Published December 5, 2023 | Version 1.0.0
Software Open

Climate Model Downscaling in Central Asia: A Dynamical and a Neural Network Approach

  • 1. Potsdam-Institut für Klimafolgenforschung eV

Description

This is the repository for the codes and input/output datasets used for the "Climate Model Downscaling in Central Asia: A Dynamical and a Neural Network Approach" paper published at Geoscientific Model Development Discussion. For a comprehensive description of the methods see (Harder et. al., 2022 and Fallah et. al., 2023 ). 

This repository contains: 

  1. A Jupyter Notebook showing the workflow of the work used in the paper "Climate Model Downscaling in Central Asia: A Dynamical and a Neural Network Approach" [Climate_Model_Downscaling_GMD-main.zip].
  2. List of analysed CMIP6 simulations  [model_lists.pdf]. 
  3. COSMO-CLM model ste-up files [cclm_setups.zip]. 
  4. Snapshot of the code as used in the paper [constrained-downscaling.zip].
  5. Input/output, as well as, trained CNN models, which could also be downloaded by the Jupyter notebook of Climate_Model_Downscaling_GMD-main.zip as following:

               -    input_test.pt, target_test.pt, input_train.pt, target_train.pt, input_val.pt, target_val.pt , : test, train and val datasets for training the model.

               -    my_own_test_generalization.zip: the required data fr generalization test.

               -    my_own_test_twc_cnn_acadd_constraints_epochs_150_lr_0.00001_alpha_0.99_test.pt: output of the models. 

       

Please note that the Jupyter Notebook will download the original code of Physics-Constrained Deep Learning for Climate Downscaling which is has the following DOI at Zenodo:  https://zenodo.org/uploads/8150694.

Given that the complete COSMO-CLM model output is of order of ~100 TB, we could provide them upon individual requests. We aim to standardise and make the model output available according to the CORDEX standards. 

Files

cclm_steups.zip

Files (14.3 GB)

Name Size Download all
md5:de6b3599bc2ef73a4fcd7bac4ed63ac5
72.8 kB Preview Download
md5:be682bb43d370886addbad316bde5860
168.1 kB Preview Download
md5:c66113bec5c7f75378e8f0fd2d2c941f
1.2 MB Preview Download
md5:b29fe5a30f37cb8dcf999cafeca07f7e
163.5 MB Download
md5:486bea71c1494d41e847d6d2b7061351
490.6 MB Download
md5:65279d9328715f482bd0848da0babc42
163.5 MB Download
md5:d6d1df8dde9b15d5de624b034ab55229
122.5 kB Preview Download
md5:1f27cc488426f13966e37c85acb2a274
420.8 MB Preview Download
md5:9703e34a76586b125b2bcdeb356c9b0d
2.6 GB Download
md5:08f6f8f8517582373bed45adcc73f340
7.8 GB Download
md5:76a041ef0f1eda10520cdeb983b524ca
2.6 GB Download

Additional details

Related works

Is described by
Preprint: arXiv:2208.05424v8 (arXiv)
Is part of
Preprint: 10.5194/gmd-2023-227 (DOI)

Dates

Available
2023-12-21