Nepal’s projected population placed on a 100 m grid, from 2026 to 2051: every ward’s projected total redistributed inside its own boundary, weighted by where WorldPop finds settlement. Ward totals are preserved exactly — the grid adds no people and moves none between wards, it only says where inside a ward they are. The change view subtracts one year from another cell by cell, which is only meaningful because every year shares an identical lattice.
Drag to pan · scroll or buttons to zoom · hover to read a value
Small-area projections are the demographic control totals; the raster only decides where inside each ward the people go. Each WorldPop pixel is assigned to the ward containing its centre, pixels are summed into grid cells, the weights are normalised within the ward, and the ward’s projected total is split across its cells by those weights. Conservation is guaranteed by construction and then checked: the national build reproduces every one of 6,743 ward totals with zero error, in each of the four years shown.
245,190 people are not on this map, 0.8% of the country in 2026.
They are Nepal’s institutional population — people in
barracks, prisons, monasteries, hostels and campuses — whom the census does not
assign to a ward. The projection carries them as one area per district, seventy-seven
in all, and the gazetteer names each one.
They are counted here and left off the map rather than spread across the wards of
their district. A proportional spread would conserve exactly and would still be
wrong: an institution stands in one place, so spreading a district’s barracks
over its wards puts soldiers in every village. Nothing in the data says where these
people are, and a map that invents it is worse than a map that says it does not know.
An earlier build of this page dropped them without saying so.
This is the ward tier; the district tier is a different page. Nepal is projected separately at district, palika and ward level. The grid uses wards, because 6,743 control totals resolve settlement far better than 77 do. For counts and structure — age, sex, dependency, ageing, by district and province — use the Nepal Population Explorer. This page answers where; that one answers how many and who.
It carries no age and no sex. The downscaling distributed total
population only, so any indicator that depends on age or sex — a dependency
ratio, a share aged 65 and over — would be constant within a ward and would look
like information it is not. popsym_grid_cell_varying() in popsymr refuses
those by design.
It is residential. These are people where they live, not where they are during the day. Workers, pilgrims, traders and travellers are counted at home.
Displayed at 200 m by summing the underlying 100 m cells, on a quantile colour scale computed once on 2026 and then held fixed. Recomputing it per year would give each map its own scale, so a square could change colour while its population stayed put and the map would show the scale moving and call it growth. A linear scale would show almost nothing: the median inhabited cell holds under three people while the busiest holds hundreds. Uninhabited cells are left blank rather than coloured, because “nobody lives here” and “one person lives here” are different statements.
projection_long.parquet, 42.75 million rows over 6,820 ward codes)
rather than from the engine’s raw output. Every ward’s projected total is
preserved exactly; the grid only decides where inside the ward it sits.
MK*.csv, whose totpop
includes the agest = −1 row. That row is the period’s
newborns — a flow, already counted in births — so every ward was inflated
by its own birth cohort, about 511,000 people nationally in 2026. The same build also
dropped the 77 residual buckets without saying so. The 2026 figure was published as
30,323,061; corrected, and with the buckets stated separately, it is
29,813,527. The contract removes that row once, in one place, so the
error cannot come back a third way.popsym_validate_grid() structurally
and under strict conservation against the ward projection.