Greater Manchester street-level quality of place, AI-scored from Google Street View imagery. Greenery · upkeep · cleanliness · activity.
MethodologyThe Neighbourhood Vitality Index (NVI) is an experimental measure of street-level quality of place. Each area is scored 0–10 on four dimensions, judged from Google Street View imagery by an AI vision model:
Greenery — trees, planting, grass and verge quality · Upkeep — building fabric, windows, absence of dereliction · Cleanliness — absence of litter, fly-tipping and graffiti · Activity — open shopfronts, benches, planters, and evidence of community use. (The activity dimension was labelled “vitality” in earlier releases.)
The NVI shown is the simple mean of the four dimensions, averaged across all valid images sampled in the area.
Areas are Lower-layer Super Output Areas (LSOAs, 2021 boundaries) or their constituent output areas. Sample points are placed randomly inside each area's polygon, snapped to the nearest available outdoor Street View panorama (within 50 m). A sample is rejected if the image is not a usable outdoor street-level scene (e.g. indoor panoramas, aerial captures, or blurred imagery).
Areas are coloured by NVI or by the Index of Multiple Deprivation (IMD) 2025 decile for comparison — IMD decile 1 = most deprived, 10 = least. “Images analysed” shows how many valid Street View images informed the score. Areas with few images have less reliable scores; unscored areas are shown grey.
The NVI shows a moderate positive association with the Index of Multiple Deprivation (IMD) 2025: across the 1,701 LSOAs scored in this release, NVI deciles correlate with IMD deciles at Pearson r = 0.50 (Spearman ρ = 0.51). The gradient is near-monotonic — the least vital decile has a mean IMD decile of 2.3, the most vital 7.5. This indicates the visual signal captured by the NVI tracks area deprivation, but the two are not the same measure: NVI reflects street-level quality of place, and areas that depart from the expected pattern (e.g. visually well-kept but deprived) may be of particular interest. The correlation is re-computed as image coverage grows and has strengthened as sparse areas have been sampled.
This is a pilot. Scores are AI-generated and have not been human-validated. A single image captures one heading and moment in time; areas are sampled sparsely (typically 2 points per LSOA in this release). Activity is systematically under-captured because Street View rarely shows people or open businesses. Scores reflect the date the panorama was captured, which may be several years old. The NVI should be treated as indicative, not definitive.
Imagery: Google Street View Static API. Boundaries and deprivation data: ONS Open Geography Portal / Ministry of Housing, Communities & Local Government (IMD 2025). Basemap: OpenStreetMap contributors. Built by The Impact Lab.