The Workshop

Took place October 28th-31st at Rydal Hall in the Lake District.

Aimed to answer the question:

“What should the next generation of environmental models look like?”

What’s a CSDMS?

CSDMS = Community Surface Dynamics Modelling System, based out of Univ. Colorado Boulder

“a diverse community of experts promoting the modeling of earth surface processes by developing, supporting, and disseminating integrated software modules…”¹

They maintain a curated repository of compliant community models.

The CSDMS Workbench is an integrated system of tools and standards for modelling:

• Interoperability of models via the Basic Model Interface (BMI) and common variable naming standards
• Language agnosticism via a Python wrapper (Fortran, C, C++, ~Java, ~Julia)
• Various tools for coupling model components exposing a BMI, setting up gridded experiments, and incorporating data

Extensions and Alternatives

• eWatercycle by the Netherlands eScience Centre (uses BMI)
  • We ran this demo notebook on Datalabs
• The Framework for Aquatic Biogeochemical Models (FABM) by Plymouth Marine Laboratory
• ‘NextGen’: a next generation National Water Model by NOAA (uses BMI)

A toybox of models

• A small group of us (including Ed Rowe and myself) discussed training and knowledge-sharing across domains in depth.

• We proposed to populate a GitHub repository with a growing collection of BMI-compliant toy models of all shapes and sizes.

• From a training perspective the common BMI framework is helpful for both novices and experienced modellers:

  • It ought to reduce the cognitive load for learners new to modelling, since you can get started and run models without needing to know too many details or going through a complicated set up.

  • It should also facilitate lateral learning since BMI serves as a map for exploring new models, potentially from other domains.

• Get in touch if you’d be interested in contributing!

Containerised, BMI-ified JULES

• I had an interesting conversation with an RSE from the Netherlands eScience Center who had containerised an old and slow Matlab model and exposed a BMI interface that could receive inputs and send outputs over HTTP.

• Initial reaction: cute but not generally a good idea for large-scale modelling (communications will become a bottleneck very quickly).

• But really this is only a problem for tightly-coupled models (think atmospheric dynamics). Meanwhile JULES is essentially a 1+1 dimensional model – decoupled in 2/3 spatial domains.

• A containerised, BMI-ified JULES could be a convenient alternative route to working with JULES, using Python and generic computing resources (i.e. without Jasmin)

Modular modelling systems

• Moving towards a system of modular models rather than large complex codebases
• Enables easier coupling of models
• Allows adaptation to ever changing environmental challenges: Easily swap one model for another depending on challenge at hand

Ethical and sustainability of models

• How sustainable are our models? - we need to balance the good the models can produce with the carbon impact of running them
• Ethical considerations - Have we considered the societal impact of our models (E.g. informing policy)? Have we involved the communities impacted in the development of models?
• Ethical considerations - Modellers often want to help after a natural disaster (E.g. flooding) but could this be perceived as help or trying to advance careers on back of disaster?

Getting modellers out of their silos

• Picking on hydrologists here but why so many different models? Should there be more community collaboration?
• Linking models requires making them easier to couple with other models - E.g. using the CSDMS BMI approach

Highlights

Presentations & workshops

• Kirsty Pringle’s Green Computing presentation

• Michael Hollaway’s DataLabs workshops

Coffee conversations

• Cohesiveness of conceptualisations around ‘the model’. What are they? Is the community even talking about comparable entities?
• How do other stakeholders (e.g. policymakers) perceive models? As facts or representations? How do we communicate what a model is?
• The role of assumptions and uncertainties in modelling. How do we document these? How to communicate these to policymakers and other researchers using and developing the model?
• How do we model society? What even is society? What assumptions would be involved?

Personal Takeaway’s and Actions

Takeaway’s

• A better understanding of the challenges and opportunities in modelling
• Challenged the concept of DRI: could modelling be a case study?
• Refocused interest in science and a social practice, the role of technology in this, as well as the role of science in policymaking

Actions

• Contribute to workshop paper on ‘Modelling and Society’
• Read the many papers recommended to me
• Continue my involvement in this community so I can learn more from them

Next steps

• Set up a community forum on Discourse (done! https://forum.csdms.io/)
• Writing a kind of proceedings / workshop outcomes paper (in progress)
• Set up Euro-CSDMS
• Developing training tools (including the toybox)
• Looking for funding for future events
• In situ hackathons + citizen science projects (solve the system)

Get involved

Join the Discourse forum!

forum.csdms.io

Further reading

• Many links available in this presentation at ukceh-rse.github.io/csdms-workshop-highlights
• Workshop website (for images): eurocsdmswork.shop/
• Sam’s blog post: www.ceh.ac.uk/news-and-media/blogs/envisioning-future-environmental-modelling