corTsgMedian.m

function [error] = corTsgMedian(hMainFig, dateMin, dateMax)
% 
% Correct the TSG salinity time series with the Water sample.
% Use the median value of TIME_WINDOWS water sample to compute the
% correction. see the documentation
% 
% Input
% hMainFig ..... Handle to the main GUI
% dateMin ...... the correction is applied between dateMin and date Max
% dateMax ...... the correction is applied between dateMin and date Max
%
% Output
% Error ........  1 everything OK
%       ........ -1 dateMax <= date Min
%
% TO DO
% corTsgMethod1.m
% 3) Test for anormal sample-tsg difference - Suppress bad points
% 4) Test on the validation code. do we apply the correction whatever the
%    is the code ?

% Get application data
% --------------------
tsg    = getappdata( hMainFig, 'tsg_data');
sample = getappdata( hMainFig, 'sample');

% Shorten the variable name
% -------------------------
TIME_WINDOWS = tsg.cst.COR_TIME_WINDOWS;

% Get PROBABLY_GOOD, PROBABLY_BAD and VALUE_CHANGED codes
% -------------------------------------------------------
PROBABLY_GOOD = get(tsg.qc.hash, 'PROBABLY_GOOD', 'code');
PROBABLY_BAD  = get(tsg.qc.hash, 'PROBABLY_BAD', 'code');
VALUE_CHANGED = get(tsg.qc.hash, 'VALUE_CHANGED', 'code');
    
% Create a structure with an NaN
% No other solution, as I can't add a structure to an empty one
% -------------------------------------------------------------
cor = struct('DAYD', NaN, 'DIFF', NaN, 'ERROR', NaN, 'NVALUE', NaN);

if dateMax > dateMin

  % Find the number of samples within the time limits.
  % --------------------------------------------------
  indSample = find(sample.DAYD >= dateMin & sample.DAYD <= dateMax);
  
  indCor = 0;
  for i=1:length(indSample)

    % Find samples within TIME_WINDOWS with Good and probably Good QC
    % ---------------------------------------------------------------
    ind = find( sample.DAYD    >= (sample.DAYD(i) - TIME_WINDOWS/2) &...
                sample.DAYD    <= (sample.DAYD(i) + TIME_WINDOWS/2) &...
                sample.SSPS_QC <= PROBABLY_GOOD);
    
    if ~isempty(ind)
      
      % detect NaN in sample.SSPS_DIF due to bad QC code in tsg.SSPS
      % ------------------------------------------------------------
      ind2 = find(~isnan(sample.SSPS_DIF(ind)));

      % Compute the median difference and error within TIME_WINDOWS
      % -----------------------------------------------------------
      if ~isempty(sample.SSPS_DIF(ind(ind2)))
        
        A = sample.SSPS_DIF(ind(ind2));
        meanA = mean(A);
        stdA  = std(A);
        
        % Standard deviation test: keep these values
        % ------------------------------------------
        ind3 = find( A >= meanA-3*stdA & A <= meanA+3*stdA);
        
        B = sample.SSPS_DIF(ind(ind2(ind3)));
        if ~isempty( B )
          indCor = indCor + 1;
          cor.DAYD(indCor)   = sample.DAYD(i);
          cor.DIFF(indCor)   = median(B);
          cor.ERROR(indCor)  = nanstd(B)/sqrt(length(B));
          cor.NVALUE(indCor) = length(B);
        end
        
        % Standard deviation test: don't keep these values
        % QC PROBABLY_BAD
        % ------------------------------------------------------
        ind4 = find( A < meanA-3*stdA | A > meanA+3*stdA);
        if ~isempty( ind4 )
          sample.SSPS_QC(ind(ind2(ind4))) = PROBABLY_BAD;
        end

      end
    end
  end

  % Eliminate the first element if NaN
  % ----------------------------------
  if isnan(cor.DAYD(1))
    cor.DAYD(1)   = [];
    cor.DIFF(1)   = [];
    cor.ERROR(1)  = [];
    cor.NVALUE(1) = [];
  end

  if ~isempty( cor.DAYD )

    % The error is maximum if the median is computed with less than 4 samples
    % -----------------------------------------------------------------------
    cor.ERROR( cor.NVALUE < 4 ) = 1;

    % The correction is applied between dateMin and dateMax
    % We attribute to dateMin the first correction computed
    % and to dateMax the last one
    %
    % Find the tsg date in the interval dateMin-dateMax
    % -------------------------------------------------
    dtTsg = find(tsg.DAYD    >= dateMin  & tsg.DAYD <= dateMax);

    if cor.DAYD(1) ~= dateMin
      cor.DAYD   = [tsg.DAYD(dtTsg(1)) cor.DAYD];
      cor.DIFF   = [cor.DIFF(1)        cor.DIFF];
      cor.ERROR  = [cor.ERROR(1)       cor.ERROR];
      cor.NVALUE = [cor.NVALUE(1)      cor.NVALUE];
    end
    if cor.DAYD(end) ~= dateMax
      cor.DAYD   = [cor.DAYD   tsg.DAYD(dtTsg(end))];
      cor.DIFF   = [cor.DIFF   cor.DIFF(end)];
      cor.ERROR  = [cor.ERROR  cor.ERROR(end)];
      cor.NVALUE = [cor.NVALUE cor.NVALUE(end)];
    end

    % The correction is applied to the TSG between dateMin and dateMax using
    % a linear interpolation only on measurements with GOOD and
    % PROBABLY_GOOD QC
    % ----------------------------------------------------------------------
    dtTsg = find( tsg.DAYD    >= dateMin  & tsg.DAYD <= dateMax &...
                  tsg.SSPS_QC <= PROBABLY_GOOD);

    tsg.SSPS_ADJUSTED(dtTsg)       = tsg.SSPS(dtTsg) + ...
                                  interp1(cor.DAYD, cor.DIFF, tsg.DAYD(dtTsg));
    tsg.SSPS_ADJUSTED_ERROR(dtTsg) = ...
                                 interp1(cor.DAYD, cor.ERROR, tsg.DAYD(dtTsg));
    
    % VALUE_CHANGED code
    % ------------------
    tsg.SSPS_ADJUSTED_QC(dtTsg)    = VALUE_CHANGED;
    
  else

    % RAZ if no correction
    % --------------------
    dtTsg = find(tsg.DAYD >= dateMin & tsg.DAYD <= dateMax);

    tsg.SSPS_ADJUSTED(dtTsg)       = NaN*ones(size(tsg.SSPS(dt)));
    tsg.SSPS_ADJUSTED_ERROR(dtTsg) = NaN*ones(size(tsg.SSPS(dt)));
    tsg.SSPS_ADJUSTED_QC(dtTsg)    = zeros(size(tsg.SSPS(dt)));
    
  end
  
  % Update tsg application data
  % ---------------------------
  setappdata( hMainFig, 'tsg_data', tsg);
  setappdata( hMainFig, 'sample',   sample);
  
  % everything OK
  % -------------
  error = 1;

else

  % DateMax <= DateMin
  % ------------------
  error = -1;
  
end