Statistical validation of modis-based sea surface temperature in shallow semi-enclosed marginal sea: A comparison between direct matchup and triple collocation

Abstract

Validating remotely sensed sea surface temperature (SST) is a fundamental step in establish-ing reliable biological/physical models that can be used in different marine applications. Mapping SST using accurate models would assess in understanding critical mechanisms of marine and coastal zones, such as water circulations and biotic activities. This study set out to validate MODIS SSTs with a spatial resolution of 1-km in the Arabian Gulf (24–30◦ N, 48–57◦ E) and to assess how well direct comparison of dual matchups and triple collocation analyses perform. For the matchup process, three data sets, MODIS-Aqua, MODIS-Terra, and iQuam, were co-located and extracted for 1-pixel box centered at each actual in situ measurement location with a time difference window restricted to a maximum of ±3 h of the satellite overpass. Over the period July 2002 to May 2020, the MODIS SSTs (N = 3786 triplets) exhibited a slight cool night-time bias compared to iQuam SSTs, with a mean ± SD of −0.36 ± 0.77◦C for Aqua and −0.27 ± 0.83◦C for Terra. Daytime MODIS SST observations (N = 5186 triplets) had a lower negative bias for both Aqua (Bias = −0.052◦C, SD = 0.93◦C) and Terra (Bias = −0.24◦C, SD = 0.90◦C). Using extended triple collocation analysis, the statistical validation of system-and model-based products against in situ-based product indicated the highest ETC-based determination coefficients (ρ2t,X≥0.98) with the lowest error variances (σ2ε ≤ 0.32), whereas direct comparison underestimated the determination coefficients and overestimated the error estimates for all MODIS algorithms. The ETC-based error variances for MODIS Aqua/Terra NLSSTs were 0.25/0.19 and 0.26/0.32 in daytime and night-time, respectively. In addition, MODIS-Aqua was relatively more sensitive to the SST signal than MODIS-Terra at night and vice versa as seen in the unbiased signal-to-noise ratios for all observation types.