Reports

Between 2020-06-12 and 2020-09-29 I cross compared PAR sensors at Vaira in preparation for removing the old PAR sensors and replacing them with newer, recently calibrated ones.  The cables on the older ones are all rotting away from sun exposure.  The cross compare ended up having four different periods of comparison:

1) 06-11 to 07-28 Two PAR sensors (sn 140455 in and sn 161073 out) were added to the Tower.  I realized later that sn 161073 is the Ameriflux standard and should not remain on the tower permanently.

2) 7-28 to 8-26 I swapped the physical positions of the two added sensors so that sn 161073 is now in and sn 140455 is now out.

3) 8-26 to 9-16 I removed sn 161073 and sn 140455 and added sn 040412 in and sn 030381 out.

4) 9-16 to 9-29 I swapped the wiring and physical mounts so that sn 040412 complete replaces the original sn 990025 in the incoming position and sn 030381 completely replaces sn 990031 in the outing position.

Below is a table with a list of all the PAR sensors and their calibration coefficients.  Several of the senors have updated coefficients and their original factory coefficients are listed for reference.

Here are the scatter plots of the four different periods.  In each plot green is the incoming PAR and blue is the outgoing PAR.  Only points with PAR greater then 25 umol/m2/s are plotted for these regressions.  You can zoom the graphs by clicking and dragging on them - note the regression equations will update using only the visible points.

Figure 1.  Comparison of the incoming and outgoing PAR sensors during the first time period.  The original Tower sensors seem to be reading 4% - 5% high.

Figure 2.  For the second time period the new calibration sensors swapped physical locations.  Almost no change in the slope of the regression.

Figure 3.  For the third time period a new pair of calibration sensors were swapped in.  Again very little change in the slopes.

Figure 4.  For the last time period the second set of new calibration sensors were moved into the permanent mounting locations and the wiring on the multiplexer was swapped.  Bigger changes in the slopes here where I would have expected less change.

Below is a table summarizing the slopes for the incoming and outgoing comparisons during the periods with different combinations of sensors.

Below are the time series plots of incoming and outgoing PAR.  You can see depressions in PAR due to smoke centered around August 21 and September 9th.

Figure 5. Time series of incoming PAR for all four time periods.  Even with the different configurations of calibration sensors, the old original PAR sensors (blue) are consistently about 5% higher (visible at midday) than the new calibration sensors (green).  On August 6th the original PAR sensor had a depressed signal perhaps from being dirty - this day of data was removed.

Figure 6. Time series of outgoing PAR for all four time periods.  Similar to the incoming PAR, the old outgoing sensors (blue) read about 5% lower than the calibration sensors (green).

An interesting consideration is that three of the calibration PAR sensors were calibrated against the Ameriflux standard (sn 161073), the fourth calibration sensor.  And the original factory calibrations of these three sensors was slightly (~2%) lower.  So if the Ameriflux standard was off then these other three would also be off.  However even using the factory calibration, the original PAR sensors on the Vaira Tower still read about 2%-3% higher.  Note, the coefficients are given as a divisor, uV/(umol/m2/s), so a smaller coefficient makes the computed PAR larger.

Figure 7.  Scatter plot for period 1 using both the current coefficient that is tied to the Ameriflux standard and the factory calibration for PAR sensor sn 140455 comparing incoming PAR.  Using the factory calibration improves the relationship but only by about half.

Conclusion:  It seems there may a smaller error (~2%) caused by the multiplexer channels but hard to quantify.  I think it best to use the calibration factors associated with the sensors.  So starting about 14:00 on 2020-09-16: PAR in = 1000/5.74 = 174.2 and PAR out = 1000/5.43 = 184.2.  Whether we should try to correct the apparent 5% drift in the old data is another question.

Figure 8. Regression of incoming PAR old sensor versus new sensor for the entire time period.  Here we are comparing mV versus Umol/m2/s to find the new coefficients.  From sn 990025 5.22 uV/(umol/m2/s) and for sn 990031 5.38 uV/(umol/m2/s).

Bonus:  There is also a Delta-T BF3 diffuse PAR sensor at Vaira that measures both total PAR and diffuse PAR.  This sensor reads about 10% low and both its total PAR and diffuse PAR values should be corrected with the relationship on the graph below.

Figure 9. Bonus.  The relationship between the Delta-T BF3 PAR sensor and the new calibration sensors for the entire time period.  Again, this plot only uses PAR values above 25 umol/m2/s.