<p>To refine the assumption that the contrast threshold of the human eye (<i>C</i>_rt) remains constant in general visibility theory, we integrated an artificial intelligence-based approach with a mechanistic model (AI-SD) to derive the equivalent contrast threshold at 488 nm (<i>C</i>₄₈₈) from remote sensing reflectance on a pixel-by-pixel basis. The derived <i>C</i>₄₈₈ was then applied to estimate Secchi depth (<i>Z</i>_sd). We trained and validated the AI-SD model using an extensive field-measured dataset (<i>N</i> = 1 577) encompassing oceanic, coastal, and inland waters and compared its performance with a traditional mechanistic model. Our findings indicate that <i>C</i>₄₈₈ theoretically ranges from 1.85 × 10⁻⁵ sr⁻¹ to 0.138 sr⁻¹ and improves the accuracy of <i>Z</i>_sd estimates from field-measured or satellite-derived remote sensing reflectance (<i>R</i>_rs) by over 10% compared to the traditional model. Furthermore, applying the AI-SD model to global <i>R</i>_rs data revealed that <i>C</i>₄₈₈ exhibits spatial and temporal variability across the world’s oceans. We linked this variability to the observed, yet traditionally unexplained, non-monotonic relationship between <i>Z</i>_sd and solar zenith angle-related water-leaving radiance. These results highlight the potential for enhancing global water transparency monitoring via ocean color satellites by incorporating accurate, pixel-level <i>C</i>₄₈₈ values.</p>