Skip to content

ACCESS-NRI Model Live Diagnostics

Welcome to the documentation and reference guide for the ACCESS-NRI Model Live Diagnostics (MLD) package, developed by the Model Evaluation and Diagnostics (MED) team. The package is installed as med-diagnostics and imported in Python as med_diagnostics. Model Live Diagnostics provides practical Jupyter-based tools for interactive monitoring and diagnostic analysis of currently running, or "live", ACCESS climate models on the Australian NCI supercomputer Gadi.

The framework is intended for use in the Australian Research Environment (ARE) and is still under active development.

What does this package do?

Model Live Diagnostics helps the ACCESS modelling community check, monitor, visualise, and evaluate the behaviour and progress of live ACCESS models on Gadi. It can also load, visualise, and compare legacy ACCESS model data with the selected live user model. It supports ACCESS-CM2, ACCESS-CM3, ACCESS-OM2, ACCESS-OM3, ACCESS-ESM1.5, ACCESS-ESM1.6, ACCESS-AM3 and MOM6 output.

A session opens an interactive interface in your notebook with four collapsible sections:

  • Load and plot user data. Builds an Intake catalogue (a searchable data index) from your live model output, so you can load any of its datasets and plot a variable as a line plot, heatmap, or an interactive heatmap over time. Sliders let you slice through extra dimensions such as ocean depth.

  • Load and plot reference models. Loads any model of the same type from the ACCESS-NRI Intake Catalogue and plots its data with the same controls.

  • Overlay user and reference models. Overlays line plots or shows heatmaps side by side for your run and as many reference models as you like, and can plot the difference between them to highlight discrepancies.
  • Run analysis recipes. Runs ready-made analyses on your data, such as the Niño 3.4 index or area-weighted regional mean timeseries, or apply your own custom analyses.

Quick-start guide

The framework only works on Gadi, and the interface requires an ARE JupyterLab session.

Prerequisites

Before starting, make sure you:

  1. Have an NCI account. See the NCI account documentation. You will also need to join a project with a compute allocation.
  2. Can access the projects containing your model data. The ACCESS-NRI Intake Catalog references data across multiple Gadi projects. See the supported project list and the NCI project access documentation.
  3. Are a member of the ACCESS-NRI xp65 project to access the ACCESS-NRI analysis3 environment.

Start an ARE JupyterLab session

  1. Log in to ARE and start a JupyterLab instance with access to the ACCESS-NRI analysis3 conda environment. Use the following settings:

    Setting Recommendation
    Compute size XX-Large (28 CPUs, 126 GB memory)
    Storage E.g. gdata/xp65+gdata/ik11 are required for the getting started tutorial

    In Advanced options, use:

    Setting Value
    Module directories /g/data/xp65/public/modules
    Modules conda/analysis3

  2. Follow the Getting started tutorial to begin a Model Live Diagnostics session.

Install the package directly

The package is pre-installed in the ACCESS-NRI analysis3 environment. Alternatively, install it from the access-nri conda channel or PyPI:

conda install -c accessnri med-diagnostics

pip install med-diagnostics

The package is still a work in progress. We value your feedback, especially in the form of reporting issues/bugs or suggesting ways to improve the framework. To do so, please open an issue here.

Next: Tutorials