In July 2020, satellite sensors orbiting 700 kilometers above Bangladesh captured something that no ground survey could: the entire country, from the Teesta basin to the Meghna estuary, in a single frame. What they recorded was water. An extraordinary amount of water. Over 2,600 square kilometers of land lay submerged under monsoon floodwater, a sprawling inland sea visible from space.
That image was not an anomaly. It was just one frame in a 40-year movie that tells a story of a country locked in an intensifying struggle with water, a struggle it is losing.
This is that story, told entirely through satellite data.
The monsoon footprint is not shrinking
Every year, Bangladesh transforms. The dry season reveals cropland, roads, settlements. Then the monsoon arrives and water reclaims territory that humans thought was theirs. Satellites have been measuring this transformation since 1990.
Flood extent time series data is under development and will be added when satellite processing is complete.
The pattern is striking. In 1990, monsoon floodwater covered 3,826 km2 of land surface. By 2015, that number reached 3,622 km2, essentially unchanged over 25 years. The 2010 monsoon was nearly as severe, at 3,459 km2.
But the averages hide the real danger. Look at the seasonal swing: the difference between dry-season water and monsoon-season water. In 2015, that swing was 3,094 km2, the largest in the dataset. That means over three thousand square kilometers of land transitioned from dry ground to standing water in a matter of weeks.
The 2020 monsoon, at 2,616 km2, was moderate by historical standards. That should not be reassuring. A "moderate" monsoon in Bangladesh still drowns an area larger than Luxembourg.
What makes this data alarming is not any single year. It is the persistence. Forty years of satellite observations show no downward trend in flood extent. If anything, the volatility is increasing: the gap between mild years (1995, at 2,038 km2) and severe years (1990, at 3,826 km2) represents a near-doubling. Communities cannot plan around that level of uncertainty.
The rivers are eating the land
Flooding is temporary. River erosion is permanent.
When a river erodes its bank, that land does not come back. Homes, farms, schools, mosques, everything on it falls into the current and is carried away. Satellite-based channel migration analysis across 16 major rivers reveals the scale of this destruction.
The Surma River in Sylhet is the most aggressive eroder in the satellite record. Between 1995 and 2005, it was consuming 486 hectares per year. That rate accelerated to 680 ha/yr in 2005-2015, and then to 734 ha/yr in the most recent decade. That is not stabilizing. That is a river accelerating its consumption of the landscape.
The Sangu River in the Chittagong Hill Tracts tells a similar story: from 14 ha/yr in 1995-2005 to 96 ha/yr in 2015-2025, a nearly seven-fold increase. The Karnaphuli, which flows through Chittagong city, has also picked up pace, reaching 18 ha/yr in the latest period after a lull in 2005-2015.
Not every river is getting worse. The Arial Khan has slowed from 79 to 28 ha/yr. The Dharla dropped from 25 to 4 ha/yr. But these improvements are dwarfed by the acceleration elsewhere.
Add the numbers up across all rivers with measurable erosion, and the picture is clear: Bangladesh is losing hundreds of hectares of land per year to river bank collapse. Each hectare lost represents families displaced, livelihoods destroyed, and tax base eroded (in every sense of the word).
The wetlands are vanishing
Bangladesh's haors, beels, and floodplains are the country's natural shock absorbers. During monsoon, they expand to hold excess water. During dry season, they contract, exposing fertile land and sustaining fisheries. They are, in ecological terms, the kidneys of the delta.
The satellites show those kidneys failing.
Chalan Beel, once the largest beel in Bangladesh at nearly 77 km2 in 1993, has collapsed to barely 11 km2 by 2023. That is an 86% decline in three decades. The causes are well-documented: encroachment for rice cultivation, siltation from upstream deforestation, and unplanned road embankments that fragment the wetland.
The Meghna Floodplain has shrunk from 186 km2 to 58 km2, a 69% loss. Hail Haor dropped from 18 km2 to under 3 km2. Even the larger haors show wild inter-annual swings that suggest degraded capacity: Hakaluki Haor measured 197 km2 in 1990 but collapsed to near-zero in 1999 and 2020, likely due to dry-season drainage for agriculture, before partially recovering.
Tanguar Haor, a Ramsar wetland site under international protection, has been more stable, fluctuating between 92 and 284 km2. But stability for a protected site is the bare minimum. The unprotected wetlands are the ones being destroyed, and they are the ones that millions of rural Bangladeshis depend on for fish, water, and flood buffering.
The arithmetic of wetland loss is simple and devastating: every square kilometer of haor that disappears is a square kilometer of floodwater that has to go somewhere else. It goes into villages.
The rain is getting wilder
The satellite and ground-station rainfall record spanning 1985 to 2023 reveals not a simple increase in total precipitation, but something more dangerous: increasing volatility.
