on loudspeakers, a head teacher, engineers rebuilding links by hand, volunteers mapping from afar.
This is not a story of technology failing. Technology always degrades in a crisis. It is a story of a system with no design for how it degrades.
The Question This Record Leaves for Policy
The people who improvised on 26 August saved lives, and this record is not a case for blame. It raises a different question.
A failure seen for the first time is a lesson. A failure seen before, written down, and seen again is a choice. So which was 26 August?
The next post, A Decade-Long Cycle: Nepal’s Emergency Communication Resilience Report Card, 2016–2026, answers that. It sets the 2026 record against the recommendations LIRNEasia and ISOC-Nepal made after the 2015 earthquake (LIRNEasia, 2016). The third post, Degrade by Design, or Collapse by Default: A Governance Choice, sets out what to build before the next one.
By Nuwan Waidyanatha (Senior Research Fellow, LIRNEasia)
Technical Appendix: Evidence Behind the Figures
Each box on Figures 1 and 2 is listed below, by the time printed on it, with what the record shows and where it comes from. Times are Nepal time on 26 August 2026 unless stated; T = 08:37, the moment of collapse.
Appendix A. Figure 1, The Clock
| Box | What the record shows | Source |
| T+0 · 08:37 | Collapse at 08:37:10 on the north side of Langtang Lirung into the Lhende Khola; energy ≈ M5.2 (USGS), Mw 5.7 (GFZ); first catalogued as an M4.4 earthquake | USGS; Kathmandu Post; ICIMOD; Severe Weather Europe |
| T+3 · 08:40 | Border gauge’s last reading, normal (1.62 m per Nepali Times and Spotlight; 3.8 m per National Herald); 10-minute reporting; surge reaches the Gyirong–Rasuwagadhi crossing ~08:44, ~20 km in ~7 min | Nepali Times; Spotlight Nepal; National Herald; CCTV |
| First losses band | 9 Armed Police personnel at Timure and 14 customs staff missing | Khabarhub |
| T+8 · 08:45 | 08:45: residents hear a roar, think it is an earthquake, run uphill | Nepali Times; Kathmandu Post |
| T+13 · 08:50 | The 08:50 transmission never arrives; upstream gauges destroyed. FFD gives the 08:40 reading as 1.62 m against a 6 m warning level | OnlineKhabar, FFD technical report; Spotlight Nepal |
| T+23 · 09:00 | 09:00: DAO Dhunche, then a Betrawati gauge staffer, phone the Flood Forecasting Division | Nepali Times; Spotlight Nepal |
| T+23 to +43 · police | Police loudspeakers at Betrawati and Dhunge; many did not believe it | Khabarhub |
| T+28 · 09:05 | 09:05 confirmation; officials’ duty to act on a DHM warning “never written down” | Spotlight Nepal; Nepali Times |
| T+33 · 09:10 | 09:10: head teacher Rajendra Dawadi moves ~900 students; flood hits ten minutes later | Mongabay |
| T+38 · 09:15 | From 09:15: SMS to 679,295 phones, urging people to “remain on high alert” | Nepali Times; The Daily Star |
| T+43 · 09:20 | Betrawati last 3.55 m against a 4.1 m warning level; silent 09:20 | OnlineKhabar, FFD technical report |
| T+58 to +77 | PM: messages sent “by 9:35”; one Nuwakot resident receives it at 09:54 | NepalNews; Mongabay |
| Last-deaths band | Homes swept in Bidur-7; students swept away in Tupche (time estimated) | Kathmandu Post |
| T+111 onwards | Galchhi +9 m in 30 min; Malekhu Red at 11:20; Devghat peaks 6.57 m, below Amber | ThePrint; Nepali Times |
Sources agree on the times (last readings at 08:40, silence from 08:50) but differ on which upstream station read 1.62 m and which read 3.8 m. This post follows the FFD technical report. The 18–19 hour dam theory is noted in the text but not used.
Appendix B. Figure 2, The Golden 72 Hours
| Box | What the record shows | Source |
| 26 Aug · T+0 to +10 h | 87 towers down (NT 60, Ncell 27); ~128 km of aerial fibre and two optical line terminals lost; 41 bridges washed away | OnlineKhabar; Radio Nepal; NDRRMA SitRep #01 |
| 27 Aug–3 Sep | Ncell loses 60+ km of fibre; temporary microwave links; equipment by helicopter and ropeway | Kathmandu Post; NepaliTelecom |
| 27 Aug · barrier lake | 27 Aug: China warns of a 1.5–2 million m³ barrier lake upstream | Kathmandu Post |
| Kathmandu box | Command post; 15 thematic emergency coordination teams; warehouse at Tribhuvan International Airport | NDRRMA SitRep #01 |
| 27 Aug · HOT and NAXA | 27 Aug: HOT and NAXA launch damage mapping, later used in NDRRMA’s assessment | OpenStreetMap Wiki; HOT |
| 28 Aug | 28 Aug: the lake overflows; rescue pauses | Al Jazeera |
| 27–30 Aug | A Kathmandu monastery shelters ~200 displaced, run by volunteers | ABC |
| 29 Aug | 29 Aug: helicopters ferry people and supplies; 669 dead (police), 2,426 missing (NDRRMA) | Kathmandu Post; HimalPress |
| 1 Sep | 1 Sep: 110 of 120 NT sites back; Timure, Chilime Khola and Trishuli towers destroyed, covered by alternatives | NDRRMA SitRep #01; NepaliTelecom |
| 9 Sep | 9 Sep: 1,093 of 1,544 fibre-to-the-home lines back, first phase | Radio Nepal |
