Performance of the CMIP6 global climate models over the Iberian Peninsula and relationships with the simulated climate system complexity

Abstract

[ES]En el presente estudio se muestra una evaluación del rendimiento de las diferentes configuraciones de los modelos climáticos globales que han aportado experimentos históricos a CMIP6 (Coupled Model Intercomparison Project 6) para la Península Ibérica (IB), utilizando un dominio similar al aplicado en iniciativas de downscalling anteriores. La evaluación se basa en los patrones típicos de circulación atmosférica regional definidos por Jenkinson y Collison (1977), que se sabe que están vinculados con un gran número de variables de la física y la química atmosféricas. Los resultados se comparan con los obtenidos de la generación anterior de los modelos (CMIP5) y con los obtenidos de un análisis hemisférico (Brands 2022a), para comprobar 1) si los modelos han mejorado con el tiempo y 2) si los resultados específicos concuerdan con los obtenidos en un dominio más grande, lo que los hace menos propensos a la propagación de errores durante períodos de tiempo no observados.

[EN]performance assessment of the global climate model configurations contributing historical experiments to CMIP6 is provided for the Iberian Peninsula (IB), using a spatial domain similar to that applied in previous downscaling initiatives. The evaluation is based on typical recurrent regional atmospheric circulation patterns as defined by Jenkinson & Collison (1977), which are well known to be linked with a large number of variables from atmospheric physics and chemistry. Results are compared to those obtained from the previous model generation (CMIP5), and to those retrieved from a hemispheric-wide analysis (Brands 2022) in order to see 1) if the models have improved over time, and 2) whether the region-specific findings agree with those obtained on a larger domain, thereby making them less prone to error propagation during unobserved time periods. It is found that the model version changes from CMIP5 and 6 lead to slight improvements, mainly associated with an increase in horizontal model resolution, but that the selection of the right model family is more important to obtain good model performance.