Remote sensing has become a valuable tool for mapping seagrass, but several limitations hinder its full potential in turbid tropical regions, being generally restricted to the intertidal zone. One key challenge is the dynamic nature of the intertidal zone, as seagrass meadows are only fully exposed during short periods of time throughout the year, and therefore may not be visible from remote sensing imagery due to the temporal and spatial limitations in satellite revisit times. Intertidal seagrasses also exhibit high spatial variability, with seagrass often consisting of sparse meadows in muddy substrate or occurring in fragmented patches that are difficult to capture with low-resolution imagery. This study aims to evaluate alternative modelling strategies to advance the field of tropical seagrass mapping, including the use of 1) imagery captured at low tides; 2) seasonal cloud-free composites to reduce pixel variability of the intertidal zone; 3) different sensors (trade-off between spatial resolution and spectral and temporal availability); and 4) different machine learning models (pixel vs object-based classification). We monitor the performance of each approach using an independent validation dataset at 3 study sites with different seagrass habitat characteristics and present the most accurate approach as a probability map of seagrass presence.