tt=readtimetable('XWinddata.xlsx');
head(tt,1)
tt=removevars(tt,{'yearhour','DoY'}); % clean so start over
tt=addvars(tt,day(tt.Timestamp,'dayofyear'),hour(tt.Timestamp),'NewVariableNames',{'DoY','HoD'},'After',{'PressurePa'});
ttHr=varfun(@mean,tt,'GroupingVariables',{'DoY','HoD'},'OutputFormat','table');
height(ttHr)
[head(ttHr);tail(ttHr)]
But, you are missing most of the odd hours...
[numel(unique(tt.DoY)) numel(unique(tt.HoD)) numel(unique([tt.DoY tt.HoD],'rows'))/2 366*24]
As the above confirms, there are only 4,486 unique DoY, HoD combinations, not the 8784 that exist in a 366-day year. That's because you have data for most days only on two-hour increments.
This also leads back to the problem that the poster in one of the earlier links had in that there are leap years and they will show up with years that do have 366 days in them and the DOY algorithm will always count them in its conversion, even if you were to remove Feb 29 data from the timetable before computing DoY.
You didn't show us the top level user code error, only the internal error trap so we can't tell for absolute certain what went wrong; I'd guess the problem was you didn't remove the 'yearhour' variable (or use 'InputVariables' named argument pair) to not try to average it as well -- being nonnumeric, you can't take the mean of it.
NOTA BENE: However, it appears as you have only a very limited number of observations for the odd hours; whether to try to interpolate to fill those first or not I'll leave to you to decide...
Says there almost two entries for each time stamp, however, and interpolation only works for unique x values so will have to average first and then fill in..
tt=removevars(tt,{'DoY','HoD'}); % clean to start over
tt=rmmissing(retime(tt,'hourly','mean')); % average remove missing hours
tt=retime(tt,'hourly','linear'); % now fill in odd hours
tt=addvars(tt,day(tt.Timestamp,'dayofyear'),hour(tt.Timestamp),'NewVariableNames',{'DoY','HoD'},'After',{'PressurePa'});
[numel(unique(tt.DoY)) numel(unique(tt.HoD))]
isleapyr=@(t)(eomday(year(t),2)==29); % leap year logical function
isleapday=@(t)((month(t)==2)&(day(t)==29)); % leap day logical function
tt(isleapday(tt.Properties.RowTimes),:)=[]; % now remove Feb29 globally
[numel(unique(tt.DoY)) numel(unique(tt.HoD))] % still 1:366 in DoY
ix=isleapyr(tt.Properties.RowTimes)&(month(tt.Properties.RowTimes)>2); % dates needing fixup
tt.DoY(ix)=tt.DoY(ix)-1; % and adjust to 365 days
[numel(unique(tt.DoY)) numel(unique(tt.HoD))] % now we're back to 365 day year w/ DoY to match
ttHr=varfun(@mean,tt,'GroupingVariables',{'DoY','HoD'},'OutputFormat','table');
height(ttHr)
[head(ttHr);tail(ttHr)]
The above does it by just throwing away Feb 29, you could alternatively perhaps collapse Feb 28-29 to Feb 28 or Feb 29/Mar 1 to Mar 1 to somehow keep the values observed in the mix, but the above is the simplest approach, certainly.
