Extraordinary long-term trends in daily temperature extremes in station and reanalysis data

Properly evaluating temporal changes in the occurrence of extreme high or low temperatures is of key importance for assessing the potential local impacts of ongoing climatic changes and estimating possible future trends. Notably, the applicability of traditional extreme value statistics to non-stationary climate data is often restricted by the available amount of data. As a possible alternative, quantile regression techniques allow estimating temporal trends in arbitrary quantiles of the distribution of observed temperatures. Here, we report corresponding results on the long-term evolution of daily mean, maximum and minimum temperatures from three different sources: a homogeneous set of German station data, the E-OBS gridded data set of European temperatures since 1950, and various global reanalysis data sets.

At the scale of individual meteorological stations across Germany, the obtained trends in very high and low quantiles reveal an extraordinary increase of high temperature extremes since the 1950s, which significantly exceeds the mean warming. In order to further assess the robustness of these results, we compare the trend values for linear and nonlinear (spline-based) quantile trends with those obtained from (linearly) time-dependent extreme value analysis and find reasonable mutual agreement in the observed mean trends. The obtained spatial patterns of trend values reveal seasonally different regionalized effects related to changes in the predominance of coastal vs. continental climatology, local geographical factors and the distribution of potential urban heat islands.

For the gridded data sets, linear quantile regression provides a full picture of (seasonal) trends in daily surface air temperature extremes across the globe. The obtained results are largely consistent among different data sets in regions with good coverage by meteorological stations, but display marked differences in poorly covered regions, expecially around the poles and within large deserts. Notably, the observed differences originate mainly from time periods without available remote sensing data, whereas the estimated trend values converge when taking only the last about 35 years into account, providing additional confidence in the robustness of the observed trends.