Find the first element that satisfies a condition on 3D data

I'm trying to use a logical expression to find the first instance of a condition being met on a 3d matrix.

I have this dummy code for a 1d array that works the way it should:

 d = randi([0 50], 100, 1);
 dd = sort(d, 'descend');
 pa = sum(dd);
 dc = cumsum(dd);
 %find row where cumsum is 50% of sum
 g = find(dc >0.5*pa,1);
    
    

However, I can't get it to work on my 3d data of longitude, latitude and time (360,150,365). I want to find the n'th timestep (in the third dimension) where SC is equal to or greater than 1/2 of the sum at each latitude-longitude pair and write that to a new matrix (2d, 360X150). My current attemp is as follows:

   prec = ncread(filename_p, 'Prec_gs'); % 360x150x365 matrix
   p_annual(:,:) = sum(prec,3,'omitnan'); % 360x150 matrix
   S = sort(prec,3,'descend'); % 360x150x365 matrix
   SC = cumsum(S,3); % 360x150x365 matrix
   N50(:,:) = find(SC(:,:,:) >= 0.5.*p_annual(:,:), 1);
   

The last row isn't working because the the dimensions are not the same (3d vs 2d), I have tried making p_annual a 3d matrix of same size as SC but still no luck.

Any thoughts on how to solve this?

[David Hill]

An example of your data input and expected output would be helpful.

[Elin Jacobs]

I have attached a sample data file (20x30x365).

Code to open/read the file:

filename = 'GLDAS_cropped.nc';
ncid = netcdf.open(filename);
prec = ncread(filename, 'precipitation');

I expect to get a 20x30 matrix with the element (a number between 1 and 365) where SC is equal to or greater than 50% of p_annual. Thanks for your help.

Answer 1

[Matt J]

   S = sort(prec,3,'descend'); % 360x150x365 matrix
   SC = cumsum(S,3,'omitnan'); % 360x150x365 matrix
   p_annual = SC(:,:,end);

   [~,N50] = max(SC >= 0.5.*p_annual,[], 3);
   

[Elin Jacobs]

This gives me the same issue I had before, dimensions not matching.

[Matt J]

I don't see any evidence of a problem:

load precipitation.mat  

   S = sort(prec,3,'descend'); % 360x150x365 matrix
   SC = cumsum(S,3,'omitnan'); % 360x150x365 matrix
   p_annual = SC(:,:,end);

   [~,N50] = max(SC >= 0.5.*p_annual,[], 3);
   
   whos

  Name           Size                  Bytes  Class     Attributes

  N50           20x30                   4800  double              
  S             20x30x365            1752000  double              
  SC            20x30x365            1752000  double              
  ans            1x42                     84  char                
  p_annual      20x30                   4800  double              
  prec          20x30x365            1752000  double              

[Elin Jacobs]

Interesting, I get the following error:

Error using >=

Number of array dimensions must match for binary array op.

I'm using R2016a, I will try a different version and see if it helps.

[Matt J]

The code assumes R2016b or higher. I do recommend upgrading, but if you can't then you can make this modification instead,

 [~,N50] = max(  bsxfun(@ge, SC, 0.5.*p_annual) ,[], 3);

[Elin Jacobs]

Yep, that works, thank you so much!

[Elin Jacobs]

To get it right, I needed to put NaN values last when sorting, but otherwise works like a charm!

S = sort(prec,3,'descend','MissingPlacement','Last');