Optimization of Neural Network Radiation Emulator using Sherpa

Pilkyo Kim, Hwan‐Jin Song

Zenodo (CERN European Organization for Nuclear Research) · 01 April 2022 · Korea Meteorological Administration (KMA)

In one sentence. Source code and weights for a neural network radiation emulator in WRF, with hyperparameters tuned by Sherpa to speed up the radiation scheme.

Abstract

These files contain main source codes for developing neural network radiation emulator and weight/bias coefficients for implementing to the WRF model. Reference: Kim, P. S. & Song, H.-J. (2022), Usefulness of automatic hyperparameter optimization in developing radiation emulator in a numerical weather prediction model, Atmosphere, 13, 721. Source : Main codes for optimizing neural network radiation emulator using Sherpa weight and bias coefficients from neural network training for two control simulations (l1, l2) and Sherpa l1_full : learning rate 0.0001 and batch size 3000000 l1_50000 : learning rate 0.0001 and batch size 50000 l1_5000 : learning rate 0.0001 and batch size 5000 l1_500 : learning rate 0.0001 and batch size 500 l2_full : learning rate 1 and batch size 3000000 l2_50000 : learning rate 1 and batch size 50000 l2_5000 : learning rate 1 and batch size 5000 l2_500 : learning rate 1 and batch size 500 sp_full : Sherpa and batch size 3000000 sp_50000 : Sherpa and batch size 50000 sp_5000 : Sherpa and batch size 5000 sp_500 : Sherpa and batch size 500 Training and validation sets are given in the following link: July: https://doi.org/10.5281/zenodo.5636127 Source codes for control simulations (l1, l2) and neural network radiation emulator for simulating the WRF model are avaliable in the following link: https://doi.org/10.5281/zenodo.5638436 Reference: Song, H.-J., Roh, S., Lee, J., Nam, G., Yun, E., Yoon, J., & Kim, P. S., (2022). Benefits of stochastic weight averaging in developing neural network radiation scheme for numerical weather prediction. Journal of Advances in Modeling Earth Systems, https://doi.org/10.1002/essoar.10508964.2

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.