IceNet: Seasonal Arctic sea ice forecasting with probabilistic deep learing

This codebase accompanies the Nature Communications paper Seasonal Arctic sea ice forecasting with probabilistic deep learning. It includes code to fully reproduce all the results of the study from scratch. It also includes code to download the data generated by the study, published on the Polar Data Centre, and reproduce all the paper's figures.

The flexibility of the code simplifies possible extensions of the study. The data processing pipeline and custom IceNetDataLoader class lets you dictate which variables are input to the networks, which climate simulations are used for pre-training, and how far ahead to forecast. The architecture of the IceNet model can be adapted in icenet/models.py. The output variable to forecast could even be changed by refactoring the IceNetDataLoader class.

A demonstrator of this codebase (downloading pre-trained IceNet networks, then generating and analysing forecasts) produced by @acocac can be found in The Environmental AI Book.

The guidelines below assume you're working in the command line of a Unix-like machine with a GPU. If aiming to reproduce all the results of the study, 1 TB of space should safely cover the storage requirements from the data downloaded and generated.

If you run into issues or have suggestions for improvement, please raise an issue or email me (tomand@bas.ac.uk).

Steps to plot paper figures using the paper's results & forecasts

To reproduce the paper figures directly from the paper's results and forecasts, run the following after setting up the conda environment (see Step 1 below):

• ./download_paper_generated_data.sh. Downloads raw data from the paper. From here, you could start to explore the results of the paper in more detail.
• python3 icenet/download_sic_data.py. This is needed to plot the ground truth ice edge. Note this download can take anywhere from 1 to 12 hours to complete.
• python3 icenet/gen_masks.py
• python3 icenet/plot_paper_figures.py. Figures are saved in figures/paper_figures/.

Steps to reproduce the paper's results from scratch

0) Preliminary setup

• I use conda for package management. If you don't yet have conda, you can download it here.

• To be able to download ERA5 data, you must first set up a CDS account and populate your .cdsapirc file. Follow the 'Install the CDS API key' instructions here.

• To download the ECMWF SEAS5 forecast data for comparing with IceNet, you must first register with ECMWF here. If you are from an ECMWF Member State, you can then gain access to the ECMWF MARS Catalogue by contacting your Computing Representative. Once registered, obtain your API key here and fill the ECMWF API entries in icenet/config.py.

• To track training runs and perform Bayesian hyperparameter tuning with Weights and Biases, sign up at https://wandb.ai/site. Obtain your API key from here and fill the Weights and Biases entries in icenet/config.py. Ensure you are logged in by running wandb login after setting up the conda environment.

1) Set up conda environment

After cloning the repo, run the commands below in the root of the repository to set up the conda environment:

• If you don't have mamba already, install it to your base env for faster conda operations: conda install -n base mamba -c conda-forge.
• For upgradeability use the versioned direct dependency environment file: mamba env create --file environment.yml
• For reproducibility use the locked environment file: mamba env create --file environment.locked.yml
• Activate the environment before running code: conda activate icenet

2) Download data

The CMIP6 variable naming convention is used throughout this project - e.g. tas for surface air temperature, siconca for sea ice concentration, etc.

Warning: some downloads are slow and the net download time can take 1-2 days. It may be advisable to write a bash script to automatically execute all these commands in sequence and run it over a weekend.

• python3 icenet/gen_masks.py. This obtains masks for land, the polar holes, monthly maximum ice extent (the 'active grid cell region'), and the Arctic regions & coastline.

• python3 icenet/download_sic_data.py. Downloads OSI-SAF SIC data. This computes monthly-averaged SIC server-side, downloads the results, and bilinearly interpolates missing grid cells (e.g. polar hole). Note this download can take anywhere from 1 to 12 hours to complete.

• ./download_era5_data_in_parallel.sh. Downloads ERA5 reanalysis data. This runs multiple python3 icenet/download_era5_data.py commands to acquire each ERA5 variable in parallel. The raw ERA5 data is downloaded in global latitude-longitude format and regridded to the EASE grid that OSI-SAF SIC data lies on.

• ./download_cmip6_data_in_parallel.sh. Downloads CMIP6 climate simulation data. This runs multiple python3 icenet/download_cmip6_data.py commands to acquire each climate simulation. This downloads the raw CMIP6 data in global latitude-longitude format and regrids to the EASE grid that OSI-SAF SIC data lies on.

