Using Bayesian statistics to detect trends in Alaskan precipitation
James H. White, John E. Walsh, Richard Thoman
Abstract
Abstract Air temperature has exhibited a clear positive trend over the past several decades throughout the arctic, including Alaska. Other variables, such as precipitation, have much more uncertain trends due to inhomogeneities in measurement and high internal variability. The use of linear regression to analyse precipitation in Alaska has resulted in often contradictory results. This paper proposes the use of Bayesian models such as the R package Rbeast to allow for the more nuanced analysis. The examples given in this paper show how Bayesian analysis can be used to detect subtle changes and better constrain the disagreement between data sources. Applied to gridded data, Bayesian analysis shows how precipitation has changed overtime across Alaska. Change has accelerated over the past decade, but only precipitation increase on the North Slope can be assigned high confidence. Overall, this analysis highlights how Bayesian techniques may be uniquely useful to climate research in regions with heterogeneous data sources and substantial internal variability.