How do you make the x-axis in imagesc using datetime values?

Hi,

I have data that I wish to display using imagesc with the y-axis being a unit of distance and the x-axis being a datetime array, however, I receive an error when I do this. Any help would be appreciated.

Thanks.

---

For example:

• data : 150 x 1200 x 3 unit8 array of data
• x : 150 x 1 array of distances that correspond to the information in "data"
• t : 1200 x 1 datetime array of times that correspond to the information in "data"

imagesc([t(1) t(end)],[x(1) x(end)],I)

Throws the following error:

Error using image
Image XData and YData must be vectors.

Error in imagesc (line 20)
    hh = image(varargin{:},'CDataMapping','scaled');

Answer 1

[Benjamin Kraus]

Starting in MATLAB R2023b, both the image and imagesc commands support datetime and duration data for the XData and YData properties.

nexttile
x = [0 1];
t = datetime(2023,9,20)+hours([0 1]);
imagesc(t,x,peaks);
nexttile
x = [0 1];
t = hours([1 23]);
imagesc(t,x,peaks);

If you are working with a release older than MATLAB R2023b you can use the workaround discussed in this answer.

Answer 2

[Walter Roberson]

In the case where the x axis is to be a duration array (rather than a datetime array):

First, plot something using an x axis in the same time units as you want the imagesc to be labeled with.

Then

hold on
h = imagesc(I);
set(h, 'XData', [1, seconds(t(end))]);

Modify the "seoonds" to the appropriate time measure. The purpose of the seconds() there is to strip off the duration class, converting the t values to pure numeric values. image() will treat the numeric values as if they were in the active duration units.

This is, of course, a hack.

If your values are datetime arrays then it appears you can do something similar. It appears that the default XData for the imagesc() will be converted into days, probably relative to the first date (but double check -- it might perhaps be relative to the month of the first date.) You cannot use days() on a datetime array to convert to number of days, but you can do

set(h, 'XData', [1 days(t(end)-t(1))]);

You might need to play with the boundary conditions a little.

[Hannah]

Thanks but I have a numeric array of seconds and this is what I am using currently, but was hoping to be able to actually use the new datetime variable type given the improvement that it is over the datenum format.

I've since discovered that datetime values also don't work with the "scatter" function. It's a shame that Matlab has implemented this new (and very functional variable type) but not made it compatible with other functions.

[Hannah]

Further to my previous comment, I realise that the problem with scatter and datetime values has been addressed in more recent versions than mine (R2016b). Can anyone comment if the issue has been addressed for datetime and imagesc? Thanks.

[xchen80]

Hi Hannah, just found a way to work around:

1. Use "datenum" to convert datetime values to a date number vector

2. Use the date number vector in image/imagesc

3. Download the "datetickzoom" function from here and use it to label your axis

https://www.mathworks.com/matlabcentral/fileexchange/15029-datetickzoom-automatically-update-dateticks

[patrick]

thanks @xchen80 !!

[Peter Perkins]

Patrick, I don't think datetime/duration support has made it to image/imagesc yet, even post-16b. I'll make a note to look into solving that.

[Damsel in Distress]

Hi,

Has the datetime/duration support made it into a released version yet? I am running 2017b and cannot get it to work (perhaps it is user error?!). If not do we know which release this is likely to be fixed in?

Thanks!

[Andrew McLean]

I've tried with Matlab R2019a it doesn't work. Does it work with R2020a, documentation suggests not?

Answer 3

[KSSV]

doc xticklabel

https://in.mathworks.com/help/matlab/creating_plots/change-tick-marks-and-tick-labels-of-graph-1.html

[Hannah]

I'm aware that you can manually change the tick marks on axis labels, but the benefit of the datetime function is that it changes as you zoom, hence my desire to use it. If I wasn't fussed about zooming, I'd use datetick

Answer 4

[Erik]

I found a way arround. It is possible to add an image produced by imagesc to an already existing axis with a duration ruler.

Your duration range in the imagesc call should be converted to numeric. As you can see in figure below, the graph keeps the duration ruler for the XAxis. .

In the code, you can find that the order of the image and other axis children (e.g. plotlines) can changed to dept, which effectively set's image below the rest. In 3D plots the order is default dept I presume.

I did not do that but made the image partly transparent to show the dots of the first plot command.

Succes Erik Esveld

function PlotFlirData(FD)
%% Plot FD data
%   FD.time       aquisition type duration 
%   FD.date       aquisition type datetime
%   FD.mean       mean temperature vector
%   FD.temp       vector of temperature bin (from T-0.5 to T+0.5)
%   FD.tempfrac   isotermal area fraction maxtrix (time,temp)
%
% 2019 [e-mail]


%% Heatmap in time plot
% we use imagesc ( heatmap itself does not work well)
% note that cdata(1,1) is nomally uppper left
% but not when ax.YDir = 'Normal' such as in plot

fi = figure;
fi.Color = 'w';

% because imagesc does not support creation of duration rulers, 
% we first use plot with duration as x data and mean value as y data
plot(FD.time,FD.mean,'.k');
hold on;

% give the ranges for the image data
timerange = [FD.time(1) FD.time(end)];
temprange = [FD.temp(1) FD.temp(end)];
fracrange = [0 0.3];
% convert x duration to num and plot transposed 
im = imagesc(gca, datenum(timerange),temprange,FD.tempfrac',fracrange);

% plot a line with the time average value
plot(timerange,repmat(mean(FD.mean),1,2),'--b');

% make image transparant to show the datapoint from the first plot
im.AlphaData = .7;
% or change sort order from chlildren to depth (image depth = [-1 1])
ax  = gca;
% ax.SortMethod= 'depth';
colormap(flipud(bone));

% set scales
% xlim(timerange);
xlim(duration({'9:37:00','9:38:00'}));
ylim([-20 0]);

% set labels
xlabel(['Time at ' datestr(FD.date)]);
ylabel('Temperature (°C)');

% set window and axes
fi.Position = [219 584 952 301];
ax.Position = [0.0683 0.1827 0.8897 0.7423];
end

[Martin Ryba]

Note, to be pendantic, you'll need to adjust the time values to be from 0.5...N+0.5 times their timestep since the image data is nominally for the centers of the facets and imagesc fixes that for you. It's part of the reason people prefer imagesc over surf/mesh/pcolor.

[Martin Ryba]

After experimenting, this seems to not work on R2021b; imagesc(gca,...) resets the axis range based on the datenums sent in which screws everything up.

Answer 5

[Martin Ryba]

See the wonderful dynamicDateTicks function. Just tried it and it is marvelous!