GK2A 자료를 이용한 약한 지도 학습 기반의 전천(All-Sky) 해수면온도 산출

Satellite-derived sea surface temperature (SST) data based on infrared imagery are crucial for weather forecasting and meteorological analysis. However, SST observations are limited in cloudy regions. This study proposes an all-sky retrieval method using an artificial intelligence (AI) model to address this issue. We trained the model using weakly supervised learning with GEO-KOMPSAT-2A (GK2A) satellite data and Global Telecommunications System (GTS) buoy SST data collected from June 1 to August 31 in 2021 and 2022, designing six different models based on experimental conditions. The analysis showed that incorporating GK2A Level 2 (L2) SST into the loss function resulted in more accurate simulation of SST distribution by latitude and reduced overall errors. Additionally, the root mean square error (RMSE) of the AI model SST for the all-sky region compared to the buoy SST is from 1.41 to 1.56°C, and the bias is from –0.17 to –0.48°C. The final model demonstrated high accuracy compared to the Operational Sea Surface Temperature and Ice Analysis (OSTIA), achieving an RMSE of 0.64°C and a bias of –0.42°C. This study successfully develops an all-sky SST retrieval technique capable of estimating SST in cloudy regions, overcoming the limitations of conventional satellite-based SST retrieval and establishing a foundation for SST data production applicable to meteorological and oceanographic forecasting research.