How well the map matches the drone surveys
What the radar saw
Mean terrain-flattened backscatter over the drone-mapped landslides and over stable ground in the same surveys, every pass of ascending track 70. The drop on 2 October is the landslide signal.
The rain that triggered it
Village tracts
Mapped landslide area by MIMU village tract. Click a row to zoom the map.
Villages closest to landslides
Where the landslides happened
Frequency ratio of mapped landslide ground in each class: 1.0 means as common as across the township, 2.0 twice as common. Bars show the ratio; the hover shows area.
Landslide sizes
Number of mapped landslides by area. The smallest class is set by the 0.2 ha minimum mapping unit.
Workflow
The previous script searched CDSE with a hard-coded box that missed the western part of the township, downloaded up to ten full GRD scenes from mixed tracks and orbit directions, and stopped there. The new workflow keeps the native 10 m grid from download to map.
Data sources
- Sentinel-1 IW GRDH, gamma0 RTC via CDSE Sentinel Hub (Copernicus DEM GLO-30 terrain flattening)
- Sentinel-2 L2A via CDSE Sentinel Hub (Jan–Mar 2026 median NDVI; post-event 30 Sep / 5 Oct)
- MIMU GeoNode: township, village tracts, villages, roads, rivers, schools, health facilities
- Copernicus DEM GLO-30 and ESA WorldCover 2021 v200 (AWS open data)
- OpenStreetMap (Overpass): buildings, roads, waterways
- Open-Meteo historical weather API: daily rainfall 1991–2026
- geonet-myanmar/launglon-landslides-2026: ten post-event drone inventories (2–8 Oct 2026), used for training and testing only
Limits. One post-event pass (2 Oct, ascending) exists so far; no descending pass has been acquired since the event. Slopes facing away from the radar (east-facing, steep) are seen at grazing angles and are mapped less reliably. Landslides smaller than 0.2 ha are mostly missed, and runout on paddy can be confused with flooding. The map ranks where to look; it does not replace a field check.