Accounting for the spatial structure of weather systems in detected changes in precipitation extremes

Abstract

The detection of changes over time in the distribution of precipitation extremes is complicated by noise at the spatial scale of weather systems. Traditional approaches for quantifying observed changes in extreme precipitation return values are often based on single-station analyses, which fail to account for the spatial coherence of individual storms and hence yield unrealistic and potentially misleading estimates of the true underlying changes in extremes. In this paper, we demonstrate how the use of a flexible statistical method that robustly accounts for the so-called “storm dependence” in measurements of daily precipitation removes a challenging source of noise and results in improved estimates of changes in the distribution of precipitation extremes. Furthermore, our analysis provides important insights into the spatial structure of seasonal extreme precipitation across increasing event rarity. Applying the methodology to long-term in situ records of daily precipitation from the central United States, we find that properly accounting for storm dependence leads to increased detection of statistically significant changes in return values as compared with existing approaches. We also find that simultaneous precipitation extremes in this region tend to organize on scales of 100–200 km for high quantile levels, which is consistent with observed spatial patterns in the NEXRAD Stage IV radar-based data set.

Keywords

• Return values
• Storm dependence
• Gaussian scale mixtures
• Asymptotic dependence
• Concurrent extremes
• Precipitation trends