Error Characterization of Satellite SSS Products Based on Extended Collocation Analysis

Abstract

Triple collocation analysis is an effective method for assessing the error of observation systems with independent random error. However, this error independence is often violated, and systems show clear error correlations. Whereas many efforts have been devoted, the cause and determination of this error correlation are still challenging. In this work, we reveal that the error correlation is actually a signal observed by some systems, and taking this signal as the error correlation between systems will bias the error estimation results. Therefore, we define it as the representativeness signal instead of the representativeness error. Based on the multiscale signal approach and extended collocation analysis, we discuss the cause of this representativeness signal and its effect on sea surface salinity (SSS) data validation by synthesized experiments, satellite, buoy, and climatology products. The results suggest that the representativeness signal may vary with the different behaviors of observed variables. For SSS, the representativeness signal lies in satellite data with medium resolutions and similar representativeness. However, the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">in situ</i> and climatology data, which have different resolutions compared with satellites, do not show apparent representativeness signals with satellite data. The data collocation procedure also affects the representativeness signal, which decreases with increasing collocation intervals. The extended collocation analysis can estimate the error of SSS products in most cases but may not provide robust estimation when the collocation data pairs are limited.