Ethylene Oxide Monitor with Part-per-Trillion Precision for In-Situ Measurements

. An Aerodyne Tunable Infrared Direct Absorption Spectrometer with 413 meter cell for the detection of ethylene oxide (EtO) is presented (TILDAS-FD-EtO). This monitor achieves precisions of <75 ppt or <0.075 ppb in 1 second and < 20 ppt in 100 seconds (1-sigma). We demonstrate precisions averaging down to 4 ppt in an hour (1-sigma precision) when operated with frequent humidity-matched zeroes. A months-long record of 2022 ambient concentrations at a site in the Eastern 10 United States is presented. Average ambient EtO concentration is on the order of 18 ppt (22 ppt standard deviation). Enhancement events of EtO lasting a few hours are observed, with peaks as high as 600 ppt. Back trajectory simulations suggest an EtO source nearly 35 km away. This source along with another are confirmed as emitters through mobile near-source measurements, with downwind concentrations in the 0.5 ppb to 700 ppb range depending on source identity and distance downwind.


Table of Contents
The basis of our EtO monitor is our commercially available (Aerodyne Research Inc., 2022a, b) dual-laser tunable infrared direct absorption spectrometer (TILDAS-FD) platform, which in this case is equipped with a single mid-infrared interband-25 cascade laser (nanoplus GmbH). The highly divergent beam of this laser is collected and focused into a long-pathlength multipass cell using a sequence of reflective optics (Mcmanus et al., 2015). A visible laser is co-aligned using a dichroic mirror to aid the optical alignment of the system.
For the system described herein, we use a multipass cell with 413 m optical pathlength and an active volume of 1.8 litres for continuous flow applications. The cell contains two mirrors with wavelength-specific high-reflectivity coating (reflectivity 30 >99.8%) to minimize reflective losses of laser power during the > 800 reflections among the two mirrors. Upon exiting the cell, the laser beam is focused on a thermoelectrically cooled photovoltaic HgCdTe detector (Judson, J19 with transimpedance amplifier).
The laser is driven by electrical current while being maintained at a constant temperature. The in-house software TDLWintel provides a voltage ramp via a digital-to-analogue converter card (National Instruments). This ramp is translated into current 35 via a low-noise laser driver (QCL-500, Wavelength Electronics). The laser temperature is maintained by an electronic temperature controller via the built-in thermoelectric element in the sealed laser housing. The detector signal is digitized using an analogue-to-digital convert card (National Instruments) and then handled by TDLWintel for processing.
The laser wavelength is scanned across a narrow wavelength interval near 3065 cm-1 approximately 1830 times per second.
For every scan, the laser is on for the first 90% of the time, followed by a brief period where it is off for the remaining 10% of 40 the time. The individual spectra are then averaged to a single spectrum every 1 second. This averaged spectrum is then processed in TDLWintel to derive the mixing ratio of EtO as well as of all other absorbers defined in the spectroscopic fit at 1 Hz in real time.
Spectra are defined by absorption signal, a polynomial spectral baseline (full light, no absorption), as well as signal during the off period (no light, complete absorption). The latter two components are used to normalize the measured spectra onto the 45 transmission scale (0 -1).
The wavelength scale of the laser scan is determined by periodically analysing the interference spectrum of a Germanium etalon. The derived wavelength scale is then further refined measuring a high-concentration ethylene (C2H4) spectrum from a reference cell built into an optional beam path in the spectrometer optics. Wavelength drift is mitigated by locking the laser to a strong H2O line in the sample spectrum using an active feedback loop controller realized in software. 50

