598,376
modelled from census totals
Source: WorldPop 100 m As of 2020
Frontier GIS · sample district
What can a policy lab learn about one district without leaving the desk? This page takes Meherpur, 720 km² in Khulna division with 3 upazilas, and reads it from public satellite archives and open map data: a population count with age structure, a building census with heights, land cover, built-up growth since 1975, the degree of urbanisation, night-time lights, surface water and radar-detected monsoon water, pre-monsoon heat, 45 years of rainfall and temperature, air quality, tree canopy height, travel time to health care, roads, bridges and facilities, landscape types and landscape change learned by an AI embedding model, and a brick-kiln inventory with a firing-season activity test.
Every number traces to a named Google Earth Engine dataset, period and resolution, and every method note says what the measurement is and what it is not. The same pipeline runs for any of the 64 districts.
598,376
modelled from census totals
Source: WorldPop 100 m As of 2020
454,284
Source: Open Buildings 2.5D Temporal As of 2023
34,863 ha
48% of the district
Source: Dynamic World As of 2025
2,088 ha
Source: GHSL R2023A As of 2020
1.09 nW/cm²/sr
Source: VIIRS DNB annual As of 2025
67 of 111 active
remote classification, not field-verified
Source: Landsat 8/9 + Sentinel-2 As of Mar 2026
90.3%
population-weighted mean 14 min
Source: Malaria Atlas Project + WorldPop As of 2019 / 2020
34.4 °C
hottest tenth above 36.7 °C
Source: Landsat 8/9 surface temperature As of Mar to May 2025
1,578 mm
1991 to 2020 mean 1,528 mm
Source: CHIRPS v2.0 As of 2025
Interactive atlas
Loading map…
The district in true colour
Per-pixel median of Sentinel-2 surface reflectance (bands B4, B3, B2) across all dry-season scenes with scene cloud cover under 40%, after masking clouds, shadows and cirrus with the Scene Classification Layer (SCL classes 2, 4, 5, 6, 7 kept).
COPERNICUS/S2_SR_HARMONIZED · 2025-12-01 to 2026-04-01 · native 10 m · 21 scenes
Natural colour: red, green and blue bands as the eye would see them.
Census from orbit
| Upazila | Area, km² | People 2020 (WorldPop) | Buildings 2023 | Built-up 1975, ha | Built-up 2020, ha | Cropland 2025, ha | Trees 2025, ha | Seasonal water, ha | Lights 2025, nW/cm²/sr | Surface heat 2025, °C | Kilns | Kilns active | To health care, min (median) | OSM roads, km | OSM bridges | OSM schools | OSM health facilities | Monsoon-only water 2025, ha | Canopy ≥ 5 m, ha | Mean building height, m |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Gangni BD40570047 | 340.4 | 282,101 | 210,938 | 98 | 908 | 15,937 | 12,030 | 359 | 0.90 | 34.3 | 61 | 37 | 15 | 547 | 12 | 4 | 1 | 48 | 29,416 | 6.2 |
| Meherpur Sadar BD40570087 | 264.1 | 228,632 | 176,535 | 124 | 859 | 13,285 | 8,178 | 461 | 1.23 | 34.3 | 35 | 24 | 9 | 571 | 12 | 12 | 9 | 66 | 22,123 | 6.1 |
| Mujibnagar BD40570060 | 113.9 | 87,631 | 66,811 | 65 | 322 | 5,641 | 3,866 | 262 | 1.37 | 34.8 | 15 | 6 | 11 | 188 | 17 | 2 | 1 | 69 | 9,465 | 5.9 |
Hectares of built surface in the district, GHSL five-year epochs.
Source: European Commission JRC, Global Human Settlement Layer R2023A (Pesaresi et al. 2023) As of R2023A
GHS-POP epochs (2025 is a projection). WorldPop 2020: 598,376.
Source: JRC GHS-POP R2023A; WorldPop 2020
Every structure the model can see at 4 m, summed over the district. Cross-check: 255,301 Open Buildings V3 polygons (193,134 at confidence ≥ 0.7) covering 1,311 ha.
Source: Google Research, Open Buildings V3 and Open Buildings 2.5D Temporal (Sirko et al. 2021, 2023)
District mean radiance of the annual VIIRS composite. Brightest upazila: Mujibnagar.
