This repository contains a Fortran implementation of the neural-network models developed in Cummins et al. (2023) to parameterise bulk turbulent fluxes of momentum, sensible heat and latent heat above ocean and/or sea ice. The methodology was originally developed in Cummins et al. (2024).
Cummins, D. P., Guemas, V., Cox, C. J., Gallagher, M. R., & Shupe, M. D. (2023). Surface turbulent fluxes from the MOSAiC campaign predicted by machine learning. Geophysical Research Letters, 50(23), e2023GL105698. https://doi.org/10.1029/2023GL105698
Cummins, D. P., Guemas, V., Blein, S., Brooks, I. M., Renfrew, I. A., Elvidge, A. D., & Prytherch, J. (2024). Reducing parametrization errors for polar surface turbulent fluxes using machine learning. Boundary-Layer Meteorology, 190(3), 13. https://doi.org/10.1007/s10546-023-00852-8
-
The code is packaged as a free-standing Fortran subroutine with hard-coded parameters and no dependencies. It has been tested using the GFortran compiler.
-
If NumPy is installed then f2py can be used to compile the code into a Python module. E.g.,
f2py -c -m fluxes_mod fluxes.f90. This will generate a shared object file that can be loaded as if it were a Python module.