Global Cities (311): The Changing Power of Vegetative Cooling (1990–2024)

Summary

  • Question. Globally, how are greenness (NDVI), land-surface temperature (LST), and vegetative cooling changing through time—and what city traits best explain where cooling still works vs. where it’s eroding?

  • What we did. Built a 1990–2024 Landsat time-series for 311 cities worldwide, quantified the LST–NDVI slope (cooling per unit greenness), and modeled spatial vs. temporal drivers (climate, physiography, built form, socio-economics).

  • Outcome. Found a global weakening of vegetative cooling—73% of cities show significant declines—amid broad urban warming and mixed NDVI trends; the biggest marginal cooling gains come from greening low-NDVI areas.

Three scientific charts titled 'Cooling scales nonlinearly with greenness,' showing relationships between NDVI and various cooling metrics. The first chart displays a scatter plot with a trend line, the second shows a line graph with city effects, and the third presents a derivative analysis with a shaded background.

Research Challenge

Cities bank on trees to cool neighborhoods, but long-term evidence at global scale has been thin. Planners need to know whether vegetation’s cooling power is weakening, where, and why—and whether adding canopy still pays off when water is limited or urban form works against cooling. The challenge: harmonize decades of satellites across climates, and disentangle spatial controls (elevation, albedo, built volume) from temporal change (warming, aridity, development).

Approach

Data

  • Landsat 5/7/8/9 Level-2 (seasonal stacks by hemisphere) → NDVI, LST, vegetative cooling (city-level LST–NDVI slope).

  • Climate & water balance: TerraClimate (P, PET, CWD, VPD), ERA5 (2-m T, LAI).

  • Physiography: SRTM elevation; albedo proxies; solar radiation.

  • Land cover & form: GLC_FCS30D (impervious/forest/sparse), GHSL Built-Up Volume (3-D intensity).

  • Demography & economy: GPW population/density; World Bank GDP/capita.

  • Context: Urban heat island intensity (Yale dataset); biomes and Köppen classes.

Methods

  • Harmonized multi-sensor reflectance; excluded Landsat-7 SLC-off years for affected scenes.

  • Computed city-median NDVI, LST, and cooling slope per year; estimated long-term trends (1990–2024).

  • Driver modeling: LASSO → stepwise OLS for spatial means and temporal trends; partial-dependence to interpret effects; nonlinearity checks (e.g., NDVI–cooling).

Tools

  • Google Earth Engine (data ingest, processing), MATLAB (modeling/plots).

Bar graphs showing trends of NDVI, LST, and vegetative cooling across different biomes and Köppen climate types, with error bars and statistical significance marked.

Key Findings

  • Cooling is eroding: 73% of cities show a significant loss of vegetative cooling through time; ~72% warmed in LST; NDVI trends are mixed.

  • Where greening pays most: While greener cities have stronger overall cooling, the largest marginal cooling gains occur when greening low-NDVI places (diminishing returns at high NDVI).

  • Top spatial drivers: Elevation, built volume, albedo dominate city-to-city differences (NDVI ↓ with built volume; LST ↓ with elevation; higher albedo ↔ weaker vegetation-driven cooling).

  • Top temporal drivers: Changes in UHI intensity, baseline cooling strength, and VPD explain much of the trend in cooling; NDVI and LST trends co-move (cities losing cooling tend to warm faster).

  • Dry-city exceptions: A small set of arid cities show anomalous positive LST–NDVI slopes (sparser, low-albedo vegetation → local warming signatures).

Three world maps showing different climate data distributions. The first map (a) shows NDVI median values with green circles indicating various levels of vegetation health. The second map (b) illustrates LST median values with orange circles indicating temperature deviations. The third map (c) depicts vegetation cooling data with blue circles representing cooling and black dots indicating points of positive cooling.

Outputs

  • Global workbook: city-level maps/tables for NDVI, LST, cooling (means & trends).

  • Diagnostics: partial-dependence plots for key drivers; climate-zone summaries.

  • Data/code: CSV/GeoTIFF exports and reproducible scripts.

  • Status: Manuscript in preparation; executive summary/preprint available on request.

Impact / How it’s used

  • Target low-NDVI corridors for the highest per-unit cooling gains; monitor equity impacts.

  • In arid/seasonally dry cities, pair canopy goals with water management (timing, species, irrigation to maintain ET).

  • Where water is limiting or built form dominates, combine greening with reflective/ shade design (cool roofs, shade structures) to lower radiant load without undermining long-term canopy plans.