Understanding seasonal cycle of daily extreme temperatures based on generalized additive model for location, scale and shape with smoothing spline

Abstract

AbstractThe seasonal cycle (SC) of surface air temperature is a substantial element in climatology. Some basic and important topics in climate change studies are based on accurate and reliable estimation of SCs, such as percentile‐based indices of extremes and probability density function (PDF) changes of daily temperatures. For both of them, most studies characterized SCs using averages of multi‐days windows, which is not smooth and accurate enough representing the extreme thresholds and climatological normal of SCs. It is necessary to construct smooth and reasonable SCs for more accurate estimation of temperature changes on extreme thresholds and PDFs. In this study, we propose a flexible method based on generalized additive models for location scale and shape and penalized b‐spline smoothing technique with respective distributions to construct smooth SCs and SC for extreme temperatures (SCETs). The accuracy of the constructed smooth SCETs is good with the estimation biases of percentiles tending to be zero. The constructed smooth SCET also exhibits good stability over time, such that the magnitude changes of temperatures on each calendar day are close to climatic changes of mean temperature when the concerned period shifts. Based on the constructed smooth SCs, climatic changes by seasons and PDFs between two periods, 1961–1990 and 1991–2020, are examined over China. The increase of thresholds for hot extremes in spring during the recent period is prominent, while the increase of thresholds for daytime cold extremes in summer over a part of central to southern China is also notable. The smooth SCs and SCETs based on our flexible statistical modelling framework can characterize daily extreme temperatures reasonably and accurately, and should be expected to have more applications for a better understanding of climate changes related to distributions and seasonal cycles.