mean value of a group of data with NaNs

I have a matrix A with the dimension of 12x360x180. It stores 12 values at each grid of the 360x180. Now I want to calculate the average value at each grid point. The problem is that there are unknown number of NaNs. Sometimes all 12 values are non-NaNs, sometimes all of them are NaNs, sometimes a portion of the 12 values are NaNs.

In this case, how do I estimate the mean values at each point of the grid? I know if there are no NaNs, the calculation is as easy as B = mean(A);

Thank you.

Answer 1

[the cyclist]

If you have the Statistics and Machine Learning Toolbox, you can use the nanmean function, which computes the mean while ignoring NaNs.

[Leon]

Many thanks! It works very well!

I happen to have Statistics toolbox on my version of the Matlab :-)

Answer 2

[Chris Turnes]

You can also just use the 'omitnan' option in "mean":

A = [1 0 0 1 NaN 1 NaN 0];
M = mean(A,'omitnan')
M =
    0.5000

[Leon]

Wow, this is refreshing! Thank you for sharing!

[Chris Turnes]

By the way, other related functions support this input argument as well: sum, var, std, median, and max and min (max and min have 'omitnan' as the default option). The mov* functions (if you have R2016a or newer) also support this argument.

[the cyclist]

I was unaware of this option! Very nice. Please accept my upvote.

Answer 3

[the cyclist]

If you do not have that Statistics and Machine Learning Toolbox, this should work:

notNan = not(isnan(A));

B = zeros(size(A));
B(notNan) = A(notNan);

sum(B)./sum(notNan)

You can do the sum over different dimensions, as required.

[Leon]

Brilliant! Good to have this alternative method for folks who do not have the Statistic toolbox and come to this page through search engine.