Most land surface models (LSMs) used in Earth System Models produce a lower ratio of transpiration (T) to evapotranspiration (ET) than field observations, degrading the credibility of Earth System Model-projected ecosystem responses and feedbacks to climate change. To interpret this model deficiency, we conducted a pair of model experiments using a three-dimensional, process-based ecohydrological model in a subhumid, mountainous catchment. One experiment (CTRL) describes lateral water flow, topographic shading, leaf dynamics, and water vapor diffusion in the soil, while the other (LSM like) does not explicitly describe these processes to mimic a conventional LSM using artificially flattened terrain. Averaged over the catchment, CTRL produced a higher T/ET ratio (72%) than LSM like (55%) and agreed better with an independent estimate (79.79 ± 27%) based on rainfall and stream water isotopes. To discern the exact causes, we conducted additional model experiments, each reverting only one process described in CTRL to that of LSM like. These experiments revealed that the enhanced T/ET ratio was mostly caused by lateral water flow and water vapor diffusion within the soil. In particular, terrain-driven lateral water flows spread out soil moisture to a wider range along hillslopes with an optimum subrange from the middle to upper slopes, where evaporation (E) was more suppressed by the drier surface than T due to plant uptake of deep soil water, thereby enhancing T/ET. A more elaborate representation of water vapor diffusion from a dynamically changing evaporating surface to the height of the surface roughness length reduced E and increased the T/ET ratio.