A rooftop unit (RTU) is a packaged heating, ventilation, and air-conditioning system mounted on a building’s roof.

Knowing where they are and how many exist supports energy-efficiency and electrification planning (e.g. heat-pump retrofits), greenhouse-gas estimation, and building-asset inventories. This work that is otherwise slow and costly to do site by site.

Analysis covers “Part 3” buildings: footprints larger than 600 m² (the larger commercial and institutional roofs where RTUs are expected). Smaller buildings aren’t analysed, so they don’t appear on the map or in search.

Addresses come from the building-footprint dataset’s attributes. The search function only covers the analysed Part 3 buildings, and not every footprint carries a complete address, so a valid street address can return nothing if its building falls outside this set or is unlabelled. Try a nearby larger building, or pan the map and click a footprint directly.

A deep-learning object-detection model draws a bounding box around each likely RTU and reports a confidence score (85–100%). Use the confidence filter under the RTU layer to show only higher-confidence detections.

> 89.4% recall, 89.7% precision (IoU ≥ 0.5)

Validated against 2,415 hand-labelled RTUs on 508 buildings using the PASCAL VOC method: 2,160 correct, 248 false positives, 255 missed (10.3% false-positive rate). The validation layer maps each case as correct true positive (TP), false positive (FP) or missed false negative (FN).

Across all analysed buildings the model detected 2,955 RTUs on 677 of 1,958 buildings.

ImageryGeoNB 2024 orthophotos, 7.5 cm resolution.

Building footprintsOpen Database of Buildings (ODB), Statistics Canada (federal).

Detection modelLicker Geospatial Consulting, deep-learning object detection.

Draft: Moncton model results. Figures and coverage may change in later versions.

build v1.06.18