Ensemble-based statistical verification of INM RAS Earth system model

Abstract Modern numerical models of the Earth system are complex and inherit its natural chaotic behaviour. The numerical results depend on various specifications of the simulation process, including computing systems, compilers, etc. Due to the chaotic behaviour, these minor differences lead to significant and unpredictable deviations. Therefore, some procedure verifying that simulation results describe the behaviour of the same physical system is of practical importance. The present paper proposes a statistical verification algorithm developed for the INM RAS Earth system model. Different ensemble generation techniques and statistical estimators are evaluated for verification suitability. The ability of the method to detect the deviations in the simulation results is demonstrated on a series of cases. Practical guidelines on how to choose the perturbation amplitude for the ensemble generation are provided for various verification cases.