Reinitialized versus Continuous Simulations for Regional Climate Downscaling

Abstract

The methodology for dynamical climate downscaling is studied using the second-generation regional climate model (RegCM2). The question addressed is, in order to simulate high-resolution details as accurately as possible, what strategy should be taken: continuous long-term integration in climate prediction mode or consecutive short-term integrations in weather forcasting mode? To investigate this problem, the model was run for 5 months in three different ways: 1) a 5-month continuous simulation, 2) monthly reinitialized simulations, and 3) 10-day reinitialized simulations. Compared to the observed precipitation, the 10-day reinitialized simulation results in the smallest error, while the continuous run shows larger error. Analysis shows that the long-term continuous simulation is contaminated by the systematic errors associated with the steep Andes Mountains and the uncertainties in the moisture processes in the planetary boundary layer near the coast. The method of 10-day reinitialization effectively mitigates the problem of systematic errors and makes a difference in the subtle precipitation processes in the regional climate model, therefore improving the accuracy in dynamic downscaling.