Source: NOAA and Colorado School of Mines Earth Observation Group, VIIRS DNB annual composites
Driest year 1982 (974 mm), wettest 2017 (2,104 mm).
Source: Climate Hazards Center, UC Santa Barbara, CHIRPS v2.0 (Funk et al. 2015)
Annual mean tropospheric NO₂ column over the district, µmol/m². A district-wide air-quality signal, too coarse to map.
Source: ESA Copernicus Sentinel-5P TROPOMI, offline Level-3 NO2
| crops | 34,863 ha | 48.4% |
| trees | 24,074 ha | 33.4% |
| built | 9,916 ha | 13.8% |
| water | 2,676 ha | 3.7% |
| flooded vegetation | 263 ha | 0.4% |
| grass | 40 ha | 0.1% |
| shrub and scrub | 2 ha | 0.0% |
| bare | 2 ha | 0.0% |
Most frequent Dynamic World class per 10 m pixel across every Sentinel-2 pass in 2025. Dynamic World is a near-real-time deep-learning classification; class areas are model estimates, not survey measurements.
Source: Google and World Resources Institute, Dynamic World V1 (Brown et al. 2022) As of 2025
Six unsupervised clusters of the 2025 AlphaEarth embedding. What each mostly contains, by Dynamic World class:
| cluster 1 | 8,365 ha | crops 81%, trees 16%, water 1% |
| cluster 2 | 16,326 ha | trees 47%, crops 46%, built 6% |
| cluster 3 | 7,369 ha | built 95%, trees 4%, crops 0% |
| cluster 4 | 12,208 ha | trees 84%, built 14%, crops 2% |
| cluster 5 | 24,401 ha | crops 81%, trees 18%, built 1% |
| cluster 6 | 3,164 ha | water 71%, crops 17%, trees 8% |
AlphaEarth Foundations produces a 64-dimensional embedding per 10 m pixel per year that summarises optical, radar, thermal and climate observations. 6,000 random district pixels were clustered with k-means (k = 6, seed 42) and every pixel was assigned to its nearest cluster. Clusters are unsupervised: the labels are numbers, and the Dynamic World composition is reported only to describe what each cluster mostly contains.
Source: Google DeepMind and Google Earth Engine, AlphaEarth Foundations satellite embedding V1 (Brown et al. 2025) As of 2025
Satellite inventory: 111 kiln structures, of which 67 show both a firing-season heat signal and yard change in 2025-26.
| active (heat and yard change) | 67 |
| heat signal only | 12 |
| yard change only | 22 |
| no activity signal | 10 |
| District administration register | 103 |
| Department of Environment, total / operating | 113 / 78 |
| Customs VAT register | 68 |
Office counts: Padma Sangbad (Meherpur), 27 January 2026, citing the three offices; press report, not verified against office records.
111 kiln structures were located by reviewing three inventories (APAD, a Stanford detection model, and an Esri imagery grid scan) and the AlphaEarth similarity search. For each site, Landsat surface temperature and Sentinel-2 shortwave-infrared reflectance were compared with a ring of surrounding land on every clear day from September 2025 to June 2026; a site is 'active' when both indicators exceed the 95th percentile of 21 control sites in the firing season. This remote classification has not been validated in the field.
Embedding search: 2.0 km² of the district scores above the kiln-similarity threshold (0.9469), the cutoff that 90% of known kilns clear. Per-site evidence, scene lists and thresholds are in the repository folder research/meherpur_brick_kiln.
Source: BDPolicyLab kiln inventory (research/meherpur_brick_kiln), USGS Landsat 8/9, ESA Sentinel-2, AlphaEarth As of firing season 2025-26
| Seasonal surface water (1 to 11 months a year) | 1,082 ha |
| Permanent surface water (12 months) | 0 ha |
| Maximum water extent, 1984 to 2021 | 2,766 ha |
| Pre-monsoon 2025 surface temperature, district median | 34.4 °C |
| Coolest tenth / hottest tenth | 32.3 / 36.7 °C |
Global Surface Water v1.4 classifies every Landsat observation since 1984 as water or not. Occurrence is the share of valid observations that were water; permanent water is present in all 12 months of the year, seasonal water in 1 to 11 months.
