Executive Summary. Bangladesh is getting greener on satellite while its forests collapse, and the second fact, not the first, should drive policy. National NDVI rose 14.5% from 2000 to 2024 even as Hansen tree-cover loss reached 246,000 hectares (2,460 km2) from 2001 to 2023, about 13.5% of the country's 2000 tree cover, because monoculture cropland and managed vegetation register the same chlorophyll signal that biodiverse forest does. The interim government began canceling forest-encroaching projects in 2025, including 3,830 hectares at Sonadia Island; whether the BNP government, in office since February 17, 2026, sustains that reversal is the open question, and enforcement capacity in the Chittagong Hill Tracts and Madhupur remains the binding constraint. The single highest-leverage move is to convert satellite canopy-loss detection into a legally binding enforcement trigger with a published annual loss-reduction target, so that the greening signal stops masking the forest emergency underneath it.
Since 2000, Bangladesh has lost forest at a pace its replanting cannot touch. Hansen Global Forest Change data, the standard 30 m Landsat-derived record, puts tree-cover loss at 246,000 hectares (2,460 km2) for 2001 to 2023, roughly 13.5% of the 18,081 km2 of tree cover the country held at the turn of the millennium. The same Hansen raster computed without a canopy-density threshold flags 4,549 km2 of loss against just 24.5 km2 of gain, a loss-to-gain ratio of about 186 to 1. The threshold choice changes the magnitude, not the verdict: replacement is negligible next to loss.
The satellite record carries a second signal that pulls the opposite way. Over the same period the Normalized Difference Vegetation Index (NDVI), the standard satellite measure of photosynthetic activity across all land surfaces, has been rising. Mean national NDVI climbed from 0.463 in 2000 to 0.530 in 2024, a 14.5% increase. The two signals are not in conflict. They are the fingerprint of a country swapping complex, biodiverse forest for monoculture cropland and managed vegetation. The satellites see chlorophyll. They do not distinguish a 100-year-old sal tree from a row of hybrid rice.
A district-by-district reduction of the same Hansen raster shows where that loss concentrates: the three Chittagong Hill Tracts districts, Bandarban, Rangamati, and Khagrachhari, account for roughly 79% of the cumulative tree-cover loss measured across all 64 districts from 2001 to 2023.
The green is shifting
The MODIS-derived NDVI record for Bangladesh spans 2000 to 2024, providing a biennial snapshot of the country's overall vegetative health. The trend is clear and consistent: the country is getting greener in aggregate.
Mean NDVI rose from 0.463 in 2000 to 0.499 in 2006, dipped slightly to 0.479 in 2008, then resumed its climb to reach 0.523 in 2016. The most recent readings hover around 0.530 (2020: 0.526, 2022: 0.536, 2024: 0.530). Maximum NDVI has followed a similar trajectory, rising from 0.718 in 2000 to 0.785 in 2024, indicating that the greenest areas of the country are getting even greener.
The drivers of this greening are well understood. First, agricultural intensification: Bangladesh has expanded irrigated boro rice cultivation into areas that were previously fallow during the dry season, adding a layer of green to the winter landscape. Second, the government's social forestry program has planted trees along roads, embankments, and homesteads. Third, aquaculture ponds, which register as moderate NDVI due to algal growth and surrounding vegetation, have proliferated across the south and southwest.
But none of this compensates for the ecological loss. A rice paddy has an NDVI of 0.3-0.5. A mature tropical forest has an NDVI of 0.6-0.8. When forest at NDVI 0.7 gives way to cropland at NDVI 0.4 while irrigated agriculture expands elsewhere, the national mean NDVI can rise even as ecological value collapses. The satellite sees photosynthesis. It does not see biodiversity.
The national average also hides where the country is greenest and where it is not. Splitting the same MODIS record by district for 2024 shows the Chittagong Hill Tracts, still under forest canopy despite the loss shown above, holding the highest NDVI in the country, while Sunamganj and Bhola, dominated by haor wetland and char cropland, hold the lowest.
What the land has become
The Dynamic World land cover dataset, derived from Sentinel-2 imagery at 10-meter resolution, provides annual snapshots of how Bangladesh's land surface is classified. The record begins in 2016 and runs through 2024.
The most striking trend is the rise of the built-up class. In 2016, built areas constituted 19.9% of the classified landscape. By 2019 that figure had surged to 25.9%, and it has stayed above 23% since, standing at 25.3% in 2024. Read literally, that makes built the largest single class in the Dynamic World output, ahead of trees (16.3%) and crops (14.0%). The literal reading is the wrong reading. As the methodology section below details, Dynamic World's built class overestimates urban extent by capturing bare soil and construction sites, and its crop class swings with image timing relative to the harvest calendar. With roughly 60% of Bangladesh's land under cultivation, the on-the-ground crop share is far higher than any single annual Sentinel-2 snapshot shows. Treat the 25.3% built figure as a classifier artifact to interrogate, not as a finding that concrete now covers more of Bangladesh than farmland.
