A methodology for deriving ensemble response from multimodel simulations

•A novel algorithm for deriving ensemble response for multimodel simulations.•An Expert Advice (EA) approach as an alternative to ensemble mean.•A generalized method applicable to different hydrologic variables. Multimodel ensembles are widely used to quantify uncertainties of climate model simulations. Previous studies have confirmed that a multimodel ensemble approach increases the skill of model simulations. However, one may need to know which ensemble member is more likely to be true, particularly when the ensemble is spread out over a wide area. Typically, ensemble response (climate response) is derived by taking the mean or median of ensemble members. However, strong similarities exist between models (members of an ensemble) which may cause biased climate response toward models with strong similarities. In this study, a model is proposed for deriving the climate response (ensemble response) of multimodel climate model simulations. The approach is based on the concept of Expert Advice (EA) algorithm which has been successfully applied to the financial sector. The goal of this methodology is to derive an ensemble response that at every time step is equal or better (less error) than the best model. The methodology is tested using the CMIP5 historical temperature simulations (1951–2005) and Climatic Research Unit observations, and the results show that the EA algorithm leads to smaller error compared to the ensemble mean.