SagarmathaIQ
·
DISASTER RESPONSE SCORECARD · UPDATED 8 SEP 2026

From the 2015 earthquake to the 2026 flood: did Nepal get better at disaster response?

Eleven years and one law separate the Gorkha earthquake from the Bhotekoshi flood. We scored both responses against a fixed rubric to find out what the Nepal Government actually learned.

Updated Sep 8, 2026. The Armed Police Force swept the Upper Trishuli 3A tunnel on Sep 4 and called off the search, resolving on day nine rather than within the 72-hour window this rubric requires. That moves search and rescue reach from level 2 to level 1 and the 2026 total from 70 to 65 percent. Casualty, financing and reconstruction figures throughout are refreshed to Sep 8; changes are noted inline.

At 8:40 on the morning of 26 August, the river gauge at Syabrubesi read 1.62 meters, comfortably below its six-meter warning line. Ten minutes later it stopped transmitting. By 9:16, emergency alerts had gone out to 679,295 people along the riverbanks of four districts downstream.

Thirty-five minutes, from a dead sensor to two-thirds of a million warnings. In 2015 no part of that chain existed. It is a fair measure of what eleven years and one law bought Nepal.

It bought nothing at all for Timure and Syabrubesi. The wall of water reached the border settlements at roughly the same minute the first message was sent. Rasuwa, closest to the source, got no warning, and Rasuwa is where most of the dead and missing come from.

That pattern holds across this whole comparison. The machinery the Nepal Government built after the Gorkha earthquake, above all the standing disaster authority created by the 2017 law, does what it was designed to do. What it was never designed to do, put local governments in charge, or stop the state from rebuilding in a river's path, remains undone.

Separating the two is the point of the exercise. A nationwide earthquake and a two-river-corridor flood are not the same test, so the useful question is not which government did better. It is which capacities improved, which did not, and which cannot be judged yet.

Two structural facts shape everything below. The 2015 quake was diffuse and hit Kathmandu directly. The 2026 flood was linear, and struck Nepal's least densely populated district first. Scale differs by roughly a factor of five in affected population, and by far more in the concentration of damage.

2015 Gorkha earthquake
60%
12 of 20 points
+5points, the net of four dimensions moving in both directions
2026 Bhotekoshi flood
65%
13 of 20 points

Scored across five dimensions where both disasters have real, undisputed evidence on both sides. Five further dimensions are excluded rather than guessed at. This measures the emergency phase only. As of this update, the 2026 flood is 13 days old, nearly 5,000 people are still missing, and reconstruction is only just starting. These scores will move.

2015
8,790
deaths (GoN Post-Disaster Needs Assessment); 8 million affected across 31 of 75 districts, 14 declared crisis-hit
2026
1,355
deaths and 6,672 injured as of Sep 5 (NDRRMA), 4,996 missing; nearly 1.6 million affected across 62 districts (NDRRMA situation reports, Sep 1–5)
2015
4 hours
to Cabinet approval of the international assistance request
2026
Same day
NDRRMA activation; 14 helicopters mobilized within hours

This piece is about the response. For how the flood itself formed and moved, a glacier collapse driving a debris flow that covered its first 20 km at roughly 180 km/h, see our companion analysis, Physics of the Bhotekoshi-Rasuwagadhi Flood.

How the scoring works

Every dimension carries equal weight and uses the same four-level scale, from 1 at worst to 4 at best. The level descriptions are fixed before either year is placed against them. Five dimensions times four levels puts the ceiling at 20 points, so the total converts straight to a percentage.

Weighting the dimensions unequally, at 30, 30, 20, 10 and 10, produces exactly the same two totals. The result does not depend on how the dimensions are weighted, which is why they are not.

Below, the bar shows all four levels for each dimension. Filled segments show where each disaster landed. Where both landed on the same level, the segment is split.

Resource mobilization

How fast the emergency apparatus activated, and how fast executive sign-off followed on major decisions.

1delayed past 72h, or no standing apparatus
2activated in 48 to 72h
3activated within 24h
4same day, sign-off in hours
2015 at level 4
NEOC activated same day; Cabinet approved the international assistance request within about four hours.
2026 at level 4
NDRRMA activated same day; helicopters mobilized within hours.

