Beyond Local Climate Zones

Continuous Morphology Archetypes for Heat Mitigation Across 66 Cities

How does the shape of a city block relate to how hot it gets? This repository holds the code and the open dataset behind a study that measures urban morphology continuously — rather than through categorical Local Climate Zone classes — and relates it to summer surface urban heat island (SUHI) intensity across 66 cities in 15 Köppen climate zones.

Emekci, S., & Emekci, H. (2026). Beyond local climate zones: Continuous morphology archetypes for heat mitigation across 66 cities. Sustainable Cities and Society, 150, 107814. doi.org/10.1016/j.scs.2026.107814

Key findings

Tree-canopy fraction is the only universal cooling lever. Its coefficient is negative in all 14 modelled climate zones and significant in 13 (standardised β = −0.64, p < 10⁻¹²²). Every other lever changes rank, or even sign, between climate regimes.

Lever	Standardised β	Direction
Tree-canopy fraction	−0.64	cooling
Water fraction	−0.58	cooling
80th-percentile NDVI	−0.31	cooling
Building-height CV	−0.18	cooling
Mean building height	−0.09	cooling
Building density	+0.06	warming
Built-up fraction	+0.31	warming
Bare-soil fraction	+0.42	warming

Morphology explains a limited but consistent share of within-city SUHI variance (marginal R² = 0.25, conditional R² = 0.39); background climate dominates, as the literature would predict. Per-zone models raise the pooled marginal R² to 0.38, reaching 0.55 in Dwa. From the 144 Pareto-efficient patches, three designer-facing archetypes emerge:

	n	UHI anomaly	Density	Tree canopy	Built-up	Mean height
A1 dense paved	81	−0.26 °C	0.54	5 %	90 %	16 m
A2 open vegetated	52	−2.95 °C	0.26	24 %	61 %	11 m
A3 tall sparse	11	−4.41 °C	0.25	8 %	62 %	27 m

The dataset

data/published/ holds the harmonised patch-level table — the component most likely to be useful independently of this study's own analysis. Each row is a 1 km² built-up patch carrying a summer surface-UHI anomaly and eight continuous morphology descriptors, comparable across all 66 cities.

File	Rows
uhi_morphology_patches_v1.parquet	33,715	full harmonised sample
uhi_morphology_patches_screened_v1.parquet	33,498	modelling sample, after the quality screen
diagnostics.json · diagnostics_perzone.json	—	model coefficients and fit statistics
pareto_summary.json · archetype_cards.json	—	Pareto fronts and the three archetypes

Every column is documented in data/published/README.md, including known missing values. Released under CC BY 4.0.

import pandas as pd
df = pd.read_parquet("data/published/uhi_morphology_patches_screened_v1.parquet")
df.shape   # (33498, 28)

Reproducing the results

git clone https://github.com/hemekci/UHI.git && cd UHI
uv venv && uv pip install -e .

mkdir -p data/features
cp data/published/uhi_morphology_patches_v1.parquet data/features/full.parquet

python run/pipeline/run_diagnostics.py          cities=full   # global mixed-effects model
python run/pipeline/run_perzone_diagnostics.py  cities=full   # per-zone models
python run/pipeline/run_pareto.py               cities=full   # Pareto fronts
python run/pipeline/run_typology.py             cities=full   # archetype clustering
python run/pipeline/make_figures.py                           # regenerate every figure

This reproduces the published coefficients exactly. Re-running the ingest stage additionally needs a Google Earth Engine account, since the Landsat, WorldCover and ERA5-Land layers are pulled from GEE; everything downstream runs from the published tables.

Repository layout

src/              analysis package — ingest, features, morphology, model,
                  pareto, typology, metrics
run/conf/         Hydra configuration
run/pipeline/     stage entry points, incl. make_figures.py
data/published/   harmonised tables and model outputs (CC BY 4.0)
docs/figures/     all 15 published figures as PNG
tests/            18 unit and smoke tests

How to cite

Please cite the article. It is the citable record for this work — the code and data are released as its supplement rather than as separately citable outputs.

@article{emekci2026morphology,
  author  = {Emekci, Seyda and Emekci, Hakan},
  title   = {Beyond local climate zones: Continuous morphology archetypes for heat mitigation across 66 cities},
  journal = {Sustainable Cities and Society},
  year    = {2026},
  volume  = {150},
  pages   = {107814},
  doi     = {10.1016/j.scs.2026.107814}
}

GitHub's Cite this repository button returns the same reference from CITATION.cff. Machine-readable metadata is in codemeta.json and .zenodo.json, and a condensed summary for automated readers is in llms.txt.

This release is archived at Zenodo under 10.5281/zenodo.19897974 for permanence and version pinning; quote that DOI if you need to identify the exact version you ran.

Scope and limitations

The dependent variable is surface UHI from Landsat land-surface temperature. It is not a measure of canopy-layer air temperature or of human thermal comfort, and the design implications are framed accordingly. Building-height coverage is uneven and absent for about half the patches, sparsest in the Global South; a height-exclusion sensitivity analysis in the article shows the conclusions do not rest on those two features. Coefficients are specific to the 1 km aggregation scale. The analysis is observational, so coefficients quantify association rather than identified causal effects.

Data sources

All inputs are public and open: Landsat 8/9 Collection 2 Level 2 (USGS), Microsoft Global ML Building Footprints, Google Open Buildings v3, the VIDA Google–Microsoft combined distribution, GlobalBuildingAtlas, GLAMOUR, ESA WorldCover, ERA5-Land (ECMWF Copernicus), WUDAPT LCZ and SRTM. Provenance and licences for each layer are in data/README.md.

Licence and contact

Code under the MIT licence; data and paper text under CC BY 4.0.

Seyda Emekci (corresponding) — Department of Architecture, Ankara Yıldırım Beyazıt University, Ankara, Türkiye — semekci@aybu.edu.tr Hakan Emekci — Institute of Informatics, Hacettepe University, Ankara, Türkiye