Trees have declined from 19.4% in 2016 to 16.3% in 2024. That trajectory aligns with the Hansen forest loss data: roughly 3 percentage points of national land cover transitioning from tree canopy to something else over eight years. Crops have fluctuated between 12.8% and 18.3%, with the variation likely reflecting differences in image timing relative to the agricultural calendar rather than real changes in cultivated area.
A second, independent land cover product confirms the same geography. ESA WorldCover, a 10-meter satellite classification for 2021, puts tree cover at 94.0% of classified area in Bandarban and 92.5% in Khagrachhari, against 12.3% in Sunamganj and 13.2% in Khulna, the lowest in the country.
Water coverage has been relatively stable at 11-13%, while flooded vegetation (a proxy for wetlands and aquaculture) has declined from 8.1% in 2016 to 5.8% in 2024. That 2.3-percentage-point drop in flooded vegetation is consistent with the wetland loss documented in Part 1 of this series: haors and beels being drained, filled, or fragmented.
The land cover transition tells a simple story: Bangladesh is converting natural and semi-natural landscapes (forest, wetland, grassland) into built-up and managed agricultural land. The net direction is toward simplification, less diverse land cover, more concrete and monoculture. This has consequences for biodiversity, for climate resilience, and for the long-term productive capacity of the land itself.
The seasons tell the truth
The annual NDVI average conceals important seasonal dynamics. Bangladesh's vegetation cycle is driven by the monsoon: pre-monsoon (March-May), monsoon (June-September), post-monsoon (October-November), and winter (December-February). Each season tells a different part of the story.
Post-monsoon NDVI is consistently the highest, reflecting the peak of the aman rice harvest and the lush residual moisture in the landscape. It has risen from 0.578 in 2005 to 0.619 in 2023. This is the season when Bangladesh looks greenest from space, and the trend confirms that agricultural productivity during the main rice season is increasing.
Pre-monsoon NDVI shows the most dramatic improvement: from 0.506 in 2005 to 0.579 in 2023, a 14.4% increase. This reflects the expansion of boro rice and dry-season irrigated agriculture. Two decades ago, much of Bangladesh lay fallow and brown between January and May. Today, irrigation has turned the pre-monsoon landscape greener, but at the cost of groundwater depletion that satellites cannot directly measure.
Monsoon NDVI, somewhat counterintuitively, is the lowest seasonal reading, ranging from 0.436 in 2005 to 0.493 in 2023. This is because monsoon flooding submerges vegetation and cloud cover interferes with satellite readings. The upward trend suggests either reduced flood extent in some years or improved agricultural management in flood-prone areas.
Winter NDVI has risen from 0.441 in 2005 to 0.517 in 2023, the largest proportional increase of any season (17.2%). This is the clearest signal of agricultural intensification: where once the land lay dormant in winter, it is now producing crops.
What the satellites cannot see
The NDVI record cannot distinguish between a diverse agroforestry system and a monoculture rice paddy. Both register as "green." It cannot measure soil organic carbon, which declines under continuous cultivation. It cannot count the species of birds, insects, and amphibians that disappear when a wetland is drained and planted with rice.
The forest loss figure, 4,549 km2 lost against 24.5 km2 of gain on the unthresholded Hansen raster, is an ecological emergency, and it still understates the degradation of the remaining canopy. The Sundarbans, the world's largest mangrove forest and Bangladesh's most critical ecological asset, still appears as dense green canopy in the NDVI record. But inside that canopy, salinity is increasing, Sundri trees (Heritiera fomes) are dying, and tiger habitat is shrinking. A forest that looks healthy from 700 kilometers up may be hollowing out from within.
A 2000 satellite baseline of mangrove extent shows just how concentrated that asset is: three southwestern districts, Bagerhat, Khulna, and Satkhira, the districts that share the Sundarbans, hold 154,524 ha, 143,090 ha, and 102,841 ha of mapped mangrove respectively, a combined 400,455 ha. The seven other coastal districts with any mapped mangrove hold 3,219 ha between them, and the remaining 54 districts have none.
The Dynamic World classification carries its own limits at national scale. The line between "crops" and "grass" can blur in Bangladesh, where fields and grazing land share the same parcels across seasons. More consequentially, the "built" class overestimates urban extent wherever bare soil or active construction sites are temporarily read as built-up, which is why its 25.3% reading should not be taken as a literal measure of how much land is paved or developed.
What policy must do
The satellite evidence demands three responses, each with a named owner and a measurable signal of success.
First, make satellite canopy-loss detection a binding enforcement trigger. Bangladesh's replanting offsets a vanishing fraction of its losses, and the Chittagong Hill Tracts and the Madhupur Sal forest keep losing canopy to shifting cultivation, illegal logging, and settlement encroachment. The Forest Department, under the Ministry of Environment, Forest and Climate Change, should be mandated to act on automated alerts generated from the same Landsat and Sentinel imagery behind this analysis: every detected loss event above a set parcel size opens a case file with a fixed response window, and confirmed illegal clearing carries statutory penalties. Success signal: total annual Hansen-measured tree-cover loss falls below the 2001-2023 trend within three years, and confirmed enforcement actions per detected loss event rise from near zero toward one-to-one. Planting ceremonies do not count toward the target.
