Roof analysis
Remote roof condition assessment from aerial, satellite and LiDAR data
HOMAI assesses the roof of a Norwegian building without anyone climbing it. A vision model reads up to seven aerial, satellite, street-level and LiDAR images, the age factor comes from the official byggeår in Matrikkelen, and plain arithmetic turns 10 risk factors into a roof risk index you can check line by line.
What is a remote roof condition assessment?
A remote roof condition assessment is a desk-based screening of a roof, made from imagery and public registers instead of a ladder. For each address, HOMAI records the roof type, the number of roof planes, pitch, material, condition, visible defects, the share of flat sections, chimneys, roof windows, vents and visible drainage. It then scores the roof on 10 weighted risk factors and writes a roof condition report with a roof risk index, where a higher number means more risk.
It suits teams that need the same assessment for many buildings, or want to know which roofs deserve a site visit first. It does not replace an on-site inspection or a certified condition report (tilstandsrapport): nobody walks on the roof, opens the attic or measures moisture.
Roof analysis is one module of the HOMAI platform, next to climate hazards and portfolio views. The roof index and the climate score are combined into one overall score for the property.
Which roof risk factors does HOMAI score?
Each factor gets a score from 0 to 100 (higher is riskier) and carries its source, provider, confidence and the basis for the value. The weights below are defaults; admins can tune them for each deployment.
| Factor | Default weight | Primary source |
|---|---|---|
| Age | 0.25 | Kartverket byggeår (official) |
| Material | 0.15 | Vision model |
| Visible defects | 0.15 | Vision model: condition plus a weight per defect |
| Geometry | 0.10 | Vision model roof shape plus Kartverket footprint |
| Penetrations (chimneys, roof windows, vents) | 0.10 | Vision model counts |
| Historical change | 0.10 | Historical orthophotos (model-inferred) |
| Flat sections | 0.08 | Vision model ratio |
| Drainage | 0.05 | Vision model |
| Pitch | 0.05 | Vision model degrees, cross-checked against LiDAR |
| Climate | 0.05 | Climate score from NVE, MET Norway and NIBIO data |
The roof risk index is HOMAI's roof condition score: the weighted sum of the factor scores, plus up to 8 points when confidence is low and a 10-point coverage penalty if fewer than three factors could be scored. A factor that cannot be assessed counts as a neutral 50 with low confidence. The health index is 100 minus the risk index.
The overall property score weights the climate score at 60% and the roof index at 40%, and falls in one of three bands: Low (under 40), Medium (40 to 69) or High (70 and above). The climate part covers flood, landslide, precipitation and storm exposure; the full formula is explained under how the score is calculated.
Which aerial and satellite images are used?
The aerial roof analysis draws on up to seven images of each roof:
- a capture of the map view
- the Kartverket orthophoto from Norge i bilder
- aerial or satellite imagery from Google or Esri
- multi-angle and street-level views
- a LiDAR shaded-relief render from Kartverket's national height model
The images are enhanced on the server before the vision model sees them. If one source has no usable image, the next one in the chain is tried: the user's own map capture, then Google, then Esri, then Kartverket's Norge i bilder.
The vision model is a multimodal AI model that runs at temperature 0 and must answer in a fixed format. It counts features quadrant by quadrant and is instructed to prefer a slight under-count to inventing a chimney or vent that is not there. It is not allowed to estimate age, footprint or area. That makes this an AI roof inspection with a narrow job: the model describes, the formula decides.
Like any satellite roof inspection or aerial survey, the assessment is only as current as its imagery. Each image is as recent as the provider's latest flight or capture, and the imagery gallery in the report shows which images the assessment used.
How does LiDAR improve the assessment?
LiDAR is airborne laser scanning. Kartverket's national height model, published through Høydedata, contains a terrain model of the bare ground and a surface model that includes buildings and vegetation (see terrain and surface models).
HOMAI uses the height model in three ways:
- Building height. Surface elevation minus ground elevation at the building.
- Terrain. Ground and surface elevation, slope and aspect for the property.
- Roof shape. A shaded-relief render of the height model is one of the images the vision model reads, and the pitch it reports in degrees is cross-checked against the LiDAR data.
A LiDAR roof inspection on its own does not show material or defects, because a height model records shape, not colour or surface. HOMAI therefore reads it together with the photographs.
Building age (byggeår) as a roof-age proxy
Age carries the highest default weight of the 10 factors (0.25). HOMAI takes it from the building's byggeår in Matrikkelen, Norway's official cadastre, published by Kartverket. The value is official, has a confidence of 0.95 and is never overridden by the AI.
Byggeår is the year the building was constructed, not the year the current roof was laid. A house from 1965 that got a new roof in 2012 still shows 1965 in the register. For an original roof the two are the same; for a renewed roof the factor overstates the roof's age.
