the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
15 Jan 2021
15 Jan 2021
New correction method of scattering coefficient measurements of a three-wavelength nephelometer
- 1Department of Atmospheric and Oceanic Sciences, School of Physics, Peking University, Beijing 100871, China
- 2School of Optics and Photonics, Beijing Institute of Technology, Beijing 100081, China
- 3State Key Joint Laboratory of Environmental Simulation and Pollution Control, College of Environmental Sciences and Engineering, Peking University, Beijing 100871, China
- 4Economics & Technology Research Institute, China National Petroleum Corporation, Beijing 100724, China
- 1Department of Atmospheric and Oceanic Sciences, School of Physics, Peking University, Beijing 100871, China
- 2School of Optics and Photonics, Beijing Institute of Technology, Beijing 100081, China
- 3State Key Joint Laboratory of Environmental Simulation and Pollution Control, College of Environmental Sciences and Engineering, Peking University, Beijing 100871, China
- 4Economics & Technology Research Institute, China National Petroleum Corporation, Beijing 100724, China
Abstract. The aerosol scattering coefficient is a significant parameter for estimating aerosol direct radiative forcing, which can be measured by nephelometers. Currently, nephelometers have the problem of non-ideal Lambertian light source and angle truncation. Hence, the observed raw scattering coefficient data need to be corrected. In this study, based on the random forest machine learning model, we have proposed a new method to correct the scattering coefficient measurements of a three-wavelength nephelometer, Aurora 3000, under different relative humidity conditions. The result shows that the empirical corrected values match Mie-calculation values very well at all the three wavelengths and under all the measured relative humidity conditions, with more than 85 % of the corrected values in error by less than 2 %. The correction method is valid to obtain scattering coefficient with high accuracy and there is no need for additional observation data.
Jie Qiu et al.
Status: final response (author comments only)
-
CC1: 'Comment on amt-2020-412', Jonathan Taylor, 18 Jan 2021
This looks like an interesting new technique. Please make your code available to make it much easier for others to try it out (and much easier for them to give you a citation!)
Many thanks
- AC1: 'Reply on CC1', Chunsheng Zhao, 11 Mar 2021
-
RC1: 'Comment on amt-2020-412', Anonymous Referee #1, 06 Feb 2021
Using in situ measurements of aerosol size distribution, black carbon etc in a few stations in China, the authors developed a nice method to correct scattering coefficient measurement. This is a very interesting research and the results sound solid, so I suggest to accept this submission after a few minor revisions.
1. L54-55, I'm a little confused why the absorption properties of particles can later the wavelength dependence of scattering
2. L81-82, I'm not confortable for this statement, CF is physically related to refractive index and particle size, SAE cannot resolve all these influences, forturnately, HBF, the simultaneous measurement with SAE, can, to some extent, provide extra information on particle size. That's it, the sentence looks like HBF is physically related to CF, but it is not, according to my understanding
3. The authors used a RF method, maybe it is necessary to talk about more why RF is much better than the ordinary regression method.
- AC2: 'Reply on RC1', Chunsheng Zhao, 31 Mar 2021
-
RC2: 'Comment on amt-2020-412', Anonymous Referee #3, 15 Feb 2021
General Comments:
The authors describe a novel method to correct measurement errors which are inherent to the use of current commercial integrating nephelometers. This method based on machine learning sounds attractive since it is said not to “need additional observation data”. Additional data are actually needed during the machine learning phase though.
The scope of applicability of the relationship between the correction factor (CF) and the two selected variables derived from the nephelometer data (the scattering Angstrom exponent, SAE, and the hemispheric backscatter fraction, HBF) is a key question that is not addressed in the manuscript. In particular, HBF is said to depend on externally mixed fraction of black carbon (Rext), while it depends also on the particle number size distribution (PNSD). Can it be demonstrated that the rule learnt by the machine to determine CF will apply at a location and/or time where different Rext and PNSD combinations lead to the same HBF for the ensemble of the aerosol? Can it be said anything about the applicability of the CF determination method described in the manuscript to nephelometer measurements performed at much less polluted locations?
Other missing elements as well as the overall organisation of the manuscript make it generally quite obscure, as described below.Specific Comments:
- The method used to determine CF cannot be understood before reading steps (1) to (8) in lines 138 – 146. It would probably be useful to have an outline of Section 2.2 at the beginning of this section.
