← Back to Research Sudhansu S. Rath — Data Notes

Applied dataset note · Urban heat

Every roof in
Phoenix has an
albedo value.

Google Research's Cool Roofs / Heat Resilience release maps building-level rooftop reflectivity for 50+ cities worldwide — including both Phoenix and Tempe. This page walks through the dataset and a proposed workflow for pairing it with a local heat-vulnerability index to prioritize cool-roof retrofits.

Building-level, 30 cm resolution Sentinel-2 × Pléiades Neo fusion RMSE 0.04 vs. airborne hyperspectral Phoenix + Tempe covered
Albedo scale — fraction of solar energy reflected 0.0 (absorbs all) → 1.0 (reflects all)
0.05–0.15Dark roofs / fresh asphalt
0.15–0.25Grass & vegetation
0.20–0.30Aged concrete
0.60–0.85White "cool" roofs

Typical surface values reported by Google Research. Every building centroid in the dataset carries a value along this scale — the lower it sits, the more a roof is contributing to local warming.

About the dataset

Beyond neighborhood averages

Free satellite-derived albedo estimates exist at roughly 10 m resolution — too coarse to tell one rooftop from the next. Google Research's approach fuses that 10 m Sentinel-2 data with 30 cm commercial imagery (Airbus Pléiades Neo) using machine learning and radiometric calibration, reconstructing a spectral reflectance profile for every urban pixel and collapsing it to a single albedo value per building centroid.

The method was validated against airborne hyperspectral measurements over Boulder, Colorado, reaching an RMSE of 0.04 against ground truth. Released June 2026 alongside a paper in Nature Communications, the expanded dataset now spans 50+ cities across 15 countries — up from a 14-city pilot in 2024 — built with the World Resources Institute.

For the Phoenix metro specifically, both Phoenix and Tempe are included as separate city extracts within the United States download, which is useful given how much the two differ in roofing stock, lot size, and canopy cover.

50+cities, 15 countries, released June 2026
30 cmeffective spatial resolution
0.04RMSE against airborne hyperspectral ground truth
2Valley cities covered: Phoenix & Tempe

Live data

Explore the Valley in the Earth Engine app

The official Heat Resilience Earth Engine App is embedded below. Pan to Phoenix or Tempe, zoom in from tract-level aggregates down to individual building centroids, and click a rooftop to read its albedo value directly.

eie-cool-roofs.projects.earthengine.app Open full screen ↗

If the panel above stays blank, Earth Engine is likely blocking embedded framing on this domain — use the "Open full screen" link instead.

Suggested application

From rooftop pixels to a cool-roof retrofit shortlist

Low-albedo rooftops are joined against Phoenix's CAP LTER Heat Vulnerability Index (358 census tracts, 2016) to rank cool-roof retrofit priority — but that index is a 2016 snapshot, so any building built afterward was never part of its accounting. Each dark roof is first matched to its Maricopa County Assessor parcel and checked against its year_built: parcels built in or before 2016 feed the tract ranking below; parcels built after 2016 are shown separately as flagged buildings — not because they matter less, but because a black roof going up today is arguably the more actionable case: new construction a utility can catch and engage before occupancy, rather than retrofit years later.

…dark roofs (albedo < 0.15) found in Phoenix
…Phoenix tracts ranked & mapped
…Phoenix buildings built after 2016 — flagged, unscored
…Tempe rooftops ranked (no HVI layer yet)
Phoenix tract (size = priority) Flagged (built after 2016, no HVI score) Tempe rooftop (darkest first)

Loading phoenix.json and tempe.json…

Data & citation

Where this comes from

Dataset

Cool Roofs — US extract (incl. Phoenix, Tempe)
Download .zip ↗

Interactive

Heat Resilience Earth Engine App
Open the app ↗

Methods paper

Fork et al., Estimating high-resolution albedo for urban applications, Nature Communications, 2026
Read the paper ↗

Vulnerability layer

Social & Heat Vulnerability Indices, Phoenix AZ, CAP LTER
View data package ↗

Citation note: Google's Broad Coverage centroids should be attributed to Google Research. If a workflow instead uses the "Open Outlines" building-footprint version, the dataset's own FAQ lists several additional required attributions (Overture, Esri Community Maps, Microsoft Global ML Building Footprints, Google Open Buildings, USGS 3DEP) — check the source's licensing terms before publishing derived maps.