What the numbers mean
- People
-
Everyone GHSL places in an area and year: its modelled residents of every 100 m cell,
built up or not.
- People per km² of built-up land
-
The people on land built up by that year, divided by the km² of that land. Under 0.25
km² of built-up land, the size of the map's 500 m square, the page says there is too
little to say, since one cell can hold thousands of people. This compares towns with
towns. Across the countryside it misleads: GHS-POP puts a district's people on the
little built-up surface it detects, so villages can read as packed as a city. The page
never ranks districts on it.
- People per km² of the whole area, or of all land
-
The people divided by all of the area's land inside the map, built up or not. Rings
and the findings about districts use it.
- People per km² in the 500 m square
-
A column's height and the live reading: GHSL's modelled residents of the 5 × 5 cells
round a cell, never-built ones included, divided by 0.25 km².
- Built-up land, in km²
-
The km² of 100 m cells over 20% built-up surface by that year. A whole cell counts, so
it is more than the surface actually built over.
- Share of the area built up
- Built-up land divided by all of the area's land inside the map.
- First built up, built up by
-
The first of GHSL's five-year epochs in which a cell is over 20%. It may have crossed
at any time before that. On the map it is read as one of six periods (below).
- Share of people on land built after a year (How new)
-
People in cells first built up after the year you choose and by the map's year,
divided by the area's people that year; 2000 unless you choose another. Nothing up to
the next epoch after that year, when no such land was built up yet.
- Grew N times, change
-
The later figure against the earlier one, between two named years. A multiple or a
percentage of a small start can be large.
- Inside the map
-
The share of a circle or shape you made that lies in the map's 28 districts, by area.
Its figures cover only that part.
The six places the map opens on
Delhi is Delhi NCT, its 13 districts as one area, because OpenStreetMap has no tehsils
there. Each of the others is one tehsil on the map's own boundaries, named by the city a
reader knows: Ghaziabad tehsil, Faridabad tehsil, Dadri tehsil (Noida, and Dadri town),
Gurgaon tehsil (Gurugram) and Gautam Buddha Nagar tehsil (Greater Noida). An area's name
in the panel gives the tehsil when you point at it.
How much of each one runs into Delhi? Here is the share of its built-up land in 2020
that belongs to one connected mass of built-up cells holding at least 1 km² of Delhi
NCT, with cells touching, then allowing a gap of 100 m, then 300 m: Ghaziabad
…; Faridabad
…; Noida
…; Gurugram
…; Greater Noida
…. So the first four run into Delhi's
built-up land, and Greater Noida's does not: it is here as the far end of the story.
Bahadurgarh (…) runs into Delhi too, and is
left out to keep the first view to six: add it with + Add an area.
The headline
"Most of the new building since 1975 happened outside Delhi itself" is about land. It
counts every 100 m square on the map whose built-up surface first passed 20% of the
square at one of the epochs from 1980 to 2020: across the map's 28 districts that is
… km² of newly built-up land, and
…% of it lies outside Delhi NCT. Moving the 20%
line between 10% and 30% gives …%. GHSL's
1975 layer is its weakest (below), so the same count is also made from 1990:
…% of the land built up from 1990 to 2020 lies
outside Delhi NCT. From either year, and at every line in that range, the share outside
is more than half at every epoch from 1990 to 2020.
Colour: when first built up
Hue means one thing only: when a cell was first built up. It comes from GHS-BUILT-S, as
the first five-year epoch in which built-up surface is more than 20% of the 100 m cell,
read as one of six periods: built by 1975, 1975–1990, 1990–2000, 2000–2015, 2015–2020
and after 2020. Built by 1975 means the cell had passed that line by the first epoch; it
may have done so earlier.
Why these six? Satellites observed built-up land in 1975, 1990, 2000, 2014 and 2018, and
the periods break at the epochs nearest those years. Between two observations GHSL fills
in the values, nearly in a straight line for most cells, so within one stretch the epoch
a cell first passes 20% says more about how fast it filled than about when it was built.
