Population whose homes are in flooded areas, Population of the raster cells the hazard footprint touches (grid: 100 m), assigned to the nearest settlement. A cell only GRAZED by the flood contributes its whole population, so this is an upper bound at any resolution, and the coarser the grid, the larger the overstatement. Two tiles: the flooded homes whose settlement still has a road route to a facility, and the ones with no facility reachable at all. Being in the water is the limiting case of losing access, so on the map the second keeps the class red; the first is painted by its travel time with a red ring.
No facility reachable on the mapped road network, outside the flooded area, Settlements on dry land with NO health facility reachable on the road network after the hazard cut it. Not 'lost the nearest one', none at all. A settlement both in the flooded area and cut off puts those people in the row above and everyone else here.
Must redirect to a different facility, The nearest reachable facility is a DIFFERENT one than before the event. The travel time may be longer, and the receiving facility may be busier.
Pushed over 30 min to care, Could reach a facility within 30 minutes before the event and cannot now, the 30-minute line is the strongest of the clinical thresholds below. Travel times that merely lengthened without crossing it are in the before/after chart, not in this category.
Facility serving more people, The travel time is unchanged, but the facility at the end of it is absorbing patients from a closed or unreachable neighbour. This is a care-quality signal, not a capacity one: the test is that the facility gained ANY patients, so absorbing ten and absorbing thirty thousand read the same here, and nothing in OpenStreetMap gives the beds or staffing to say whether either exceeds what the place can handle. It is also the lowest level, so it counts only those for whom crowding is the ONLY harm, people who are additionally rerouted or delayed are counted above, many of them at these same facilities, which makes this figure a floor.
"Facilities likely experiencing surge" DOES NOT add to the rest and is the one figure on this page that deliberately overlaps them. Every other number follows people whose own travel time changed; that one follows the people whose travel time did not, same facility, same trip, a waiting room now serving a catchment it did not have last week. It counts the WHOLE prior catchment, so anyone in it who was also rerouted or delayed is already counted above. Two cautions: catchment here means nearest-facility assignment, not registration, so it is where people would go rather than where they are enrolled; and there is no capacity in it, so the same percentage means different things at a referral hospital and at a health post. NOR IS THERE CAPABILITY: facilities come from OpenStreetMap amenity=hospital/clinic, which records no specialism, so a settlement is assigned its nearest facility whether or not that facility treats what it needs. Rows whose name suggests a specialism are marked; none is reassigned. A facility showing NO prior catchment is not a facility that served nobody: catchments are assigned settlement by settlement to a single nearest facility, and this event has more facilities than settlements, so most carry a zero baseline either way. Those are reported separately, with no percentage, and the effect on their existing patients is UNKNOWN rather than zero.
Each person is counted once at their worst level, so the figures add up. A settlement both in the flooded area and cut off contributes those people to the first level and everyone else to the second.
Vulnerable-group figures are SHARES from epoch 2022 survey indicators applied to the affected population, an estimate of proportion, not a headcount of identified people.
Settlement population is the population raster summed over the settlement's own polygon, zonal sum of npl_GHS_POP_E2025_R2023_54009_100.tif over NPL_spl_settl_w5_t0.01_2025_54009.shp.
Travel times assume FREE-FLOW speeds, there is no congestion model Every surviving road is routed at its normal speed, even where it now carries traffic diverted from a road that was cut. Nothing slows down: routing runs before the caseload surge is known and the surge never feeds back. The travel times here are therefore CONSERVATIVE. Real journeys on a diverted route will take longer, by an amount this page does not estimate, so the counts of delayed and cut-off people are the fewest the event can account for, not the most. The surge figures show the demand that is not being modelled: a facility absorbing tens of thousands of extra patients implies the roads to it are carrying them.
A road inside the flood footprint is assumed impassable That is the primary rule: the water extent cuts the network directly. Reported damage to bridges and roads COMPLEMENTS it, it can only add cuts the footprint does not already make, for structures damaged outside the observed water. It never replaces the footprint rule and never removes a cut.
