Mangroves are a group of woody plants that occur in the dynamic tropical and subtropical intertidal zones. Mangrove forests offer numerous ecosystem services (e.g. nutrient cycling, coastal protection and fisheries production) and support costal livelihoods worldwide. Rapid environmental changes and historical anthropogenic pressures have turned mangrove forests into one of the most threatened and rapidly vanishing habitats on Earth. Yet, we have a restricted understanding of how these pressures have influenced mangrove abundance, composition and functions, mostly due to limited availability of mangrove field data. Such knowledge gaps have obstructed mangrove conservation programs across the tropics. This thesis focuses on the plants of Earth’s largest continuous mangrove forest — the Sundarbans — which is under serious threat from historical and future habitat degradation, human exploitation and sea level rise. Using species, environmental, and functional trait data that I collected from a network of 110 permanent sample plots (PSPs), this thesis aims to understand habitat preferences of threatened mangroves, to explore spatial and temporal dynamics and the key drivers of mangrove diversity and composition, and to develop an integrated approach for predicting functional trait responses of plants under current and potential future environmental scenarios. I found serious detrimental effects of increasing soil salinity and historical tree harvesting on the abundance of the climax species Heritiera fomes. All species showed clear habitat preferences along the downstream-upstream gradient. The magnitude of species abundance responses to nutrients, elevation, and stem density varied between species. Species-specific density maps suggest that the existing protected area network (PAN) does not cover the density hotspots of any of the threatened mangrove species. Using tree data collected from different salinity zones in the Sundarbans (hypo-, meso-, and hypersaline) at four historical time points: 1986, 1994, 1999 and 2014, I found that the hyposaline mangrove communities were the most diverse and heterogeneous in species composition in all historical time points while the hypersaline communities were the least diverse and most homogeneous. I detected a clear trend of declining compositional heterogeneity in all ecological zones since 1986, suggesting ecosystem-wide biotic homogenization. Over the 28 years, the hypersaline communities have experienced radical shifts in species composition due to population increase and range expansion of the disturbance specialist Ceriops decandra and local extinction or range contraction of many endemics including the globally endangered H. fomes. Applying habitat-based biodiversity modelling approach, I found historical tree harvesting, siltation, disease and soil alkalinity as the key stressors that negatively influenced the diversity and distinctness of the mangrove communities. In contrast, species diversity increased along the downstream – upstream, and riverbank — forest interior gradients, suggesting late successional upstream and forest interior communities were more diverse than the early successional downstream and riverbank communities. Like the species density hotspots, the existing PAN does not cover the remaining biodiversity hotspots. Using a novel integrated Bayesian modelling approach, I was able to generate trait-based predictions through simultaneously modelling trait-environment correlations (for multiple traits such as tree canopy height, specific leaf area, wood density and leaf succulence for multiple species, and multiple environmental drivers) and trait-trait trade-offs at organismal, community and ecosystem levels, thus proposing a resolution to the ‘fourth-corner problem’ in community ecology. Applying this approach to the Sundarbans, I found substantial intraspecific trade-offs among the functional traits in many tree species, detrimental effects of increasing salinity, siltation and soil alkalinity on growth related traits and parallel plastic enhancement of traits related to stress tolerance. My model predicts an ecosystem-wide drop in total biomass productivity under all anticipated stress scenarios while the worst stress scenario (a 50% rise in salinity and siltation) is predicted to push the ecosystem to lose 30% of its current total productivity by 2050. Finally, I present an overview of the key results across the work, the study’s limitations and proposals for future work.