European Urban Heat Island Trends Since 2000: No Open CSV/Parquet Dataset Covers the Full Period, Partial Pieces Exist

Interpretation: I read "urban heat island intensity" as either surface UHI (satellite land surface temperature, urban minus rural) or canopy-layer UHI (2 m air temperature, modelled or observed), and "changed since 2000" as needing a multi-year series, not a single snapshot. Search coverage was limited (details under Coverage), so "none found" below means none found in this run, not proven absent.

Summary

I found no open-licence resource, in CSV or Parquet, that gives a European city-level UHI intensity series from 2000 to today. The datasets that are open and in tabular form cover one summer or three years. The one longer record I found (2003 to 2020) is global and, per its paper, available on Google Earth Engine (so not CSV or Parquet unless you export it). Code exists, but the repositories I checked either lack a licence file or cover one city and one summer.

Item	Resource	Type	Link	What it covers	Format/size	Access steps	License	Maintenance/activity	Quality caveats	Fit
UHI intensity, European cities, trend	UrbClim "Climate variables for cities in Europe from 2008 to 2017" (C3S)	Dataset	CDS record	100 European cities, hourly 2 m air temperature at 100 m, plus rural-urban masks, January 2008 to December 2017	NetCDF-4, size not found	CDS page shows login/register; whether download needs an account: unchecked	CC-BY	Update date shown 2026-05-12; title marked "Deprecated 10-05-2024"; no updates expected	Not CSV/Parquet; starts 2008 and ends 2017; model output, not observations	Medium
UHI intensity, European cities, trend	Huang and Manoli UHI mortality data (Zenodo 7986841)	Dataset	Zenodo API record	85 European cities, 2015 to 2017, built on UrbClim	CSV tables of about 10 KB each; NetCDF time series; zips up to about 1.7 GB	Direct download	CC-BY-4.0	Record modified 2024-05-30	Only three years; CSVs are mortality and cost tables, urban-rural temperature difference is in NetCDF	Low to medium
UHI intensity, European cities, trend	Schwaab, urban trees and LST (Zenodo 5526674)	Dataset	Zenodo API record	City-level LST differences (urban fabric vs trees, forests, pastures), hot-day and June to August mean	CSV, 95,904 bytes	Direct download	CC-BY-4.0	Version 1.0.0, 2021, modified 2021-10-08	Period covered and city count not found in record or readme; a cross-section, not a trend	Low

Datasets in detail

Best pick for data: UrbClim (C3S). It is the only source I verified that gives hourly, 100 m, city-scale air temperature with an explicit rural-urban mask for 100 European cities, so UHI intensity can be computed directly. It covers January 2008 to December 2017 and is under a CC-BY licence. It fails two of your constraints: the format is NetCDF-4, and it stops in 2017. The EEA catalogue also publishes a raster of average summer UHI intensity from the same simulations (that record's licence was not read). Because both are one model decade, they can show variability within 2008 to 2017 but cannot show change since 2000.

Best CSV option: Zenodo 7986841. The record is CC-BY-4.0, derived from UrbClim for 85 European cities over 2015 to 2017. The CSV files (for example uhi_mortality_table_ci_cityage.csv, 9,343 bytes) hold mortality and cost estimates tied to UHI, not a temperature series. The urban minus rural series is in data_urbanruralavg_timeseries.nc (112,815,341 bytes). You would convert that NetCDF to CSV or Parquet yourself.

Cross-section only: Zenodo 5526674. This is a small CSV with city coordinates, region and LST differences. It answers a different question (cooling effect of trees), so I rate it low.

Only multi-year record from 2003: Mentaschi et al. (global, on Google Earth Engine). The paper builds a global daytime SUHI dataset and reports 2003 to 2020 results, including that SUHI extremes averaged worldwide are 1.04 K or 31% higher than in 2003. That is a global figure, not a European one. The analysis ran on Google Earth Engine, and The paper's conclusions state that the dataset is available on the Google Earth Engine. I did not locate the asset itself, its licence, or its format, so I could not verify it at origin. It is the strongest lead for a 2003 onward European trend: clip it to European cities and export to CSV or Parquet.

Code

Resource	Link	What it does	License	Activity	Caveats
hkatty/Paper_UHI_mortality_Europe	GitHub	Core code for Huang et al. 2023, Nature Communications 14, 7438: UrbClim preprocessing, attribution in R, urban-rural time series, figures	None stated (no LICENSE file in the root listing)	4 commits, 0 open issues; last commit date not found	Needs UrbClim and other inputs downloaded separately; outputs also on Zenodo
urbes-team/2023_UHI_mortality_Europe	GitHub	Same README and file list as above	None stated (no LICENSE file in the root listing)	4 commits, 0 open issues, 1 star	Appears to duplicate the first repository; relationship not verified
TALEA-platform/suhi	GitHub	Earth Engine notebook computing LST, NDVI, albedo and composite indices for Bologna from Landsat 8/9 and MODIS, summer 2024	MIT	63 commits, 0 open issues; last commit date not found	One city, one summer; outputs are GeoTIFF and GeoJSON, not CSV

Best pick for code: the Huang repository, because it ties to a published paper and its data record lists a code DOI (10.5281/zenodo.8429209) that I did not open. Without a licence file, reuse terms are unclear. The TALEA repository is the better template for a multi-year surface UHI workflow in Earth Engine, since swapping the 2024 window for a 2000 to present loop is a small change (my inference, untested).

Near misses

Zenodo 17644687, 6,022 cities, summer 2019: CC-BY-4.0, but one summer only and an .xlsx file, so it fails period and format.
Zenodo 8032910 (Copernicus-based SUHI method, 10 CURE cities): CC-BY-4.0, but the only file is a PDF paper, so no separate data file is listed.
Mentaschi global dataset: right period (2003 to 2020), but global rather than European, and its asset, licence and format are unverified.

Gaps and notes

Gap, period: nothing verified spans 2000 to 2007 for European cities. UrbClim begins in 2008.
Gap, format: every verified gridded product is NetCDF or raster, so a conversion step is needed to meet CSV or Parquet.
Gap, observations: I did not verify any station-based canopy-layer UHI series.
Licence note: the two GitHub repositories showed no LICENSE file in the root listing. That is not the same as proving none exists, and licence compatibility is yours to judge.
Not checked: Hugging Face, Kaggle, UCI, Dryad, Figshare, GitLab, data.europa.eu, NASA Earthdata and Copernicus Land portals were not searched in this run.

Do next

Open the CDS record, register, and download a few UrbClim cities to compute summer UHI (urban mean minus rural mask) per year for 2008 to 2017.
Find the Mentaschi SUHI asset in Google Earth Engine (the paper says it is available there), check its licence, and export European city values to CSV or Parquet; the TALEA notebook is a scaffold for the export code.
Open Zenodo DOI 10.5281/zenodo.8429209 to confirm the licence of the Huang code.
Convert data_urbanruralavg_timeseries.nc to Parquet with xarray if you want the 2015 to 2017 series in your preferred format.
If the Earth Engine asset cannot be found, ask the paper's corresponding author for its catalogue ID; I did not contact anyone.

Coverage

Reached: Zenodo API records, CDS dataset page (via a fallback extractor), EEA catalogue, GitHub repository pages, one journal PDF. Not reached or failed: Zenodo HTML pages (bot check, API used instead), the ECMWF Confluence documentation (login wall), the CDS page through the main fetcher (TLS error), the MDPI paper on European LST developments (page body did not carry its data statement). Access date for all: 5 October 2026.