How to align data series along x-Axis to achieve besst fit?

by cripcate   Last Updated April 15, 2019 09:19 AM

I have 4 datasets that I am trying to align to each other so I can so some further analysis. I asked a similar question earlier, that didn't get much attention, so I am trying again (this time with different data).

My data is available here. In the reference set, d18O is the "reference" Y-Axis, rel_scale is the "reference" x-Axis, and the Date column is what ultimately want to assign to my other data. In the data-to-align set, distance is the raw measurement distance, rel_scale is the distance relatively scaled to yearly growth periods -> the x-axis to align to the reference. d18O is the dependent variable again.

data

the first dataset are measured temperatures, and salinity of a location over time (so a time series), which are displayed in red and blue in the upper plot.

For the same time period I measured chemical parameters (here d18O) in 3 bivalve shells (datasets 2-4) and I am trying to align it to the instrumental temperatures. They are displayed in the 3 lower plots. For that reason I already calculated a d18O from the instrumental temperatures folloiwng some equation (black line upper plot).

The reason I want to align modeled d18O and measured d18O in bivalves is, that I want to assign a timestamp to each position in the shell. Right now the x-Axis fo the lower 3 plots really is Distance between measurements in mm. Because the shells grow faster and slower depending on the season, I want to detrend this data to refer to actual calendar dates. So I want to transform my x-Axes to achieve best fit with the black plot in the top curve.

I read a bit into DTW and similar methods, but I just can't find a solution because my knowledge on math and statistics like this is limited. My big problem is that sampling period and frequency are varying.

If I put the curves on top of each other it looks like this:

enter image description here

which Is already a good fit but still not optimal (see the seasonal bumps that dont quite align. Another reason I want to detrend this data for growth rate is that I could not the "displacement factor" of each measurement and possibly reconstruct how fat or slow the bivalves grow during each month.

I hope someone can point me to methods / resources / procedures to use for this problem, or alternatively show me where I am overcompicating things and point me to easier methods for detrending fluctuating sampling rates. I'm always here to answer questions you might have about my data.

Best regards.



Related Questions


Updated January 30, 2019 16:19 PM

Updated October 19, 2017 15:19 PM

Updated January 08, 2018 06:19 AM

Updated March 27, 2018 11:19 AM