Given the spatial resolution of these phenomena, high-resolution mesoscale models are required to accurately predict the onset of and the dynamical circulations associated with sea breeze events. Obtaining realistic sea breezes in numerical models is contingent on the accurate representation of land and sea surface temperature (SST) magnitudes and spatial gradients. Sea surface temperature spatial gradients can be highly complex in the coastal region, with a much finer spatial variability than is captured by the initial conditions typically used in mesoscale models. The sensitivity of simulated sea breeze events to the resolution of SST fields used in numerical models is unknown.
This study aims to use the Weather Research and Forecasting (WRFv3.6.1) model to evaluate the influence of SST gradients on the physical processes associated with sea breezes. The 21 August 2013 sea breeze event over coastal Connecticut (CT) was selected due to its robust land-ocean temperature contrast of 15oC during the event. Simulations use a 9km/3km/1km triply nested domain centered on the CT coastline, with the North American Regional Reanalysis (NARR; 32 km resolution) for initial and boundary conditions. Given the relatively coarse resolution of the NARR, the coastal boundaries along Long Island Sound are not well resolved. Interpolating the relatively coarse NARR initial conditions onto the high-resolution WRF grid causes land information (i.e. skin temperature) to be projected onto the western Long Island Sound, impacting the evolving sea breeze circulation. We hypothesize that using a correct representation of the coastline may be as significant in obtaining an accurate sea breeze circulation as using a high resolution SST product.
In the current study, NARR SST initial conditions over western LI Sound were modified to obtain a more realistic water body. Preliminary results from this technique will be discussed, including the impact on the evolving environmental variables and the resulting sea breeze circulation, as well as comparisons with the control simulation (i.e. no modification). Sensitivity of the initialization time on the simulation will be discussed as well, which may have implications for numerical forecasting.