Abstract. Parameterizations for physical processes in weather and climate models are computationally expensive. We use model output from a set of simulations performed using the Weather Research Forecast (WRF) model to train deep neural networks and evaluate whether trained models can provide an accurate alternative to the physics-based parameterizations. Specifically, we develop an emulator using deep neural networks for a planetary boundary layer (PBL) parameterization in the WRF model. PBL parameterizations are commonly used in atmospheric models to represent the diurnal variation of the formation and collapse of the atmospheric boundary layer – the lowest part of the atmosphere. The dynamics of the atmospheric boundary layer, mixing and turbulence within the boundary layer, velocity, temperature, and humidity profiles are all critical for determining many of the physical processes in the atmosphere. PBL parameterizations are used to represent these processes that are usually unresolved in a typical numerical weather model that operates at horizontal spatial scales in the tens of kilometers. We demonstrate that a domain-aware deep neural network, which takes account of underlying domain structure that are locality specific (e.g., terrain, spatial dependence vertically), can successfully simulate the vertical profiles within the boundary layer of velocities, temperature, and water vapor over the entire diurnal cycle. We then assess the spatial transferability of the domain-aware neural networks by using a trained model from one location to nearby locations. Results show that a single trained model from a location over the midwestern United States produces predictions of wind components, temperature, and water vapor profiles over the entire diurnal cycle and all nearby locations with errors less than a few percent when compared with the WRF simulations used as the training dataset.