Examining spin-up behaviour within WRF dynamical downscaling applications.

Spin-up is the period after initialization when a model transitions away from its dependence on initial conditions toward a dynamic equilibrium between driving boundary conditions and its own internal dynamics. Regional climate models (RCMs) are often used to simulate conditions over several decades to inform local adaptation and resilience activities. The spin-up period represents added cost to already resource-intensive simulations, and it is often infeasible to use a spin-up period that produces complete model equilibrium. Therefore, a pragmatic compromise is desired to minimize the effects of spin-up. Here, two overlapping dynamically downscaled simulations (31 year and 11 year integrations) using the Weather Research and Forecasting model over the contiguous US (CONUS) are used to explore convergence associated with model spin-up. The shorter simulation is initialized 20 years after initialization of the longer (reference) run, and the runs are analysed over the period covered by both simulations, giving the reference simulation a 20 year period to attain spin-up prior to comparison. After initialization, the shorter run features cooler surface and near-surface temperatures and greater soil moisture compared to the reference simulation. Differences between the runs decrease in magnitude over the first 3 months as autumn transitions to winter; however, these differences re-emerge and reach a secondary peak during the proceeding spring and summer. During this warm season, evaporation and accompanying evaporative cooling increase and temperature differences between the simulations re-emerge. These results support using at least one year of spin-up time in RCM applications to account for the seasonality of spin-up behaviour. Results from some regions of the CONUS indicate that spin-up durations of 1-4 years are needed to exclude spurious behavior in top-layer soil moisture, which exhibits prolonged spin-up compared with other near-surface variables examined here.