CO2 sensor accuracy, network comparisons

[spestana]

I wanted to put some thoughts down here about accuracy and comparisons of Ribbit Network observations with other sources of distributed CO2 observations. I'm putting this in the dashboard repo since this is sort of in the realm of post-processing data.

The SCD-30 NDIR CO2 sensor has a listed accuracy of about +/- 30ppm, with some dependence on temperature/humidity and possibly some drift away from its original calibration over time. While averaging measurements over some time interval could help cut down on noise due to the sensor's limited accuracy, that wouldn't help correct for drift or bias in the measurements. One thing that could be interesting to try out is to set up multiple Ribbit Netowork sensors at the same spot to see how well they agree with one another, and try to characterize how they might drift apart, though it would be hard to tell if they all drift together in the same direction).

@keenanjohnson and I were talking about this, and a method of correcting for bias could be to "calibrate" the Ribbit Network observations with other nearby CO2 observations (that are perhaps using higher-accuracy sensors, or data that has been QC'd and/or assimilated into an atmospheric model). One possible source of data could be NOAA's Global Greenhouse Gas Reference Network, either direct observations or from one of their data products. However, this wouldn't be a simple comparison to make, since their observations (while globally distributed) are not spatially dense - so the likelihood of a measurement being taken nearby a Ribbit Network site isn't good, and a single point Ribbit Network observation can't be expected to match a 1x1degree grid mean value.

Also might be of interest: How NOAA's Global Monitoring Laboratory uses its network of observations to come up with global mean CO2