Exploring Extended Warm Periods in an Observational Large Ensemble of Historical European Temperature

Contextualizing extended periods of extreme regional heat, such as the persistently mild Siberian winter of 2020, is challenging within the order 100-year observational temperature record. A robust understanding of these events amid historical climate variability, crucial for climate risk assessment, necessitates the use of model simulations to increase the temporal sample. Models and observations can combine to form alternative sequences of historical climate; these "observational large ensembles" have been generated using a model-derived forced signal and observationally-generated "climate noise".

Here, we create an observational large ensemble of European Surface Air Temperature (SAT) over the historical period (1950-2014) using model-derived climate noise and an observationally-generated signal. The components are computed using constructed circulation analogue dynamical adjustment (Deser et al. 2016) applied to 8 large ensembles in the Multi-Model Large Ensemble Archive and the Berkeley Earth Surface Temperature observational record. Noise is typically computed in large ensembles by subtracting the ensemble mean from each member, a method with no observational equivalent. Dynamical adjustment, however, allows any SAT record to be empirically separated into a dynamic component (associated with atmospheric variability) and a residual, or climate noise estimate and signal, respectively.

We assess to what extent dynamic components from different models are distinguishable using out-of-sample testing and document systematic biases between model and observational dynamic components. Sets of models are selected such that observed statistical properties are maintained at each grid point. Finally, model-derived climate noise is combined with the observationally-generated signal, and the resulting observational large ensemble is evaluated via a multi-month extreme SAT frequency metric against the state-of-the-art observational large ensemble developed by McKinnon et al. 2018.