Fast Neural Network Emulation of a Planetary Boundary Layer Parameterization in a numerical weather forecast model

Description

This study describes an emulator developed using deep neural networks for a planetary boundary layer (PBL) parameterization in a numerical weather/climate 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 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 climate/weather models that operates at horizontal spatial scales in the 10’s of kms. Parameterizations in weather/climate models are computationally expensive. We use model output from a set of simulations performed using the Weather Research Forecast (WRF) model to train a series of deep neural network algorithms to evaluate if deep neural networks can provide an alternative to the physics based parameterizations. We demonstrate that certain neural networks can fairly successfully simulate the vertical profiles within the boundary layer of velocities, temperature and water vapor over the entire diurnal cycle. We develop the neural network based on several locations that represent a variate of surface conditions and climate zone over continental United States. We also assess the spatial transferability of the developed neural network for nearby locations. It produced the U, V components of the velocity, temperature and water vapor profiles over the entire diurnal cycle and all locations with errors less the a few percent when compared to the WRF simulations.