Thesis topic: estimation of ice thickness using inverse methods

fmaussion commented 5 years ago

Motivation

OGGM is able to compute the volume of glaciers using only surface properties. See this paper for a good overview of ice thickness estimation methods, and this paper for a global application.

That said, OGGM's way of doing things is quite simple and has several caveats: it doesn't work well for large ice fields (ice caps), and rely on a very strong assumption: that glaciers are in steady state. Therefore, we started to explore better ways to deal with the problem. In 2018, we started a new project which is computing ice thickness with a completely new approach: COMBINE, written by @phigre and documented in his thesis.

In this new thesis topic, I propose to build upon and extend this work in two possible ways (all listed below).

Specific goals

I suggest to address one of these goals, which are largely independant:

Apply COMBINE to real world ice caps. In his thesis, Philipp used a theoretical, idealized environment to develop and test the method. The next natural step would be to try it out in the wild! This will come with a bunch of new challenges ;-)
Release the steady state assumption in COMBINE. While there is no steady state assumption hard coded in COMBINE, in practice we are still using steady state cases. It would be a great step forward to actually test if COMBINE can work in non-steady-state. This will imply the development of a "1D version" of COMBINE in order to be able to create simplified flowline experiments first, then define idealized retreating / advancing cases and see if COMBINE can deal with them.

1 is more practical, 2 is more theoretical, both require quite a big investment in inverse theory, machine learning and its algorithms.

Profile of the student

high maths and programming skills
willingness to deal with new concepts hardly taught at ACINN
interested in glaciology and machine learning
demonstrated good to very good grades in the climate / glaciology / programming lectures

Fabien's assessment of the topic

Highly rewarding, high profile topic, only suitable for very motivated students able to work / develop on their own. An excellent entry point in the world of machine learning.

fmaussion commented 4 years ago

@pat-schmitt will start to work on topic 2! Welcome Patrick!

phigre commented 4 years ago

Great! Welcome Patrick. I think, we should get in touch for all the details. And of course I am also highly interested in your work.

pat-schmitt commented 4 years ago

@phigre, thank you for you offer. This week I start with the work and i will come back to you when the first questions arise.

OGGM / oggm