← Return to Flood ITNEPAL · EXPERIMENTAL TERRAIN SCREENING
What this map can tell you.
Flood IT explores how a hypothetical water surface could connect from a selected river reach to surrounding terrain and mapped residential land. It is a screening prototype, with a visible assessment boundary and reproducible inputs.
Not locally calibrated. Not an operational forecast.This release does not establish flood probability, safe locations, evacuation routes, arrival time, velocity or damage. Exposure in an assumed scenario is different from a complete flood risk assessment. Local hydraulic modelling and specialist review are needed before operational planning decisions.
Local assessments across Nepal
Select a natural river or stream anywhere in the modelled atlas, either on the terrain or in the river finder. Choose up to 6, 12 or 20 km around that location, wait for the local elevation and source mapping, set a hypothetical offset, then select Flood It. Use the west/east region navigator to move across Nepal. The original Narayani example remains available as a precomputed benchmark.
Nationwide inputs retain 55,331 original river/stream segments, 45,256 water polygons and 86,288 residential polygons from the same Nepal OSM extract. Unnamed channels are searchable by OSM ID. Segments are not distinct rivers; polygon counts are not households or people. Missing mapping remains unknown. Source import recorded one unreadable waterway and one unreadable area. Results are limited to source coverage, which follows the Nepal extract boundary, including some border context.
| National calculation | Treatment and limits |
|---|
| Reach selection | Original centrelines join only at an exactly shared endpoint when there is one unambiguous compatible continuation. Different named branches are not silently merged. Reaches stop at unresolved junctions; some are shorter than the requested maximum. Channels shorter than 300 m are not assessed. Canals, drains and ditches remain visible but are excluded from this natural-channel method. |
| Elevation | Native floating-point Copernicus GLO-30 surface elevations are read on demand through bounded byte requests from the public AWS source. The decoder uses the full-resolution image, aligned on 1 arc-second pixel centres, including shared tile edges. No coarser display grid substitutes for the analysis. The local Three.js surface also uses this grid; mobile rendering may thin vertices without changing calculations. |
| Local domain | The selected reach bounds receive a 3 km computational margin. Reporting excludes the outer three grid cells and cells assigned to the first/last four reference stations. Invalid or unreachable cells and cells outside the mapped extract have no result. Hatching marks this domain limit. A warning appears when estimated inundated land meets the reporting edge. This is a local window, not a basin-wide simulation; flood effects can continue beyond it. |
| Reference and connectivity | The same local 3 × 3 minimum, 11-station median and downstream-decreasing isotonic reference used in the pilot are applied at approximately 30 m stations. Direction is estimated from endpoint elevation groups; little measured drop is labelled uncertain. Four-neighbour minimum-spill connectivity determines 25 scenarios from +0 to +12 m. These offsets are not today's river level, rainfall, discharge or gauge readings. |
| Data checks | Missing channel elevations block a calculation. A mean profile correction above 10 m or any correction above 45 m also blocks it; these are conservative implementation guards, not calibrated accuracy criteria. A mean correction above 3 m is flagged. Try a shorter reach where appropriate. Narrow gorges, steep headwaters and vegetation can make the surface model unsuitable even when source pixels exist. |
| Exposure | Original residential polygons, including holes, intersect native cell footprints in the local UTM zone (usually 44N or 45N). Only the affected portion turns red. Mapped water polygons are excluded from land exposure. Overlapping residential land is unioned for total area, while distinct source polygons retain separate rows. Full source polygon area is the percentage denominator, and assessed coverage is reported separately. Zero mapped areas never establishes absence of settlement or flood risk. |
| Performance and reproducibility | Calculations run in a cancellable worker on the user's device. Each window is bounded to 1.2 million native cells. Changing a reach or its length clears the previous result and comparison. JSON reports include the selected source IDs, bounds, stage, source timestamp, elevation tiles, native Float32 grid SHA-256 hash, method version and area results. Source-service failures show a retry message; there is no synthetic or distance-buffer fallback. |
National method version: connected-stage-national-v2. Download original national source tiles and catalogue · National analysis source and verification tests. The 3D exploration models remain available across all atlas regions; suitability of a flood calculation is checked separately for each selected reach.
The original Narayani benchmark
The original precomputed example is a 21.38 km Narayani centreline from Devghat through Bharatpur / Gaindakot. Choosing the original example preserves the pilot's fixed grid, bounds and 25 results. Its data and detailed procedure below remain available for reproduction.
| Input | Source and treatment |
|---|
| Surface elevations | Copernicus GLO-30 Public on AWS, tile N27 E084. Native 1 arc-second grid, approximately 27.4 × 30.8 m here. This is a surface model, including vegetation, infrastructure and buildings; it is not surveyed bare earth or river bathymetry. Pixel spacing is not a statement of elevation accuracy. Floating-point source elevations are used for analysis. |
| River reach | OSM way 25708647. The centerline geometry is retained without display simplification. Side channels and tributary inflows are not separate hydraulic sources. OSM topology does not establish measured discharge or bankfull stage. |
| Residential land | 1,311 original OSM landuse=residential polygons intersect the pilot rectangle. Source: Geofabrik Nepal extract, snapshot 6 September 2026 at 20:21:35 UTC. These are mapped land-use areas, not individual buildings, households or population. Names such as “Residential area 001” are interface identifiers for unnamed polygons; OSM IDs are retained. |
| Mapped water | OSM water polygons are removed from estimated inundated land and residential impact calculations. Missing, seasonal or outdated water mapping can affect the result. |
| Coverage | Analysis uses a buffered window (84.24–84.49° E, 27.61–27.82° N). Reported results are limited to 84.27–84.46° E, 27.64–27.79° N and exclude cells assigned to the first/last approximately 100 m of the channel reference. Hatched areas have no assessment. A polygon may be only partly assessed. |
How the original benchmark is calculated
- The source centerline is projected into UTM 45N and sampled at approximately 30 m stations. A local three-by-three-cell minimum reduces narrow surface artifacts. An 11-station median filter followed by a downstream-decreasing isotonic fit defines an estimated channel surface profile. Its elevations range from approximately 182 m upstream to 162 m downstream.
