One problem is that setting XLimMode to 'manual' can have the effect of changing the XLim.
Example:
% create a figure, axes, and plot:
figure()
ax = axes();
plot(ax,0:10,0:10)
% at this point XLim *should be* [0 10], and setting XLimMode to manual
% *should* not change them ...
ax.XLimMode = 'manual';
get(ax,'XLim') % ... but in fact the XLim are [0 1] now
Obviously you could set the XLim before setting XLimMode, but I assume you want to let MATLAB pick the XLim for you (and then you're setting XLimMode to 'manual' to have them be fixed in place).
Interestingly, merely querying the XLim (not setting them) before setting XLimMode to 'manual' seems to give the expected behavior:
figure
ax = axes();
plot(ax,0:10,0:10)
get(ax,'XLim') % now XLim are [0 10]
ax.XLimMode = 'manual';
get(ax,'XLim') % and XLim remain [0 10] after setting XLimMode
You could also insert a drawnow before setting XLimMode:
figure
ax = axes();
plot(ax,0:10,0:10)
drawnow();
ax.XLimMode = 'manual';
get(ax,'XLim') % and XLim remain [0 10] after setting XLimMode
With that observation in mind, including a few drawnow() commands at strategic points in the code seems to fix all the problems. See explanations in the comments in the code below:
% %reshape 4D arrays of data to 2D arrays
% y1=reshape(A,1,12);
% y2=reshape(B,1,12);
% y3=reshape(C,1,12);
%
% %reshape 4D arrays of error to 2D arrays
% y1_err=reshape(A_err,1,12); %y1
% y2_err=reshape(B_err,1,12); %y2
% y3_err=reshape(C_err,1,12); %y3
% some data, since I don't have your data:
y1 = randi(100,1,12);
y2 = randi(20,1,12);
y3 = randi(70,1,12);
y1_err = 10*rand(1,12);
y2_err = 2*rand(1,12);
y3_err = 7*rand(1,12);
% %set x axis
% x = datetime(2002,1:(16*12),1,'Format','MMM-yyyy');
% set x-coordinate of lines (I'm not sure how your x was 1-by-192
% when your y1, y2, etc., were all 1-by-12):
x = datetime(2002,1:12,1,'Format','MMM-yyyy');
%plot y1,y2
figure
ax1=axes;
yyaxis left;
y1plot=errorbar(x,y1,y1_err,'color',[0.8500,0.3250,0.0980]); %y1 on left
ax1.YColor=[0.8500,0.3250,0.0980];
ylabel('y1');
% this drawnow() allows ax1 properties to update before
% setting their modes to manual:
drawnow();
ax1.XTickMode='manual';
ax1.YTickMode='manual';
ax1.YLim=[min(ax1.YTick), max(ax1.YTick)];
ax1.XLimMode='manual';
grid(ax1,'on')
ytick=ax1.YTick;
yyaxis right
y2plot=errorbar(x,y2,y2_err,'-b'); %y2 on right
ax1.YColor='b';
ylabel('y2');
% this drawnow() allows ax1.Position to update before using it for
% ax2.Position, so that the axes overlap properly
drawnow();
%create 2nd transparent axes & plot y3
ax2 = axes('position',ax1.Position);
y3plot=errorbar(ax2,x,y3,y3_err,'-k'); %y3 on left
ylabel('y3');
ax2.Color='none';
% grid(ax2,'on'); % Note: ax2 grid is not necessary, because ax1 has the same grid
% (the x-limits of ax1 and ax2 are the same, and ax2.YTick
% is set below to correspond to ax1.YTick)
ax2.YAxis.TickLabelFormat='%.2f';
%horizontally scale the y axis to align the grid
ax2.XLim=ax1.XLim;
ax2.XTick=[];%ax1.XTick; % Note: I set ax2.XTick to [] (no ticks) to avoid duplicating
% the xticklabels that already exist in ax1
% this drawnow() allows ax2.YLim to update before setting ax2.YLimMode
drawnow();
ax2.YLimMode='manual';
% Note: I'm calling this variable "y_limits" instead of "y1" because I
% don't want to overwrite the data variable "y1". (I could use "yl",
% with a lower-case L, but that's difficult to distinguish from "y1"):
y_limits=ax2.YLim;
ax2.YTick=linspace(y_limits(1),y_limits(2),length(ytick));
%horzontally offset y tick labels
% ax2.YTickLabel=strcat(ax2.YTickLabel,{' '});
% Note: I removed a few spaces to be able to see the
% YTickLabels of ax2 here in the forum environment
ax2.YTickLabel=strcat(ax2.YTickLabel,{' '});
As an alternative to placing several drawnow commands in code, you could rearrange the code so that all the plotting and axes creation is done first, and then all the axes appearance modification (setting labels, ticks, limits, etc.) is done afterward, but that's a little tricky because of the use of yyaxis.
