Partitioning uncertainty components of mean climate and climate change in a large ensemble of European regional climate model projections

Abstract

A study of seasonal mean temperature, precipitation, and wind speed has been performed for a set of 19 global climate model (GCM) driven high-resolution regional climate model (RCM) simulations forming a complete 5 × 4 GCM × RCM matrix with only one missing simulation. Differences between single simulations and between groups of simulations forced by a specific GCM or a specific RCM are identified. With the help of an analysis of variance (ANOVA) we split the ensemble variance into linear GCM and RCM contributions and cross terms for both mean climate and climate change for the end of the current century according to the RCP8.5 emission scenario. The results document that the choice of GCM generally has a larger influence on the climate change signal than the choice of RCM, having a significant influence for roughly twice as many points in the area for the fields investigated (temperature, precipitation and wind speed). It is also clear that the RCM influence is generally concentrated close to the eastern and northern boundaries and in mountainous areas, i.e., in areas where the added surface detail of e.g. orography, snow and ice seen by the RCM is expected to have considerable influence on the climate, and in areas where the air in general has spent the most time within the regional domain. The analysis results in estimates of areas where the specific identity of either GCM or RCM is formally significant, hence obtaining an indication about regions, seasons, and fields where linear superpositions of GCM and RCM effects are good approximations to an actual simulation for both the mean fields analysed and their changes. In cases where linear superposition works well, the frequently encountered sparse GCM-RCM matrices may be filled with emulated results, leading to the possibility of giving more fair relative weight between model simulations than simple averaging of existing simulations. An important result of the present study is that properties of the specific GCM-RCM combination are generally important for the mean climate, but negligible for climate change for the seasonal-mean surface fields investigated here.

Balanced estimate and uncertainty assessment of European climate change using the large EURO-CORDEX regional climate model ensemble

“…and for RCM/GCM interactions, particularly in summer. Christensen and Kjellström (2020) share these results (see their Fig. S14), RCM/GCM interactions contributing to 5%-20% of the total variance for summer precipitation over all Europe and the Mediterranean sea.…”

“…Except for the evolving aerosol effect described in section 7, it is difficult so far to explain the RCM individual behaviour as the most influencing RCMs of the current study have not always been included or detected in previous articles dealing with EURO-CORDEX projections (e.g. Sørland et al, 2018;Fernández et al, 2019;Boé et al, 2020;Christensen and Kjellström, 2020). We consider however that the identification of strong RCM individual effects in our study motivates the definition of a standardized indicator of regional climate model sensitivity mimicking the ECS or TCR indicators for GCMs.…”

“…t-tests of WRF simulations in Laux et al, 2017), test the significance of the components of an ANOVA decomposition (e.g. F-test in Christensen and Kjellström, 2020).…”

Supplementary material to "Balanced estimate and uncertainty assessment of European climate change using the large EURO-CORDEX regional climate model ensemble"

“…The last remaining variability is related to interactions RCM/GCM/RCP, which can be estimated using different replicates of the simulation chains (Yip et al, 2011) or by making additional assumptions, e.g. by considering residual terms corresponding to successive years as replicates (Christensen and Kjellström, 2020). Northrop and Chandler (2014) note that direct estimates of these interactions are often biased.…”

A Climate Service for Ecologists: Sharing pre-processed EUROCORDEX Regional Climate Scenario Data using the eLTER Information System

“…bias adjustments), and the impact model; the models have structure and parameter uncertainty, bias-adjustment methods may be prone to instationarities of their parameter settings, and natural variability also contributes to total uncertainty (e.g., Bosshard et al, 2013;Christensen and Kjellström, 2020;Her et al, 2019;Kundzewicz et al, 2018). Decomposing the total uncertainty of a GCM-RCM dynamical downscaling modelling chain into its different sources and contributions, for example Christensen and Kjellström (2020) show that the GCM has a large influence on the climate change signal, while the choice of 100 RCM impacts results e.g. in mountainous areas considerably; Sørland et al (2018) show that modelling chain errors are not additive, but that RCM can reduce biases inherited by the GCM.…”