API Documentation

What still answers

Status values on existing records

Results of existing reports keep their original shape: riverine and coastal depths per return period under the scenarios and horizons they were produced with, plus their quality flags. Read them as a historical record of the retired method, not as current guidance.

One coordinate resolves into a site bundle: a bare-earth elevation grid, building footprints and road centrelines over a square box centred on the point. CDEM computes nothing — no ratings, no scores, no hydraulics. It emits data.

Endpoints

Submit a site

curl -X POST https://api.continuuiti.com/api/v1/cdem/analyze/ \
  -H "X-API-Key: sk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"lat": 17.4245, "lon": 78.3804, "half_km": 1.0, "label": "Site A"}'

Async only. A bundle takes roughly 30 seconds to 3 minutes; most of that is queueing. Poll /jobs/{job_id}/ until terminal, then fetch /results/{job_id}/. Jobs also complete unattended — you can submit and come back later.

Job status

Layer status — read before consuming a layer

Each of the three layers carries its own status, and three of the four are not "ok". An empty building layer and a failed building layer are the same empty array and mean opposite things — one is farmland, the other a dense district you would otherwise model as bare earth.

Geometry conventions

• Cells are rectangular and increasingly so with latitude. Read dx_m and dy_m separately; assuming square cells skews the site.
• origin is the centre of the north-west cell, not a box corner. row_order is north-to-south.
• Elevations are metres above EGM2008, bare earth. Buildings are not stamped in.
• NoData is null, never a sentinel — a −9999 would read as a very deep pit.
• Buildings and roads may extend beyond the terrain box: footprints overscan by 50 m and roads keep whole segments.

Batch

curl -X POST https://api.continuuiti.com/api/v1/cdem/batch/analyze/ \
  -H "X-API-Key: sk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"locations": [{"lat": 17.42, "lon": 78.38, "label": "A"}], "half_km": 1.0}'

Maximum 10 sites per batch, for real and sandbox runs alike. Each site is a separate bundle and 30 seconds to 3 minutes of work, so a batch is a handful of sites rather than a portfolio sweep. half_km is batch-level and applies to every site.

The submit path is /cdem/batch/analyze/, the same shape as every other batch module. A batch status payload counts PARTIAL sites separately under progress.partial (progress.usable = completed + partial).

Retention

Bundles are retained for 180 days, then deleted. The summary is kept indefinitely, so an expired report still tells you what was there. Re-running the same coordinate rebuilds the bundle.

Licensing

Licences travel per layer, not per bundle. Terrain (CC BY 4.0) and buildings (CC BY 4.0 or CDLA Permissive 2.0) are attribution-only. Roads are ODbL 1.0, which is share-alike on derived databases: redistributing a derivative of a road-bearing bundle carries that obligation. Attribution strings are in the payload; please surface them.

One coordinate resolves into two present-day hazard grids over a 6 km box: the JRC GloFAS riverine depth field at its native ~93 m and the WRI Aqueduct coastal inundation field at its native ~928 m, each for the 1 in 10, 50, 100 and 500 year return periods. Values are raw and unmasked, with a per-band audit of what is masked, exactly zero, the source's fill value or a real depth. The service computes nothing: no ratio, no scenario, no horizon, no flags, no projection of any kind.

Endpoints

Submit a site

curl -X POST https://api.continuuiti.com/api/v1/flood-field/analyze/ \
  -H "X-API-Key: sk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"lat": 23.6220, "lon": 58.2498, "label": "Muscat coastal site A"}'

Async only. A site takes seconds to tens of seconds (two Earth Engine reads). Poll /jobs/{job_id}/ until terminal, then fetch /results/{job_id}/. Jobs also complete unattended.

Job status

Layer status

Each of the two layers carries its own status. An empty grid and a failed read are different things: an inland box where the coastal source holds nothing is a real reading, a read error is not a reading at all.

Geometry and value conventions

• Each layer is sampled in its own native grid (EPSG:4326). row_order is north-to-south; origin is the centre of the north-west cell.
• dx_deg and dy_deg are carried so you can georeference without recovering them from the rounded metre sizes or the request box.
• pin (row, col, idx) and site_cell per return period are in the summary, so a list page needs no bundle fetch.
• Riverine: null is "outside the modelled floodplain"; cells valued exactly 1.000 m are a tile fill value on permanent water, counted in exact_1p000_cells and excluded from real_cells and real_max, never masked. valid_cells + masked_cells + outside_footprint_cells equals the cell count.
• Coastal: the sea is null, dry land is a real 0.0. exact_zero_cells is a first-class number; a 0.0 maximum is a value. Never test a depth for truthiness.
• Band arrays are flat, row-major, width × height long, with null for masked cells. Audit maxima are rounded (3 decimals riverine, 4 coastal); the bands are the raw source values.

Batch

curl -X POST https://api.continuuiti.com/api/v1/flood-field/batch/analyze/ \
  -H "X-API-Key: sk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"locations": [{"lat": 23.62, "lon": 58.25, "label": "A"}], "half_km": 3.0}'

Maximum 100 sites per batch, for real and sandbox runs alike. An optional half_km in the request body applies to every site (Tools stamps it on each item upstream; the default 3.0 is used when absent). A batch status payload counts PARTIAL sites separately under progress.partial (progress.usable = completed + partial). Batch results are paginated: page_size defaults to 100 and is capped at 200.

Retention

Bundles are retained for 180 days, then deleted; the job status and the results body carry expires_at. The summary is kept indefinitely, so an expired report still tells you what was there. Re-running the same coordinate reads the fields again.

Licensing

Licences and attribution travel per layer in the payload. The JRC datasets are available without restriction on use or distribution (Baugh et al. 2024, European Commission Joint Research Centre). WRI Aqueduct Flood Hazard Maps V2 are CC BY 4.0 (World Resources Institute). Please surface the attribution strings.