A period that cut through a stretch would mostly show density.
The colours run from blue through purple, magenta and red to orange and amber, and get
lighter the more recent the period, so early and late land differ in lightness too. That
holds with colour-vision deficiencies as well, though there neighbouring periods are
closer than in full colour. To read one column's period, turn live reading on.
Time
Years between the five-year epochs are filled in between GHSL's values, and 2025 and
2030 are GHSL's projections made in 2023.
GHSL measures built-up surface from satellite images for 1975, 1990, 2000, 2014 and
2018; its other epochs are its own estimates. On the charts the line is dashed after
2020. A new column grows in over the five years up to the epoch it first appears in, and
while time moves it is drawn taller and lighter.
Height: people per km² in its 500 m square
A column's height says how many people live around that spot, counted over the 500 m
square around it. It is GHSL's modelled people per km² in the 500 m × 500 m square (5×5
cells) centred on it, from GHS-POP, drawn ten times taller than true scale, the same in
every view and at every distance:
The sum runs over the full population grid, so
neighbouring cells that never became built up still count; at the map's edge the square
is cut off, and the people in the part of it on the map are divided by that part's area.
A column is a 100 m cell, not a building. Perspective makes near columns look taller
than far ones, so read heights from the live reading or the panel, not by eye. Lightness
carries nothing: a roof is its ramp colour, and walls are shaded by the light, darker at
the foot of a column that stands above its neighbours, to show depth.
Districts, tehsils and places
Districts and tehsils are today's OpenStreetMap boundaries, used for every year. A
district or tehsil is the 100 m cells whose centres fall inside it. The map's edge is
the outer boundary of its 28 districts, so no district or tehsil is cut and the findings
compare whole districts. A circle or shape you make can cross that edge: it says how
much of it is inside, and its figures cover only that part. OpenStreetMap has no tehsils
for Delhi's districts. A place's figures are about the land within 1 km of its point.
The place lists rank OSM cities, towns, suburbs and neighbourhoods whose 1 km circle
lies at least 80% inside the map, with no two rows within 2 km of each other; "Fewer
people than in 2000" needs at least 10,000 people there in 2000. Place names are OSM
place points, named as points, not as areas.
Comparing
A comparison holds up to eight areas. Values within 5% of each other count as level: the
census and the model differ by about that much at district level. Compared as two
groups, a group counts each 100 m cell once, however many of its areas hold it. Each
area keeps its number and colour until it is taken out. GHSL observes built-up surface
in 1975, 1990, 2000, 2014 and 2018, so a change dated between other years rests on its
estimates.
Rings
Rings count people per km² of all land in rings 1 km wide, because how far people thin
out has to count the land they are not on. They are centred on the spot you chose, or on
an area's 500 m square with the most of its people in 2020, counting only people inside
it. The page says people thin out when a steady percentage fall with distance fits the
rings: R², which runs from 0 to 1, of 0.8 or more over at least 4 rings. That describes
the rings. It is not a test, and it does not show that a spot is a centre: on a copy of
this map scrambled to have no centres at all,
…% to
…% of spots fall steadily by the same measure,
because rings share their neighbours' people. Real places here do so
…% to …% of
the time. "About level" means the rings with the most and the fewest people are within
35% of the rings' average, a choice of wording, not a measured limit.
Early years in small towns
Take Khurja: in 1990 the 1 km circle around it holds about
… people, while the 1991 census counted 80,305 in
the whole town. Why so many? GHS-POP spreads each census area's people over the
residential built-up volume GHSL estimates for that year, and in 1975 and 1990 Landsat
detected little of the villages, so in towns away from Delhi the early counts run high.
