Is there added value in the EURO‐CORDEX hindcast temperature simulations? Assessing the added value using climate distributions in Europe

Abstract

Regional climate simulations with high horizontal resolutions are becoming increasingly common and although model development has continually enhanced the representation of atmospheric phenomena, the model improvements are variable, region and time scale‐dependant. The high computational costs of increasingly smaller grid‐spacing underline the need for a robust assessment of the benefits or losses associated to the dynamical downscaling of coarser resolution models (reanalysis, global climate models or ~tenths km runs), that is, quantitative added value evaluation. In the current study, a probability density function (PDF) matching score is used to determine the distribution added value (DAV) of the EURO‐CORDEX maximum and minimum temperatures from the hindcast simulations at 0.44° and 0.11° resolutions. The gridded maximum and minimum temperatures from European Climate Assessment & Dataset (E‐Obs) were used as benchmarks for the matching scores and DAVs were determined for the 0.44° and 0.11° simulations individually against ERA‐Interim as well as against each other. Temperature added value against ERA‐Interim is difficult to find due to the assimilation of surface temperatures in the reanalysis. Nonetheless, there is positive added value in Tmax for half of the models at the yearly time scale, and the increase in resolution also implies positive DAV in half of the models, with DAVs between 0.1 and 3.8%. More importantly, the benefits of downscaling are significantly visible in the extreme end tail of Tmax, where all models have high added value at all resolutions and seasons except in summer (from ~1 to ~21%). In minimum temperature there is added value in winter and autumn at both resolutions. The Mediterranean and the British Isles are the regions where larger positive DAV values for both variables and resolutions are identified. In Tmax, added value is also associated to the increase in resolution (from 50 to 10 km) for half of the models in all regions. Scandinavia, the Alps, and eastern Europe are regions where negative added value is clearly linked with the highest mountain peaks for both variables.

The added value of high‐resolution EURO‐CORDEX simulations to describe daily wind speed over Europe

“…The improvement in the representation of small-scale processes are related to a better performance of the drag generated by the unresolved terrain in the low-resolution models and the possibility of a speed-up of the flow over the mountains (Jiménez and Dudhia, 2012), and the higher spatial variability of the gusts over complex terrain produced by a better representation of the main orographic structures (Kunz et al, 2010). Previous works for maximum temperature (Vautard et al, 2013;Careto et al, 2021;Cardoso and Soares, 2022) and precipitation (Prein et al, 2016;Careto et al, 2022) also display added values in coastal areas, mainly due to a better representation of the differential warming associated with the land-sea thermal contrasts. Besides, Torma et al (2015), Fantini et al (2018), Soares and Cardoso (2018) and Ciarlo et al (2020) show that the EURO-CORDEX RCMs provide larger added value in representing extreme precipitation events.…”

“…In the end, DAV represent the percentage of added (positive) or lost (negative) value associated with the higher resolution in relation to the lower resolution, with values from −100 until . To assess the DAV regarding extreme wind speed, we also compute the added value for the PDF section above the 95th percentile of the observations (), following (Soares and Cardoso, 2018; Cardoso and Soares, 2022). Although the percentile selection is somewhat arbitrary, the 95th percentile has been selected following previous studies (Frank et al ., 2020; Outten and Sobolowski, 2021).…”

“…As shown in Soares and Cardoso (2018) and Cardoso and Soares (2022), the DAV method can be applied to aggregated regions. Here, the DAV are computed for the PRU-DENCE subdomains to analyse the performance of the EURO-CORDEX models in the representation of the full PDFs of the wind speed (Figures 5, S38-S40).…”

“…(2017) found considerable differences in the spatial pattern of precipitation changes between resolutions in southwestern Germany, although convection‐permitting models do not seem to bring added value to assessments of climate change in precipitation in near future (Fosser et al ., 2017). For temperature, added value was found in the use of finer resolution simulations, mainly associated with land–sea contrast and small‐scale topographical features, in different domains from North America (Di Luca et al ., 2013a), to Europe (Cardoso and Soares, 2022) and in the Iberian Peninsula (Careto et al ., 2021).…”

Critical analysis ofCMIPspast climate model projections in a regional context: The Iberian climate

“…The representation of the annual cycles and long‐term trends is also analysed. The agreement between the CMIPs and the reference datasets is carried out focusing on the bias (Equation ()), the PDF‐Score (S; Perkins et al, 2007; Boberg et al, 2009; Brands et al, 2011; Soares et al, 2018; Lima et al, 2019; Equation ()), and the distribution added value (DAV; Soares and Cardoso, 2018; Lemos et al, 2020; Cardoso and Soares, 2022; Careto et al, 2022a, 2022b; Equation ()) metrics, defined respectively as: In Equation (), i corresponds to each data entry, to the length of the time‐series, and and to the set of GCMs per scenario and reference datasets (IB01, E‐OBS and ERA5), respectively. In Equation (), the terms and correspond to single‐variable outputs from different CMIPs, intercompared through their performance ().…”