How reliable is automatically generated building data? To answer this question, we compared the nPro Building Atlas against an independent official reference dataset in seven study areas – ranging from dense inner-city blocks to scattered rural settlements.
Reference dataset and approach
The reference is the heat demand model for North Rhine-Westphalia published by LANUV (the state agency for nature, environment and consumer protection). It was developed for municipal heat planning, is resolved at building level and builds on official cadastral, building and statistical data – making it one of the most detailed publicly available datasets on building heat demand in Germany.
The building data of the nPro Building Atlas was derived without using the LANUV data, so reference and test dataset are independent of each other. The comparison covers the total heat demand of each area (space heating and domestic hot water), the net floor area as the key input to the demand calculation, and the distribution of construction periods, building heights and specific heat demand.
Results
For district heating network planning, the decisive question is whether the total heat demand of an area is captured correctly, since it drives network sizing, plant capacity and economic viability. The study areas were deliberately chosen to cover very different building structures.
| Study area | Area type | nPro Building Atlas | LANUV reference | Deviation |
|---|---|---|---|---|
| Cologne, city centre | urban | 4.77 GWh | 4.66 GWh | +2.5 % |
| Dortmund, city centre | urban | 71.67 GWh | 71.91 GWh | −0.3 % |
| Cologne-Rodenkirchen | suburban | 21.61 GWh | 24.81 GWh | −12.9 % |
| Bonn, outskirts | suburban | 22.13 GWh | 21.94 GWh | +0.8 % |
| Düsseldorf, outskirts | suburban | 30.59 GWh | 30.13 GWh | +1.5 % |
| Rheinbach | rural | 23.69 GWh | 22.34 GWh | +6.1 % |
| Thier (Wipperfürth) | rural | 6.02 GWh | 5.34 GWh | +12.7 % |
| All areas | – | 180.48 GWh | 181.13 GWh | −0.4 % |
The mean absolute deviation of the total heat demand is 5.3 %, and in five of the seven areas it stays below 7 %. Across all areas combined, the difference to the reference is 0.4 % – a figure that benefits from over- and underestimations partly cancelling each other out, which is why the deviation per area is the more meaningful measure.
Where the deviations come from
The differences are not random but trace back to a small number of systematic causes:
- Definition of net floor area: differing assumptions about heated areas, attics and basements shift the total floor area systematically.
- Number of storeys: deriving storeys from building height produces jumps for unusual storey heights and for buildings with pitched roofs.
- Construction period: the year of construction determines the assumed energy standard, so different class assignments act directly on the specific space heating demand.
- Building type and use: non-residential and mixed-use buildings are harder to classify than purely residential ones.
- Geometry and building delineation: outbuildings, garages and merged structures are delineated differently across sources, which strongly affects individual building pairs.
- Data vintage: new construction, demolition and retrofits are recorded at different points in time in the underlying sources.
Limitations
- The comparison is made against a model, not against metered consumption. Both datasets may deviate in the same direction.
- Only annual demands are compared. Peak loads, simultaneity and load profiles are modelled separately in nPro and are not part of this analysis.
- The reference dataset is available for North Rhine-Westphalia only. Results cannot be transferred directly to other regions with a different data situation.
- Seven study areas are not a representative sample. They cover a broad range of settlement structures but do not support a statistically robust statement about all area types.
What this means for your project
- Reliable area-level balances without your own data research: heat demands at district and area level agree well with the official reference model.
- Fast pre-screening of heat network areas: feasibility, heat density and connection potential can be assessed before effort goes into a detailed survey.
- Consistent data across area types: urban, suburban and rural areas are modelled with the same methodology and therefore remain comparable.
- Full control over the values: all building data can be reviewed in nPro and adjusted or replaced with measured consumption before the calculation.