Facilities inside the footprint are treated as CLOSED which is binary: a partially working facility is modelled as gone.
Times are DRIVING times on the Overture Maps / OpenStreetMap road network at a speed set by road class and surface (paved or unpaved), not by posted limits, which cover under 1% of Nepal's roads and are not used; walking for the off-road last mile (Tobler's hiking function on the leg's average SRTM gradient, land-cover adjusted, gentle to ridge crossings, so remote absolute times are conservative), no public transport, congestion, night driving or weather. The network is Overture and OSM as mapped: informal paths and post-event repairs are absent until mappers add them.
Routes stay inside Nepal destinations and transit both, except where the only road briefly weaves across the border, which stays open. Care across the border is the separate purple what-if scenario, never part of these numbers.
A destination is any mapped hospital or clinic that could admit an acute casualty animal hospitals and facilities that do not admit an acute general case (eye, dental, skin, physiotherapy, dialysis, fertility, psychiatry, ayurvedic or homeopathic, pharmacies and dispensaries, hospices, diagnostics-only labs and imaging) are excluded by rule; maternity, children's, polyclinics, rehabilitation and general hospitals and clinics stay in, and a name that suggests a specialism is marked, not reassigned.
People are counted at home GHS population (2025) allocated to settlements; each settlement travels from one representative point, and displacement is not modelled.
Everyone is modelled at their NEAREST open facility by network travel time, the standard geographic-access assumption. Real patients bypass for quality, cost or referral, and nothing here models capacity, so who-goes-where is indicative rather than predictive; the 2nd- and 3rd-nearest views show the alternatives, and a settlement counts as cut off only when no facility is reachable on the mapped network.
The cells are the same model at 100 m Every populated 100 m cell of the population grid is routed on its own to its nearest, 2nd- and 3rd-nearest open facility, before and after the event. The map's hexagons aggregate those cells: each goes to the facility most of its people go to, and carries the population-weighted mean of their travel times, so a hexagon and the settlement tiles can differ, they count different units. A cell is in the flooded area when its centre falls on mapped water; a hexagon is highlighted when any of its people are.
The timeline shows each date as it was observed Each stop routes on the footprints current at that date, a later version of the same product replacing the earlier one, so the page reports the state at the last observation of each area, not the union of every observation. Cells, settlement classes and travel times follow the stop; routes, facility marks and the tiles describe the latest state. A date that appears in a product's name is the day the satellite looked only when the source says so; a release date is labelled as one.
Cuts combine across footprints A settlement that every footprint on its own still leaves a road to can have none once all the cuts apply together. That check runs on the union of cuts and is what the cut-off count uses; where a dated stop has not had it computed, a note under the dial says so.
Copernicus' own count of people in the water sits beside ours theirs is the population in the mapped water from the activation's summary tables; ours counts every 100 m cell the water touches, so ours is the upper bound and the two are not expected to agree.
Travel-time thresholds are clinical, not arbitrary WHO's Global Health Observatory reports geographic access to emergency obstetric care within 2 HOURS, one of its 100 core health indicators. That 2-hour figure comes from a 1987 conference estimate of time-to-death and its evidence base is weak; a 2026 meta-analysis (Roder-DeWan et al., J Glob Health 16:04136) finds 30 minutes or 1 hour does better for perinatal survival, odds ratio 3.25 for interfacility transfers under 30 minutes. All three bands are shown because a reader who sees only the reported standard cannot tell it is the weakest of them.
Facility counts are an UPPER BOUND on EmOC availability The UN standard (WHO/UNICEF/UNFPA/AMDD 2009) is five emergency-obstetric-care facilities per 500,000 people, at least one of them comprehensive. That is defined on what a facility can DO, and our facilities come from OpenStreetMap, which does not record capability. Every facility is therefore counted as if it qualified: the figure can show an area FAILING the standard, never passing it.
Health access only This page does not account for the death toll, displacement, access to clean water, power or communications, so it will read LOWER than a humanitarian caseload for the same event. That is a difference of definition, not a disagreement.
