Extended Reconstructed Sea Surface Temperature, Version 6 (ERSSTv6). Part I: An Artificial Neural Network Approach
Boyin Huang, Xungang Yin, Tim Boyer, Chun‐Ying Liu, Matthew J. Menne, Yuhan Douglas Rao, Thomas M. Smith, Russell S. Vose, Huai‐Min Zhang
Abstract
Abstract NOAA’s Extended Reconstructed Sea Surface Temperature (ERSST) is an operational global SST product based on in situ observations, which has been widely used in monitoring and assessing global ocean climate particularly El Niño–Southern Oscillation (ENSO) events. ERSSTv5 and its predecessors, however, encountered two shortcomings: 1) low SST spatial variabilities in the data-sparse regions before the 1950s and 2) low-performance scores against in situ observations after the 1970s. The first problem has been mitigated in this Part I study of ERSSTv6 by removing a 3-month running average and applying an interpolation method using an artificial neural network (ANN). The improvements of the ANN method over an empirical orthogonal teleconnection (EOT) method used in previous versions were assessed against validation, testing, and observation datasets. In comparison with ERSSTv5, the spatial correlation coefficient (SCC) with reference to observations increases by 5%, and root-mean-square difference (RMSD) with reference to observations decreases by 0.03°C in ERSSTv6. The improvements of SCC and RMSD are more pronounced in the tropical Pacific and the Southern Hemisphere oceans between 60° and 30°S. The second problem has been addressed separately in our Part II study. Significance Statement ERSSTv6 has been upgraded from ERSSTv5 by implementing an interpolation method using an artificial neural network (ANN). In comparison with ERSSTv5, ERSSTv6 has higher spatial coherence and lower root-mean-square error in the global oceans from 1850 to 2021.