Assuming you have a vector named temperature that stores the temperatures and a vector named dt that stores the datetime stamps (in datetime format), you can use findgroups() to group the data into decades and then splitapply() to calculate the group means.
% Create demo data
dt = datetime(1950,1,1):datetime(2019,12,15);
temperature = rand(size(dt)) * 100;
% Group by decade
[groups, groupID] = findgroups(floor(year(dt)/10)*10);
% Compute decade means
decMeans = splitapply(@mean,temperature,groups);
% Put results into a table
results = table(groupID.', decMeans.','VariableNames',{'Decade','MeanTemp'});
View results
results =
7×2 table
Decade MeanTemp
______ ________
1950 49.348
1960 50.959
1970 50.13
1980 49.847
1990 49.808
2000 49.471
2010 49.404
[update]
If there are NaN values you'd like to ignore,
decMeans = splitapply(@(x)mean(x,'omitnan'),temperature,groups);