Search and rescue reach

Time to reach or resolve the single hardest documented access problem in each disaster: the epicenter zone in 2015, the trapped hydropower-tunnel workers in 2026.

1unresolved past 72h
2resolved in 48 to 72h
3resolved in 24 to 48h
4resolved under 24h
2015 at level 3
The epicenter area was reached by helicopter the day after the quake.
2026 at level 1
Rescuers spent three days locating the entrances to flooded tunnels holding roughly 900 hydropower workers, then five more before controlled drilling and explosions opened the Upper Trishuli 3A tunnel. On Sep 4, day nine, an Armed Police Force team swept every accessible section and the operation was called off: two of the site's 42 workers had been pulled out alive that morning, one body was recovered, and the other 39 were left unaccounted for under mud and debris. That is well past the 72 hours this rubric sets for "resolved." The prior version of this score, a level 2 with a flagged caveat that a stricter reading would put it at 1, is now settled by the outcome. As of Sep 7 the Nepali Army still counted about 121 hydropower workers unlocated across the flood-hit tunnels generally.

Caveat: this dimension scores time, not difficulty. 2015's problem was reaching a helicopter landing zone; 2026's was extracting workers from a flooded, roadless tunnel, a harder technical job. The rubric doesn't distinguish between them, on purpose: levels were fixed before either outcome was known, so scores can't be re-argued after the fact.

Last-mile relief

How fast relief funds reached a sub-national distribution point, and whether local governments were meaningfully involved or bypassed.

1months of delay, corruption reporting, local structures bypassed
2within a month, local govt substantially bypassed
3within a week, local govt partly involved
4within 72h, local govt leading
2015 at level 1
The National Reconstruction Authority, the body meant to disburse funds, wasn't created for eight months. A year on, most victims had received nothing beyond a small initial payout. Tarpaulin-procurement corruption was documented.
2026 at level 3
Rs 67.5 million reached 15 local units within six days, and by Sep 8 the government had released about Rs 1 billion to Rasuwa district alone from a Prime Minister's relief fund that had grown to Rs 11.77 billion. But local governments were reported unable to lead relief in the worst-hit areas, and 35 civil-society figures accused the federal government of running relief through a centralized "one-door system" that sidelined elected local officials. The government partly reversed course on Sep 1, forming six-party political coordination committees in Rasuwa and Nuwakot to route relief through local governments.

International coordination

Engagement with the standing multinational search-and-rescue architecture, and whether financing became one coordinated instrument.

1minimal engagement, financing inadequate even per capita
2narrow bilateral only, financing fragmented
3partial multinational engagement, financing partly coordinated
4full UNDAC engagement and one coordinated instrument
2015 at level 3
A full UNDAC-run coordination center registered 386 international teams, and a single $415 million appeal moved money fast. The openness itself created real congestion at the airport and command center.
2026 at level 2
We found no reporting of a UNDAC deployment. Assistance arrived through narrow bilateral deals after a disputed initial refusal. Financing fragmented across separate agency asks, and on the government’s own affected-population figure it was thinner per person than 2015.

Victim identification and compensation

Whether a systematic identification protocol exists, and how fast the compensation channel actually moved money rather than announcing it.

1undocumented process, compensation announced but not disbursed for months
2ad hoc identification, some early disbursement
3systematic protocol, overwhelmed; channel moving within weeks
4majority identified in days, compensation disbursed in weeks
2015 at level 1
A peer-reviewed forensic review called the recovery process poorly organized, with no documented recovery location for many remains. Most compensation still hadn't reached people a year later.
2026 at level 3
A systematic DNA-and-numbered-tag protocol is in place despite overwhelmed capacity: of 1,068 bodies recovered as of Sep 8, about 400 are partial remains, and only around 60, roughly 5.6%, had been matched and returned to families. A 19-member Indian forensic team joined Nepali counterparts at the national forensic labs to speed up DNA matching. Rs 67.5 million had already moved to local units within six days, over payment rails that did not exist in 2015. That tranche sits inside a far larger flow: the Prime Minister's relief fund had taken in roughly Rs 11.77 billion by Sep 8, up from Rs 7.9 billion on Sep 2.
Level scored on each of five equally weighted dimensions for the 2015 earthquake and 2026 flood responses, totalling 12 and 13 of 20 points, or 60 and 65 percent.
Dimension20152026
Resource mobilization44
Search and rescue reach31
Last-mile relief13
International coordination32
Victim identification and compensation13
Total of 201213
As a percentage60%65%

Because the lowest level still scores 1, the scale floors at 25 percent rather than zero. Scored 0 to 3 instead, the same placements give 47 percent and 53 percent. The five-point gap between the two responses holds either way.

What isn't scored, and why

Foreign assistance acceptance. Disputed. Reporters said Nepal turned foreign rescue teams away; the foreign minister denied that two days later, citing terrain. Undisputed: none were working in the first 48 hours. We score what is settled.
Early warning, infrastructure restoration, financial accountability, long-term resilience, command visibility. Each fails on at least one side. An earthquake gives no lead time. Visibility is not performance. Reconstruction has barely begun. That is 45 of a fuller framework's 100 points, left out rather than spread across the rest and called measured.
Eleven years of technology, not just of institutions. The largest confounder here, and it favors 2026 throughout. Mass SMS alerting, daily satellite revisit and digital payment rails did not exist usefully in 2015; better tools alone would have lifted these scores. Discount the 2026 lead accordingly. It cuts the other way once: with better hazard data than 2015 ever had, rebuilding in the same flood path reads worse, not better.

Reading the score against two very different disasters

Raw totals would mislead when one event is five times larger than the other, and a misleading number here would distort what the response score above actually says. Three quantities are worth normalizing, not to compare the disasters for their own sake, but to make sure the response comparison stays fair: the money that came in, the damage left behind, and the dead.

International financing

Raw international financing favors 2015 by roughly eight to one. Normalizing by affected population cuts that lead sharply, but it does not reverse it. An earlier version of this piece said it did, on a far lower affected-population estimate than the government has since published.

2015 earthquake $415 million
2026 flood $49.6 million

A single coordinated UN appeal in 2015 against fragmented agency asks in 2026. On absolute dollars, 2015 raised roughly eight times more.

The correction matters for how the international coordination score reads. Less money moved in 2026, and on the government’s own figure it moved for a comparable order of people, not for a tiny fraction of them. The fragmentation critique stands on its own; a per-person defense of the 2026 figure does not.

Damage, and what recovery will cost

Reconstruction isn't in the five-dimension score above; the emergency phase is barely over and rebuilding has not been judged. But both governments sized their rebuild the same way, which makes this the cleanest read on where the response goes next. The 2015 Post-Disaster Needs Assessment put reconstruction at $6.7 billion, roughly 30 percent of GDP. Finance Minister Swarnim Wagle puts the 2026 rebuild at $4 to $5 billion, close to a tenth of the economy. NDRRMA's own preliminary loss assessment, released Sep 3-4, came in lower still, at roughly Rs 387.5 billion, or about $2.9 billion. The two government numbers, a bottom-up loss tally and a top-down rebuild estimate, have not been reconciled.

So the fiscal burden the government faces is about a third of 2015's, on the GDP-share measure. That is the one dimension on which this response starts with an easier hand, and it matters, because fiscal capacity decides whether reconstruction finishes.

Restoration has moved fast where it could: by Sep 4, Nepal Telecom had restored 116 of 120 damaged towers and Ncell 76 of 78, most within two weeks. Three sites, at Trishuli in Nuwakot and Tiru and Timure in Rasuwa, remain destroyed. Roads and power are months away, and reconstruction of the worst-hit corridor only began in the second week.

Two caveats on reading a smaller bill as an easier job. The cost is concentrated, not spread, and it is still climbing as assessment reaches the hardest-hit corridor: road and bridge losses, an initial Rs 15 billion estimate in early September, had risen to about Rs 30 billion by Sep 8, with reconstruction of that sector alone projected near Rs 50 billion. Of 63 motorable bridges between Rasuwa and Chitwan, 41 were destroyed outright and 4 more damaged, leaving 18 intact. Damaged hydropower is over 12 percent of national generating capacity. And 2015 is a warning on delivery, not on cost: that rebuild took a decade, overran its own estimate by a third, and is still unfinished in places. A smaller bill only makes this government's job easier if it is spent faster than the last one was.

