Characterization of a commercial lower-cost medium precision NDIR sensor for atmospheric CO 2 1 monitoring in urban areas

12 13 CO2 emission estimates from urban areas can be obtained with a network of in-situ instruments 14 measuring atmospheric CO2 combined with high-resolution (inverse) transport modeling. Because 15 the distribution of CO2 emissions is highly heterogeneous in space and variable in time in urban 16 areas, gradients of atmospheric CO2 (here, dry air mole fractions) need to be measured by 17 numerous instruments placed at multiple locations around and possibly within these urban areas. 18 This calls for the development of lower-cost medium precision sensors to allow a deployment at 19 required densities. Medium precision is here set to be a random error (uncertainty) on hourly 20 measurements of ±1 ppm or less, a precision requirement based on previous studies of network 21 design in urban areas. Here we present tests of newly developed NDIR sensors manufactured by 22 Senseair AB performed in the laboratory and at actual field stations, the latter for CO2 dry air mole 23 fractions in the Paris area. The lower-cost medium precision sensors are shown to be sensitive to 24 atmospheric pressure and temperature conditions. The sensors respond linearly to CO2 when 25 measuring calibration tanks, but the regression slope between measured and assigned CO2 differs 26 between individual sensors and changes with time. In addition to pressure and temperature 27 variations, humidity impacts the measurement of CO2, all of these factors resulting in systematic 28 errors. In the field, an empirical calibration strategy is proposed based on parallel measurements 29 with the lower-cost medium precision sensors and a high-precision instrument cavity ring-down 30 instrument during 6 months. The empirical calibration method consists of using a multivariable 31 regression approach, based on predictors of air temperature, pressure and humidity. This error 32 model shows good performances to explain the observed drifts of the lower-cost medium precision 33 sensors on time scales of up to 1-2 months when trained against 1-2 weeks of high-precision 34 instrument time series. Residual errors are contained within the ±1 ppm target, showing the 35 feasibility to use networks of HPP3 instruments for urban CO2 networks. Provided that they could 36 be regularly calibrated against one anchor reference high-precision instrument these sensors 37 could thus collect the CO2 (dry air) mole fraction data required as for top-down CO2 flux estimates. 38