A Deep Learning Parameterization for Ozone Dry Deposition Velocities

The loss of ozone to terrestrial and aquatic systems, known as dry deposition, is a highly uncertain process governed by turbulent transport, interfacial chemistry, and plant physiology. Recent work has shown that current theoretical parameterizations struggle to accurately predict the correct values of ozone dry deposition, with biases frequently greater than 40%. We demonstrate the value of using Deep Neural Learning in predicting ozone dry deposition velocities. We find that a feedforward deep learning neural network trained on independent observations can predict hourly ozone dry deposition velocities at a mixed forest site in New England far more accurately than modern theoretical models, with a substantial reduction in the Normalized Mean Bias of (0.05 as compared to 0.32) while maintaining similar skill in capturing the observed variability (r = 0.35). This accuracy can be achieved with just 6 months of training observations, and represents an opportunity for improving surface-atmosphere process representations in atmospheric models.