Improve Accuracy of CO2 Sensor Automatic Self Calibration

[keenanjohnson]

In #23 we discussed the tradeoffs in calibration methods for the CO2 sensor. We decided to pursue the ASC calibration method, but there will still be some inaccuracies introduced by using this method.

Let's keep the discussion going on strategies to mitigate this @eren-rudy and @daveb377.

[Derek-Gutheil]

Hey, I was looking into sensor accuracy today. I stumbled upon co2meter.com and after browsing their site for some time, sent them a question:

"
Hello!

My name is Derek Gutheil, and I am working on an open source project (https://www.ribbitnetwork.org/) with the goal of creating an global dataset of outdoor CO2 levels.

Seeing as we are measuring CO2 levels in the 400 to 800 ppm range, and I hoping you can give us some advice on how to accurately measure this low level of CO2. I have looked into your sensors and it seems like the most accurate sensors that you have achieve +-30ppm and 3% MV. As you show in your example (https://www.co2meter.com/blogs/news/how-to-measure-carbon-dioxide), this gets an effective range of 358ppm to 442ppm for a 400ppm sample.

So I have a few questions:

1. Are there other sensors in development or on the market that you know of that can get down to +-5-10ppm accuracy?
2. Are there industry standard techniques to get +-5-10ppm accuracy out of +-30ppm sensors?
3. Finally, are there more expensive sensors with higher accuracy that we can use to calibrate the sensors in our devices?

Thanks so much for the help,
-Derek Gutheil
"

Overall, the lack of accuracy and precision on these sensors is definitely concerning. Have there been any updates on your side recently?

[keenanjohnson]

Hey @Derek-Gutheil! Great job reaching out! I haven't heard any new updates and I'm not aware of any sensors at a relatively inexpensive pricepoint with a higher accuracy, unfortunately. I think this is an area that deserves continuous focus and development though!

[Derek-Gutheil]

They just responded and linked me to one of their sensors, so no luck going down that road unfortunately.

I am definitely concerned about this and I think we need a real plan to handle calibration in the field. The only thing worse than no data is bad data.

[keenanjohnson]

Indeed! I share the concern!

[leckerschmecker38]

HI all .. i am professionally involved in gas sensing .. i look into this project here for a littel bit and wonder if anyone has done the research to actually determine the sensor specifications needed for the purpose of the project ..

i understood this here is about a global CO2 sensor network ,,..

but what do you want to see / detect / qauantify ..
A) global variations in background Co2 due to industry .. e.g combustion
B) local variations in urban / slightly polluted areas ( Cities in developed countries )
C) local variations of co2 near indstrial areas or cities in developping countries with proably strongly polluted air

case a ) you will need to detect around 400 ppm .. but the global average CO2 rises currently with 2,5 ppm PER YEAR ! you will need >> the sensors with best stability and the best accuracy .. i would guess you will face quite some problem if you try this with a sensor with a resolution of 30 ppm .. not sure what the stabiliyt of the SCD30 is over a year

case B) CO2 will be between 400 and 800 .. .. there will be daily ups and downs as soon as you get some how near traffic .. at night co2 will be domintaed by other sources an during rush hours you will find big peaks ... so depending on where you place your sensor , what fetch it has and near which emmiters you ar e you will get highly variable Co2 fluctuations an those will be mixed and modified by transport ( caused by wind , temperature , pressure and humidiuty )
so here you will have less problems with 30 ppm resolution .. but you will face an interpreatation problem if your co2 fluctuationsa are caused by the senosr or by your sources and especially by which sources

case c ) polluted cities even higher co2 values and higher dynamics ,, but how do you want to do the source attribution ?

human disturbances ,,
everone using the senosr should be made aware that your breath contains up to 50.000 ppm of co2 .. so if the sensor get contact with any of your exhaled air or from someone nearby you can also cause very big fluctioations and even saturation

other disturbances
the Co2 NDIR sensoor uses light and some sort of light filter to become specific ..
if there is any other gas absorbing in the selected light you will get cross dependencies .,
in particular in urban outdoor areas , near industry etc - particularly in developping countries with bad airquality you have to be aware of cross sensitivities of your sensor to other gas species ,, like hydrocarbons , VOCs etc pp

back to the sensor and sensor system

for case A) i you need a stability and accuracy of about 1 ppm CO2 to detect and annual change of 2.5 ppm .. for this i think you need a lab based study to detemine the perfromacne limits of your sensor ( alone) and as the sensor systems ( incl your electronics) sofar there is a factor of 10 or more missing to be a solution for case a) .. and there needs to be a valdiation strategy to proove this performance and you need some sort of reference to check you sensors .. you will also need a site selection and valiidation stategy .. in order to select a site suitable for background Co2 determination

case B) and c)
yoiu still need a applicaiton derived definition of the required sensor performance
and you need a method to quantify your sensor performance

also you need additional infomration to provide some informaiton on the transport problem ( where does your gas come from which goes in the sensor) and you should better have some infomration on the dominating sources which are in the fetch of your sensor .. those things will take a lot of work to inquire

if you dont have this infomration you will to put it bluntly just have a fluctuation sensor signal .. but what infomration do you want to derive from this

i am aware that my comments here might be a bit overwhleming or even spoiling the story ..
but it takes a lot of research and background knowledge ( on environment , sensor , transport processes etc etc )
to design -
setup -
contruct -
operate -
read out
validate
a sensor network and its elements
and in particular
to interpret the signals and output

i am also fascinated by the low cost , size weight and power requirment of these intergrated sensors
but
the big question is .. how can the ambient / environmental signals / patterns / sources / sinks and gas transport so you are looking really for be extracted form the lower qualit signals of the lower cost sensors ? or is the infomration so disturbed by the noise and drift and problems which are caused by the lower costs that they cant be extraxted

and another question is ., can you answer the above question without emplyoing some sort of reference sensor ( of much higher performance and much higher cost ) ?

looking forward to discuss this

and if i find enough time i will try to provide some answer or at least informaitions to help you in tackling this questions

best regards

leckerschmecker38