As villages are detected in later years their people move out of the town cells, which
can read as a decline that did not happen. Early figures for such towns are the model's
spread, not a count. For the same reason, outside the cities people added can read as
all on new land, with the old land flat or falling: the page says those counts only for
an area you choose, and draws no finding from them.
How people and land are counted
Totals count everyone GHSL places, in every 100 m cell and every year, built up or not.
The 20% line decides only what is built-up land: what is drawn as a column, how much is
built, how packed it is. A drawn shape counts the cells whose centres fall inside it.
These are GHSL's modelled populations, census counts placed by each cell's estimated
residential built-up volume. Land never built up holds
…% of these districts' people in 2020 (…% in Delhi NCT, …% in the five neighbours).
Its people travel at the full 100 m as whole people per cell, with the rounding carried
from cell to cell, so totals are exact and a small shape is off by under one person for
each row of cells it crosses.
Finding a place
The search knows the names in OpenStreetMap, GeoNames and Wikidata, joined so that each
place appears once, at OpenStreetMap's point where they disagree. On a random sample
from sources it does not use, the government's list of villages and India Post's list of
offices, it finds …% of the names in the
countryside and …% in the towns. A place by a name
it does not know can still be found on the map. A suggestion's context comes from
boundaries: its district, its tehsil, and for a neighbourhood the nearest town of its
district.
How precise these figures are
Every people figure is modelled, not counted. GHSL's published validation (Pesaresi et al. 2024, International Journal of Digital Earth) compares GHS-POP with census grids in 24
countries, not including India: R-accuracy 0.68 to 0.74 by year, 1995 to 2020. For
built-up surface at 100 m against dated building footprints worldwide it reports 0.25 in
1975 rising to 0.43 in 2020, so one cell's figure is weak and totals over many cells are
firmer. This map's Delhi NCT totals are within about
…% of the censuses of 1981 to 2011, as they should
be, since the censuses are the model's input;
… neighbouring districts at 2011 differ by
…% to
…%. Read a percentage here as good to a few
points at best.
Noida was founded in 1976, yet within 2 km of its point GHSL's 1975 layer already has
…% of the 2020 built-up surface; Rohini,
developed from the 1980s, reads …%. That layer
is GHSL's weakest, built from 1970s Landsat images, and it can be wrong either way: it
also misses villages (above), which works the other way. So the headline's count is also
made from 1990 (above), and no finding about how packed a place is rests on 1975. An
area's chart of people by build period does rest on each cell's first-built date: a cell
wrongly shown as built up early counts its people as on land built up early.
Map context
The ground under every year is today's, as a historical map lies over a present-day base
map. What covers the land is ESA WorldCover's map of 2021 at 10 m: tree cover, scrub,
grass, cropland, bare ground and marsh, each in a quiet tone of its own; cropland
carries a texture of fields along the grid the satellite picture shows, a way of drawing
the class, not real parcels. Close in, the trees stand where Meta and WRI's canopy
height map sees a crown 2 m high or more, each at its own place, height and width; where
WorldCover says tree and that map sees no canopy, the land has the tree class's tone and
no tree is drawn. The water is drawn from a water map made for this one: the rivers'
channels as Sentinel-2 saw them in the dry season of 2020, lakes, ponds and wetlands
from OpenStreetMap, flowing downhill; beside the rivers the ground is darker and greener
where water was seen more often from 1984 to 2024 (JRC Global Surface Water), which is
not a floodplain or a flood risk. Grass, reeds, crops and trees sway in a light breeze
and the water flows; both are still with reduced motion.
Roads, rail, canals and river banks, the Delhi NCT
boundary and state borders are present-day OpenStreetMap layers: they are drawn the same
in every year, including years before they existed. Close in, the map adds every street
and drain OpenStreetMap has, on the ground like every line, so a neighbourhood's own
streets can be traced, and landmarks named by rule, in English. The Satellite view shows
what the ground looked like in a year.
