Separating Storm Intensity and Arrival Frequency in Nonstationary Rainfall Frequency Analysis

Abstract

Nonstationary Rainfall frequency analysis (RFA) is used to assess how climate change is impacting the likelihood of extreme storms. A key limitation of covariate-based approaches to nonstationary RFA is that without a physical basis, models selected based on the quality of fit to historical data cannot be reliably projected to estimate future quantiles. Here we propose to improve the physical representation of rainfall processes by using a peaks-over-threshold approach to separate the processes of storm intensity (impacted by thermodynamic drivers related to changes in atmospheric moisture) and storm arrival frequency (impacted by dynamic drivers that lead to changes in regional weather systems). Through stochastic experiments we demonstrate that quantiles can only be accurately projected beyond the observed climate when nonstationary models reflect the underlying nonstationary process. Through a case study we demonstrate how climate model projections of rainfall can be utilized to deduce nonstationary model structures, showing that changes in both the storm intensity and storm arrival frequency are needed to accurately estimate future quantiles. While here we propose a single simple physically informed approach for storm intensity, structuring the arrival frequency component requires a detailed understanding of atmospheric dynamics in the region of interest.