Second, map and legally designate the remaining flooded vegetation. The decline from 8.1% to 5.8% of national land cover between 2016 and 2024 is the ongoing destruction of Bangladesh's wetland heritage, the same haors, beels, and floodplains discussed in Part 1, whose flood buffering, fisheries, and groundwater recharge are irreplaceable. The Department of Environment should complete a national inventory of every remaining wetland larger than 1 km2, gazette each as protected, and publish an annual satellite-monitored status report. Success signal: a gazetted, publicly mapped wetland register completed within 18 months, after which annual net flooded-vegetation loss inside designated sites is held at or below zero.
Third, audit the groundwater cost of the winter greening. The 17.2% rise in winter NDVI is the signature of irrigated agriculture pushing into a season that was historically fallow, overwhelmingly powered by groundwater extraction, with water tables dropping in the northwest (Rajshahi, Rangpur). The Bangladesh Water Development Board, working with the Barind Multipurpose Development Authority, should link satellite-derived dry-season crop maps to monitored well levels and publish a district groundwater balance. Success signal: an annual district-level extraction-versus-recharge balance published within two years, with stressed districts assigned dry-season cropping limits where the balance turns negative.
Bangladesh's vegetation story is one of substitution: trading ecological complexity for agricultural simplicity, biodiversity for caloric output, and groundwater for a green landscape in seasons it was never meant to be green. The satellites see the green. Policy must see what is behind it.
The Satellite Bangladesh series. Part 1: Bangladesh is Drowning. Part 2: Bangladesh is Glowing. Part 3: Bangladesh is Greening (and Browning) (this part). Part 4: Bangladesh is Eroding. Part 5: Bangladesh is Dividing. Part 6: Bangladesh is Trading.
Sources
- MODIS NDVI: NASA Terra MOD13A2, 1 km resolution, 2000-2024. https://lpdaac.usgs.gov/products/mod13a2v006/. Internal output:
bd_gis/outputs/vegetation/ndvi_timeseries.csv,seasonal_ndvi.csv. - Hansen Global Forest Change v1.11: University of Maryland, Google, USGS, NASA. 30 m Landsat-derived, 2000-2023. https://glad.earthengine.app/view/global-forest-change. Internal raster computation (no canopy-density threshold):
bd_gis/outputs/vegetation/forest_stats.csv= 4,549 km2 loss, 24.5 km2 gain. - Tree-cover loss 246,000 ha (2,460 km2), 2001-2023, ~13.5% of 2000 cover: Global Forest Watch / Hansen GFC, Bangladesh country dashboard (30% canopy-density threshold). https://www.globalforestwatch.org/dashboards/country/BGD/
- Bangladesh 2023 tree-cover loss ~17,806 ha: Global Forest Watch, Bangladesh country dashboard, 2023. https://www.globalforestwatch.org/dashboards/country/BGD/
- Dynamic World land cover: Google / World Resources Institute, Sentinel-2, 10 m, 2016-2024. https://dynamicworld.app/. Internal output:
bd_gis/outputs/landcover/dynamic_world_timeseries.csv. - 2025 government reversals (Sonadia 3,830 ha): Mongabay, "Bangladesh retreating from development activities planned in forest lands," August 27, 2025. https://news.mongabay.com/2025/08/bangladesh-retreating-from-development-activities-planned-in-forest-lands/
- District NDVI (2024): NASA MODIS MOD13A1, 500 m, 16-day composite, annual mean, computed per district on Google Earth Engine (a separate product from the 1 km MOD13A2 national series above). https://lpdaac.usgs.gov/products/mod13a1v061/. Internal output:
data/geo/gee/modis_annual_ndvi_district.parquet. - District tree-cover loss (2001-2023): Hansen Global Forest Change v1.11, 30 m, per-district sum computed independently of the national figures above. Internal output:
data/geo/gee/hansen_forest_loss_district.parquet. - District tree cover (2021): ESA WorldCover v200, 10 m resolution. https://esa-worldcover.org/en. Internal output:
data/geo/gee/esa_worldcover_2021_district.parquet. - District mangrove extent (2000 baseline): Giri, C. et al. (2011), "Status and distribution of mangrove forests of the world," Global Ecology and Biogeography, via the Landsat-derived
LANDSAT/MANGROVE_FORESTSEarth Engine asset, 30 m. Internal output:data/geo/gee/gmw_mangroves_district.parquet. - Administrative boundaries: geoBoundaries ADM2, William and Mary geoLab, used for the district map rendering. https://www.geoboundaries.org. The four district datasets above were computed against FAO GAUL 2015 level-2 boundaries on Earth Engine; district names match the geoBoundaries set used for rendering (64 of 64).
- District-level Earth Engine boundaries: FAO GAUL 2015 level-2, used for the per-district computations behind the district maps.