Physical climate risk data for banks and lenders

Start with the collateral. A mortgage book is, in practice, a list of addresses, and physical climate risk sits in each building: whether it lies in a flood zone, on ground mapped as landslide-prone, or where heavy rain and strong gusts are common. Hazard maps describe areas. A credit or risk team needs the answer for the building that secures the loan.

HOMAI checks each address against national hazard maps and weather statistics and returns hazard data for each collateral property: four exposures banded 1–5 and combined into a climate score from 0 to 10. Run the same analysis for every address and you can describe the climate risk of the whole mortgage book, retail or commercial, in the same terms.

HOMAI covers physical risk only. Transition risk, the effect of policy, technology and market changes, is outside its scope.

The roof risk index adds the roof condition of the collateral, read remotely from up to seven images, including a LiDAR shaded-relief image from Kartverket's national height model. By default the climate score counts for 60% of the overall property score and the roof index for 40%. Scores fall into three bands: Low (under 40), Medium (40–69) and High (70 and above).

Every derived value carries a provenance label (official, calculated, AI-estimated, AI-inferred or estimated), the provider, a confidence from 0 to 1 and a basis. Official values such as byggeår are used as recorded and never re-estimated. If something cannot be derived, the report says "Not derivable" instead of guessing.

Supervisors set the expectations for how banks handle climate risk; in Norway, that is Finanstilsynet. HOMAI does not interpret regulation or produce supervisory reports. What it can give credit, risk and sustainability teams is building-level evidence that someone else can check:

• which public dataset each hazard value comes from;
• which method turned the inputs into a score, with the formula and the numbers used;
• how confident each value is, and which values are missing;
• the same result when the same property is analysed again with the same imagery and weights.

The sources are Norwegian public bodies: NVE for flood and landslide, MET Norway's Frost API for precipitation and wind, and Kartverket for property data.

Import the collateral addresses as a CSV file, or add properties by address search, map click or a drawn area. Each analysis runs as a background job and can take minutes.

The portfolio view then shows risk bands across the loan book: how many properties fall into Low, Medium and High, which ones score highest, and the analysis history of each. Administrators control users, roles, categories and which users see which data.

The figures summarise the latest analysis of each property. HOMAI does not monitor the book continuously or send alerts; you re-run an address when you need a fresh view.

HOMAI's score is calculated by explicit code, not by a trained model, and CUATRO trains no machine-learning models of its own. A multimodal AI model is used only to describe what is visible on the roof and to write prose. It is not allowed to estimate age, footprint or area, and the narrative may not introduce any number, hazard or date that is not in the computed facts.

• Each report includes a methodology section with the formula and the actual values substituted.
• Default weights and constants are documented. Administrators can tune them per deployment, so record the settings you use.
• Same property, same imagery, same weights: same score.
• Byggeår from Matrikkelen is authoritative and never overridden by the AI.

That makes a HOMAI result one documented input to a climate risk assessment for banks, next to your own credit data. The full method, data sources and limits are described in one place: see the auditable scoring method and the data sources and methodology.

• It does not estimate probability of default (PD), loss given default (LGD) or the effect of climate risk on collateral value. There is no monetary or valuation model.
• It does not cover transition risk.
• It does not run forward-looking climate scenarios. The method uses current NVE hazard maps and historical MET Norway statistics, so on its own it is not a climate stress test.
• It does not score snow load. The four exposures are flood, landslide, precipitation and storm.
• It does not monitor properties or send alerts.
• It does not replace a valuation, a site inspection or a tilstandsrapport.
• It covers Norwegian addresses only and publishes no accuracy percentages; each value carries a confidence instead.

Financing and ownership are two views of the same buildings. For acquisition and asset questions, see HOMAI for real-estate investors, or see all industries.