A raw, homogenization-free global dataset of air-temperature urban heat-island intensity from GHCN-Daily, 1941–2020

Global urban heat-island (UHI) assessments rely predominantly on satellite land-surface temperature, while station-based studies typically use homogenized air-temperature series in which breakpoint adjustment can remove the gradual urban signal of interest. We release a raw (quality-controlled but explicitly non-homogenized) air-temperature UHI dataset derived from the Global Historical Climatology Network–Daily (GHCN-Daily) archive. From 40,800 candidate temperature-reporting stations, 8,666 clear our daily-quality and monthly/annual completeness thresholds; we aggregate their raw daily maximum/minimum to monthly and annual means ourselves, and compute UHI as the elevation-corrected difference between an in-city station and a contamination-screened rural reference, for both mean (TAVG) and nocturnal (TMIN) temperature. The release comprises: (i) an annual station-level temperature panel; (ii) a city-level UHI panel for up to 1,664 cities nominally spanning six continents but concentrated in North America and Europe (Africa and South America together contribute under 10 % of cities and are too sparse to characterize on their own; Usage Notes), 2000–2020, plus a full-record climatological set (1941–2020, mean/nocturnal/daytime), a balanced long-record subset, and an epoch-resolved within-city panel (1975–2020); (iii) the city–station matching, for both the broad and long-record working sets, with all distance and quality metadata; (iv) the analysis-ready city covariates and the co-located satellite panel used for the surface-versus-air comparisons; and (v) the complete construction and analysis code, including a build script that rebuilds the analysis-ready tables from the raw station record alone and verifies the rebuild numerically against reference coefficients tabulated in this descriptor. Technical validation includes internal consistency, sensitivity to the rural-reference radius, a station-siting stratification, an independent cross-check against satellite surface UHI, and a demonstration that homogenization removes most of the signal — cutting the global-mean intensification by ~60 % and North America’s by ~83 % (+0.21 → +0.04 °C). The dataset supports urban climate attribution, UHI scaling analysis, heat-exposure modelling, and evaluation of homogenization effects, and is intended for use at the level of aggregate patterns and relationships, for which it is robust, rather than precise per-city values.