Assessing surface air temperature variability using quantile regression

Abstract:

Many researches in climate change currently involve linear trends, based on measured variables. And many of them only consider trends in mean values, whereas it is clear, that not only means, but also whole shape of distribution changes over time and requires careful assessment. For example extreme values including outliers may get bigger, while median has zero slope.

Quantile regression provides a convenient tool, that enables detailed analysis of changes in full range of distribution by producing a vector of quantile trends for any given set of quantiles.

We have applied quantile regression to surface air temperature observations made at over 600 weather stations across Russian Federation during last four decades. The results demonstrate well pronounced regions with similar values of significant trends in different parts of temperature value distribution (left tail, middle part, right tail). The uncertainties of quantile trend estimations for several spatial patterns of trends over Russia are estimated and analyzed for each of four seasons.

For temperature trend estimation over vast territories, quantile regression is an effort consuming approach, but is more informative than traditional instrument, to assess decadal evolution of temperature values, including evolution of extremes.

Partial support of ERA NET RUS ACPCA joint project between EU and RBRF 12-05-91656-ЭРА-А is highly appreciated.