As the global leader in catastrophe risk modeling, we are excited to announce the latest update to KatRisk’s Canadian Flood Model. This update improves the model in two key ways: a higher-resolution digital terrain model covering the entirety of Canada, and enhanced terrain correction powered by machine learning.
Flooding is the most common and costly natural disaster in Canada. Annual average insured losses from floods have run near $800 million in recent years, over 1.5 million Canadian homes sit in high flood-risk areas, and roughly 80 percent of cities are at least partly on floodplains. 2025 marked the 10th costliest year on record for weather-related insured losses in the country, and insured annual weather losses have nearly tripled over the last decade compared with the 2006–2015 period. The July 2024 Toronto flash flood was a reminder of how quickly that exposure turns into claims.
Water moves across terrain in ways that statistical approximations struggle to capture, and the quality of a flood model depends heavily on the quality of the elevation data it is modeling water across. That is the heart of what changed in the latest updates to the Canadian Flood Model.
“Better flood modeling starts with better ground truth,” said Brandon Katz, EVP of Strategy at KatRisk. “By updating the terrain model for the entire country to 10 meters, we are giving Canadian insurers a far clearer picture of where water actually goes.”
A single, unified terrain model for the entire country
The headline change is the digital terrain model. The updated Canadian Flood Model is built on KatDEM, a single unified 10m x 10m DTM covering the entirety of Canada.
Where it exists, KatRisk uses Canada’s 1m/2m lidar elevation data (HRDEM), which covers approximately 95 percent of all populated areas. In areas where lidar is not present, KatRisk uses an enhanced version of the Copernicus DEM (~30m), built with previously created algorithms alongside peer-reviewed machine learning processes. In Canada, that 30m data is downscaled to 10m. Every source is normalized into one continuous 10m surface across the country, so the model is not stitching together inconsistent resolutions.
Three correction processes do the work of turning raw elevation into hydrologically sound terrain:
- River artifact removal cleans elevation noise introduced along watercourses in the lidar data.
- Forest and building removal strips vegetation and structures out of the Copernicus surface so the model sees bare earth.
- River hydro-conditioning corrects the Copernicus surface so water routes through river channels the way it does in reality.
The before-and-after comparisons across Dawson Creek, Prince Albert, Vegreville, Vancouver, and Winnipeg show the same pattern: river channels resolve more sharply, and floodplain detail that earlier DTMs smeared or missed becomes visible.
Updated physics for mountainous terrain
Terrain is only half the story. The updated model runs an updated physics engine that corrects for issues in mountainous areas, paired with KatRisk’s proprietary DTM correction methodology. Steep terrain has long been one of the hardest settings for flood physics, and the updated steep-area modeling produces more realistic water behavior through valleys and high-relief regions than the prior version.
Updated defenses
Flood defenses have been updated using data from CanVec and regional data sets, and large infrastructure projects are accounted for, including the Red River Floodway and the Springbank Off-stream Reservoir. Defended areas now reflect the protection that is actually in place, which matters for any portfolio concentrated behind major flood infrastructure.
What sits behind the model
Running physics-based calculations at this resolution across an entire country takes significant computational resources. Calculation resources for the update were supplied by Oak Ridge National Laboratory under the National Center for Computational Sciences (NCCS) User Agreement Number IP-26-1672, on the Frontier system.
The model retains the characteristics KatRisk flood users rely on: a full-scale financial model with 50,000 simulated years, loss analytics that run up to 30x faster than the competition, full transparency into the data and assumptions used in calculations, climate variability driven by global sea-surface signals such as El Niño and La Niña, and native climate change scenarios for simulating both current and future conditions. Full technical documentation accompanies the release.
Why it matters for insurers
For insurers and reinsurers writing Canadian risk, terrain resolution is not an abstract technical detail. It determines whether a property on the edge of a floodplain reads as in or out, and whether flood depth at a given address reflects the real shape of the land. A unified 10m surface, cleaned and hydro-conditioned across the whole country, gives underwriting and exposure management teams a more consistent and physically grounded view of flood risk at property, account, and portfolio scale.
To see the latest version of KatRisk Canada Flood Model, request a demo here.
