A Novel Machine Learning Method for Accelerated Modeling of the Downwelling Irradiance Field in the Upper Ocean

Abstract

The downwelling irradiance is a crucial part of the air-sea heat flux and a major energy source for the photosynthesis processes in the upper ocean. The modeling of the irradiance field is often time-consuming in conventional Monte Carlo (MC) simulations. We propose an accelerated model based on machine learning to significantly reduce the computational cost. By introducing a generalized beam spreading function, we transform the raw data generated with the MC simulation into training data of reduced dimensions, which is then used to develop an artificial neural network. To validate the machine learning model, we use the MC model to simulate the irradiance field under a broadband wave field. For both clear and turbid seawater, the irradiance field predicted by the machine learning model agrees well with the MC simulation result. Compared with the MC simulation that is computationally expensive, our machine learning model is O(1,000) times faster.