• ./rotate_wind_data_in_parallel.sh. This rotates the ERA5 and CMIP6 wind vector data onto the EASE grid in parallel using icenet/rotate_wind_data.py.

• ./download_seas5_forecasts_in_parallel.sh. Downloads ECMWF SEAS5 SIC forecasts. This runs multiple python3 icenet/download_seas5_data.py commands to acquire 2002-2020 SEAS5 forecasts for multiple lead times in parallel via the ECMWF MARS API and regrid the forecasts to EASE. The forecasts are saved to data/forecasts/seas5/ in the folders latlon/ and EASE/.

• python3 icenet/biascorrect_seas5_data.py. Bias corrects the SEAS5 2012+ forecasts using 2002-2011 forecasts. Also computes SEAS5 sea ice probability (SIP) fields. The bias-corrected forecasts are saved as NetCDFs in data/forecasts/seas5/ with dimensions (target date, y, x, lead time).

3) Process data

3.1) Set up IceNet's custom data loader

• python3 icenet/gen_data_loader_config.py. Sets up the data loader configuration. This is saved as a JSON file dictating IceNet's input and output data, train/val/test splits, etc.The config file is used to instantiate the custom IceNetDataLoader class. Two example config files are provided in this repository in dataloader_configs/. Each config file is identified by a dataloader ID, determined by a timestamp and a user-provided name (e.g. 2021_06_15_1854_icenet_nature_communications). The data loader ID, together with an architecture ID set in the training script, provides an 'IceNet ID' which uniquely identifies an IceNet ensemble model by its data configuration and architecture.

3.2) Preprocess the raw data

• python3 icenet/preproc_icenet_data.py. Normalises the raw NetCDF data and saves it as monthly NumPy files. The normalisation parameters (mean/std dev or min/max) are saved as a JSON file so that new data can be preprocessed without having to recompute the normalisation. A custom IceNetDataPreProcessor class

• The observational training & validation dataset for IceNet is just 23 GB, which can fit in RAM on some systems and significantly speed up the fine-tuning training phase compared with using the data loader. To benefit from this, run python3 icenet/gen_numpy_obs_train_val_datasets.py to generate NumPy tensors for the train/val input/output data. To further benefit from the training speed improvements of tf.data, generate a TFRecords dataset from the NumPy tensors using python3 icenet/gen_tfrecords_obs_train_val_datasets.py. Whether to use the data loader, NumPy arrays, or TFRecords datasets for training is controlled by bools in icenet/train_icenet.py.

4) Develop the IceNet model

4.1) OPTIONAL: Run the hyperparameter search (skip if using default values from paper)

• Set icenet/train_icenet.py up for hyperparameter tuning: Set pre-training and temperature scaling bools to False in the user input section.
• wandb sweep icenet/sweep.yaml
• Then run the wandb agent command that is printed.
• Cancel the sweep after a sufficient picture on optimal hyperparameters is built up on the wandb.ai page.

4.2) Train IceNet

• Train IceNet networks with python3 icenet/train_icenet.py. This takes hyperameter settings and the random seed for network weight initalisation as command line inputs. Run this multiple times with different settings of --seed to train an ensemble. Trained networks are saved in trained_networks/<dataloader_ID>/<architecture_ID>/networks/. If working on a shared machine and familiar with SLURM, you may want to wrap this command in a SLURM script.

5) Produce forecasts

• python3 icenet/predict_heldout_data.py. Uses xarray to save predictions over the validation and test years as (2012-2020) as NetCDFs for IceNet and the linear trend benchmark. IceNet's forecasts are saved in data/forecasts/icenet/<dataloader_ID>/<architecture_ID>/. For IceNet, the full forecast dataset has dimensions (target date, y, x, lead time, ice class, seed), where seed specifies a single ensemble member or the ensemble-mean forecast. An ensemble-mean SIP forecast is also computed and saved as a separate, smaller file (which only has the first four dimensions).

