Monthly sum function with nansum not being used correctly

I use the function below to perform the monthly precipitation accumulator, however, when there are days without NaN measurements in a given month, the result for that month is zero. But this result is wrong, the correct series for this month is NaN and not zero.

Another problem seen in the function is the result of months. From January 2000 to December 2020 there are 252 months. when I use the function, the result is only 251 months, that is, one month to go

Below is:

1 - the function used for the monthly sum.

2 - rainfall data, in column 1 the days of the year and in columns 2, 3, 4 and 5 the rainfall (INMET_Diario.mat);

function Y = monthly_sum(time_X,X)

[l,c] = size(X);


grp = [];
for y = 2000:2020 % Change period here
    for m = 1:12
        grp = [grp datenum(y,m,1)];
    end
end
    

X_hour = zeros(length(grp)-1,c);
for i = 1:length(grp)-1
        if(mod(i,500)==0)
                disp(i)
        end
        for j = 1:c
            X_hour(i,j) = nansum(X(time_X>=grp(i) & time_X<grp(i+1),j)); 
        end

end


Y = [grp(1:length(grp)-1)', X_hour];

Best Regards,

AP

[Mathieu NOE]

hello

why not use this kind of code ?

here the data are organized as [date (yyyymmdd) rain fall data]

This is a typical example where you can use table instead of matrix.

A = [19260702    0.026   0.000   NaN 1.175
     ...
     19260816    0.042   0.007   NaN 25.928];

% Matrix to table conversion
T = array2table(A,'VariableNames',{'DATE','S1','S2','S3','S4'});
% We add a new column "month"
T.MONTH = floor(T.DATE/100);
% varfun can apply a custom function to your table and group the result according 
% to one (or more) variable(s)
Result = varfun(@sum,T,'InputVariables',{'S1','S2','S3','S4'},'GroupingVariables','MONTH')

[Augusto Gabriel da Costa Pereira]

If my goal is to AVERAGE, replace @sum with @mean on the last line?

[Mathieu NOE]

yes that works also

A = [19260702    0.026   0.000   NaN 1.175
     19260705    0.036   0.002   NaN 2.175
     19260715    0.044   0.003   NaN 3.175
     19260816    0.042   0.007   NaN 25.928];
% Matrix to table conversion
T = array2table(A,'VariableNames',{'DATE','S1','S2','S3','S4'});
% We add a new column "month"
T.MONTH = floor(T.DATE/100);
% varfun can apply a custom function to your table and group the result according 
% to one (or more) variable(s)
Result_sum = varfun(@sum,T,'InputVariables',{'S1','S2','S3','S4'},'GroupingVariables','MONTH')

Result_mean = varfun(@mean,T,'InputVariables',{'S1','S2','S3','S4'},'GroupingVariables','MONTH')

Answer 1

[Jan]

function Y = monthly_sum(T, X)
grp = [datenum(2000, 1:252, 1), Inf];
ngrp = numel(grp) - 1;

c = size(X, 2);
X_hour = nan(ngrp, c);
for i = 1:ngrp
    for j = 1:c
        TX = X(T >= grp(i) & T < grp(i+1), j);
        if ~isempty(TX)
           X_hour(i,j) = nansum(TX); 
        end
    end
end
Y = [grp(1:ngrp).', X_hour];
end

[Augusto Gabriel da Costa Pereira]

thank you Jan

problem fully resolved

[Jan]

You are welcome.

datenum(2000, 1:252, 1) is tricky, but working fine.