TL;DR — In Campania, elevation alone predicts summer land surface temperature remarkably well. Coastal Naples–Salerno forms a significant hot cluster; Avellino–Benevento interior forms a cool cluster. Accounting for spatial autocorrelation (SAR/SEM/SDM/GWR) doesn't erase the elevation effect — it sharpens it.
1. The question
Urban Heat Island literature often emphasises land cover and density. For a heterogeneous Mediterranean region like Campania — sea to 1,800 m Apennines in <60 km — does elevation dominate the summer thermal pattern at comune scale? And is the heat spatially clustered or random?
2. Data & method (reproducible)
Summer 2023 mean LST from MODIS (1 km), aggregated to 36 comuni representative of coastal, plain and interior Campania. Covariates: elevation (DEM), NDVI, impervious surface share. Models tested: OLS → SAR → SEM → SDM → GWR. Spatial weights: queen contiguity, row-standardised. Diagnostics: Moran's I, LM tests, AIC, cross-validated GWR bandwidth.
python docs/campania_spatial_lite.py --input docs/campania_spatial_dataset_SAMPLE.xlsx --weights queen --models ols,sar,sem,sdm,gwr --out results/
Full pipeline, sample data and thesis PDF are downloadable on the homepage — docs/campania_spatial_lite.py + campania_spatial_dataset_SAMPLE.xlsx.
3. What the numbers say
| Model | Elevation coeff. | R² / pseudo-R² | AIC | Note |
|---|---|---|---|---|
| OLS | −0.0041 °C/m ** | 0.75 | — | Baseline, residuals spatially autocorrelated |
| SAR (lag) | −0.0036 ** | 0.78 | lower | ρ significant |
| SEM (error) | −0.0039 ** | 0.79 | lowest | λ significant, best fit |
| GWR (median) | −0.0038 | 0.81 (local) | — | Stronger on coast, milder inland |
** p < 0.01 | Queen weights, n=36. See thesis Ch. 4 for full tables.
Reading: Every 100 m of elevation ≈ 0.41 °C cooler summer surface temperature. The effect survives spatial controls — spatial error model actually fits best, meaning unobserved spatially-correlated factors matter but don't confound elevation away.
4. Clustering: not random
Moran's I = 0.28 (p = 0.004) — significant positive spatial autocorrelation. LISA maps show:
- Hot spot: Naples–Salerno coastal corridor (High-High)
- Cool spot: Avellino–Benevento interior, Apennine foothills (Low-Low)
- Plains (Caserta) mixed — local land-cover effects emerge after controlling for elevation.
Implication: municipality-level heat mitigation can't be uniform — coastal densification + sea-breeze blocking vs interior greening need different levers.
5. Limitations & next steps
- n=36 (thesis sample) — regional but not exhaustive of all 550 comuni. Expansion to full Campania is next.
- Single summer snapshot (2023) — multi-year robustness in progress (see Climate Trends 1985–2025 project).
6. Reproduce it
All artefacts from the thesis are on the site:
- Thesis PDF — Ch. 3–4 detail specs & diagnostics
- campania_spatial_lite.py — runnable pipeline (OLS/SAR/SEM/SDM/GWR)
- Sample dataset (.xlsx) — 36 comuni, LST + covariates
pip install pandas numpy pysal spreg mgwr openpyxl python docs/campania_spatial_lite.py --help
Cite: Pentu, M. (2026). Elevation drives summer LST in Campania — Spatial regression on UHI. Daily Research #01, mahi0104.github.io/research/2026-09-20-ndvi-lst-cooling.html
Previous: — | Next: All research → | Follow daily on LinkedIn