Data Downloads

Every dataset behind this site is available below.

Lake areas, growth rates, dam types, slopes and settlement distances in these files are simulated (data/generate_data.py), as are every hazard score, risk class and change figure derived from them. Population and building counts are real. See the Methodology page for a full provenance table before reusing any of this data.

Lake risk GeoJSON

One Point feature per lake (WGS84 / EPSG:4326) with hazard score, risk class, dam type, elevation and basin.

lake_id · lake_name · area_km2 · area_growth_rate · dam_type · slope_downstream · distance_to_settlement_km · risk_score · risk_class · basin · district · elevation_m · last_updated
Download (15 KB)

Lake time-series CSV

Annual area values for all 25 lakes from 2000 to 2024 (625 rows).

lake_id · lake_name · year · area_km2 · centroid_lat · centroid_lon · basin · district
Download (37 KB)

Flood corridors GeoJSON

8 downstream LineString corridors for the highest-risk lakes, digitised from valley topography.

lake_id · lake_name · risk_class · geometry (LineString)
Download (4 KB)

Buffered corridors GeoJSON

All 25 lakes with ±2 km Polygon corridors (8 real LineStrings buffered, 17 synthetic from lake centroids).

lake_id · lake_name · data_source · geometry (Polygon)
Download (77 KB)

Population exposure JSON

Pre-computed population and building counts within each lake's flood corridor, derived from WorldPop Nepal 2020 and OpenStreetMap. Real data.

lake_id · lake_name · corridor_area_km2 · population_at_risk · buildings_at_risk · data_source
Download (5 KB)

Lake area cache (2016–2024)

One JSON file per lake, bundled as a zip. Each file records its own source: the cache shipped here is generated from the simulated area series, not from Sentinel-2.

lake_id · lake_name · last_updated · source · scenes[ year · date · area_km2 · cloud_pct ]
Download (10 KB)

GLOF event catalogue

Confirmed GLOF events used to train the classifier, curated from the ICIMOD GLOF database. Coverage is the wider Hindu Kush Himalaya — Nepal, Bhutan, Tibet and Sikkim — not Nepal alone. The events are real; the per-event attributes are unverified estimates.

lake_name · year · area_km2 · area_growth_rate · dam_type · slope_downstream · distance_to_settlement_km · elevation_m · glof_occurred
Download (1 KB)

Trained ML model

Random Forest classifier (scikit-learn, joblib). Cross-validation scores are omitted deliberately: the classes come from separately authored files, so any score reflects that split rather than predictive skill.

joblib pickle — loading a pickle executes code; only load if you trust the source
Download (160 KB)

PDF summary report

Auto-generated report with key statistics and a risk table for the top 10 lakes.

PDF
Download (2 KB)

WorldPop Nepal 2020 raster (~100 MB)

The WorldPop 2020 population raster is used to compute population counts within flood corridors. It is too large to bundle here — fetch it directly, or let the offline script do it:

https://data.worldpop.org/GIS/Population/Global_2000_2020/2020/NPL/npl_ppp_2020.tif

pip install -r requirements-offline.txt
python data/compute_exposure.py

Source: WorldPop (www.worldpop.org) — School of Geography and Environmental Science, University of Southampton. Licence: CC BY 4.0.