Earth’s climate and water cycle are highly dependent on the latent heat flux (LE) associated with terrestrial evapotranspiration. While the widely-used Penman-Monteith LE model is useful to explore vegetative controls on LE, land-atmosphere interactions are difficult to interpret due to the complex role of biological controls on underlying physical processes. Here, we present a novel LE model that defines LE as a combination of diabatic (heat-driven) and adiabatic (relative humidity (rh) gradient-driven) processes using only abiotic variables. This approach yields new insights on the fundamental characteristics of LE. Here we show that the ratio of LE to available energy is mainly controlled by vertical rh gradients, but the spatiotemporal variability in rh gradients are small due to equilibration at the land-atmosphere boundary. Consequently, the global mean vertical rh gradient is near zero, implying land-atmosphere equilibrium at the global-scale. As a result, the spatiotemporal variability of LE is largely determined by the diabatic term, which can be readily determined by standard meteorological measurements. Our proposed model and findings provide a fundamental benchmark for LE predictions. By demonstrating how land surface conditions become encoded in the atmospheric state, our model will also help to improve our understanding of Earth’s climate system and water cycle.