Mean annual rainfall across Bangladesh ranged from 2,130 mm (1989, a relatively dry year) to 3,148 mm (2017, the wettest in the record). That is roughly a 1.5-to-one ratio between the driest and wettest years. For a country whose agriculture, infrastructure, and settlement patterns were designed around a relatively narrow rainfall band, this volatility is a crisis in itself.
The 2017 spike is worth pausing on. At 3,148 mm mean rainfall, with a spatial maximum of 7,419 mm, that year was an outlier by any measure. It coincided with devastating floods in Sylhet, Sunamganj, and the northern districts, displacing millions.
The monsoon share of rainfall shows a similar pattern of unpredictability. Monsoon precipitation has oscillated between 1,293 mm (1992) and 1,931 mm (2004), with no clear directional trend but substantial year-to-year swings. In 2022, monsoon rainfall dropped to 1,355 mm, only to rebound to 1,864 mm in 2024.
Meanwhile, land surface temperature data from MODIS satellites shows the thermal backdrop against which all of this is playing out. Daytime land surface temperatures have remained stubbornly high, averaging 26.6 to 27.8 degrees Celsius across the country. Night-time temperatures show a slight upward drift over the 2000-2024 period, from around 19.4 degrees to above 20.9 degrees. Warmer nights mean more atmospheric moisture capacity, which means more intense rainfall events when they do occur.
What the satellites cannot see
The satellite record is powerful but incomplete. It measures area, not depth. It captures extent, not velocity. It shows where water was, not how many people it displaced.
But the data it does provide leads to an inescapable conclusion: Bangladesh's water crisis is not a future threat. It has been unfolding for four decades, visible from orbit, measured in square kilometers and hectares, and it is getting worse in several critical dimensions simultaneously.
The monsoon flood footprint is not shrinking. River erosion is accelerating in the northeast and southeast. Wetlands that once buffered the worst impacts are being destroyed. And rainfall is becoming more volatile at exactly the moment when the country's natural and built infrastructure is least able to absorb shocks.
What policy must do
The satellite evidence points to three urgent priorities.
First, protect the remaining wetlands. Chalan Beel's collapse from 77 km2 to 11 km2 is not a natural process. It is the result of policy failures: unregulated encroachment, infrastructure that fragments ecosystems, and agricultural incentives that reward draining wetlands for rice. Every haor and beel needs the legal protection and enforcement that Tanguar Haor receives. The Ramsar model works. Scale it.
Second, invest in erosion-resistant riverbank infrastructure along the Surma and Sangu corridors. These rivers are accelerating. At 734 ha/yr, the Surma is consuming land faster than any resettlement program can relocate people. This requires hard engineering: revetments, geobags, strategic dredging. But it also requires land-use policy that stops allowing settlement on actively eroding banks.
Third, redesign flood preparedness around volatility, not averages. The difference between a 2,038 km2 monsoon (1995) and a 3,826 km2 monsoon (1990) is the difference between manageable and catastrophic. Current preparedness plans tend to be calibrated to recent experience. The satellite record shows that the system can and does swing to extremes within a single decade. Early warning systems, evacuation infrastructure, and crop insurance programs must be designed for the worst case the satellite record has observed, not the average.
Bangladesh did not choose its geography. It sits at the bottom of the world's largest river delta, receiving the runoff from 1.7 million square kilometers of the Himalayas. But geography is not destiny. The Netherlands sits below sea level and thrives. The difference is policy.
The satellites have been watching for 40 years. The evidence is clear. The question is whether anyone is acting on it.
This is Part 1 of the "Satellite Bangladesh" series. Part 2 will examine urban expansion and the collision between cities and floodplains.
Data Sources
- Flood extent: Landsat 5/7/8/9 surface water classification (NDWI/MNDWI/AWEI with Otsu thresholding), 1990-2020. USGS/NASA.
- River erosion: Landsat-derived channel migration analysis across 16 major rivers, 1995-2025. USGS/NASA.
- Wetland area: Landsat-based haor/beel delineation using dry-season composites (Nov-Feb), 1990-2023. USGS/NASA.
- Rainfall: CHIRPS (Climate Hazards Group InfraRed Precipitation with Station data), 5.5 km resolution, 1985-2023. UCSB/USGS.
- Monsoon rainfall: CHIRPS monsoon-season (Jun-Sep) precipitation, 1990-2024. UCSB/USGS.
- Land surface temperature: MODIS MOD11A2 (daytime and nighttime LST), 1 km resolution, 2000-2024. NASA.
- Poverty proxy / nightlights: VIIRS Day/Night Band (NASA), DMSP-OLS (NOAA), WorldPop population density, GHSL built-up area (EC JRC), MODIS NDVI.
- Administrative boundaries: geoBoundaries (William & Mary).
- All satellite processing: Google Earth Engine.