On the use of Bayesian quantile regression method to explore the historical trends in extreme precipitation and their connections with large-scale climate patterns over the contiguous USA

Abstract

The probability distribution function (PDF) of precipitation is expected to change under a warmer climate. The existing studies have not been able to detect trends in different thresholds of the PDF of extreme precipitation series for the contiguous USA. In this study, a fresh statistical approach (the Bayesian quantile regression method) was employed to analyze trends of annual daily maximum (ADM) and monthly maximum (MM) precipitation indices at different quantile levels and their teleconnections with large-scale climate patterns over the contiguous USA. Historic precipitation time series over the period of 65 years (1950–2014) for 1108 sites was used for the analysis. To investigate the relations between precipitation and large-scale climate patterns, two major oscillations [the El Nino–Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO)] were used as covariates in the Bayesian quantile regression model. Our results show that changes in upper quantiles of the distributions of the extreme precipitation (for both ADM and MM) have occurred at a much higher rate than previously believed as compared to the lower quantiles mainly in the coastal regions in the eastern half of the country. Variability in the spatial distribution of significant trends was observed when either ENSO or NAO index was used as a covariate in the Bayesian quantile regression model. However, for many regions of the USA, a stronger teleconnection between climate indices and precipitation indices for upper quantile level was observed as compared to lower quantile levels for both ADM and MM. The trends detected using the Bayesian quantile regression method were undetected or overlooked by other previously used approaches for trend detection. Such results are particularly useful for water managers who are more concerned with extreme values rather than the averaged one.