Chapter 5 of 6
The average hides a real spread
Every other chapter on this site uses one number for the whole city: a single ERA5 grid cell, 247 times too coarse to see a single neighborhood. This chapter uses a different, sharper instrument: satellite land-surface temperature at roughly 100-meter resolution, measured separately for each of Atlanta's 247 official neighborhoods. The spread between the coolest and hottest of them is real, and it tracks tree cover far more tightly than it tracks income.
The scatter that explains it
Each dot is a neighborhood: canopy on one axis, land-surface-temperature anomaly on the other. The negative slope is the whole story of tree equity in one frame.
What the spread is actually made of
Canopy cover explains almost all of it. Across all 247 neighborhoods, tree canopy over three meters and land-surface-temperature anomaly move almost in lockstep in opposite directions (r = -0.91); built surface (roads, roofs, pavement) moves just as tightly in the other direction (r = 0.90). The tenth of neighborhoods with the most canopy average 75 percent tree cover and run about 7.6°F cooler than the citywide mean. The tenth with the least average 20 percent canopy and run about 9.2°F warmer.
Poverty rate tracks the same pattern, but loosely, not tightly: the coolest tenth of neighborhoods average a 15 percent poverty rate, the hottest tenth 23 percent, a real gap but a modest one (r = 0.21 against land-surface temperature, versus canopy's 0.91). Some of the hottest neighborhoods in the city, Atlantic Station and the Marietta Street Artery among them, are dense, low-canopy redevelopment with poverty rates below the city average. Low canopy predicts heat far more reliably than low income does; the two overlap often enough to matter, but they are not the same map.
LST anomaly is a 3-warm-season composite (2023-2025, June-August) vs. the city-wide mean, ~100m Landsat resolution, a real intra-city measurement, unlike the ERA5-based chapters elsewhere on this site. Poverty rate is crosswalked from Census tracts to neighborhoods by centroid-in-tract assignment (a simple approximation, not full area-weighted interpolation): each neighborhood gets the poverty rate of whichever single tract contains its centroid. For any neighborhood larger than one tract, or with disjoint parts, this applies one tract's value to the whole area rather than reflecting real internal variation, not just an edge case near a boundary, but the norm for larger or irregularly-shaped neighborhoods. Treat as indicative, not precise. Canopy-change is at NPU (Neighborhood Planning Unit) granularity, coarser than the 247-neighborhood granularity of every other layer: multiple neighborhoods share one NPU value.
LST coverage: 244 of 247 neighborhoods have a valid satellite composite. A handful had no clear-sky scene in the composite window, a real gap, not hidden. See Methods for how the ERA5-based chapters differ from this one.