Median of Landsat 8 and 9 Collection 2 Level-2 surface temperature (band ST_B10, scaled to °C) over all pre-monsoon scenes with scene cloud cover under 60%, after masking cloud, cloud shadow, cirrus and dilated cloud with QA_PIXEL. Surface temperature at satellite overpass (about 10:30 local), not air temperature.
Source: European Commission JRC, Global Surface Water v1.4 (Pekel et al. 2016); USGS Landsat 8/9 Collection 2 Level-2 Surface Temperature
People and services
| Group | District |
|---|---|
| Under 5, % | 7.0 |
| 5 to 14, % | 15.3 |
| 15 to 64, % | 72.0 |
| 65 and over, % | 5.7 |
| Females per 100 males | 101.3 |
| Dependency ratio, % | 38.8 |
| People | 598,364 |
WorldPop 2020 age-sex grids (five-year bands by sex, 100 m) summed over each upazila. WorldPop applies census age-sex proportions from the administrative level available, so differences between upazilas reflect those input proportions and the gridded totals, not an independent local enumeration. In this run the three upazilas carry identical shares, which is why only the district column is shown.
Source: WorldPop (University of Southampton) age and sex structures, 2020 As of 2020
| Nearest health facility, population-weighted mean | 14 min |
| People within 30 minutes of care | 90.3% |
| People within 60 minutes of care | 100.0% |
| Farthest 1 km cell from care | 65 min |
| Nearest city of 50,000+, district median | 12 min |
| People in urban centres or clusters (Degree of Urbanisation) | 65.9% |
Malaria Atlas Project friction surfaces give the least-cost travel time, by motorised transport, to the nearest health facility (2019, facilities from a global compilation) and to the nearest city of 50,000 or more (2015). Population weighting uses WorldPop 2020. The facility list behind the 2019 surface is not verified against DGHS records.
GHS-SMOD applies the UN-endorsed Degree of Urbanisation to 1 km cells of built-up surface and population: urban centres (at least 50,000 people at 1,500 per km2), urban clusters (5,000 at 300 per km2) and rural grades. Population per class is WorldPop 2020 summed inside each class.
Source: Malaria Atlas Project, Weiss et al. 2018 and 2020; European Commission JRC, GHS-SMOD R2023A V2.0
2,212 road segments and 109 facilities or bridges from OpenStreetMap, OSM database as of 2026-09-10T17:35:21Z. Toggle them on the map above.
| Road class | km |
|---|---|
| primary | 72 |
| secondary | 54 |
| tertiary | 101 |
| residential or unclassified | 1,063 |
| track or path | 18 |
| All mapped roads | 1,307 |
| Bridges | 41 |
| Facility | Gangni | Meherpur Sadar | Mujibnagar |
|---|---|---|---|
| Schools and colleges | 4 | 12 | 2 |
| Hospitals, clinics, doctors | 1 | 9 | 1 |
| Pharmacies | 10 | 2 | 0 |
| Marketplaces | 0 | 3 | 1 |
| Places of worship | 1 | 14 | 2 |
| Banks | 0 | 6 | 0 |
Every OpenStreetMap way tagged highway, every bridge, and every school, college, hospital, clinic, doctor, pharmacy, marketplace, place of worship and bank inside the district, fetched from the Overpass API and assigned to the upazila containing its midpoint. Road length is summed by class. OpenStreetMap is volunteer-mapped: coverage is uneven, and a low facility count means unmapped, not absent.
Source: OpenStreetMap contributors, ODbL 1.0
Change, hazard and structure
| Water in the dry season (Feb to Mar 2026) | 522 ha |
| Water only in the monsoon (Jul to Sep 2025) | 183 ha |
| People (WorldPop 2020) on monsoon-only water pixels | 411 |
| Scenes, monsoon / dry | 36 / 20 |
Sentinel-1 C-band radar sees through monsoon cloud. The median VV backscatter of all monsoon scenes and of all dry-season scenes was smoothed over 30 m and classed as water below -17 dB, a fixed threshold that is not calibrated for this district. 'Monsoon only' is water in the monsoon median but not the dry-season median, which under-reads short floods and over-reads flooded rice. People on those pixels are WorldPop 2020.
