Observations of the minimum temperature (Tmin) over 24 hours at 264 stations worldwide since 1950 were expressed as anomalies, relative to the period 1961–90 where possible. Coverage of Tmin data was good north of 20° N, in Australasia and in the western tropical Pacific, but poor in Africa, South America, Antarctica and parts of southern Asia. Reanalysed4 daily-average near-surface wind components were used to classify the Tmin anomalies into ‘windy’ (upper tercile) and ‘calm’ (lower tercile) conditions. Daily average wind speeds were used because the timings of temperature extremes are not known. For stations between 140° E and the dateline, Tmin — which occurs most frequently in the early morning — was matched with the previous day's speed. This is because the early morning in terms of universal time (equivalent to Greenwich Mean Time) is still in the previous day in the Far East.

Annual and seasonal anomalies of Tmin were gridded on a 5° × 5° resolution for windy, calm and ‘all’ conditions. Coverage was at least 200 grid boxes (equivalent to more than 27% of global land area) in 1958–99. For 1950–2000, the trends of global annual average Tmin for windy, calm and all conditions were identical (0.19 ± 0.06 °C per decade; Fig. 1a). So, urbanization has not systematically exaggerated the observed global warming trends in Tmin. The same can be said for poor instrumental exposure and microclimatic effects, which are also reduced when instruments are well ventilated5.

Figure 1: Anomalies in Tmin for windy (red) and calm (blue) conditions.
figure 1

a, Annual global data; b, winter data (December to February) for Northern Hemisphere land north of 20° N; c, summer data (June to August) for Northern Hemisphere land north of 20° N. The linear trend fits, and the ±2σ error ranges given in the text, were estimated by restricted maximum likelihood10, taking into account autocorrelation in the residuals. As expected from the reduced stratification of the boundary layer, Tmin is, on average, warmer on windy nights than on calm nights.

When the criterion for ‘calm’ was changed to the lightest decile of wind strength, the global trend in Tmin was unchanged. The analysis is therefore robust to the criterion for ‘calm’. To assess the effect of time differences between the reanalysis4 daily-average winds and Tmin, I repeated the analysis using 26 stations in North America and Siberia that have hourly or six-hourly reports of simultaneous temperature and wind. Again, windy and calm nights warmed at the same rate, in this case by 0.20 °C per decade.

Because a small sample was used, I compared global trends for the reduced period 1950–93 with published all-conditions trends for that period based on a sample of over 5,000 stations6. All differences were within ±0.02 °C per decade. This robustness arises because of the spatial coherence of surface temperature variations and trends7.

The global annual result conceals a relative warming of windy nights in winter in the extratropical Northern Hemisphere (Fig. 1b), mainly in western Eurasia. The observed tendency to an increased positive phase of the North Atlantic Oscillation8 implies that the windier days in western Eurasia had increased warm advection from the ocean9, yielding greater warming. In summer in the extratropical Northern Hemisphere (Fig. 1c), there was no relative change in Tmin on windy nights. At that time of year, atmospheric circulation changes are less influential, but an urban warming signal is still absent. In the tropics, calm nights warmed relative to windy nights on an annual average, but only by 0.02 ± 0.01 °C per decade, which is much less than the overall tropical warming in Tmin (0.16 ± 0.03 °C).

This analysis demonstrates that urban warming has not introduced significant biases into estimates of recent global warming. The reality and magnitude of global-scale warming is supported by the near-equality of temperature trends on windy nights with trends based on all data.