| # Example script to show the steps needed to downscale ACE | |
| # using HiRO | |
| # Provided are two inference configurations used in the | |
| # manuscript: | |
| # 1.) Global inference with ACE for 2023 then downscaled | |
| # for most of the global (-66S to 70N) using HiRO | |
| # | |
| # 2.) Global inference with ACE for 2014 - 2023 then | |
| # the Pacific NorthWest (PNW) only downscaled with HiRO | |
| # First run coarse ACE2S | |
| python -m fme.ace.inference ace2s_inference_config_global.yaml | |
| # Second downscale ACE2S using HiRO | |
| # HiRO is more computationally costly then ACE2S | |
| # for faster through put more GPUs may be required | |
| NGPU=1 | |
| # This will downscale a subselected region around the Pacific Northwest. | |
| torchrun --nproc_per_node $NGPU -m fme.downscaling.inference hiro_downscaling_ace2s_pnw_output.yaml | |