Bias-Aware Numerical Surface Temperature Prediction System in Cheonsu Bay during Summer and Sensitivity Experiments

Abstract

A real-time numerical prediction system was developed to predict sea surface temperature (SST) in Cheonsu Bay to minimize damages caused by marine heatwaves. This system assimilated observation data using an ensemble Kalman filter and produced 7-day forecasts. Bias in the temperature forecasts were corrected based on observed data, and the bias-corrected predictions were evaluated against observations. Using this real-time numerical prediction system, daily SSTs were predicted in real-time for 7 days from July to August 2021. The forecasted SSTs from the numerical model were adjusted using observational data for bias correction. To assess the accuracy of the numerical prediction system, real-time hourly surface temperature observations as well as temperature and salinity profiles observed along two meridional sections within Cheonsu Bay were compared with the numerical model results. The root mean square error (RMSE) of the forecasted temperatures was 0.58°C, reducing to 0.36°C after bias-correction. This emphasizes the crucial role of bias correction using observational data. Sensitivity experiments revealed the importance of accurate input of freshwater influx information such as discharge time, discharge volume, freshwater temperature in predicting real-time temperatures in coastal ocean heavily influenced by freshwater discharge. This study demonstrated that assimilating observational data into coastal ocean numerical models and correcting biases in forecasted SSTs can improve the accuracy of temperature prediction. The prediction methods used in this study can be applied to temperature predictions in other coastal areas.