UCLA

Abstract There is considerable interest in determining whether recent changes in the temperature distribution extend beyond simple shifts in the mean. The authors present a framework based on quantile regression, wherein trends are estimated across percentiles. Pointwise trends from surface station observations are mapped into continuous spatial fields using thin-plate spline regression. This procedure allows for resolving spatial dependence of distributional trends, providing uncertainty estimates that account for spatial covariance and varying station density. The method is applied to seasonal near-surface temperatures between 1979 and 2014 to unambiguously assess distributional changes in the densely sampled North American region. Strong seasonal differences are found, with summer trends exhibiting significant warming throughout the domain with little distributional dependence, while the spatial distribution of spring and fall trends show a dipole structure. In contrast, the spread between the 95th and 5th percentile in winter has decreased, with trends of −0.71° and −0.85°C decade−1, respectively, for daily maximum and minimum temperature, a contraction that is statistically significant over 84% of the domain. This decrease in variability is dominated by warming of the coldest days, which has outpaced the median trend by approximately a factor of 4. Identical analyses using ERA-Interim and NCEP-2 yield consistent estimates for winter (though not for other seasons), suggesting that reanalyses can be reliably used for relating winter trends to circulation anomalies. These results are consistent with Arctic-amplified warming being strongest in winter and with the influence of synoptic-scale advection on winter temperatures. Maps for all percentiles, seasons, and datasets are provided via an online tool.