Interpolation not-equally spacing data

Hi.

I'm working with the following data:

Name           Size              Bytes  Class 

  X          136x1               544  single              
  Y          32x136            17408  single              
  var        32x136            17408  single   

Where X is equally spaced data but Y is not. Y represent depth in meters (negative values) and I'm trying to interpolate to a finner grid. I have use two approaches:

• Trying with INTERP1. In this case, I first remove depths lower than 20m, then interpolate "column by column":

  for i1 = 1:size(Y,2)
    DEP = Y(:,i1);
    % If the column depth is bigger than -20m
    if min(fix(DEP)) < -20;
        zx = -20:1:0;
    else
    % If the column depth if smaller than -20m
        zx = min(fix(DEP)):1:0;
    end
    out_var(1:size(zx,2),i1) = interp1(DEP,var(:,i1),zx,'nearest','extrap');
    new_dep(1:size(zx,2),i1) = zx;
  end

However, the results are not very good. The upper (lower) panel in the figure show the original (interpolated) data.

I need a better interpolation in the right side because of the gradient. I already have tried with the different method in interp1 but non of theses have work as I need it. I also have tried smooting this result but I get a worse result.

• Then, I tried with interp2 (In the x-data I keep the same spacing as X):

[x,y] = meshgrid(x(1):0.0333:x(end),-60:1:0);
new_var = interp2(X,Y,var,x,y );

But I get the following error:

Error using interp2>makegriddedinterp
Input grid is not a valid MESHGRID.
Error in interp2
F = makegriddedinterp(X, Y, V, method,extrap);

Is there another aproach to get a better interpolation?

Thanks in advance,

[Jan]

It is a mathematical difference, if you interpolate in 1 or 2 dimensions. You have to decide for the method, with matchs your model. You cannot simply choose the method, which produces the nicer graphics.

It matters if you choose "nearest" or "linear" interpolation. You need a physical argument to decide for the matching method.

You post 2 graphics. Which on is the interpolated one?

[user20912]

The second one is the interpolated one, sorry for not indicate it. I have edit the post to clarify this.

I agree with you that the method should be chosen with a physical argument. However, the graphics is a way to check the results, right?

EDIT: In my approach, what I'm trying to do. Is to reproduce my var over a more fine grid. I don't know if this is the right place to ask this, but anyhow... How should I decide which method to use?

Answer 1

[Star Strider]

I generally use griddata for these sorts of problems.

I cannot determine if it will produce the desired result since I don’t have your data to experiment with. It has generally produced satisfactory results when I’ve used it.

[user20912]

I'll check that function. Thank you

[Star Strider]

My pleasure!