A quarter century after the first global coral bleaching event in 1998, reports differ on the relative importance of anthropogenic influences, local environment and bleaching recurrence in determining the resilience of coral reefs. While life history traits largely determine how corals respond to temperature anomalies, it is unclear if these traits also determine how corals fare over time.  From 1998 to 2022, we tracked compositional changes in reefs across the Lakshadweep Archipelago to explore how global El Niño events, and local environment (wave climate and depth) influenced coral responses to repeated mass bleaching. From the 1998 to the 2016 bleaching event, the magnitude of coral mortality reduced overall, particularly at deeper reefs (shallow: -38% to -3%; deep: -18% to -0.45%). Post-bleaching recovery correlated positively with higher wave exposure, linked to the creation of stable structures for coral settlement and survival. Across bleaching phases, recovery was initially slow (6-7 years post-mortality), but, given time, showed a much steeper increase, led by space-occupying genera like Acropora. However, recurring mass bleaching maintained coral cover low (~15% across all sites). These broad trends mask dynamic compositional patterns. Genera such as Porites, Pocillopora, and Favia declined less through time compared to Acanthastrea, Turbinaria, Psammocora and Plesiastrea among others. We identified six community clusters that describe contrasting long-term responses to local and global factors, mediated by depth and wave exposure. Interestingly, genera with different functional traits cluster together indicating that bleaching susceptibility interacts with depth and exposure, creating a spatial mosaic of coral assemblages. These clusters serve as a predictive, site-specific framework to understand the dynamically shifting but declining assemblage of Lakshadweep reefs. While local management could help maintain this changing composition, urgent global action is needed to secure the long-term ecological integrity of tropical reefs.

We sampled each site at deep (~16 m) and shallow (~8 m) depth classes, where we estimated coral cover in 1x1 m quadrats, placed at a minimum of 10-15 m of distance from one another. NAs pertaining to the depth column are for years when the depth measurement was not noted and instead characterised by deep/shallow. Each depth class therefore had, on an average, 18 quadrats, corresponding to 18 m² of area. Each quadrat is assumed to be independent of one another and from our unit of sampling. Data were recorded either directly in-situ (1998-2003) or with digital photographs (2007 onwards). Photoquadrats were analysed with ImageJ (Schneider et al., 2012), where the percent cover of broad benthic categories were estimated in overlaid 10 x 10 cm grids. NAs in the 'Photo_ID' column indicate inwater surveys where photography was not possible. This is largely restricted to the years 1998 & 1999. Live coral was identified to the genus, based on Veron’s (2000) classification (Supplementary Table 1 for new classification; surveys in 1998 and 1999 did not record genus-level information). All samples were collected within the months of January – March, before the peak summer season in the Indian subcontinent. We analyse changes in coral composition from 1998 to 2022 and present data for 13 years across this 24-year period (Supplementary Table 2 for details on sampling period).