The construction AI Pack is aimed at capturing information about the built environment, and how it changes over time. When initially launched, it included a single "Construction Site" AI Layer. Since
gen4-building_storm-2.0, this has been expanded to include three additional layers to provide further granularity around the nature and stage of the construction. Combined with the real-time capabilities of the AI Feature API , and the strategy of processing AI on every survey we capture, this makes the construction AI Pack a powerful means of identifying new building work and changes often before they are even completed.
Gen 4 - Vector outlines of construction vehicles on a work site.
Construction includes a group of four AI Layers:
Characteristics and Recommended Use
Unlike many of our AI Layers, Construction Site does not have a clear and unambiguous boundary, and can indicate anything from bare earth with no active construction, to active earthworks, foundations, a building going up, or even minor renovations or an in-progress roof replacement.
Building Under Construction is more specific, and constrained to the building itself once the walls start to come up from the foundations. While there is no formal notification about the phase of the construction, several detailed stages can be inferred from the data.
- If the parcel is vacant (no building):
- Has it been vacant for long (covered by vegetation)
- Has it been prepared (with dirt/sand/gravel and Construction Site)
- Has construction underway (construction vehicles present)
- If the construction site is in a preparatory state (Construction Site, but not yet a Building Under Construction)
- If it is a major construction project (large area and presence of Construction Cranes)
- If the building is complete, but general works such as landscaping are ongoing (no Building Under Construction, but Building Present and other construction types).
Note that from Gen 2 data onwards, Construction Site recognition obtained a substantial boost in performance.
When we state there is a construction site on a parcel, the incorrect cases are made up almost entirely of "construction like" areas, such as landscaping work, rocky beaches with waterfront properties, or other areas that are visually similar to construction. The second highest cause of error is incorrect parcel boundaries, where the construction site overlaps a neighbouring property. Of all the construction sites we miss, the majority are either bare earth (where there is little evidence of actual construction) or very minor construction (such as repairs on a small part of a roof). Effectively, we miss only a small portion of parcels where clear, active construction is ongoing.
For a precise changelog, refer to the AI Generations pages. Here is a summary to describe changes to this AI Pack over time:
"Construction Site" layer only.
Major improvements in AI Layer raster quality for Construction Site.
Ability to retrieve Construction Site polygons from AI Feature API.
- Improvements in AI Layer raster quality for Construction Site.
- Addition of 3x new construction layers. Full set of 4x layers: