Projecting the potential impacts of LULC (Land Use/Land Cover) change on watershed hydrological response is critical for water management decisions in a changing environment. An improved representation of vegetation dynamics is needed to improve the capability of several hydrological models to produce reliable projections of these impacts. Here we in troduce a modification in the plant growth module of SWAT (Soil Water Assessment Tool) to improve the representation of the bimodal seasonality of LAI (Leaf Area Index), which is particularly important for tropical watersheds with bimodal precipitation regimes. The new SWAT-Tb variant that we propose here reproduces not only observed streamflow, but also the bimodal seasonal pattern of LAI in a tropical mountain watershed of the Andes. In contrast, standard SWAT is inherently unable to reproduce this bimodality, although it can be calibrated to reproduce streamflow. Differences between models in the representation of LAI seasonality can lead to significantly different results about LULC change impacts on streamflow. SWAT-Tb results show that deforestation impacts on streamflow are more pronounced for seasonal than for annual streamflow, and indicate that forests can play a crucial role in enhancing water availability during dry seasons. The seasonality of streamflow anomalies is switched due to forest-to-pasture conversion, implying that while forest expansion increases water availability in dry seasons, forest conversion into pasture decreases it. Due to its poor representation of LAI seasonality, standard SWAT largely underestimates this role of forest, which can be misleading for decision making about water security and forest conservation