Getting Started#
Getting Started with ICESEE
Configure and run your first ensemble data assimilation workflow through CryoStack using supported models, filters, observations, and computing backends.
ICESEE, the Ice Sheet State and Parameter Estimator, is the data assimilation application within CryoStack. It combines a forecast model, an ensemble Kalman filter, and observations to estimate model states and uncertain parameters.
This guide runs a first experiment — Lorenz-96 in Local mode — and points to the Remote and Cloud paths.
Before You Begin#
You need:
a modern web browser;
access to the CryoStack platform;
for Remote runs: your own access to an HPC system (username, allocation, and an SSH key you can register) — the same access CryoLauncher uses;
for Cloud runs: an AWS account you can connect to CryoStack.
A first Lorenz-96 experiment needs none of the Remote or Cloud prerequisites.
Open ICESEE#
Open https://cryostack.eas.gatech.edu/icesee-gui/.
The interface has two areas:
Run settings — the Local / Remote / Cloud tabs, the example, filter, ensemble size, seed, output, and the example’s parameter sections.
Workspace — Runs, Files, Run Log, and Results.
Choose Where to Run#
Local Supported — a single process on the CryoStack server, with no scheduler or MPI. For Lorenz-96 and small tests.
Remote Supported — a Slurm batch job on an HPC system, under your own HPC identity, using ICESEE-Spack or a container. Required for ISSM.
Cloud Lorenz-96 only — AWS Batch on your own AWS account, through the same Connect AWS Account → Prepare cloud → Review & Launch flow as CryoLauncher. Verified for exactly one configuration: Lorenz-96 at a single process. Other examples, or more processes, are refused at Review.
Select an Example#
The Example menu shows each example’s status:
Example |
Status |
Where it runs today |
|---|---|---|
Lorenz-96 |
fully runnable locally |
Local and Cloud (one process); Remote is available |
ISSM (ISMIP_Choi) |
fully runnable in Remote |
Remote |
Flowline |
under development |
not yet an end-to-end workflow |
Icepack |
under development |
not yet an end-to-end workflow |
See Examples and maturity in the User Manual for details.
Configure the Experiment#
Preset — only Default (the example’s own configuration) is offered.
Filter — EnKF, DEnKF, EnTKF, or EnRSKF. Keep the example’s default for a first run.
Ensemble size — 1 to 200 members (default 30). Larger ensembles represent uncertainty better and cost more.
Seed — the random seed; keep it fixed to repeat an experiment exactly.
Output — which result set the report reads: true-wrong (the demonstration output) or EnKF.
Generate report — run the example’s results notebook after a successful Local run.
The example’s parameter sections (physical-parameters,
modeling-parameters, enkf-parameters) hold everything else, including the
observation settings and the parallel settings. Review them, but leave them
unchanged for a first run.
Run Your First ICESEE Experiment#
Open ICESEE and select the Local tab.
Choose Lorenz-96 in the Example menu.
Keep the default filter, ensemble size, and seed.
Leave Output at true-wrong and Generate report ticked.
Click Run.
Follow the Run Log: configuration, ensemble initialization, forecast and analysis cycles, then report generation.
When the run finishes, select it in Runs and open Results.
Lorenz-96 is a lightweight synthetic model, so this completes quickly and shows the whole assimilation cycle: a true state and synthetic observations are generated, the ensemble is initialized and advanced, and the filter updates it at each observation time.
View Results#
Results shows the figures the run produced and lists its HDF5 result files; Download results downloads them. With Generate report ticked, the report notebook is executed into the run folder.
Next Steps#
Read the ICESEE User Manual for Remote and Cloud runs, the parameter sections, and the parallel settings.
Review ICESEE Resources for repositories, publications, models, and data assimilation references.
Open CryoLauncher for model runs without data assimilation.