Ice Sheet Emulator (ISE)

ISE is a Python package for training and analyzing ice sheet emulators, including ISEFlow — a hybrid flow-based neural network emulator for improved sea level projections and uncertainty quantification.

ISEFlow supports emulation for both the Antarctic Ice Sheet (AIS) and the Greenland Ice Sheet (GrIS), producing projections of ice volume above flotation (IVAF) changes driven by ISMIP6 climate forcings. Uncertainty is decomposed into epistemic (model) and aleatoric (data) components.

This codebase has been used in peer-reviewed research, including:

• "A Variational LSTM Emulator of Sea Level Contribution From the Antarctic Ice Sheet"
• "ISEFlow: A Flow-Based Neural Network Emulator for Improved Sea Level Projections and Uncertainty Quantification"
• "Emulator-expanded projections reveal structure in Antarctic sea level uncertainty" (in review)

Manuscript-specific code is archived on Zenodo — see Manuscript code archives below for DOIs.

Documentation: https://ise.readthedocs.io/

Installation

Install from PyPI:

pip install ise-py

Pretrained weights are hosted on HuggingFace Hub and downloaded automatically on first use. No manual download required.

For development (editable install):

git clone https://github.com/Brown-SciML/ise.git
cd ise
pip install -e ".[dev]"

Or with uv:

uv venv
uv pip install -e ".[dev]"

Project Structure

ise/
├── examples/                   Example scripts for using ISEFlow
├── ise/                        Main package
│   ├── data/                   Forcing/grid loading, feature engineering, dataset classes
│   │   ├── anomaly.py          AnomalyConverter: raw forcing → ISMIP6 anomalies
│   │   ├── dataclasses.py      EmulatorDataset, PyTorchDataset, TSDataset, ScenarioDataset
│   │   ├── feature_engineer.py FeatureEngineer: split, scale, lag, outliers
│   │   ├── forcings.py         ForcingFile: load/process climate NetCDF data
│   │   ├── grids.py            GridFile: sector boundary definitions
│   │   ├── inputs.py           ISEFlowAISInputs, ISEFlowGrISInputs
│   │   ├── process.py          ProjectionProcessor, DatasetMerger, sector helpers
│   │   ├── scaler.py           PyTorch StandardScaler, RobustScaler, LogScaler
│   │   └── utils.py            convert_and_subset_times()
│   ├── evaluation/             Metrics
│   │   └── metrics.py          Point, probabilistic, and distribution metrics
│   ├── models/                 Model architectures
│   │   ├── iseflow.py          ISEFlow, ISEFlow_AIS, ISEFlow_GrIS
│   │   ├── deep_ensemble.py    DeepEnsemble
│   │   ├── lstm.py             LSTM
│   │   ├── normalizing_flow.py NormalizingFlow
│   │   ├── training.py         CheckpointSaver, EarlyStoppingCheckpointer
│   │   ├── loss.py             WeightedGridLoss, WeightedMSELoss, and variants
│   │   ├── _experimental/      Legacy models from prior manuscripts (deprecated)
│   │   └── pretrained/         Pretrained weights (v1.0.0, v1.1.0)
│   └── utils/                  Data helpers and tensor utilities
│       ├── functions.py        get_X_y, get_data, to_tensor, unscale_output, …
│       └── io.py               check_type() runtime type validation
├── manuscripts/                Research paper scripts
├── tests/                      Unit tests
├── pyproject.toml
└── uv.lock

Usage

Loading and Running a Pretrained ISEFlow-AIS Model

import numpy as np
from ise.models import ISEFlow_AIS
from ise.data.inputs import ISEFlowAISInputs

year = np.arange(2015, 2101)  # 86 annual timesteps

# Option A: use a known ISMIP6 ISM configuration shortcut
inputs = ISEFlowAISInputs(
    year=year,
    sector=np.ones(86, dtype=int),      # sector 1 of 18
    pr_anomaly=np.zeros(86),
    evspsbl_anomaly=np.zeros(86),
    smb_anomaly=np.zeros(86),
    ts_anomaly=np.zeros(86),
    ocean_thermal_forcing=np.zeros(86),
    ocean_salinity=np.zeros(86),
    ocean_temperature=np.zeros(86),
    model_configs="AWI_PISM1",          # loads all ISM config fields automatically
    ice_shelf_fracture=False,
    ocean_sensitivity="medium",
    ocean_forcing_type="standard",
    standard_melt_type="local",
)

# Option B: provide all ISM parameters individually
inputs = ISEFlowAISInputs(
    year=year,
    sector=np.ones(86, dtype=int),
    pr_anomaly=np.zeros(86),
    evspsbl_anomaly=np.zeros(86),
    smb_anomaly=np.zeros(86),
    ts_anomaly=np.zeros(86),
    ocean_thermal_forcing=np.zeros(86),
    ocean_salinity=np.zeros(86),
    ocean_temperature=np.zeros(86),
    initial_year=1980,
    numerics="fd",
    stress_balance="ho",
    resolution="16",
    init_method="da",
    melt_in_floating_cells="floating condition",
    icefront_migration="str",
    ocean_forcing_type="open",
    ocean_sensitivity="low",
    ice_shelf_fracture=False,
    open_melt_type="picop",
    standard_melt_type=None,
)