• Compute IceNet's ensemble-mean temperature scaling parameter for each lead time: python3 icenet/compute_ensemble_mean_temp_scaling.py. The new, ensemble-mean temperature-scaled SIP forecasts are saved to data/forecasts/icenet/<dataloader_ID>/<architecture_ID>/icenet_sip_forecasts_tempscaled.nc. These forecasts represent the final ensemble-mean IceNet model used for the paper.

6) Analyse forecasts

• python3 icenet/analyse_heldout_predictions.py. Loads the NetCDF forecast data and computes forecast metrics, storing results in a global pandas DataFrame with MultiIndex (model, ensemble member, lead time, target date) and columns for each metric (binary accuracy and sea ice extent error). Uses dask to avoid loading the entire forecast datasets into memory, processing chunks in parallel to significantly speed up the analysis. Results are saved as CSV files in results/forecast_results/ with a timestamp to avoid overwriting. Optionally pre-load the latest CSV file to append new models or metrics to the results without needing to re-analyse existing models. Use this feature to append forecast results from other IceNet models (identified by their dataloader ID and architecture ID) to track the effect of design changes on forecast performance.

• python3 icenet/analyse_uncertainty.py. Assesses the calibration of IceNet and SEAS5's SIP forecasts. Also determines IceNet's ice edge region and assesses its ice edge bounding ability. Results are saved in results/uncertainty_results/.

7) Run the permute-and-predict method to explore IceNet's most important input variables

• python3 icenet/permute_and_predict.py. Results are stored in results/permute_and_predict_results/.

8) Generate the paper figures and tables

• python3 icenet/plot_paper_figures.py. Figures are saved in figures/paper_figures/.

Misc

• icenet/utils.py defines IceNet utility functions like the data preprocessor, data loader, ERA5 and CMIP6 processing, learning rate decay, and video functionality.
• icenet/models.py defines network architectures.
• icenet/config.py defines globals.
• icenet/losses.py defines loss functions.
• icenet/callbacks.py defines training callbacks.
• icenet/metrics.py defines training metrics.

Project structure: simplified output from tree

.
├── data
│   ├── obs
│   ├── cmip6
│   │   ├── EC-Earth3
│   │   │   ├── r10i1p1f1
│   │   │   ├── r12i1p1f1
│   │   │   ├── r14i1p1f1
│   │   │   ├── r2i1p1f1
│   │   │   └── r7i1p1f1
│   │   └── MRI-ESM2-0
│   │       ├── r1i1p1f1
│   │       ├── r2i1p1f1
│   │       ├── r3i1p1f1
│   │       ├── r4i1p1f1
│   │       └── r5i1p1f1
│   ├── forecasts
│   │   ├── icenet
│   │   │   ├── 2021_06_15_1854_icenet_nature_communications
│   │   │   │   └── unet_tempscale
│   │   │   └── 2021_06_30_0954_icenet_pretrain_ablation
│   │   │       └── unet_tempscale
│   │   ├── linear_trend
│   │   └── seas5
│   │       ├── EASE
│   │       └── latlon
│   ├── masks
│   └── network_datasets
│       └── dataset1
│           ├── meta
│           ├── obs
│           ├── transfer
│           └── norm_params.json
├── dataloader_configs
│   ├── 2021_06_15_1854_icenet_nature_communications.json
│   └── 2021_06_30_0954_icenet_pretrain_ablation.json
├── figures
├── icenet
├── logs
│   ├── cmip6_download_logs
│   ├── era5_download_logs
│   ├── seas5_download_logs
│   └── wind_rotation_logs
├── results
│   ├── forecast_results
│   │   └── 2021_07_01_183913_forecast_results.csv
│   ├── permute_and_predict_results
│   │   └── permute_and_predict_results.csv
│   └── uncertainty_results
│       ├── ice_edge_region_results.csv
│       ├── sip_bounding_results.csv
│       └── uncertainty_results.csv
└── trained_networks
    └── 2021_06_15_1854_icenet_nature_communications
        ├── obs_train_val_data
        │   ├── numpy
        │   └── tfrecords
        │       ├── train
        │       └── val
        └── unet_tempscale
            └── networks
                ├── network_tempscaled_36.h5
                ├── network_tempscaled_37.h5
                :

Acknowledgements

Thanks to James Byrne (BAS) and Tony Phillips (BAS) for direct contributions to this codebase.