Bayesian maximum entropy interpolation of sea surface temperature data: A comparative assessment

ABSTRACT

Sea surface temperature (SST) is an important oceanography attribute that has been used to study ocean climatic conditions, ocean dynamics and air-sea interactions. In the present work, the Bayesian maximum entropy (BME) method was used to interpolate a FY-3 C /VIRR satellite dataset in the region with coordinates 120°-150°W and 45°-60°S and during January 2020. A novel approach of constructing valuable soft data was developed by combining BME interpolation with highly correlated SST day and night data differences. The BME interpolation accuracy was assessed by cross-validation, and the results showed that the average RMSE (root mean squared error) was 0.700 and the average bias was 0.441°C. Furthermore, using the Argo data as a basis of comparison, the coverage and accuracy of the BME interpolation of the FY-3 C/VIRR satellite SST dataset were compared numerically with those of the Ordinary Kriging (OK) interpolation of the FY-3 C/VIRR satellite SST dataset and the Optimum Interpolation Sea Surface Temperature (OISST) of the AVHRR satellite SST dataset. It was found that BME had the best SST interpolation performance among the three methods with the lowest average bias and the largest correlation coefficient. Although OISST had a full product coverage rate overall (due to its use of more perfect treatment means), BME’s coverage rate (97.5%) considerably improved that of the FY-3 C/VIRR SST data. Also, both the BME and OK products maintained a 12hrs temporal resolution and a 0.05 decimal degrees longitude/latitude spatial resolution, which is an improvement over OISST data with a 24hrs time resolution and a 0.25 decimal degrees longitude/latitude spatial resolution. Another advantage of BME is that because of its broad theoretical support its performance in practice can be improved further as more knowledge sources become available (which can only be incorporated by BME).

Acknowledgements

This work was supported by the China Postdoctoral Science Foundation (2020M681825).

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the China Postdoctoral Science Foundation [2020M681825].