- Figure 2 shows intermediate results (indicating that satisfactory CF values cannot be obtained based on the SAE only), but there is no figure showing CF values eventually determined by the novel method vs CF values calculated using the Mie theory in dry conditions (a vast majority of nephelometers are operated in in dry conditions, in accordance with WMO-GAW recommendations).
- Simple processes are described in details (e.g. particle hygroscopic growth) while unobvious logical steps are not precisely explained (see examples in Technical Comments).
- How much is the CF assessment learning depending on the assumptions about the aerosol mixing state, i.e. the fraction of purely scattering, purely absorbing, and mixed particles, including fully internally mixed aerosol?
- It is not stated if the CF determination method described would apply to TSI 3563 instruments, nor how measurements performed with a TSI 3563 (at least in campaign #5) were used in the machine learning process regarding the Aurora 3000 instrument.Technical Comments:
Line 23: “… the aerosol direct radiative forcing varies …” across what, as a function of what? Is this a range of uncertainty or variability?
Line 25: Aerosol direct radiative forcing also depends on the aerosol HBF and vertical profile (or at least the integrated aerosol optical depth). The 4 variables (aerosol single scattering albedo, extinction coefficient, aerosol scattering coefficient, and absorbing coefficient) are equivalent. Knowing 2 of them is enough. Since this manuscript is about integrating nephelometers (which measure scattering), I would suggest to stick to “scattering and absorbing coefficients”
Line 50: suggestion: “Bond et al. (2009) found that SAE is also affected by the particle refractive
Index.”Line 50 – 56: please consider streamlining: the sentences referring to Bond et al (2009) are redundant.
Line 70: suggestion: “our number size distribution measurements cover a wide range of 10-1000 nm, …”
Line 71: Table 1: which nephelometers were used in campaigns 1-4?
Line 88: “…three types of particles:”. The composition (i.e. chemical composition) controls the refractive index. What was the refractive index selected for the absorbing material?Lines 97-100: this section is confusing. The variations in the SAE as a function of the particle diameter directly result from the Mie theory, and does not support the last sentence starting with “Therefore” (line 99): on which basis are particles in the size range 100-200 nm stated not to contribute to the overall SAE values? Is it meant that they do not contribute much to SAE variations?
Line 107-109: “aerosol particles show a noticeable feature of HBF decreasing with the increment of particle size. However, when the particle becomes larger than 300 nm, the HBF is almost unchanged.” is again a direct consequence of the Mie scattering theory. And again the logical connection with the last sentence “HBF can represent the size information of particles smaller than 300 nm” is unclear. It was probably meant that HBF variability is mostly sensitive to the concentration of particles in the 100-300 nm size range.
Line 123: rather HBF is sensitive to Rext
Line 149: RF is not defined.
Figure 5: the diagram omits to mention that Mie calculations are performed using both the actual nephelometer light source characteristics and an ideal light source, which is essential for determining CF.
Line 167-168: RH instead of RH’ in the denominator
Line 171: “f (RH) and fb(RH) values”
Line 175: C(RH) is not precisely defined. It can be guessed afterwards that C = CF.
Line 231: information is missing to support the statement “which improves the accuracy of the CF estimation in the dry state”: “improves” compared to what?
Line 249-250: this is quite obvious since increasing RH and increasing hygroscopicity have the same effect on the particle sizes (increased diameters).
Line 271: suggestion: “essential” rather than “significant”
Line 273: The sentence “The scattering correction factor (CF) relating to the aerosol size and chemical composition is thus put forward” is unclear. Is it meant: “The correction factors (CF) to be applied depend on the aerosol particle number size distribution and chemical composition.”?
Line 274-285: The description of the Mie calculation method is obscure. However, a clear and concise description is needed, since it is the basis for the machine learning process.
Line 286: Suggestion: “SAE and HBF provide information on the aerosol particle size distribution for different size ranges (…)”. The size ranges should be specified in brackets.
Line 293: The first sentence should state that this paragraph regards “humidified nephelometer measurements”.
- AC3: 'Reply on RC2', Chunsheng Zhao, 31 Mar 2021
-
RC3: 'Comment on amt-2020-412', Anonymous Referee #2, 16 Feb 2021
General comments:
The authors introduce a new correction method to correct for the truncation error of the Aurora3000 nephelometer. This method uses the Angstrom exponent and also the hemispheric backscattering coefficient and is based on training a random forest machine learning model. To the reviewer's knowledge, the method is new and could be a step forward.
