Making an equidistant latlon grid from a non equidistant latlon grid

I have the following type of ascii dataset: A 3 vector array with gridded latlon data with size [3,62880] in which the first column represents longitude values, the second column latitude values and the third height values. This consists of a full latlon grid made onedimensional. The latitude and longitude gridsizes are 240 and 262 respectively. (So this data can be easily put into a grid for plotting purposes). But (coming from a UTM conversion) the lat and lon values are not equidistant (and unevenly spaced) and I want to get an equidistant grid. So far I have tried to it like this:

   x = data(1,:); %longitude
  y = data(2,:); %latitude
  z = data(3,:); %height
  xmin = min(x); ymin = min(y);
  xmax = max(x); ymax = max(y);

nlat = 240;
mlon = 262;

xv = linspace(xmin, xmax, mlon);
yv = linspace(ymin, ymax, nlat);
[X,Y] = meshgrid(xv,yv); 
[X,Y,Z] = griddata(x,y,z,X,Y);
ID = isnan(Z);
Z(ID) = -9999;

Is this the correct way to do it? I obtain quite a bunch of missing values (which I set to -9999 for further processing) with trying different interpolation (should I use something else than griddata and meshgrid?) methods, and I want to minimize that as much as possible. Besides, I feel like some of the data is lost with this method. Could somebody recommend me the best way to convert a non equidistant latlon grid to a equidistant grid with correctly interpolated height values?

I hope I have provided enough information for you to help me, if not, ask me to provide more. I feel I might be doing something wrong with the conversion from the 1D vectors to the 2D grid as well (but find it hard to write down how it is originally put into the current array) but I doubt it has something to with the missing values.

Thanks in advance!

(for people known with NCL this is the way I usually read this data (which might explain the format)):

lon2d = onedetond(x,(/nlat,mlon/)) lat2d = onedtond(y, (/nlat,mlon/)) z2d = onedtond(z,(/nlat,mlon/))

[Walter Roberson]

Dang, no wonder I could never locate these kinds of questions to refer to them: I kept reading the term as "lation" instead of "latlon". Time for new glasses, I think!

Answer 1

[Bjorn Gustavsson]

Look at the matlab function: griddata, (there is nowadays a similar function with a new name that eludes me at the moment.) or John D'Errico's gridfit function: http://www.mathworks.se/matlabcentral/fileexchange/8998-surface-fitting-using-gridfit, that provides some smoothing as well, while griddata does a triangulation-interpolation.

HTH,

[JMvanwessem]

Thanks. But again: I want to know which is the most accurate, but I might just try to figure that out myself. Now I have some methods to compare. griddata,TriScatteredInterp and gridfit :)

Answer 2

[JMvanwessem]

Any useful replies? anyone?

I am now getting about to use TriScatteredInterp to just process the entire rows of vectors. I can get it all to work, but I really wonder which one is the most accurate.

Answer 3

[Walter Roberson]

MATLAB File Exchange, John D'Errico's inpaint_nans

[JMvanwessem]

Seems a bit of a workaround, but probably gets the job done. You advise this over the standard matlab functions?

[Walter Roberson]

It depends what you want to do with the missing values. inpaint_nans would be for the case where you want the values to be interpolated to fill in the holes, taking in to account that multiple adjacent values might be missing; John's routine takes in to account curvatures of the area around, whereas most other routines that try to fill in for nans just use linear interpolation.