Model pertubation dm for lsrtm

[kerim371]

Hi,

I can see in the examples the model pertubation is computed as a difference between true model and its approximation: dm = vec(m0 - m).

But in practice we don't know the true model.
I think we can get dm as a difference between some estimations from FWI but I'm not sure.

What are the ways to to get dm model pertubation in a field case?
Could you recommend some source where to read about model pertubation?

[mloubout]

This is just a plain example to show how to use the function, this is a function for iterative process (LSRTM) so you start with dm = 0 usually and iteratively update it. The goal is to recover the true one, ie you are estimating dm such that you can get m from m0

[kerim371]

Thank you, but still have some misunderstanding.

In that case I can't understand how can I update both model and image at the same time.
Let's suppose m0 is some estimation of true model (probably after FWI).
Could you please make the following code more correct:

function objective_function(dm)

    fval, gradient = lsrtm_objective(model0, q, d_obs, dm, options=jopt)

    return fval, gradient
end

sol = spg(objective_function, model0.m.data)

I suspect the gradient should be referred to the reflectivity (imaging) updates rather than model update.

[mloubout]

lsrtm_objective is for LSRTM, i.e min_dm || J*dm - data ||' not FWI which is || F(m) * q - data ||which should usefwi_objective`

[kerim371]

Ok, I will try some experiments to better understand that.
I have some lack of knowledge.