Summary

“Electrochemical sensors onboard a Zeppelin NT: In-flight evaluation of low-cost trace gas measurements” evaluates the performance of a suite of sensors installed in a hatch box within the bottom of an airborne platform. The focus here is on NO and NO2 measured by Alphasense electrochemical sensors. Six units are flown together underneath the Zeppelin and intercompared. Other aspects, including results from a reference mid-infrared MIRO instrument, were described in a related paper by the authors. Many papers have been published in recent years on low-cost gas and particulate matter sensing, along with their calibration and correction for spurious environmental dependencies. In my opinion, the manuscript is well written and scientifically sound, although the scope is fairly limited. The most novel aspect is airborne deployment in Zeppelin flights, which took place in Germany.

Main

My main comment is that it feels like the reference MIRO data from the flights is underutilized in terms of validating the performance of the electrochemical sensors, especially since laboratory test data is not included. Some thoughts on this point:

- While the intercomparison of the six setups in Figure 4 is interesting, is the conclusion about setup #2 performing best also supported by comparing against MIRO MGA? What do those results look like?

- Corrections were derived to remove dependences on T, AUX, and dRH/dt without using MIRO, and compared against manufacturer recommended corrections. This is advantageous in avoiding requiring use of a reference instrument. However, can MIRO be used independently here to evaluate how well this correction approach works?

- What can be said about stability of the ECS sensors and derived calibration during or between flights.

Minor

I am not completely sure what data was used to generate the figures. On L265 it talks about only showing setup #2. L160 talks about excluding #4, #6, and partly #5. Which sensor setups are included in Fig 3, 5-7?

Similarly, Figures 4 and 6 seem to be showing aggregate data over all the flights. This could be clearer. How many individual flights and hours of data are included? Is any data excluded?

Figure 1(a) what is the height and width of the hatch box?

L160 “From this correlation analysis it is evident that the sensors of setups #4, #6 and partly #5 provide erroneous data.”

This is not obvious to me. Figure S2 in particular is referenced in the sentence before, but is hard to read both in terms of the font size within the figure and having a fairly brief figure caption.

Figure 6; is top panel with y-axis label ‘Accuracy / ppb’ showing the +/- 2 standard deviations as shown in the bottom panel? This should be clearer.

“Evidently, with the manufacturer’s correction, amount fractions in the low ppb range cannot be quantified (Fig. 6) predominantly due to the high offset of -19.76 ppb.”

I don’t understand this point since an offset affects low ppb as well as high ppb measurements equally and does not determine the sensitivity of the measurement. It could be an issue if readings are filtered to be above zero.

Acknowledgements: WRF is not directly mentioned in the main text, although perhaps it was run to generate the wind field in Figure 7. EURAD-IM is not mentioned.

I think it is a really good platform for low-cost sensors to measure the boundary layer. Please find some comments and questions below about the manuscript and methods that were used.

Time synchronization

What is LTE?

Did you apply the Savitzky-Golay filter before or after you had applied your temperature correction? How noisy were the raw data versus after this filter had been applied. Did the analyte peaks stand out prominently from the noise?

Determination of amount fractions

Not quite sure how you reached a window size of 31 seconds for the t₉₀ value. Did you run the EC sensors in the laboratory under controlled conditions and then change the [NO] and [NO₂] to concentrations that you’d expect to see on the Zepplin and calculate it from there?

Was Mead et al 2013 also using these sensors on a Zepplin or aircraft? Was the concentration range in that paper the same as you’d expect for your deployment?

You use the fact that Mead et al 2013 found a t₉₀ value of 21 sec for NO₂. What was Mead et al t₉₀ value for NO? The manufacturer provides vastly different t₉₀ for NO₂ and NO. Also do we expect that the EC sensor technology hasn’t changed in ten years so you and Mead et al have the same versions of EC?

Sensor signal dependances

So you assume that an increase in NO or NO₂ from 0 ppb to 2 ppb does not cause any increase of EC. Did you check this in the lab?

Why did you not find the median NO signal (of the six NO EC) and subtract that from each NO sensor to get the ‘signal primarily influenced by other factors’? I think that would be more appropriate rather than assuming a step change of the analyte concentration doesn’t impact the EC output at all.

 Why did you correlate WE to the change in humidity and not the humidity? Or if you wanted to use the differential, why not use that for temperature also, dT/dt? Changing the way these are plotted makes it difficult to see quickly the difference between the temperature and the humidity dependance. 

Validation of ECSs performance

I think there needs to be a timeseries of the 6 NO/NO2 sensors showing how they all behaved over the duration of the flights. The slopes look good but did they start out correlated and then drift apart as they all respond differently to each interfering variable? Or were they all slightly offset and this stayed consistent throughout? Maybe add temperature and humidity onto this too, with a shared time axis. 

Comparison of ECSs with the MIRO MGA

Accuracy plots are nice but please add the x-axis. Is this over time? In which case the accuracy decreases with the T,AUX and dRH/dt correction towards the end of the experiment. Or is it a shared axis with the plot below – so accuracy decreases at higher NOx (MIRO) concentrations? That seems a little odd. Why would the accuracy decrease if the reference instrument is measuring more than 125 ppb NOx? I think if this is the case a little description of why that might happen is needed.

Both the EC NO and NO₂ concentrations are overestimated (even at the higher concentrations you mention a 30 % overestimation). Was this a consistent overestimation throughout? Did you zero the EC amount fractions to the MIRO at the beginning of the experiment? It would be interesting to see if this overestimation changed over a 20 hour flight or over the entire campaign. I was under the impression the EC drift apart over a matter of hours – days.

I know you mention that the AUX electrode is supposed to remove all impacts of variables other than temperature but I think you could write a few sentences about the NOx EC response to other gases. Did you look at how the NOx EC responded to cross interferences? Was the signal influenced by any of the other compounds measured by the MIRA? i.e. did changes in the ozone concentration cause a response in NOx EC?

Conclusions

Deviations of the sensor values to MIRO decreased from an average of 27.3 to 3.5 ppb. What is the accuracy of the MIRO? The concentrations of NOx observed by this reference instrument seem to be pretty low; the Weisweiler timeseries shows that for this flight the majority of the NOx was around 5 ppb.  A deviation of 3.5 ppb is quite a lot for concentrations in this range.

I think you successfully show that you can improve the accuracy of the NO NO₂ EC sensors by correcting for temperature and humidity but I am not sure you could reliably deploy EC sensors in the Zepplin with no reference device and trust the output. It does seem to qualitatively pick up the NOx emissions  from the power plant and industrial regions but there is still large amounts of uncertainty (below 5 ppb deviations between MIRO and EC increased up to 300 % and -600 %; the MIRA reported ambient NOx mixing ratios of 5 ppb for the majority of the time).

On page 11, you also say that you only ended up using sesor set up #2 for the noise plots and intercomparison with MIRA. Really, this means that 1 out of 6 EC sensors can be used on a UAV platform without a reference instrument. Or would you still advise having 6 NOx EC and then taking the best performing one?