Where the border rivers moved, and how much ground they took or gave back.
Eleven reaches of the Brahmaputra, Ganga and Padma, Teesta, Torsa, Mahananda, Ichamati, Raimangal, Kushiyara, Feni, Sutlej, and Ravi, compared pixel by pixel from the first Landsat years to 2020, then checked against Sentinel-2 images of the 2025-26 dry season. For frontier planners, survey liaison, and the disaster authorities who deal with erosion and the people it displaces.
Figures load from the latest export.
Border rivers do not stay put. Assam's Water Resources Department records that "more than 4.27 Lakh Hectares of land was already eroded away by the river Brahmaputra and its tributaries since 1950", 7.40 % of the state, and puts the average loss at nearly 8,000 ha a year (Government of Assam). On the Ganga in West Bengal, more than 350 km² of Murshidabad district was lost in four decades and more than 80,000 people were displaced across the state (Prasad and Sen, EGU 2017).
When a channel moves, roads, embankments, outposts, and whole villages end up on the other bank or under water, and new chars rise where boats used to pass. Planners usually learn of it from the ground, one report at a time. The satellite record of every reach, back to 1984, is public but hard to use.
- Measures, for each reach, how much ground became permanent water (erosion) and how much former channel became land or seasonal bar (emergence), in square kilometres and per 10 km of river.
- Finds every connected patch of change of 10 ha or more, with its area, its exact location, and its distance and direction from the nearest town.
- Shows the change as map layers drawn from the satellite's 30 m pixels: channel change, all ten JRC classes, how often water was seen, and the river's water in the first year, in 2020, and now, side by side.
- Counts only a corridor around each river, built from the water itself, so ponds, canals, and wetlands in the same box do not inflate the numbers.
- Rates each reach's confidence from how many pixels wide its channel is, and states why.
- Adds an indicative view of change since 2020 from Sentinel-2 images of the 2025-26 dry season: new water on ground that was dry land in 2020, and 2020 channel that is now dry.
| Reach | Erosion km² | Emergence km² | Permanent water km², first year → 2020 | New water since 2020 km² | 2020 channel now dry km² |
|---|---|---|---|---|---|
| Loading… | |||||
Inside each reach's corridor. The first year is each pixel's first year with enough observations, 1984 or later. The last two columns compare Sentinel-2 water in the 2025-26 dry season with JRC water in 2020; they are indicative, because river levels differ between years.
- Read the record. JRC Global Surface Water classes every Landsat scene since March 1984 as water or not, pixel by pixel, month by month. NADI reads only the windows under its eleven reaches.
- Find the river. Trace each reach through its own water and keep only that river's belt and banks.
- Count the change. Sum exact pixel areas by class: permanent and seasonal water, first year against 2020.
- Find hotspots. Group touching pixels of erosion or emergence; keep patches of 10 ha or more.
- Check since 2020. Map water in the clearest Sentinel-2 passes of the last dry season on the same grid, and compare it with 2020.
- Publish. Map layers, tables, and CSV exports in the console, each with its source and time.
The full method, with formulas, thresholds, agreement checks between JRC products, and known failure modes, is in the methods note and on the console's methods page.
| Source | What | Licence | Status |
|---|---|---|---|
| JRC Global Surface Water v1.3 (Pekel et al. 2016) via Microsoft Planetary Computer | Transitions, occurrence, change intensity, seasonality, and maximum extent, 1984-2020, 30 m | Copernicus: free of charge, without restriction of use; cite Pekel et al. | In use |
| Sentinel-2 Level-2A (ESA Copernicus) via Microsoft Planetary Computer | Dry-season water 2025-26 from NDWI and MNDWI, compared with JRC 2020: change since 2020 (indicative) | Copernicus Sentinel data, free and open; contains modified Copernicus Sentinel data 2025-2026 | In use |
| Trinetra frontier reference places | Public town locations for describing where a hotspot is | Trinetra | In use |
Nothing in NADI is simulated. The JRC record stops in December 2020: JRC's later releases mix Landsat collections with a co-registration offset that can reach one pixel, the size of the change being measured. The Sentinel-2 layer fills the gap with an indicative dry-season view.
- Frontier planners: which reaches are moving, and where a road, outpost, or fence line sits close to a channel that has already shifted.
- Survey liaison: a common, dated picture of where the river was and is, to bring to joint inspections.
- District and state disaster authorities: where erosion is concentrated along each reach, to set priorities for protection works and relief.
- Researchers and journalists: the numbers, the method, and CSV exports.
- River morphology only. NADI does not draw, use, or infer any international boundary, and nothing in it is a statement about one.
- Places and conditions only. NADI holds no data about people.
- Advisory only. Pages name the State Disaster Management Authority, the state water resources department, the Central Water Commission, the Brahmaputra Board, and the Ganga Flood Control Commission as the authorities to follow.
- Every number shows its source and retrieval time; the method, checks, and limits are published.
- Runs from local exports. Once the JRC tiles are copied, nothing needs a foreign cloud to operate.
- The JRC record ends in December 2020. The Sentinel-2 view since then is one dry season, from a different sensor, at whatever river level that season had; read it as indicative.
- 2020 was a major flood year on the Brahmaputra in Assam, which inflates seasonal water gains there.
- Channels under about 100 m wide, such as the Feni at Sabroom, are only a few pixels across; their figures are indicative.
- Aquaculture ponds and other human water bodies near a channel can appear as water gains.
- In the Sundarbans, water extent depends on the tide at each image time.
We can add other border reaches on request, and run a pilot with a district or state disaster authority on recent imagery. contact@trinetra.in