However, the reviewer has major concerns about the description of the model and the presentation and interpretation of results.
The role of the field measurements in this manuscript is not clear. In the absence of a complete, albeit simple, classification of the field data using SAE and SSA, it is questionable whether the data provide a sufficient basis for initialising the model. For example, it is not clear how strong the light absorption of the simulated aerosols are. Any information on single scattering albedo or the imaginary part of the refractive index are missing.
Furthermore, the reviewer finds it difficult to distinguish between measurement (PNSD) and speculative assumptions (refractive indices or kappa) in both models (dry and different RH conditions). For simplicity, the simulation study could also have been carried out with synthetic data for clearly defined aerosol types. The description of the model calculations are often imprecise, as important parameters such as refractive indices used from etc. are not specified.
Chapter 3 points out the performance of the new model. Why are results of the new model only shown for data from the Gucheng measurement campaign? Why were experimental data not used to show the performance of the new algorithm for dry conditions by a closure experiment of the light scattering coefficient? And more important, why haven’t the authors shown how their approach compares to the simple linear parameterization shown in Mueller et al., (2011)?
Specific comments:
Line 36: What parameter is mentioned?
Line 39: What methods have been proposed in Mueller et al (2011)?
Line 56: Figure 5 in Mueller et al (2011) suggests that a simple linear function is not sufficient. Unfortunately, this was not discussed further in Mueller et al. (2011).
Chapter 1: In general, the description of the state of the knowledge is little vague. How large are uncertainties when using the simple parameterizations of Anderson (1998) and Mueller (2011)?
Line 70 and Figure 1: Just taking a large set of total number concentrations as an argument that a large number of possible aerosol types have been covered is not sufficient. Furthermore, no evidence of a coarse mode particle can be seen in the particle size distributions. The large range of scattering Angström exponents (see Figure 2) suggest that could be are cases with a significant coarse mode volume fraction.
Line 89: A core radius of 35 nm might be too small to represent internally mixed aged particles. Furthermore, a constant core size also means that the volume fraction of absorbing material and the single scattering albedo decreases with increasing particle size. What does this mean for the interpretation of Figure 3? What range of single scattering albedos is covered with this model?
Line 88: What refractive indices are used for absorbing and scattering materials?
Line 91 and throughout the manuscript: Replace “band“ by „wavelengths”.
Line 92: Mie model: The description of the optical model should include how large the truncation angles were and how the imperfect Lambertian light source was taken into account. Are calculated values shown in Figures 3 and 4 for in ideal nephelometer or simulating the output of Aurora3000?
Figures 3 and 4: The reviewer believes that all measured size distributions (referred to as 'bulk' in Figure 3) served as the basis for the calculations. This should be mentioned in the text. Furthermore, it is not clear how the ratio of size resolved to total scattering is calculated. Was the size resolved scattering calculated for a constant size interval on linear scale or constant on logarithmic scale?
Line 123: Can the authors explain why Rext is sensitive to HBF?
Figure captions 3 and 4 : “absorbing particles (b)”
Line 141: Please specify “Conditions of nephelometer light source”
Line 149: “RF predictor” not defined.
Line 150: Why are the model results just checked for data from Gucheng and not for the other stations?
Line 170: Specify “assumed size distributions of kappa”
Figure 6: y-axis, “CRH” should be C(RH)?
Figure 6: Do not split legend to subplot (a) and (b)
Figure 6: How can f(RH) and Fb(RH) be negative for low RH?
Lines 230: The reviewer can not follow the conclusion on the strength of the absorption. The authors missed to give any information on the strength of absorption like single scattering albedo or complex refractive index.
Line 310: The reviewer thinks it would be better to reword the paragraph, since the study could also be done with synthetic datasets. With synthetic data, also simulations of e.g. desert and marine aerosol types could be done.
- AC4: 'Reply on RC3', Chunsheng Zhao, 31 Mar 2021
Jie Qiu et al.
Jie Qiu et al.
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 293 | 81 | 14 | 388 | 2 | 4 |
- HTML: 293
- PDF: 81
- XML: 14
- Total: 388
- BibTeX: 2
- EndNote: 4
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1