Impact of Assimilation on Hindcast and Forecast Performance in Small Nearshore Domains

Thomas G Almeida, Arie L Reath and David T Walker, SRI International, Ann Arbor, MI, United States
Abstract:
Data assimilation techniques can be used to obtain enhanced estimates of the ocean state by combining observational data with physics-based hydrodynamic models. In general, the two most significant results of applying such techniques are a hindcast that accurately depicts the ocean state during the observational period, and a nowcast that can be used to initialize a forecast. In this work, variational inverse modeling has been applied to estimate nearshore circulation over a relatively small area of interest (AOI) using velocity data. The observational data used are from the High-Frequency Radar Network (HFRNet) hosted by Scripps Institution of Oceanography’s Coastal Observing Research and Development Center. A high-resolution, fully three-dimensional hydrodynamic model (Delft3D) is used to model the nearshore ocean state, including buoyancy. An initial estimate of the circulation is obtained via nesting Delft3D in the Navy’s Southern California regional NCOM model. The AOI is a 22km by 16km region encompassing San Diego Bay, CA. We wish to minimize the error between modeled and observed velocity during a hindcast period, subject to the constraint that the estimated velocity field is a valid solution to the governing equations. We then use that hindcast to initialize a forecast of the ocean state. The time period for this study is 48 hours beginning on June 01, 2014. The hindcast period (during which the data are collected) is the first six hours, and the remainder of the period is the forecast. While the increased fidelity in the high-resolution nested model results in a better comparison to the data (~10% reduction in RMSE relative to NCOM), the assimilation further reduces the error by ~20% in the hindcast period; this impact rapidly decays into the forecast period over the first 3-6 hours. The forecast is also analyzed to examine the impact and persistence of the data assimilation on the forecast by comparing the modeled circulation obtained through simple nesting and through assimilation, where localized differences in flow behavior persist longer (~6-12 hours). We further discuss the impact of using a larger region to conduct the assimilation in an effort to enact a more significant influence on the smaller high-resolution forecast. Supported by ONR Contract Number N00014-12-C-0237.