Comparing regional precipitation and temperature extremes in climate model and reanalysis products

Abstract

A growing field of research aims to characterise the contribution of anthropogenic emissions to the likelihood of extreme weather and climate events. These analyses can be sensitive to the shapes of the tails of simulated distributions. If tails are found to be unrealistically short or long, the anthropogenic signal emerges more or less clearly, respectively, from the noise of possible weather. Here we compare the chance of daily land-surface precipitation and near-surface temperature extremes generated by three Atmospheric Global Climate Models typically used for event attribution, with distributions from six reanalysis products. The likelihoods of extremes are compared for area-averages over grid cell and regional sized spatial domains. Results suggest a bias favouring overly strong attribution estimates for hot and cold events over many regions of Africa and Australia, and a bias favouring overly weak attribution estimates over regions of North America and Asia. For rainfall, results are more sensitive to geographic location. Although the three models show similar results over many regions, they do disagree over others. Equally, results highlight the discrepancy amongst reanalyses products. This emphasises the importance of using multiple reanalysis and/or observation products, as well as multiple models in event attribution studies.

On the nonlinearity of spatial scales in extreme weather attribution statements

“…4c). over North America and parts of Asia, which were found to be biased in favour of 288 being overly weak (Angélil et al, 2016). Extremes in all of the current generation HadAM3P-N96, which corresponds to the PRs for hot extremes in Fig.…”

Toward Calibrated Language for Effectively Communicating the Results of Extreme Event Attribution Studies

“…Numerous studies have demonstrated a spatiotemporal-scale dependence of FAR/RR results, in addition to the specific metric being explored (as discussed in Angélil et al, 2016;Cattiaux & Ribes, 2018;Harrington, 2017;Uhe et al, 2016). First, it is well documented that definition of an extreme event is a critical facet of the EEA result.…”

The stationarity of two statistical downscaling methods for precipitation under different choices of cross‐validation periods

“…We used Wilcoxon signed-rank tests to test if significantly higher RMSEs were obtained from the downscaled precipitation using ERA-Interim data (Hollander et al, 2015). JRA-55 shares a considerable amount of input data with ERA-Interim (Kobayashi et al, 2015), and exhibits some similarity in the temperature and precipitation data for the study region (Angélil et al, 2016;Dhanya and Villarini, 2017). This similarity may produce a smaller increase in RMSEs than when Rglimclim is transferred from JRA-55 to a generic climate model.…”