Simultaneous inversion for rheology_B and basal friction?

rueckamp

Hello,

I would like to run the inversion for MaterialsRheologyBbar in the floating part and for the FrictionCoefficient in grounded part. If I run it separately, it works fine

But if I run it simultaneously only the friction coefficient is updated. I have setup the model like:

...

pos_grounded=find(md.mask.groundedice_levelset>0.);

pos_floating=find(md.mask.groundedice_levelset<=0.);

md.inversion.cost_functions_coefficients(pos_grounded,1)=200.;

md.inversion.cost_functions_coefficients(pos_floating,1)=1.;

md.inversion.control_parameters={'MaterialsRheologyBbar','FrictionCoefficient'};

md.inversion.min_parameters(:,1)=10^6*ones(md.mesh.numberofvertices,1);

md.inversion.max_parameters(:,1)=2*10^9*ones(md.mesh.numberofvertices,1);

md.inversion.min_parameters(:,2)=1.*ones(md.mesh.numberofvertices,1);

md.inversion.max_parameters(:,2)=250.*ones(md.mesh.numberofvertices,1);

md.inversion.gradient_scaling(:,1)=1*10^8*ones(md.inversion.nsteps,1);

md.inversion.gradient_scaling(:,2)=3.*ones(md.inversion.nsteps,1);

…

So, I would like to know if its general possible to run an inversion for several parameters? If so, what can I do wrong?

best,

Martin

mathieumorlighem

Hi Martin,

First, I highly recommend that you use m1qn3inversion() rather than inversion() for md.inversion. It is much more efficient and requires fewer parameters. To convert your existing class, you can use md.inversion = m1qn3inversion(md.inversion).

Second, even though it is theoretically possible to invert for both B and alpha in ISSM, we HIGHLY discourage it because you have an infinite number of solution. The effect of softer ice (decreasing B) and reducing basal friction has the same effect on the cost function.

I can look more into it after AGU.

Best,

Mathieu