Comprehensive assessments of coastal biodiversity in complex coral communities are crucial but challenging, particularly under unfavorable conditions such as poor underwater visibility in urbanized and eutrophic environments. Here we aim to examine the scope of underwater diversity detection and environmentally driven community shifts across environmental gradients in Hong Kong SAR, a highly urbanized coastal city with limited underwater visibility. We employ and compare two methods: 12S rRNA eDNA metabarcoding coupled with custom built reference database and simultaneous extensive underwater visual census (UVC) surveys. eDNA detected a higher species richness per site. Yet, each survey method featured a distinct species profile, with 98 (32.3%) and 120 (39.6%) species found exclusively by UVC and eDNA only respectively. eDNA featured species from diverse habitats and evolutionary distances, including cryptic and large fishes with high mobility. Even for the concurrently detected species, half of them showed distinct detection probabilities between methods as revealed by occupancy models. eDNA also recorded 90 putative species that had never been recorded in 7 year long UVC, with a few prospective new species to the territorial waters. Both methods captured similar patterns of community spatial structure across environmental gradient, while only eDNA detected an expected positive correlation between large species richness and distance from the closest fishing homeport. This suggests inshore overfishing but also incapability of UVC in surveying large mobile species in turbid environments. Considering the discrepancies between two methods, we highlight the importance of complementing both UVC and eDNA metabarcoding survey for a complete overview of local biodiversity, especially under unfavored underwater conditions in an urbanized seascape.