Source: ESA Copernicus Sentinel-1 GRD, via Google Earth Engine
| Median / 90th / 99th percentile embedding distance | 0.041 / 0.084 / 0.186 |
| Area above 0.186 | 718 ha |
| of which now crops (Dynamic World 2025) | 279 ha |
| of which now water (Dynamic World 2025) | 213 ha |
| of which now trees (Dynamic World 2025) | 187 ha |
| of which now flooded vegetation (Dynamic World 2025) | 19 ha |
Each 10 m pixel has a unit-length 64-dimensional AlphaEarth embedding per year. The map is one minus the dot product of the 2018 and 2025 embeddings: 0 means the pixel looks the same to the model, larger values mean it changed. The cutoff is the district's own 99th percentile, so 'changed' means the 1% of pixels that moved most in embedding space; the Dynamic World 2025 class of those pixels says what they became.
Source: Google DeepMind and Google Earth Engine, AlphaEarth Foundations satellite embedding V1
| Mean building height, 2023 | 6.1 m |
| Gangni | 6.2 m |
| Meherpur Sadar | 6.1 m |
| Mujibnagar | 5.9 m |
| Mean canopy height over vegetated pixels, 2020 | 9.3 m |
| Area with canopy of 5 m or more | 61,004 ha |
| Elevation, min / mean / max (SRTM) | -7 / 16.6 / 32 m |
Open Buildings 2.5D Temporal estimates a height for every building pixel from Sentinel-2 time series. The presence-weighted mean height is the sum of height over the sum of building presence at the 16 m pyramid level. Heights are model estimates with metre-level error and are best read comparatively.
ETH Zurich fused GEDI lidar canopy heights with Sentinel-2 imagery in a deep model to map canopy height globally at 10 m for 2020. Means are over pixels of at least 1 m; the 5 m area counts pixels of at least 5 m, which picks out homestead groves and orchards in a district with almost no closed forest.
Source: Google Research, Open Buildings 2.5D Temporal (Sirko et al. 2023); Lang et al. 2023, ETH Zurich Global Canopy Height 2020; NASA/USGS Shuttle Radar Topography Mission, SRTMGL1 v3
Annual mean of ERA5-Land 2 m temperature. 1981 to 2010 mean 25.3 °C; 2016 to 2025 mean 25.6 °C.
Source: ECMWF ERA5-Land reanalysis (Muñoz Sabater 2019), Copernicus Climate Change Service
Not built here
These are within reach with the same free archives and tooling but are not on this page, either because they need field labels, a free NASA Earthdata login, or heavier compute than one export run. None of the numbers above depend on them.
Method and limits
Sources
Source: The district in true colour: ESA Copernicus Sentinel-2 Level-2A, via Google Earth Engine (COPERNICUS/S2_SR_HARMONIZED, 2025-12-01 to 2026-04-01, 10 m)
Source: Land cover, 2025: Google and World Resources Institute, Dynamic World V1 (Brown et al. 2022) (GOOGLE/DYNAMICWORLD/V1, calendar 2025, 10 m)
Source: Built-up surface, 1975 to 2020: European Commission JRC, Global Human Settlement Layer R2023A (Pesaresi et al. 2023) (JRC/GHSL/P2023A/GHS_BUILT_S, 1975 to 2020, five-year epochs, 100 m)
Source: Night-time lights, 2025: NOAA and Colorado School of Mines Earth Observation Group, VIIRS DNB annual composites (NOAA/VIIRS/DNB/ANNUAL_V21 and ANNUAL_V22, 2013 to 2025, 500 m)
Source: Surface water, 1984 to 2021: European Commission JRC, Global Surface Water v1.4 (Pekel et al. 2016) (JRC/GSW1_4/GlobalSurfaceWater, March 1984 to December 2021, 30 m)
Source: Pre-monsoon surface heat, 2025: USGS Landsat 8/9 Collection 2 Level-2 Surface Temperature (LANDSAT/LC08/C02/T1_L2 and LC09/C02/T1_L2, 2025-03-15 to 2025-06-01, 30 m)
