Application of optimal interpolation to vertical current data in the Ieo Island of Korea using unstructured grid ocean numerical model(FVCOM)

초록

The data assimilation in the field of oceanography is an estimation problem for producing model parameters or initial states for ocean predictions. Optimal interpolation (OI) is a least square method to estimate the initial conditions by calculating the weight where the error covariance between model values and observations becomes minimized. OI is a statistical data assimilation method and thus is influenced by the number of observation data. Although theoretical framework of OI application in ocean modeling is now relatively well established, OI is still quite useful methodology in ocean modeling, because OI can be applied to any local region of the model domain, requiring relatively less computational cost compared to vibration and ensemble-based methods. In this study, we assimilate the vertically distributed ocean current data at Ieo Island and South-western part of Korea via OI technique into the Finite Volume Coastal Ocean Model (FVCOM), which is the unstructured grid ocean numerical model. FVCOM is first setup for tidal simulation in the region of Korea peninsula (See Fig. 1). The complex and irregular geographic characteristics of the coastline of Korea were complemented with unstructured grid, and the 30 second grid water depth (KorBathy30s) of the seas near Korea provided by Korea Ocean Research & Development Institute (KORDI), and NAO99 was used for tidal boundary conditions. The verification of tide model was conducted based on the surface measurement data at sea level observation station operated by National Oceanographic Research Institute (NORI). Before applying the data assimilation to the model domain, the various kinds of numerical experiment were conducted to calculate the weight for the model and the observation data calculated previously in the observation station. The scaling factor and correlation radius based on the numerical experiment were used to calculate the weight of observation and model data in the domain. Based on the estimated weig