Definitions of tropical dry forest (TDF) and tropical moist forest (TMF) typically rely on coarse spatial resolution climate envelopes that are unable to capture the detailed landscape dynamics essential for understanding their conservation status and protection needs. In addition, TDF and TMF are often separated by short distances, complicating their spatial delimitation using coarse spatial resolution methods. TDF typically exhibits marked seasonality prompted by dry months that determine its deciduous leaf phenology, contrasting starkly with the evergreen phenology of TMF. We present a methodology that leverages those distinct leaf phenologies to map TDF and TMF in Colombia. Our methods use a training dataset derived from a coarse climate envelope representing the conditions where TDF and TMF are expected to occur. This dataset is then used to train a machine learning model in which spatio-temporal features derived from the Continuous Change Detection and Classification algorithm (CCDC) applied to 20-year Landsat time series are used as predictors to capture the differences in leaf phenology between these two forest types. The resulting dataset shows the annual probability of pixels with tree cover being TDF or TMF over the last two decades. We validated our methodology using field plot data and ancillary datasets. Our methodology is flexible to allow accommodating custom parameters to define sub-regional climate envelopes or incorporating data from different optical sensors (e.g. Sentinel-2, Harmonized Landsat Sentinel data).