Open source · Canadian open data

Anywhere in Canada, draw and run

Draw a boundary on the map and get a complete, ready-to-run EPA SWMM stormwater model — rainfall, terrain, land cover, soil, and the city's real storm and sanitary pipes, assembled from open data. No hunting across data portals, no manual setup.

Try the live beta View on GitHub Read the preprint

MIT licensed
EPA SWMM 5.2
35 real-network cities
Release v0.5.0

swmmcanada

› draw a boundary on the map
  · ECCC rainfall + design storms
  · 1 m LiDAR terrain where available
  · land cover · soil · storm + sanitary pipes
✓ model.inp ready · runs in EPA SWMM 5.2
✓ exports · MIKE+ · InfoWorks ICM

How it works

Draw, fetch, build

Draw

Draw (or upload) a boundary anywhere in Canada in the web app, pick your simulation dates, and click Build SWMM model.

Fetch

SWMMCanada pulls the Canadian open data for that spot — rainfall, terrain (1 m LiDAR where available), land cover, soil, ECCC design-storm intensities, and the city's storm and sanitary pipes.

Build

A complete, validated model.inp you can open and run in EPA SWMM 5.2 — plus a shareable datastore and export packages for MIKE+ and InfoWorks ICM.

The web app

A real city network, ready to download

The SWMMCanada web app after building a downtown Victoria area: the storm network in blue and the sanitary network in brick with flow arrows, per-system layer toggles, and a click-to-inspect card showing subcatchment attributes

Downtown Victoria after a build: the storm network (blue) and the sanitary network (brick) with flow arrows and diameter-scaled pipes, per-system layer toggles, and click-to-inspect attributes — with the model package (SWMM, MIKE+, ICM) ready to download.

Validation

Checked against a real flow gauge

Automatic model building is easy to claim and hard to prove. Over the summer of 2024, an uncalibrated model built automatically over the official 22 km² Water Survey of Canada basin for Graham Creek, Ottawa reproduced all 12 rainfall events on the correct days against the measured record, with a total volume bias of −22.3%. The remaining gaps are structural and explained rather than tuned away.

Daily rainfall and daily mean flow at Graham Creek, Ottawa, summer 2024: the observed WSC gauge record in black and the uncalibrated SWMMCanada model in blue, with a same-day simulated response under every rainfall pulse — Graham Creek, Ottawa (WSC 02KF015), June–September 2024. Black is the gauge; blue is the uncalibrated model. No parameter was adjusted toward the observations.

See the results & validation

Two modes, picked automatically

Real networks in 35 cities, synthesized anywhere else

Real network

35 cities

Uses the city's published storm pipes — real inverts, diameters, manholes, and outfalls — in 34 Canadian municipalities from Victoria to Moncton, plus Reykjavík, Iceland, the first international city. Where a city publishes parcels, subcatchments follow real lot lines; where it publishes a sanitary sewer, the model carries it as a second tagged system with its own dry-weather flow.

Synthesize

Anywhere else in Canada

Builds a realistic network from the street map and open data — DEM-delineated subcatchments where the terrain earns it, and pipes sized by the rational method with real ECCC IDF design-storm intensities.

The ecosystem

The first half of a closed modeling loop

SWMMCanada turns an area into a complete, runnable model. The second half, Agentic SWMM, takes that model the rest of the way — it runs EPA SWMM, calibrates against observations, post-processes, and quantifies uncertainty.

Downstream · runs your model

Agentic SWMM

Hand a SWMMCanada package to Agentic SWMM and the loop closes: open data → model → calibrated results with uncertainty. It does the downstream work through plain natural-language chat, and every step stays auditable and transparent.