Contents

function [UserVar,InvStartValues,Priors,Meas,BCsAdjoint,RunInfo]=...
    DefineInputsForInverseRun(UserVar,CtrlVar,MUA,BCs,F,l,GF,InvStartValues,Priors,Meas,BCsAdjoint,RunInfo)

Note: This m-file is just an example of how to define inputs for an inverse run. You will need to modify to fit your own problem.

What you need to define are:

Note: When doing an inverse run, presumably a good start is to copy this file from the source directory to you own run director.

persistent FuMeas FvMeas FerrMeas  % keep scattered interpolants for the data in memory.

get measurments and define error covariance matrices

if isempty(FuMeas)

    % Here I'm assuming the user has defined the field
    %
    %   UserVar.SurfaceVelocityInterpolant
    %
    % which is the name of a mat file containint surface velocity data interpolants.
    %
    %

    fprintf('Loading interpolants for surface velocity data: %-s ',UserVar.SurfaceVelocityInterpolant)
    load(UserVar.SurfaceVelocityInterpolant,'FuMeas','FvMeas','FerrMeas')
    fprintf(' done.\n')
end

% Now interpolate the data onto the nodes of the mesh
Meas.us=double(FuMeas(MUA.coordinates(:,1),MUA.coordinates(:,2)));
Meas.vs=double(FvMeas(MUA.coordinates(:,1),MUA.coordinates(:,2)));
Err=double(FerrMeas(MUA.coordinates(:,1),MUA.coordinates(:,2)));

% Here I set any NaN values to 0. The assumption here is that these NaN values represent missing data and I set these values to 0. This
% may, or may not, be a good idea. But the important thing is to set the errors where data is missing to some really high value. Here I
% set the errors to 1e10.
MissingData=isnan(Meas.us) | isnan(Meas.vs) | isnan(Err) | (Err==0);
Meas.us(MissingData)=0 ;  Meas.vs(MissingData)=0 ; Err(MissingData)=1e10;

% The data errors as specified by these covariance matrices.
% The data errors are assumed to be uncorrelated, hence we are here using diagonal covariance matrices.
usError=Err ; vsError=Err ;
Meas.usCov=sparse(1:MUA.Nnodes,1:MUA.Nnodes,usError.^2,MUA.Nnodes,MUA.Nnodes);
Meas.vsCov=sparse(1:MUA.Nnodes,1:MUA.Nnodes,vsError.^2,MUA.Nnodes,MUA.Nnodes);

Not enough input arguments.
Error in DefineInputsForInverseRun (line 32)
    fprintf('Loading interpolants for surface velocity data: %-s ',UserVar.SurfaceVelocityInterpolant)

Define Priors

Priors.AGlen=AGlenVersusTemp(-10);
Priors.n=F.n;

% Here the priors are defined by a rough estimate of C based on the measured data.
% Depending on the situation better priors might be available.
switch CtrlVar.SlidingLaw

    case {"Weertman","Tsai","Cornford","Umbi"}

        % u=C tau^m

        tau=100 ; % units meters, year , kPa
        MeasuredSpeed=sqrt(Meas.us.*Meas.us+Meas.vs.*Meas.vs);
        Priors.m=F.m;
        C0=(MeasuredSpeed+1)./(tau.^Priors.m);
        Priors.C=C0;


    case {"Budd","W-N0"}

        hf=F.rhow.*(F.S-F.B)./F.rho;
        hf(hf<eps)=0;
        Dh=(F.s-F.b)-hf; Dh(Dh<eps)=0;
        N=F.rho.*F.g.*Dh;

        MeasuredSpeed=sqrt(Meas.us.*Meas.us+Meas.vs.*Meas.vs);
        tau=100+zeros(MUA.Nnodes,1) ;
        C0=N.^F.q.*MeasuredSpeed./(tau.^F.m);
        Priors.C=C0 ;
        Priors.m=F.m ;

    otherwise

        error("Ua:DefineInputsForInverseRund:CaseNotFound","Sliding law prior for this sliding law not implemented")
end

Define Start Values

This is only used at the very start of the inversion. (In an inverse restart run the initial value is always the last values from previous run.)

InvStartValues.C=Priors.C ;
InvStartValues.m=F.m ;
InvStartValues.q=F.q;
InvStartValues.muk=F.muk ;
InvStartValues.AGlen=Priors.AGlen;
InvStartValues.n=F.n ;

end

Published with MATLAB® R2021a