Sources and credits
Today's ground: streets, water and landmarks, ©
OpenStreetMap contributors, ODbL. Land cover: © ESA WorldCover project 2021 / Contains modified Copernicus
Sentinel data (2021) processed by ESA WorldCover consortium,
CC BY 4.0; Zanaga et al. (2022),
doi:10.5281/zenodo.7254221. Trees: Meta and World Resources Institute (WRI), High Resolution Canopy Height Maps,
CC BY 4.0; Tolan et al. (2024),
doi:10.1016/j.rse.2023.113888. Rivers' channels: contains modified Copernicus Sentinel data (2020). Water seen 1984
to 2024: Source: EC JRC/Google; Pekel, Cottam, Gorelick and Belward (2016),
doi:10.1038/nature20584. Landmarks' names and how widely they are written about:
Wikidata (CC0).
GeoNames and Wikidata, place names for the search:
GeoNames,
CC BY 4.0; Wikidata, CC0.
Changed here: joined with OpenStreetMap's places so that each appears once.
GHSL, built-up surface and population, © European Union, 2023. Reuse is
authorised under
CC BY 4.0, provided the source is acknowledged: Pesaresi and Politis (2023),
GHS-BUILT-S R2023A; Schiavina, Freire, Carioli and MacManus (2023),
GHS-POP R2023A; European Commission, Joint Research Centre,
Global Human Settlement Layer, its five-year epochs from 1975 to 2030. Changed here: resampled to a 100 m UTM 43N
grid, and first-built periods and the figures on this page derived from it. Validation:
Pesaresi et al. (2024).
OpenStreetMap, boundaries, rivers, roads, railways, canals, land cover and place
names, ©
OpenStreetMap contributors. The data are available under the
Open Database License.
Satellite pictures, 1975 to 2015: Landsat imagery courtesy of the U.S. Geological
Survey. 2020 and 2025, Sentinel-2: contains modified Copernicus Sentinel data 2020;
contains modified Copernicus Sentinel data 2025. Each picture is a cloud-free composite
of February and March scenes, and some take scenes from the years around their own,
which the map names beside the year: 1980's are of 1977 and 1980, 1990's mostly of 1991,
2005's mostly of 2002 and 2003, and 1995's, 2000's and 2010's of three years each. There
is none for 1985, since the free archives hold no picture of this area from 1982 to
1988, so from 1985 to 1990 the map shows its columns alone. A year between two pictures
shows the earlier one, and 2030 the latest, of 2025, by its own year. Nothing is
stretched or blended to make a year. Close in, the pictures of 2000, 2005 and 2015 carry
the detail of Landsat's 15 m panchromatic band, drawn at 10 m, and those of 2020 and
2025 have the blur of Sentinel-2's own optics taken back out, by its published
measurements; no detail is invented. Seen whole, some years keep a faint regional haze
where every look of the ground was hazy, most in 1990 and across the south in 2020; it
is not ground, and close in it does not show.
NASADEM, terrain: NASA JPL (2020),
NASADEM Merged DEM Global 1 arc second V001, NASA Land Processes Distributed Active Archive Center, openly shared without
restriction; read from the
OpenTopography
public mirror.
Geographic and population scope
The map covers 28 whole districts around Delhi, about 26,500 km²: the 13 of Delhi NCT,
and Ghaziabad, Gurugram, Gautam Buddha Nagar, Faridabad, Sonipat, Baghpat, Jhajjar,
Meerut, Hapur, Palwal, Rewari, Rohtak, Nuh, Bulandshahr and Khairthal-Tijara. That is
not the official NCR administrative boundary. Their outer boundary is the map's edge,
and nothing beyond it is data. Columns stand only on cells that pass the 20% built-up
line in at least one epoch. The people of cells that never do stand in no column, but
every total here counts them (How people and land are counted). The Delhi NCT boundary
and state borders are reference layers; they do not define the extent.
Write to: hello@truemanifolds.com
If this piece was worth something to you,
you can support the next one.