At 8:40 in the morning on 26 August, the automatic gauge on the Bhote Kosi at Nepal’s border with China reported the river at 1.62 metres. The reading was low, and it was falling.
It was the last reading that station sent. Devesh Belbase, a Kathmandu-based water policy and governance expert, records in Nepali Times that a wall of water coming down the Lhende Khola destroyed the equipment before the next reading was due. When the flood arrived at Devghat late in the afternoon, at the far end of the monitored river, it peaked at 6.57 metres, below the 7.3-metre warning level the Department of Hydrology and Meteorology publishes for that station.
Three stations went out below their thresholds
The gauge on the Bhote Kosi at Syaphrubesi, 15 kilometres downstream of the border, died when the water reached 3.8 metres. Betrawati was knocked out at 3.55 metres. Belbase records that all three stations went out below their prescribed warning thresholds, and the department’s own published levels bear him out: 6 metres at Rasuwagadhi, 5.5 metres at Syaphrubesi on the Bhote Kosi, 4.1 metres at Betrawati on the Trishuli.
Those figures are stage heights, and three metres at one gauge is not three metres at another: a stage height only means something against the datum and the thresholds of the station that measured it.
Belbase calls 6.57 metres “the highest point measured during this flood event”. That comparison does not survive the company of his own sources: the rapid hazard assessment filed on the disaster by eight scientists under the HiRISK banner records the Kalikhola crossing a danger threshold of 12.1 metres in the early afternoon, a higher number against a higher line.
The far end of the monitored river
The flood peaked at Devghat at 4pm, on the Narayani, 168 kilometres from the Tibet-Nepal border by the assessment’s impact table.
The Flood Forecasting Division’s own findings, published the next morning, gave the peak flow at Devghat as 5,850 cubic metres per second, the figure the assessment also carries. Belbase’s reading of that number is that it describes a kind of flood that comes once every two to three years, so that what he calls Nepal’s deadliest and most destructive flood in recent history arrived at the last measuring station on the river as a fairly ordinary event.
There is no mystery in the arithmetic, and Belbase spells it out: over its course to Devghat the river had been joined and diluted by the Budi Gandaki, the Marsyangdi, the Seti and the Kali Gandaki and other tributaries.
The question a gauge asks
Nepal’s flood gauges are designed to warn about monsoon rains that drain a catchment and raise river levels to seasonal heights, Belbase writes, and the warning levels are set from that history, for which purpose he says the system works well. This flood did not behave that way. It began as a rock and ice avalanche off the Lirung massif and then, in his phrase, “entered a gorge as a wall of debris and spent its height as it travelled”.
His account of the afternoon is that by the time the flood reached that end of the river the water was not especially high. It was rising very fast. The Narayani came up almost two metres within two hours, on a day it had been falling. His summary of the design problem is that the gauges ask a river one question, how high are you, and do not ask how quickly are you rising.
The speed is already in the data: every flood gauge station, he writes, reads its river six times an hour. That is why the record shows the water at Malekhu, 100 kilometres downstream of the border, rising almost four metres between one ten-minute report and the next before crossing its red alert mark.
Belbase times that at 11:20 and calls it the first threshold crossed anywhere on the network, though his own account also records the Phalakhu Khola, which joins the Trishuli at Betrawati, crossing its warning level at ten past nine the night before and still rising. The assessment’s timeline records a crossing at Furke, Malekhu, at 11:43.
Belbase’s proposal is a second trigger beside the height thresholds, one that fires when a river rises abnormally fast, and his estimate is that with it Malekhu could have raised the alarm before 11:30 and Devghat by mid-afternoon. That is a judgement about a warning that was not issued, and no source read for this piece tests it.
What the warning system did that morning
The alerting that day did not begin with a gauge, and it did not begin at nine in the morning either. At eight, an hour before the telephone rang, the Flood Forecast Bulletin for 25 August had gone out listing Rasuwa, Nuwakot, Dhading and Gorkha, along with four other districts, at high flash-flood risk on small rivers, and forecasting a notable rise on the Trishuli at Betrawati. Belbase records that the districts in the flood’s path were Code Orange that morning, and that the 26 August bulletin put 30 districts and adjoining ones under some level of risk.
His objection is not that no warning existed. It is his own line that “a warning issued every day, carrying no responsibility for anyone, is not a warning. It is a weather forecast.”
At nine the Flood Forecasting Division learned of the flood by telephone, from the District Administration Office in Dhunche and then from an observer at Betrawati. The assessment’s timeline logs the Department of Hydrology and Meteorology activating early warning at 9:15 to 9:16, declaring high alert the length of the Trishuli corridor down to Mugling, and sending 632,295 text messages; its impact table splits that total as 432,635 through Nepal Telecom and 243,660 through Ncell, which come to 676,295. Belbase reports 679,295, so more than 600,000 is the figure all three readings support.
At 11am, five hours before the peak at Devghat, the timeline logs a high risk declaration urging evacuation along the Bhotekoshi and Trishuli corridor in Rasuwa, Nuwakot, Dhading, Gorkha and Chitwan. That last district is where Devghat sits, and where the assessment’s impact table records 139 of the 469 deaths it had confirmed by 28 August, the largest number it assigns to any named district. Belbase’s judgement is that the alerts stayed ahead of the water down to Devghat, and he records a school principal in Bidur moving 1,643 students to higher ground before the debris arrived.
The assessment’s own verdict splits the valley in two. Early warning systems had recently been set up in the region, its authors write, focused on several glacial lakes in China, but none was operational in this place for glacier-related hazards.
Given the speed of the avalanche they judge that warning from conventional in-channel systems would have been minimal and likely insufficient for successful evacuation at Rasuwagadhi. Further downstream, they write, with warning times of 10 minutes or more, comprehensive and well-capacitated systems could have saved a significant number of lives.
The numbers in this account are still moving
Almost everything quantitative here is provisional, and some of it disagrees with itself. The assessment’s impact table records 469 confirmed deaths, 977 people out of contact and 3,253 helicopter rescues. The footer of its own event timeline, in the same document on the same date, says 359 deaths and 1,976 rescues, and its running text says more than 1,000 out of contact.
Those were the figures as of 28 August. Nepal’s National Disaster Risk Reduction and Management Authority put the toll at 1,114 dead and 3,916 missing a week after the flood, with hundreds of people believed trapped in hydropower tunnels. The confirmed-death count has been revised sharply upward since, and the assessment calls its own list of pre-event indications hindsight.
The green reading rests on one source. The assessment reports the same 6.57-metre peak at four in the afternoon, but that is not a second measurement: the source line under its own timeline credits a DHM technical report and an NDRRMA situation report, so the peak and the discharge figure trace back to the same government reporting.
The department’s flood monitoring page does publish the threshold, listing the Narayani at Devghat with a warning level of 7.3 metres and a danger level of 9. What rests on Belbase alone, in a signed commentary rather than a wire report, is his statement that the station “stayed green throughout the flood of 26 August”.
It also overstates matters to say the network cannot see a rising river at all. The same page sorts stations into nine states, combining three level bands with three trends, and one of those states is below warning level and rising, so the direction is on the display. The threshold that turns a station amber reads the level. Nothing in it reads the rate.
In the hours before the peak, that is where the last gauge on the Narayani sat, in the category the dashboard keeps for a river under its warning line and climbing. What reached Devghat ahead of the water had come from a telephone call, a daily bulletin and thresholds crossed at gauges upstream. It never came from the gauge at Devghat.