Upgrade of the HadGEM3-A based attribution system to high resolution and a new validation framework for probabilistic event attribution

We present a substantial upgrade of the Met Office system for the probabilistic attribution of extreme weather and climate events with higher horizontal and vertical resolution (60 km mid-latitudes and 85 vertical levels), the latest Hadley Centre atmospheric and land model (ENDGame dynamics with GA6.0 science and JULES at GL6.0) as well as an updated forcings set. A new set of experiments designed for the evaluation and implementation of an operational attribution service are described which consist of pairs of multi-decadal stochastic physics ensembles continued on a season by season basis by large ensembles that are able to sample extreme atmospheric states possible in the recent past. Diagnostics from these experiments form the HadGEM3-A contribution to the international Climate of the 20th Century Plus (C20C+) project and were analysed under the European Climate and Weather Events: Interpretation and Attribution (EUCLEIA) event attribution project as well as contributing to the Climate Science for Service Partnership (CSSP)-China programme. After discussing the framing issues surrounding questions that can be asked with our system we construct a novel approach to the evaluation of atmosphere-only ensembles intended for event attribution, in the process highlighting and clarifying the distinction between hindcast skill and model performance. A framework based around assessing model representation of predictable components and ensuring exchangeability of model and real world statistics leads to a form of detection and attribution to boundary condition forcing as a means of quantifying one degree of freedom of potential model error and allowing for the bias correction of event probabilities and resulting probability ratios. This method is then applied systematically across the globe to assess contributions from anthropogenic influence and specific boundary conditions to the changing probability of observed and record seasonal mean temperatures of four recent 3-month seasons from March 2016-February 2017.