- Each terrain cell is assigned the reference elevation of its nearest station along the selected reach. The user-selected offset (0–12 m, in 0.5 m increments) is added to this varying profile. The reference is not a measured riverbed, normal-flow level, bankfull level or gauge datum. “+0 m” is the assumed reference and may already touch mapped land; it does not represent today's river conditions.
- A four-neighbour minimum-spill calculation finds the smallest stage at which each cell can connect to the selected channel over the intervening terrain. A depression behind a higher bank does not connect until the bank's relative height is exceeded. Invalid cells and excluded endpoints are barriers. This is a terrain-connectivity rule, not a distance buffer.
- The resulting native-grid cell footprints are intersected with original residential polygons in metres (UTM 45N), preserving polygon holes and partial intersections. Percentages use each full source polygon as the denominator; assessed coverage is shown separately. Total residential land exposure uses the union of polygons so overlapping land is counted once. Area counts refer to distinct source polygons and may overlap geographically.
- The Three.js overlay displays these same masks on a separate terrain mesh. Camera position, pan/orbit, display colours and vertical exaggeration do not participate in the analysis. Red fill marks only the affected portion of a residential polygon; a red outline helps locate any polygon with an affected portion.
This method is not HAND and is not a hydrodynamic solver. It uses a nearest-station channel profile with terrain connectivity. It does not conserve flow volume or momentum, route a hydrograph, model infiltration, simulate tributary discharge or reproduce backwater effects. Nearby flow systems may interact in reality. Broad, long or branching floodplains can produce misleading connectivity and reference assignments.
Building coverage
The independent building layer shows original OSM footprints when zoomed in and building-point density at broader scales. This corrects missing built neighbourhoods that were absent from the residential land-use layer. These footprints are visual context only and are not individually assessed for flood exposure. Buildings outside mapped residential polygons are absent from residential exposure totals. Neither a pale footprint nor a blank area means safe. The building dataset and source notes describe its snapshot, precision, holes and incomplete coverage. House labels represent named places; they do not represent individual buildings or population.
Read the colours
| Map treatment | Meaning |
|---|
| Blue | Estimated inundated land in the selected scenario, excluding mapped water polygons. |
| Red residential fill / outline | A residential polygon is potentially exposed; only its intersecting portion is filled red. A small intersection does not mean the entire settlement is flooded. |
| Amber fringe | Additional terrain reached at one metre above the selected scenario. The associated lower/upper counts use ±1 m, clipped at 0 and 12 m. This is a parameter sensitivity check, not a confidence interval, measured DEM error band or flood probability. |
| Neutral | Outside the selected scenario within assessed coverage. This does not mean safe or low risk. |
| Hatching | Outside the assessment domain. Missing residential mapping is also absent from the counts. |
| White comparison outline | The extent from the pinned scenario. It is another hypothetical stage, not a historical observation. |
| House symbol | A mapped settlement location. It is not a building footprint and is not used to infer the exposure of the whole town. |
Validation status
Automated checks cover native elevation registration across source-tile corners, original vector loading, river selection, bank barriers, disconnected depressions, invalid cells, diagonal leakage, monotonic stages, polygon holes, partial coverage, source geometry preservation, mask registration and residential intersections. These checks verify implementation consistency; they do not establish real-world flood accuracy.
Sentinel Asia's August 2017 Nepal flood collection includes an IWMI map from Sentinel-1 observations on 13 August 2017 covering the wider Chitwan/Nawalparasi area. It was reviewed as regional context. No matching surveyed channel stage or georeferenced validation extent was established for this prototype; no quantitative event validation or local specialist review has been completed. The 2017 map is not used to tune the scenario stages or claim predictive accuracy.
The terrain can miss embankments, small drains, culverts and narrow channels. Surface vegetation and buildings can create artificial barriers. River morphology and settlements change, and the elevation acquisition and OSM snapshot dates differ. Real flood planning requires suitable surveyed terrain and channel geometry, discharge/boundary conditions, calibration and independent review. See USGS flood inundation mapping science for the role of calibrated water-surface modelling and terrain. Current official Nepal river information is available from DHM River Watch; Flood IT is not connected to those gauges.
Original benchmark data and reproducibility
Original pilot OSM geometryScenarios, polygons & provenanceChannel reference profileConnected-stage grid
The stage grid is row-major, north to south, 900 × 756 little-endian unsigned 16-bit values in centimetres (rounded up); 65535 means not assessed. The exact scenario masks were computed with floating-point elevations before encoding and are listed in the manifest. Each grayscale PNG mask has identical geographic bounds; white cells are included and black cells are excluded. Use the separate assessed mask to distinguish no result from outside the scenario.
Analysis version: connected-stage-v1. Source DEM SHA-256 is recorded in the manifest. Preprocessor · Connectivity algorithm · Verification tests. To regenerate, place these scripts in the source project's scripts directory and run the preprocessor with the Nepal PBF and the N27 E084 Copernicus tile. Dependencies: NumPy, SciPy, Shapely, PyProj, Rasterio, PyOsmium and Pillow. Delete the cached source.geojson when changing the PBF snapshot.