Comparison of Two Spatiotemporal Reconstruction Methods for Spaceborne Sea Surface Temperature Data at Multiple Temporal Resolutions

The satellite remote sensing sea surface temperature (SST) plays a crucial role in global climate change and ocean–atmosphere interactions. With a notably severe issue of missing data due to clouds and rainfall, data reconstruction methods have been developed to effectively enhance the spatiotemporal completeness of satellite-derived SST data products in recent years. However, few studies have focused on performance comparisons between these different data reconstruction methods, which limits further improvement and application of reconstructed data products. In this study, two representative methods, the data interpolating empirical orthogonal functions (DINEOF) and a spatiotemporal geostatistical method of Bayesian maximum entropy (BME), were used to reconstruct satellite SST data in four regions, and their reconstruction performance under various temporal resolutions (hourly, daily, and monthly) and missing data rates were evaluated and compared. Our results demonstrate that BME consistently outperforms DINEOF. As the missing data rate increases from 10% to 90%, especially when it exceeds 70%, DINEOF reconstruction results exhibit significant increasing noises and reconstruction errors, while BME demonstrates stable precise reconstruction results. Compared with DINEOF method, the results of BME method are less influenced by missing data rates, spatiotemporal resolutions, temporal length, and regions of input data series by different remote sensing sensors, rendering it more applicable and robust in reconstructing multisensor SST data with different temporal resolutions. The BME method holds promising implications in reconstructing high-quality gap-filled data using noisy and high-missing-rate multisensor data in regional areas with high dynamics.