Consider real world coordinates in 2D ERT

[Boorhin]

Hi,
I am inverting some field 2D ERT data and I would like to load the saved vtk into a real world 3D environment.
I can reproject the whole final mesh but I was wondering if there was any alternative while I do the data loading.
I tried to give the sensors a xyz

for s in range(len(data[f"ERT{i}"].sensors()))
  data[f"ERT{i}"].setSensorPosition(s,[xy[s,0],xy[s,1],z[s]])

but this has sent some confusion as it is expecting a distance and a z if I understand well. I tried to understand the documentation without success.
Thanks for your help in advance,

[halbmy]

Rotating the data will lead to 3D data and this does not help, but rotating the final mesh should do it. But I agree there could be an example and some helping function for 2D-3D conversion of models and data.

[Boorhin]

any trick to rotate the final mesh without doing univariate interpolation or the sort?
thanks

[halbmy]

The function model2Dto3D(mesh, txy) converts (by interpolation) a result into the 3D space defined by polygonal.

An existing profile with 3D coordinates can be converted into 2D by data3Dto2D(data), also returning the interpolation

I've created an example where all this is used on https://github.com/gimli-org/notebooks/blob/main/ERT/2din3d/ert2dinv3d.ipynb

The functions will be further simplified and embedded in pyGIMLi.

[Boorhin]

Thanks, 
that's kind of you.
I have done similar exercises and in theory the distance is correct and you can rotate your data. However in real field examples, I have had to reproject the profile from a d-z space to a xyz space of local coordinates from sinuous setups.
Because the surveys are never straight, the only method I know is to work with curvilinear abscissa in d vs each 3 space variables. 
If I follow your fx()  def model2Dto3D(mesh, xy):

I suggest the following function that will accurately restore the 3D geometry of a profile from the D space into a 3D mesh surface.

def modelDztoxyz(mesh, d,x,y,z, deg=3):
    """Convert Dz-coordinates mesh to geographic coordinates by curvilinear interpolations.

    Parameters
    ----------
    model2d : pg.Vector
        2D model vector.
    d : ndarray
        Distance of sensors from start (x in 2D mesh)
    x : ndarray
        easting geographic coordinates of sensors at known distance
    y : ndarray
        northing geographic coordinates of sensors at known distance
    z : ndarray
        altitude geographic coordinates of sensors at known distance
    deg : int
        degree of curvilinear interpolation. A minimum of 3 is recommended
    Returns
    -------
    projected mesh
    """
    import scipy.interpolate as SC
    #padding the spline could help having a better result as well as working on a smoothing factor
    x_spline = SC.InterpolatedUnivariateSpline(d, x, k=deg)
    y_spline = SC.InterpolatedUnivariateSpline(d, y, k=deg)
    z_spline = SC.InterpolatedUnivariateSpline(d, z, k=deg)
    md= pg.x(mesh)
    mz = pg.y(mesh)
    out = mesh.copy()
    out.setDimension(3)
    for i in range(len(out.nodes())):
        xp,yp=float(x_spline(md[i])), float(y_spline(md[i]))
        zp=float(z_spline(md[i]))+mz[i]
        out.nodes()[i].setPos([xp, yp, zp])
    return out

if you find the function of interest you can integrate it. My experience was pretty good with it in particular when acquisition lines are crooked due to e.g. obstacles.
It is untested in your setup

[Boorhin]

while testing it I ran in a funny error

ArgumentError: Python argument types in
    Node.setPos(Node, list)
did not match C++ signature:
    setPos(GIMLI::Node {lvalue}, GIMLI::Pos pos)

It seems that what comes out of scipy is akind of float that the C++ side doesn't like.
I tried with x,y,z = 1,2,3 and it works... I will look at a solution

[Boorhin]

ok spline outputs an array so that was the trap. I edited my function accordingly.