S1.2 Flow system
Sample gas is drawn through the multipass cell at a reduced pressure of 20 Torr (27 hPa) using a vacuum pump downstream 55 of the instrument and an upstream pressure controller. This reduced pressure is used to sharpen the absorption lines in the spectrum and provide the best compromise of spectroscopic selectivity and sensitivity. The gas-flow rate is typically around 3 -5 slpm (slpm: standard litres per minute), resulting is a gas exchange rate of 1 Hz or better.
An overblow port is set up to deliver calibration gas or ultra-zero air. To prevent pressure disruptions, this port is tied into the inlet ~6 inches from the tip using a union tee. For humidity-matched zeroes, a parallel flow path is set up, with a length-60 matched piece of tubing and scrubber cartridge isolated by two solenoid valves. These valves actuate at the same time, pulling ambient air through the scrubber and providing near-humidity-matched zero air free of EtO.
Spectral backgrounding (or autobackgrounding) is done by intermittently and regularly measuring air free of EtO. A background spectrum is acquired and used to divide subsequent sample spectra, reducing the impact of drift due to instrumental effects like optical fringes and spectral baseline effects. Each background takes about 1 minute at 3 -5 SLPM in order to flush 65 out the cell, acquire clean spectra, and return to sampling.

S2 Calibration
The EtO-TILDAS instruments reports dry air mixing ratios that have been mathematically corrected for the dilution effects of water vapor, as well as for empirical water broadening effects. At the ambient humidity measured, these effects are expected to be < 3%. 70 In a typical calibration, 50-500 sccm of an EtO standard at 1 ppm EtO is delivered via an Alicat mass flow controller to a ½" overblow line connected via a T-fitting 6 inches from the inlet tip. Calibrations are done either by standard addition to humid ambient air, with a known total inlet flow rate (3 -10 SLPM) or via dilution into a known flow (3-10 SLPM) of ultra-zero air delivered by a second Alicat mass flow controller. Humid standard additions are preferred, as they most closely resemble sampling conditions. 75 Standard concentrations were calculated based on the known calibration tank concentrations and the known system flows.

S3.1 Statistics for Ambient Measurements
Averages of the full time series, and of winter/spring (Feb 2022 -April 30 2022) and summer (July 1, 2022 -Aug 4, 2022) 95 averages of the 1hr data shown in Figure 3 of the Manuscript are listed in Table S3 below. To understand whether the Winter/Spring and summer averages are significantly different, we compute the standard error of the mean (SEM), where SEM = SD/sqrt(N). We also include and propagate through a 5% error due to calibration uncertainty, yielding propagated error bars of 1-2 ppt for the averages. The upper and lower confidence limits at 95% confidence (UL and LL) are listed for winter/spring and summer averages. The winter UL of 14 ppt does not overlap the summer LL of 31 ppt, leading us to conclude, using these 100 Gaussian statistics, that the averages are significantly different at the 95% confidence level.
Since these statistics do include the small number of plumes observed, we also show histograms and Gaussian fits of the data in Figure S4. The Gaussian peaks occur at 12 ppt for the winter/spring and 32 ppt for summer, comparing well to the full averages computed in Table S3 of 12 ppt and 33 ppt, respectively, and do not alter the conclusions. Table S3. Averages and statistics for 1hr ambient data shown in Figure 3 of the Manuscript, all values in ppb. "Winter/Spring" spans from February to April 30, 2022. "Summer" spans July to August 4, 2022. The standard Deviation (SD), number of 1hr averages (N), Student's T statistic at 95% confidence (t), 95% error bars using the SD are listed. Additional statistics for the mean are included: the standard error of the mean (SEM), 95% error bars for the average using the SEM, estimated error bars assuming a 5% calibration uncertainty, the resulting propagated error, and the resulting 95% Lower Limit (

S3.2 Hysplit Back Trajectories 115
The back trajectory analysis was computed using the National Oceanic and Atmospheric Administration (NOAA) Air Resources Laboratory (ARL) Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT). The work depicted here in the SI used the web-based model with default options. The parameters for the run are noted in the figure legend. This work involved straightforward back trajectory calculation. Although HYSPLIT is capable of a rigorous source-receptor analysis, the physical correspondence noted below has been performed 'by eye', not via a quantitative attribution. Details of 120 the NOAA/ARL HYSPLIT model (Rolph et al., 2017;Stein et al., 2015) can be found in the cited sources or at the primary website (https://www.arl.noaa.gov/hysplit/).