Lately I have been spending quite some time working on the implementation of tools for the visual inspection of out data as they are being produced by the TROPOMI L01b calibration pipeline. As we are heading (in rather quick leaps) towards the on-ground calibration measurements (OCAL), preparations are under way to inspection and judgment of the measured data in a timely fashion (such that feedback can be given to e.g. adjust or even repeat measurements). As it turns out simply starting to write some (Python) code for this might not be the best option, because when trying to arrange multiple objects (plots in this case) on a page (without some graphical drag-and-drop tool), things quite easily can get messed up.

After getting overlapping frames and the like a number of times, I decided that it probably would be a good idea to step back from the code at all for a while to address the layout first; once there would be reasonably stable description of what exactly to plot and how to arrange the individual plots, I could then go about implementing this.

Sketch of the quicklook page

The good thing about a sketch as the above is, that this will give enough information to start with the computation of the geometry. Obviously every element on the page has a certain position and size – so using the sketch as an aid it (mainly) is a matter of patience to work out the lower-left point of a subplot and its size. Starting off with the size of each plot …

  dim_scatter    = [500, 500]
  dim_hist       = [200, dim_scatter[1]]
  dim_lightcurve = [dim_scatter[0]+self.fill[0]+dim_hist[0], 200]
  dim_projection = [(dim_scatter[0]+dim_hist[0]-self.fill[0])/3.0,
  dim_figure     = [2.0*self.fill[0]+dim_lightcurve[0],

… we are able to define the bounding boxes for each of the sub-plots:

  rect = {}
  # Total size of the figure
  rect["figure"] = dim_figure
  # Bounding box for the lightcurve of the laser intensity
  rect["lightcurve"] = [self.fill[1]/dim_figure[1],
  # Bounding box for the 3D scatter plot of the laser track
  rect["scatter"] = [(2*self.fill[1]+dim_lightcurve[1])/dim_figure[1],
  # Bounding box for the histogram of the laser intensities
  rect["hist"] = [(2*self.fill[0]+dim_lightcurve[1])/dim_figure[1],

Once all the bounding boxes are defined (and are returned in the form of a dictionary), we finally can go about creating the plots themselves: first we define the figure axes

  axes = self.axes_scanning_quicklook (data.shape)
  fig  = plt.figure(figsize=(axes["figure"][1]/100, axes["figure"][0]/100))

  axis_scatter    = plt.axes(axes["scatter"], projection='3d')
  axis_lightcurve = plt.axes(axes["lightcurve"])
  axis_hist       = plt.axes(axes["hist"])

… and then have Matplotlib display the data as scatter plot or otherwise:

  axis_scatter.scatter(xs=track["x"], ys=track["y"], zs=track["z"],
  axis_scatter.set_xlabel("\nColumn / Wavelength")
  axis_scatter.set_ylabel("\n\nRow / Swath")
  axis_scatter.text2D(0.9, 0.95,
                      "Track points ({})".format(track["method"]),

All of this put together finally yields the overview page as envisioned in the original sketch:

Implementation of the quicklook page

As can be guessed the whole process took a while from beginning to end; a good fraction of the time I have been struggling with fine-tuning the output of Matplotlib. While indeed there are quite a few examples available, I am still missing a more systematic demonstration of how exactly the various plot configuration settings influence the final outcome.