A deep learning approach to fast radiative transfer

• A statistical regression approach for fast transmittance modeling is proposed. The structure of the line-by-line radiative transfer process outlined. Two deep learning, hidden-layer neural networks are compared as models according to their performance. Fitting accuracy and computation speed shown be excellent. Due sheer volume data, leveraging satellite instrument observations effectively in a data assimilation context numerical weather prediction or remote sensing requires model an observation operator that both accurate at same time. Physics-based (RT) fulfil requirement accuracy, but too slow costly computational terms operational applications. Therefore, methods were developed able perform RT calculations using techniques such spectral sampling pre-computed look-up tables. currently calculate absorption scattering coefficients from atmospheric state cloud profiles. As novel solution this problem, work investigates learning replace models. selection network configurations trained against profile computed by performance evaluated advantages disadvantages discussed.