regriding an irregular matrix

Dear all,

I have a 3D irregular matrix in time and position below :

A = 3000 x 3

Column1 = time (vary from 2002-2005)

Column2 = Longitude (vary from 40-50)

Column3 = Value

A = [2Jan2002 40 65

   6Jan2002 45 22

   8Jan2002 43 45

   8Jan2002 49 64

   30Jan2002 42 78

   4Feb2001 43 56

   6Feb2001 47 67

       :     :  :

   23Nov2005 44 76

   3Dec2005 42 89

   5Dec2005 48 67

   9Dec2005 44 78

   13Dec2005 41 89

   19Dec2005 49 34

   23Dec2005 45 67

   24Dec2005 43 88  

   31Dec2005 47 34];

I want to regridding this irregular data in time and position into regular matrix based on monthly (2002-2005; so 48 month) and 1 degree lon position (40-50; so 10 points), in summary the output mtrix will be 10x48.

Some points longitude will have NaN value cause no record in this months at x longitude.

Anyone can help?

Best regards,

[Cedric]

This is not a valid array as it mixes strings and numeric values. How are your data truly stored at this point? If you have time, longitude, and value, you just need a 2D array, as you are mentioning later in the question when you define the dimension of your final array. Do you have in fact something like a 3 columns CSV or Excel file with the original data? If so, assuming that we can treat "31Dec2005"-like timestamps correctly, how do you want to summarize values on a per month basis; is it a simple average?

Answer 1

[Andrei Bobrov]

If you have data as text file - yourdata.txt (eg):

   2Jan2002 40 65
   6Jan2002 45 22
   8Jan2002 43 45
   8Jan2002 49 64
  30Jan2002 42 78
   4Feb2002 43 56
   6Feb2002 47 67
  23Nov2005 44 76
   3Dec2005 42 89
   5Dec2005 48 67
   9Dec2005 44 78
  13Dec2005 41 89
  19Dec2005 49 34
  23Dec2005 45 67
  24Dec2005 43 88  
  31Dec2005 47 34

please try this is code:

f=fopen('yourdata.txt');
c = textscan(f,'%s %f %f');
fclose(f);
[y,m] = datevec(c{1});
[m1,y1] = ndgrid(1:12,2002:2005);
[i2,i2] = ismember([y,m],[y1(:),m1(:)],'rows');
[j2,j2] = ismember(round(c{2}),40:50);
out = accumarray([i2,j2],c{3},[48,10],@mean,nan); % [EDITED]

[m0xty Wilopo]

This is work! Thanks Andrei

Answer 2

[m0xty Wilopo]

Yes, the date is datenum precisely, I just want to show that the first column is date in string. It is in numeric.

Answer 3

[Cedric]

A general approach could be the following, assuming that you have this matrix A with datenums in the first row.

 [y, m] = datevec(A(:,1)) ;

 dateId = ((y - min(y))*12 + m).' ;
 nYears = range(y) + 1 ;

 lonId  = A(:,2) - min(A(:,2)) + 1 ;
 nLon   = range(A(:,2)) + 1 ;                 % You might want to fix this 
                                              % instead of having it flexible.
 dataArray = nan(nLon, nYears*12) ;
 ind = sub2ind(size(dataArray), lonId, dateId) ;
 dataArray(ind) = A(:,3) ;

Now, as mentioned in my comment, we should improve it a bit if you can have cases where there are multiple values for the same pairs dateId/lonId. Just let me know and we can extend it using ACCUMARRAY.

Answer 4

[m0xty Wilopo]

Dear Cedric and Andrei :)...Many THANKS! As usual it was really helpful. However if I have data at same longitude, I want to averaging it (mean) how to do that ie:

   2Jan2002 40 65
   6Jan2002 45 22    ---> So I will have value 24 instead of 48
   7Jan2002 45 26
   8Jan2002 43 45
   8Jan2002 43.1 51 ---> result will be 48 at lon 43
   8Jan2002 49 64
  30Jan2002 42 78
   4Feb2002 43 56
   6Feb2002 47 67
  23Nov2005 44 76
   3Dec2005 42 89
   5Dec2005 48 67
   9Dec2005 44 78
  13Dec2005 41 89
  19Dec2005 49 34
  23Dec2005 45 67
  24Dec2005 43 88  
  31Dec2005 47 34

How to do that? Andrei giving answer using accumarray in my previous question but it was for another data set in same longitude. But now I want averaging it at multi longitude/position.

[Andrei Bobrov]

Please see my answer after edited.