Publications

Gao, ZK; Jiang, YT; He, JY; Wu, JP; Christakos, G (2023). Comparing eight remotely sensed sea surface temperature products and Bayesian maximum entropy-based data fusion products. SPATIAL STATISTICS, 54, 100741.

Abstract
Sea surface temperature is an important oceanographic physical attribute. In this work, 5 kinds of infrared SST data sets (in-cluding FY3C-VIRR, MetopB-AVHRR3, MODIS-Aqua, MODIS-Terra, SNPP-VIIRS) and 3 kinds of microwave SST data sets (including AMSR-2, GMI, WindSat) were considered for fusion purposes. The area with geographical coordinates ranging from 100 degrees W to 180 degrees W and from 10 degrees N to 10 degrees S was chosen as the study area. Using the Bayesian maximum entropy (BME) method, four SST fusion products were generated from January 1 to January 31, 2020, with temporal resolution of 1 day, spatial resolution of 0.25 decimal degrees, and coverage rate of 100%. The Holdout cross -validation method was employed to evaluate the performance of BME for SST interpolation purposes by using two data sets, i.e., WinSat SST and MODIS-Terra SST data; and the valida-tion results showed that the mean absolute error (MAE) values of the WindSat SST data and the MODIS-Terra SST data were 0.117 mg/m3 and 0.191 mg/m3, respectively, and the root mean square error (RMSE) values were 0.158 mg/m3 and 0.321 mg/m3, respectively. In addition, four SST interpolation products were compared with the Optimum Interpolation Sea Surface Temper-ature (OISST) and the Group for High-Resolution Sea Surface Temperature (GHRSST) products by combining Argo data with Buoy data treated as in-situ measured data sets. Three (MW-WindSat SST product, ALL-Terra SST product and ALL-WindSat SST product) of the four interpolation products performed better than the OISST and GHRSST data sets. Among them, the ALL-WindSat SST product considering WindSat SST data and other 7 SST data as hard and soft data exhibited the best performance in terms of all three indicators: the coefficient of determination value (0.744), the MAE value (0.82 mg/m3) and the RMSE value (1.082 mg/m3). In sum, the present work made full use of the inclusiveness and integration abilities of the BME method for various types of data, and demonstrated its high accuracy in a number of real world data, thus demonstrating its considerable potential in the context of marine data interpolation and fusion. (c) 2023 Elsevier B.V. All rights reserved.

DOI:
10.1016/j.spasta.2023.100741

ISSN: