Long Simulation of Regional Climate as a Sequence of Short Segments

Abstract

Regional climate simulations are long time integrations over an open system where the atmosphere over part of the model domain (boundary zones) is updated periodically. Model reinitialization after a long period of integration can allow several segments of a long simulation to be run in parallel and also minimize possible drift caused by accumulated model errors. However, the spinup problems introduced by each additional restart must be addressed. The necessity and feasibility of subdividing long integrations is investigated by means of a series of experiments in which the authors examine the effects of reinitialization frequency and the relative importance of surface forcing and atmospheric forcing. It is found that for integrations that continued without reinitialization, locations of specific meteorological features drifted downstream because simulated winds were too strong, implying the need for periodic reinitialization of the model. The results indicate also that when the model reinitialization interval is relatively long, simulated domain-averaged variables, including rainfall, were not very sensitive to model reinitialization since they are largely constrained by transient boundary conditions, suggesting the feasibility of dividing long regional climate simulations into a set of shorter ones that could be run in parallel.