Thesis topic: Distributed glacier modelling in OGGM - "OGGM-2D"

[fmaussion]

Context

OGGM uses simplified glacier geometries ("flowlines") to represent the flow of ice. This works fine for certain applications (valley glaciers and glacier shrinkage scenarios), but much less for others. For example arctic ice caps (largest contributors so sea-level rise) or past mountain glaciations are poorly represented by the OGGM default ice dynamic model.

Recently, a new model of glacier evolution (the Instructed Glacier Model IGM) was released and circumvents one of the main obstacle towards large-scale glacier modeling using gridded (distributed) models: computational costs. IGM is an "emulator" of ice flow trained using state-of-the art ice flow models using neural networks. IGM is coded in python and therefore readily compatible with OGGM.

IGM uses OGGM as a "preprocessor" to fetch data necessary for its workflow. IGM also implements a simple class that follows OGGM's syntax for glacier evolution models: https://github.com/jouvetg/igm/blob/main/igm/instructed_oggm.py

Objectives

The goals of this thesis are multiple and can be combined / adapted / separated in separate work packages and multiple dissertations:

1. maintain and update "instructed OGGM" a compatibility layer between OGGM and IGM allowing IGM to be run instead or in parallel to the existing 1D evolution model.
2. using this compatibility layer, define and run comparison exercises where IGM output is compared to OGGM output for future projections of glacier change. This task will involve defining a workflow similar to OGGM to initialise and calibrate the model before running the simulations.
3. Task 2 can be done in various ways:
   • with static calibration (the default in IGM), we intialise the model with existing ice thickness products and run the model from this initial state. This is however unlikely to yield satisfying results.
   • Instead IGM should be initialised like OGGM by taking into account the mass-balance and dynamical state of the glacier (dynamical spinup)
   • ... or it could be inialised with ice velocities (data assimilation) and run from there (see this paper for a start).

Depending on interest and motivation, these tasks can be addressed one at a time. The main objective is to test the suitability of IGM for operational OGGM runs, replacing the simplified glacier representation in OGGM.

Profile of the candidate

Interest in machine learning, climate science and glacier modelling, as well as good programming skills.

[bowenbelongstonature]

Do you have idea about how long will this take?

I'm quite interested in this since ice caps among northern Tibet Plateau(TP) is hard to simulate by OGGM with flowlines. Plus, I know something about PISM cause I tried to use PISM last year for simulating early Holocene glaciers on northern TP, but this project stopped since we think climate constraints in early Holocene have low confidence and glaciers' heritages are rare for flatten topography(and dating them is hard).

[fmaussion]

Do you mean that you are interested in working on this?

[bowenbelongstonature]

Yes, I think maybe I have more than a whole year to do this.

[fmaussion]

OK - are you available to join the OGGM community meeting on Wednesday?

[bowenbelongstonature]

Of course!

[fmaussion]

Until further notice, this topic is still open for students!