Temperature Trends in the Free Atmosphere: Calculations Using the Quantile Regression Method

Abstract

Results of calculations of temperature trends in the free atmosphere (troposphere and lower stratosphere) using the quantile regression apparatus are considered and analyzed. In traditional techniques used in climatology, trends are estimated by use of regression based on the least squares method. Quantile regression, in contrast to these techniques, makes it possible to estimate regression parameters for each quantile of predictand values in the quantile range from zero to one. Using quantile regression to estimate climate changes results in a detailed picture of the dependence of the climate trend on the variation range of meteorological parameters in the quantile range of these parameters from zero to one. In particular, climate trends can be estimated for meteorological parameter values close to extreme. This paper uses the global radiosonde data array from which the stations are selected if the completeness of their data meets the requirements stated. Using the radiosonde data from the selected stations, the dependences of climatic trends of temperature on isobaric surfaces on values of quantiles (so-called process diagrams), as well as vertical quantile cross sections of climate trend values, are calculated, plotted, and analyzed. For thirteen high-latitude stations in the Northern Hemisphere among the selected ones, temperature trends are estimated both using radiosonde data and based on the ERA 5/ERA 5.1 reanalyses. An analysis of the results allows one to note the nonuniform character of tropospheric warming trends in the range of quantile variation, which is more apparent in the winter season. The nonuniform (for the range of quantile variation) character of tropospheric temperature trends is due to the fact that the tropospheric warming rate in the “cold” part of the quantile range is higher than that in its “warm” part. This agrees with the results obtained previously by analysis of surface temperature trends using the quantile regression method (QRM). The nonuniform character of cooling trends in the lower stratosphere is noted for the range of quantile variations. In winter and, to a lesser extent, in spring, the rate of stratospheric cooling decreases in absolute magnitude with an increase in quantile values at some stations in northern latitudes. Moreover, for the quantiles close to 1.0, negative trends can change sign. This can be both due to incomplete data on lower stratospheric temperature, which is particularly inherent in the high-latitude regions of the Northern Hemisphere, and due to the influence of more frequently occurring sudden stratospheric warmings (SSWs) on the temperature trend structure that is detailed within the range of quantile values. In is noted that the detailed structures of climate temperature trends that are obtained on the basis of radiosonde data proved to be very similar to those obtained based on arrays of ERA 5/ERA 5.1 reanalysis.