A separate factor, historical change (default weight 0.10), is scored from historical orthophotos in Norge i bilder. It is inferred by the model and describes change over time; it is not used to date the roof. HOMAI never estimates age from images.
Read the age factor as a starting point, and take your own records into account if you know a roof has been replaced.
What happens when imagery is poor or missing?
- No usable imagery. The report becomes a climate and registry report with a clear "Roof not yet assessed" banner. No roof values are invented to fill the gap.
- Weak imagery. Counts the model is unsure of are shown with a tilde, for example ~3 roof windows, and carry low confidence.
- Values that cannot be derived. The field shows "Not derivable" with low confidence. Estimates are prefixed with ~ and state the method used.
- Unscored factors. They count as a neutral 50 with low confidence, and the confidence drag and coverage penalty push the index up, not down.
Gaps are shown in the report rather than hidden in the score.
What does the roof section of the report contain?
The roof section is part of the full HOMAI report, next to registry data, terrain, planning zones and climate hazards. It includes:
- roof facts: type, planes, pitch, material, condition, defects, flat-section share, penetrations and drainage
- the factor table with score, weight, source, provider, confidence and basis for each factor
- the roof risk index and health index, and the overall property score with its band
- an imagery gallery with the images behind the assessment
- a methodology section that shows the formula with the property's own numbers
Around it, the report adds an executive summary, an insurance classification with recommendations, a maintenance plan and data-reliability rows. These come from a synthesis step that uses official values as they are and marks anything it cannot derive. The risk summary and the score explanation are written by a text model only from the computed fact sheet: it may not invent any number, hazard, date or measurement, and if it fails, the deterministic text is kept. The report exports as an A4 PDF with the sections admins choose.
How does it compare with an on-site or drone inspection?
| Aspect | HOMAI remote assessment | On-site or drone inspection |
|---|---|---|
| Site visit | None; the analysis runs from the desk | Someone travels to the building |
| What it sees | The roof surface from above and from street level, plus registry and terrain data | Close-up detail; on site, also the attic, underlay and moisture |
| Who sets the result | A fixed formula over 10 weighted factors | The inspector's professional judgement |
| Scale | The same method for one address or a whole portfolio | One building per visit |
| Output | Roof risk index, factor table with sources, report and PDF | Inspection notes, photos or a condition report |
The two work best together. Use the remote roof inspection to screen many buildings the same way, then send an inspector to the roofs where the factors point to a problem. In Norway, a condition grade (TG0 to TG3) is set by a professional on site, not by HOMAI.
Who uses roof analysis?
For insurers and underwriters
Roof facts, an insurance classification and recommendations for each address, with the source of every value.
For banks and lenders
The roof condition of collateral properties, next to their climate exposure.
For real-estate investors
A roof check before an acquisition, with the same method for every building.
For property managers and housing associations
Roof data and maintenance plan fields for deciding which buildings to inspect first.
See the roof analysis in a report
Book a demo and we will walk you through a HOMAI report: the roof facts, the factor table and the methodology section.
Frequently asked questions
- Can HOMAI tell how old a roof is?
Not directly. The age factor uses the building's byggeår from Matrikkelen, which is the year the building was constructed, not the year the current roof was laid. A separate historical-change factor is inferred from historical orthophotos, but it does not date the roof, and HOMAI never estimates age from images. Treat the age factor as a proxy, especially for roofs you know were replaced.
- Does a remote roof assessment replace a condition report?
No. HOMAI is remote screening. It does not replace a physical inspection or a certified condition report (tilstandsrapport), because nobody checks the attic, underlay or moisture on site. What it gives you is the same first assessment for many roofs, with the source of each value, so you can decide where an inspector should go first.
- How accurate is HOMAI's roof condition rating?
HOMAI does not publish an accuracy percentage. Each value carries a provenance label, a confidence from 0 to 1 and the basis it rests on, and low confidence is added to the risk index instead of being hidden. Uncertain counts are marked with a tilde, and missing roof data shows as not assessed. Scores are repeatable: the same imagery and weights give the same result.
- Which buildings can be assessed?
Any building registered in Matrikkelen at a Norwegian address can be looked up. How much of the roof can be read depends on the imagery. The model classifies gable, hip, flat, mansard, shed and complex roofs and marks counts it is unsure of. If no usable image exists, the report covers climate and registry data and states that the roof is not yet assessed.
- Does HOMAI assess snow load on roofs?
No. HOMAI does not calculate snow load. Its climate score covers four exposures: flood, landslide, precipitation and storm. On the roof itself, it scores geometry, pitch and the share of flat sections, which are three of the 10 risk factors. If you need a snow-load calculation for a building, ask a structural engineer.
- How long does a roof assessment take?
A full analysis, roof included, can take minutes. HOMAI collects registry and terrain data, fetches and enhances up to seven images, runs the vision model and then computes the score. Vision and narrative results that are already cached come back faster. There is no site visit to schedule.