Fig. 1: A schematic of WenHai’s architecture.
From: Forecasting the eddying ocean with a deep neural network
The surface atmosphere and ocean variables on the day \(t+l\) are first combined to compute the air-sea momentum, heat, and freshwater fluxes based on the bulk formulae (green block), where t is an arbitrary date index and l is the forecast lead time (indexing in 1-day intervals). Then, the ocean variables and air-sea fluxes are sent to a deep neural network (red block) that forecasts temporal tendency between ocean variables on the days \(t+l\) and \(t+l+1\). The tendency field is added to the ocean variables on the day \(t+l\) to yield the ocean variable forecast on the day \(t+l+1\)Â (blue block). Finally, the above processes are iterated to generate a sequence of forecasts. Maps created with Cartopy65. Background land image provided by NASA Earth Observatory.
