Hot droughts are becoming increasingly more frequent across tropical regions, and the future of tropical terrestrial ecosystems will largely depend on how plants of different sizes and functional groups respond to climate extremes. Plants across moist and dry forests have a diverse range of carbon allocation strategies that must balance between allocation to growth and to drought tolerance. One important drought tolerance strategy is drought deciduousness, which reduces the risk of hydraulic failure during long dry seasons, at the expense of allocating more carbon to leaves every time plants enter the growing season. Here, we use an extensive suite of publicly available trait observations across the American tropics, combined with field observations of ecosystem functioning and remote sensing measurements of seasonal variability in leaf phenology to calibrate and benchmark the Functionally Assembled Terrestrial Ecosystem Simulator (FATES) across a very broad precipitation gradient and forest types in the American tropics. We defined multiple tropical tree plant functional types that vary in both leaf phenology (evergreen and drought deciduous) as well different levels of shade tolerance. Current simulation results with FATES qualitatively represent the variability of above-ground biomass, albeit the model tends to underestimate the total carbon stocks at moist tropical forest sites. To reduce biases in the timing of leaf abscission and flushing in deciduous-dominated sites, we are carrying out FATES simulations with fixed forest structure but active phenological cycles, and comparing with time series of gross primary productivity from eddy covariance towers and vegetation greenness based on the European Space Agency’s Sentinel-2 platform. The results so far provide a promising pathway for advancing understanding of PFT coexistence across the tropics using observed traits, in situ observations and remote sensing; however, they suggest that additional data and model process development are needed to quantitatively improve the model predictions.