yaoyichen / radiationNet

Overall Architecture

This repo is the implementation of the article "Physics-Incorporated Framework for Emulating Atmospheric Radiative Transfer and the Related Network Study". We propose a physics-incorporated framework for the radiative transfer DL model, in which the physical relationship between fluxes and heating rates is encoded as a layer of the network. It is also found that the prediction accuracy was improved with the physic-incorporated layer. In addition, we trained and compared various types of deep learning model architectures, including fully connected (FC) neural networks (NNs), convolutional-based NNs (CNNs), bidirectional recurrent-based NNs (RNNs), transformer-based NNs, and neural operator networks, respectively.

Code Structure

The code description is as follows:

• main program: main.py
• ./models/ network structure folder
• ./utils/ utils folder
• ./checkpoints/ saved checkpoints
• ./climart extra utils that helps to train on the climart dataset, not required in this paper

How to train the model

The following is an example script of running the LSTM model.

./scripts/run_script.sh

The specific script file is as follows. The model structure can be selected by the argument "model_name", and the avialable model structures are "FC","RES","UNET","LSTM","GRU","FNO","ATT" etc, and the detailed description of each model can be found in our paper.

python main_fullyear.py \
--main_folder "FullYear" \
--sub_folder "FullYear_V01" \
--train_file "./dataset/fullyear_trainset_240_10_useful.nc" \
--train_point_number 15943 \
--test_file "./dataset/fullyear_trainset_240_10_useful.nc" \
--test_point_number 15943 \
--prefix "fullyear_V01" \
--dataset_type "FullYear" \
--loss_type "v01" \
--learning_rate  1e-3  \
--batch_size 10 \
--model_name "LSTM" \
--num_workers 0 \
--num_epochs 600 \
--save_mode "True" \
--save_checkpoint_name "model" \
--save_per_samples 10000 \
--load_model "False" 

The figures shown in the paper can be generated using the code in the paper_plot folder. The vertical profiles of the statistics can be generated by plot_height_distribution.py. The vertical profiles of the liquid water distribution of the 3 typical case:no liquid cloud, single-layer liquid cloud , and multi-layer liquid cloud can be generated by plot_instaneous.py

License

This project is under the CC-BY-NC 4.0 license. See LICENSE for details.