Calibrations at last!

Last week marked the end of week 7 for me. My calibration system was designed first and foremost with the ECO Triplet sensors in mind, as those will be the ones taking measurements for CMOP next year. However, because the order hasn't yet come in, I attached another fluorometer from the same manufacturer, WET Labs. There were a few design details and kinks to work out with this other fluorometer, but by Wednesday, I was ready to test for linearity using dye.

Unfortunately there was a bit of a glitch with the dye. Our sensor was calibrated at WET Labs ten years ago using Coproporphyrin III tetramethyl ester disolved in 6 M HCl. YIKES! WET Labs has since quit using copro dye to calibrate its equipment because the dye solution is so unstable and dangerous to work with. Though it would have been nice to use the exact same dye as the manufacturer, we ruled that it was just not going to be possible. Good thing someone invented food coloring! I ran a ten point dilution series with some green/blue food coloring, and a regression analysis of the normalized voltage output versus the normalized concentration of dye came out with an R squared value of 0.999657, meaning we can assume the sensor is linear.

Friday I calibrated the sensor with a sample of T. weissflogii. I'm still waiting on the chlorophyll analysis from Oceanography to determine a scale factor, but I was able to do another linear regression of the data. This one wasn't quite so good. With an R squared of only 0.9202, there was some significant source of error that had to be unrelated to the linearity of the instrument (which was proven to be basically perfect on Wednesday). I noted a small leak on the 4th dilution, and a plot of the deviation from linearity with time confirmed my suspicions. The data points were getting increasingly more linear until that 4th dilution. Suddenly, there was a spike in the deviation from the linear plot and the points started slowly moving closer and closer towards linearity once again. It is somewhat unfortunate that the leak occured, because it could have easily been prevented by adding weight to the chamber so that the O-ring sealed tightly on the bottom. This is a good lesson for next Monday's field calibrations of Nitzschia, and hopefully I will still be able to salvage some of the data points and get a decent linear regression when I throw out the points that were severely impacted by the leak.