Interpolate timeseries of a 4D matrix (4th D as time)

Hey experts, please could you help me to figure this out?

I am trying to interpolate a 4D matrix, in which the 4th dimension is time. It has 12 records (monthly), but I want it each 3 days, so it'll be 121 records).

For this I am trying both ways below:

load data.mat

dtStr = ["15/01/2016","15/02/2016","15/03/2016","15/04/2016","15/05/2016","15/06/2016",...
    "15/07/2016","15/08/2016","15/09/2016","15/10/2016","15/11/2016","15/12/2016"];
dtvA=datevec(dtStr,'dd/mm/yyyy');
dttA = datetime(dtvA);

timeS = datestr((time/86400)+datenum(1958,1,1));
dttB = datetime(timeS);
tmtb = timetable(dttA,Nit2);

NO3_clim_3daily = retime(tmtb,'regular','linear','TimeStep',days(3));
% Or
NO3_clim_3daily = retime(tmtb,dttB,'linear');

But it give me the error: "Error using timetable/retime (line 140). Interpolation requires at least two sample points in each dimension".

Why am I getting this error, if I have a matrix of 31x11x37x12?

I still could not figure it out and how to solve it. Please, could someone help me?

Thanks!

Answer 1

[Stephen23]

Avoid using deprecated DATENUM & DATESTR, using DATETIME is much better.

S = load('data.mat')

S = struct with fields:
    Nit2: [31x11x37x12 double]
    time: [121x1 double]

dttA = datetime(2016,1:12,15);
dttB = datetime(S.time(:),'convertfrom','epochtime','Epoch',datetime(1958,1,1));

"Why am I getting this error, if I have a matrix of 31x11x37x12?"

Note that for a TIMETABLE the first dimension (i.e. rows) must correspond to the time: in contrast the 4th dimension of your array has size 12, thus seems to correspond to the 12 months you specified. So we would need to permute that array so that it fits the TIMETABLE definition:

tmtb = timetable(dttA(:),permute(S.Nit2,[4,1,2,3]))

tmtb = 12x1 timetable
       Time              Var1       
    ___________    _________________

    15-Jan-2016    1x31x11x37 double
    15-Feb-2016    1x31x11x37 double
    15-Mar-2016    1x31x11x37 double
    15-Apr-2016    1x31x11x37 double
    15-May-2016    1x31x11x37 double
    15-Jun-2016    1x31x11x37 double
    15-Jul-2016    1x31x11x37 double
    15-Aug-2016    1x31x11x37 double
    15-Sep-2016    1x31x11x37 double
    15-Oct-2016    1x31x11x37 double
    15-Nov-2016    1x31x11x37 double
    15-Dec-2016    1x31x11x37 double

after that RETIME is exactly as you showed:

NO3_clim_3daily = retime(tmtb,'regular','linear','TimeStep',days(3))

NO3_clim_3daily = 113x1 timetable
       Time              Var1       
    ___________    _________________

    15-Jan-2016    1x31x11x37 double
    18-Jan-2016    1x31x11x37 double
    21-Jan-2016    1x31x11x37 double
    24-Jan-2016    1x31x11x37 double
    27-Jan-2016    1x31x11x37 double
    30-Jan-2016    1x31x11x37 double
    02-Feb-2016    1x31x11x37 double
    05-Feb-2016    1x31x11x37 double
    08-Feb-2016    1x31x11x37 double
    11-Feb-2016    1x31x11x37 double
    14-Feb-2016    1x31x11x37 double
    17-Feb-2016    1x31x11x37 double
    20-Feb-2016    1x31x11x37 double
    23-Feb-2016    1x31x11x37 double
    26-Feb-2016    1x31x11x37 double
    29-Feb-2016    1x31x11x37 double

NO3_clim_3daily = retime(tmtb,dttB)

NO3_clim_3daily = 121x1 timetable
       Time              Var1       
    ___________    _________________

    02-Jan-2016    1x31x11x37 double
    05-Jan-2016    1x31x11x37 double
    08-Jan-2016    1x31x11x37 double
    11-Jan-2016    1x31x11x37 double
    14-Jan-2016    1x31x11x37 double
    17-Jan-2016    1x31x11x37 double
    20-Jan-2016    1x31x11x37 double
    23-Jan-2016    1x31x11x37 double
    26-Jan-2016    1x31x11x37 double
    29-Jan-2016    1x31x11x37 double
    01-Feb-2016    1x31x11x37 double
    04-Feb-2016    1x31x11x37 double
    07-Feb-2016    1x31x11x37 double
    10-Feb-2016    1x31x11x37 double
    13-Feb-2016    1x31x11x37 double
    16-Feb-2016    1x31x11x37 double

[Luís Henrique Bordin]

Thank you so much, Stephan,

I know, if I do not permute, the timetable don't work, I did that before writing the code here to ask for help, and should have forgot to include it in the example.

Well, the data I gave you is a subset of the real matrix, since the file was too big to attach.

When trying to apply to my real matrix, it didn't work, and just then I realized the issue it wasn't working, it was because the real matrix had NaNs, and the subset I gave to you does not. By replacing the NaNs by zero, it worked.

Anyway, I learned from your code, so thank you very much for that!

Best,

Luis