Why the death rate isn't in the score

This is the place a reader is likeliest to mistake the response score for a verdict on the disaster itself, so it is worth separating the two directly. Rates per 100,000 are the standard humanitarian denominator for comparing unlike events, and applied here they show why a faster, better-run response and a deadlier disaster can both be true at once.

On confirmed deaths alone, the 2026 flood now looks only modestly less lethal than 2015: about 85 deaths per 100,000 affected as of Sep 5, against roughly 110 per 100,000 in 2015 — a gap that has narrowed as the confirmed toll rose from 1,114 to 1,355. But 2015 ended with very few people unaccounted for, while 2026 has 4,996 missing against those 1,355 confirmed dead, most of whom will not be recovered. Counting the missing, 2026 runs at roughly 397 per 100,000 affected, close to three-and-a-half times the 2015 rate. None of that moves the 65 percent above, up or down: it describes the disaster, not either government's handling of it.

A government can run a faster, better-organized emergency response and still preside over a far deadlier disaster. Speed of response and lethality of event are different measurements, and only the first is what this rubric scores.

The road that washed out three times

One finding sits outside the comparison and matters more than its weight would suggest. In July 2025, a flood from the same Tibetan source destroyed the Syabrubesi-Rasuwagadhi road and the Nepal-China Friendship Bridge, under Prime Minister Oli. China funded a full rebuild, still under construction in July 2026. The August 2026 flood destroyed it again, under Prime Minister Shah.

Rebuilt in the same flood path, across a change of government. That's the strongest evidence in this piece that one specific failure, rebuilding without updating the risk model, isn't about who holds office.

What the numbers say

The Nepal Government got better at what a law can fix. 2026 wins last-mile relief and victim identification by wide margins, and both trace to plumbing that did not exist in 2015: a disbursement channel that no longer takes eight months to stand up, and a documented identification protocol. How much of that is competence and how much is the calendar, we cannot say.

2015 wins search and rescue and international coordination outright, and search and rescue by a wider margin than we first scored: the Upper Trishuli 3A tunnel search ran nine days, not the 24 to 48 hours it takes to match 2015's mark. Not because that government was sharper, but because it reached a hard site faster and plugged into a multinational system 2026 either lacked or declined to use.

Resource mobilization ties exactly, which is itself the finding. Activation speed was never Nepal's failure mode, in either disaster.

What did not improve is what requires moving power rather than writing procedure. Relief still routes through a centralized chain. Local governments were still sidelined. And the state still rebuilds in flood paths it has watched wash away, now with better hazard data than 2015 ever had. Where technology flatters the rest of the scorecard, it indicts that one.

A better-run response is also not a better outcome. Counting the missing, this flood killed about three-and-a-half times as many people per 100,000 affected as the earthquake did. The rubric scores how the government responded, not how much it prevented.

And it scores only the opening days. Nearly 1.6 million people are affected. Close to 5,000 are missing, most of whom will not be found. Rasuwa and Nuwakot have lost roads, bridges, power and telecommunications. Thousands of families have no house to return to.

2015 is the warning. Its response phase was over in weeks, but reconstruction ran a decade, and the failures that mattered came in the decade: the eight-month wait for the reconstruction authority, compensation that never arrived, codes written but unevenly enforced. A good early response was not what determined how 2015 turned out.

So 65 percent is not a verdict on this disaster. It is a reading of the response taken while the water is barely down. Whether displaced families are housed, whether the missing are identified and their families paid, whether the Rasuwagadhi corridor is rebuilt where the river cannot reach it: that is the part that will matter, and none of it can be scored yet. We will come back when it can.

Score the 2026 response yourself

The rubric above is fixed, but placing a disaster against it involves judgment. Score the 2026 flood on the same five dimensions and see how your reading compares with ours, and with other readers'.

This is a reader poll, not evidence. People who reach the bottom of an analysis like this are not a sample of anything. Treat the accumulated results as a record of what this piece's readers thought, not as an independent assessment. Our own scores stay where they are regardless of how the poll runs.