# Option C: if you have raw (non-anomaly) forcing values, use from_absolute_forcings()
inputs = ISEFlowAISInputs.from_absolute_forcings(
    year=year,
    sector=10,
    pr=pr_array,
    evspsbl=evspsbl_array,
    smb=smb_array,
    ts=ts_array,
    ocean_thermal_forcing=otf_array,
    ocean_salinity=sal_array,
    ocean_temperature=temp_array,
    aogcm="noresm1-m_rcp85",           # or custom_climatology={...}
    model_configs="AWI_PISM1",
    ice_shelf_fracture=False,
    ocean_sensitivity="medium",
    ocean_forcing_type="standard",
    standard_melt_type="local",
)

# Load the pretrained v1.1.0 model (weights auto-downloaded from HuggingFace Hub on first use)
model = ISEFlow_AIS(version="v1.1.0")
predictions, uncertainties = model.predict(inputs)

print(predictions.shape)           # (86, 1)  — SLE in mm, 2015-2100
print(uncertainties["epistemic"])  # epistemic uncertainty per timestep
print(uncertainties["aleatoric"])  # aleatoric uncertainty per timestep
print(uncertainties["total"])      # total uncertainty (epistemic + aleatoric)

Running the Pretrained GrIS Emulator

import numpy as np
from ise.models import ISEFlow_GrIS
from ise.data.inputs import ISEFlowGrISInputs

inputs = ISEFlowGrISInputs(
    year=np.arange(2015, 2101),
    sector=np.ones(86, dtype=int),      # basin 1 of 6
    aST=np.zeros(86),                   # surface temperature anomaly
    aSMB=np.zeros(86),                  # SMB anomaly
    ocean_thermal_forcing=np.zeros(86),
    basin_runoff=np.zeros(86),
    model_configs="AWI_ISSM1",
    ice_shelf_fracture=False,
    ocean_sensitivity="medium",
    standard_ocean_forcing=True,
)

model = ISEFlow_GrIS(version="v1.1.0")
predictions, uncertainties = model.predict(inputs)

Training ISEFlow from Scratch

from ise.models import ISEFlow, DeepEnsemble, NormalizingFlow

nf = NormalizingFlow(input_size=93, output_size=1, num_flow_transforms=5)
de = DeepEnsemble(input_size=93, num_ensemble_members=5, output_sequence_length=86)
model = ISEFlow(deep_ensemble=de, normalizing_flow=nf)

# X: (N, n_features), y: (N,) — pre-scaled ISMIP6 data
model.fit(
    X_train, y_train,
    nf_epochs=100,
    de_epochs=100,
    X_val=X_val,
    y_val=y_val,
    early_stopping=True,
    patience=15,
)

model.save("./ISEFlow/", input_features=list(X_train.columns))

Evaluating Model Performance

from ise.models import ISEFlow
from ise.evaluation import metrics as m
from ise.utils import functions as f

model = ISEFlow.load("./ISEFlow/")

predictions, uncertainties = model.predict(X_val)
y_val_unscaled = f.unscale_output(y_val.reshape(-1, 1), "./ISEFlow/scaler_y.pkl")

mse = m.mean_squared_error(y_val_unscaled, predictions)
print(f"MSE: {mse:.4f}")

Contributing

We welcome contributions! To get started:

1. Fork the repository on GitHub.
2. Create a new branch for your feature or bugfix.
3. Submit a pull request (PR) for review.

Run tests before submitting:

pytest tests/

Manuscript code archives

The current ise-py package on PyPI is the actively maintained, versioned successor to the manuscript-specific codebases. Frozen snapshots of the code used in each publication are archived on Zenodo at the DOIs below:

• "A Variational LSTM Emulator of Sea Level Contribution From the Antarctic Ice Sheet" Peter Van Katwyk, Baylor Fox-Kemper, Helene Seroussi, Sophie Nowicki, Karianne Bergen. DOI: 10.5281/zenodo.10416633

• "ISEFlow: A Flow-Based Neural Network Emulator for Improved Sea Level Projections and Uncertainty Quantification" Peter Van Katwyk, Baylor Fox-Kemper, Sophie Nowicki, Helene Seroussi, Karianne Bergen. DOI: 10.5281/zenodo.14908114

• "Emulator-expanded projections reveal structure in Antarctic sea level uncertainty" (in review) Peter Van Katwyk, Baylor Fox-Kemper, Sophie Nowicki, Helene Seroussi, Karianne Bergen. Data & code DOI: 10.5281/zenodo.19355381

For new work, install the maintained package (pip install ise-py); use the Zenodo archives when you need to reproduce results from a specific paper.

Contact & Support

Developed by Peter Van Katwyk (Ph.D., Brown University).

• Email: pvankatwyk@gmail.com
• GitHub Issues: Report a bug

If you use ISE in research, please consider citing our work. See CITATION.md for details.