Source: Buildings counted from space, 2023: Google Research, Open Buildings V3 and Open Buildings 2.5D Temporal (Sirko et al. 2021, 2023) (GOOGLE/Research/open-buildings-temporal/v1 and open-buildings/v3/polygons, 2016 to 2023 (temporal), 2023 release (v3 polygons), 4 m)
Source: People per hectare, 2020: WorldPop (University of Southampton) Global Project 100 m, 2020; JRC GHS-POP R2023A (WorldPop/GP/100m/pop and JRC/GHSL/P2023A/GHS_POP, 2020 (WorldPop), 1975 to 2025 (GHS-POP epochs), 100 m)
Source: Landscape types learned by AI, 2025: Google DeepMind and Google Earth Engine, AlphaEarth Foundations satellite embedding V1 (Brown et al. 2025) (GOOGLE/SATELLITE_EMBEDDING/V1/ANNUAL, 2025 annual embedding, 10 m)
Source: Where the district looks like a brick kiln: Google DeepMind and Google Earth Engine, AlphaEarth Foundations satellite embedding V1 (GOOGLE/SATELLITE_EMBEDDING/V1/ANNUAL, 2025 annual embedding, 10 m)
Source: Travel time to the nearest health facility, 2019: Malaria Atlas Project, Weiss et al. 2018 and 2020 (projects/malariaatlasproject/assets/accessibility/accessibility_to_healthcare/2019 and accessibility_to_cities/2015_v1_0, 2019 (healthcare), 2015 (cities), 1000 m)
Source: Degree of urbanisation, 2020: European Commission JRC, GHS-SMOD R2023A V2.0 (JRC/GHSL/P2023A/GHS_SMOD_V2-0, 2020 epoch, 1000 m)
Source: Monsoon water seen by radar, 2025: ESA Copernicus Sentinel-1 GRD, via Google Earth Engine (COPERNICUS/S1_GRD, monsoon Jul to Sep 2025 vs dry Feb to Mar 2026, 10 m)
Source: Where the landscape changed most, 2018 to 2025: Google DeepMind and Google Earth Engine, AlphaEarth Foundations satellite embedding V1 (GOOGLE/SATELLITE_EMBEDDING/V1/ANNUAL, 2018 vs 2025 annual embeddings, 10 m)
Source: Building heights estimated from space, 2023: Google Research, Open Buildings 2.5D Temporal (Sirko et al. 2023) (GOOGLE/Research/open-buildings-temporal/v1, mid-2023, 4 m)
Source: Tree canopy height, 2020: Lang et al. 2023, ETH Zurich Global Canopy Height 2020 (users/nlang/ETH_GlobalCanopyHeight_2020_10m_v1, 2020, 10 m)
Source: Annual rainfall, 1981 to 2025: Climate Hazards Center, UC Santa Barbara, CHIRPS v2.0 (Funk et al. 2015) (UCSB-CHG/CHIRPS/DAILY, 1981 to 2025)
Source: Nitrogen dioxide over the district, 2019 to 2025: ESA Copernicus Sentinel-5P TROPOMI, offline Level-3 NO2 (COPERNICUS/S5P/OFFL/L3_NO2, 2019 to 2025)
Source: Air temperature, 1981 to 2025: ECMWF ERA5-Land reanalysis (Muñoz Sabater 2019), Copernicus Climate Change Service (ECMWF/ERA5_LAND/MONTHLY_AGGR, 1981 to 2025)
Source: Brick kilns and their 2025-26 firing season: BDPolicyLab kiln inventory (research/meherpur_brick_kiln), USGS Landsat 8/9, ESA Sentinel-2, AlphaEarth
Source: Age and sex structure, 2020: WorldPop (University of Southampton) age and sex structures, 2020 (WorldPop/GP/100m/pop_age_sex, 2020)
Source: Elevation: NASA/USGS Shuttle Radar Topography Mission, SRTMGL1 v3 (USGS/SRTMGL1_003, February 2000)
Source: Roads, bridges and facilities mapped by volunteers: OpenStreetMap contributors, ODbL 1.0 (OpenStreetMap via Overpass API, OSM database as of 2026-09-10T17:35:21Z)
Source: Administrative boundaries: BBS/OCHA Bangladesh administrative boundaries COD-AB v03 (valid 2023-05-21)
Source: Processing: Google Earth Engine (Gorelick et al. 2017); pipeline scripts/meherpur_space/export_layers.py in the BDPolicyLab repository As of 2026-09-10