The Development and Application of Sea Surface Temperature Prediction Service System in East-Asia Using Synthesis Data Derived from Multi-Satellite Sensors: Supporting Fishing Activities Around the Korean Peninsula

The sea surface temperature (SST), which is a factor of climate change, is an important oceanic variable that affects the oceanic eco system. This variable is also used for various research fields about numerical forecasting and long-term fluctuation characteristics. Depending on the changes in sea surface temperature, the activeness of the fish changes sensitively, and this information is one of most important information for anglers increasing number every year. Also, anglers need the accurate and precise sea surface temperature information in advance to find the best fishing spot. In this study, therefore, we developed the sea surface temperature prediction service system for providing the accurate and precise sea surface temperature data nearby sea area around the Korean peninsula into anglers. This system also is able to provide the future sea surface data in advance using the artificial intelligence (AI) technology. This system contains 5 subsystems: 1) data collection, 2) synthesis data production, 3) sea surface temperature prediction, 4) production management, and 5) production service. First, the data collection subsystem collects data from 6 different infrared sensors-equipped satellites (COMS/MI, GK-2A/AMI, NOAA-20/VIIRS, Suomi-NPP/VIIRS, Terra/MODIS, and Aqua/MODIS), 3 different types of microwave sensors-equipped satellites (GCOM-W1/AMSR2, GPM/GMI, and WindSAT). Also, this subsystem collects in situ data from 3 different Korean agencies (Korea Hydrographic and Oceanographic Agency, Korea Meteorological Administration, and National Institute of Fisheries Science) through file transfer protocol (FTP) or open application programming interface (OpenAPI). Additionally, this subsystem has the quality control module to correct the collected data in real-time. Second, the synthesis data production subsystem produces the sea surface temperature synthesis products with 1 km spatial resolution in East Asia based on the OI (Optimal Interpolation) synthesis algorithm. Third, sea surface temperature prediction subsystem estimates sea surface temperature data after from 1 to 10 days using the deep learning model based on LSTM (Long Short-Term Memory) algorithm. sea surface temperature data for the past 10 years were employed to train this deep learning model. Fourth, The production management subsystem was built using parallel/distributed processing techniques to distribute the predicted sea surface temperature data as quickly as possible. Finally, production service subsystem provides sea surface temperature data as the javascript object notation (JSON) file format. Here, anglers or end-users should send request parameters (i.e., longitude, latitude, date, time) to the OpenAPI server in order to receive sea surface temperature data. To evaluate the accuracy of the synthesis data production subsystem, the produced synthesis data were compared with in situ buoy data from KMA for the past 10 years (2011–2020) in terms of root mean square error (RMSE). The result of RMSE was 0.96 around the sea area located in 37.45°N and 131.11°E. Also, the accuracy of sea surface temperature prediction subsystem was compared as similar methodology above. The result of RMSE for the 1-day prediction was 0.63 around the sea area located in 37.45°N and 131.11°E. Meanwhile, large-capacity satellite image processing and high-speed calculation are possible through high-performance server (128 core) and Infinite band network configuration, and a stable and accurate operating system has been built to synthesis and prediction data production. Therefore, the produced data is able to transmitted to the end-users in real-time. For the future work, we will provide additional angler-aided products such as tide, bottom material, water depth, and so on.