the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 20 Dec 2023
Merging holography, fluorescence, and machine learning for in situ, continuous characterization and classification of airborne microplastics
Abstract. The continued increase in global plastic production and poor waste management ensures that plastic pollution is a serious environmental concern for years to come. Because of their size, shape, and relatively low density, plastic particles between 1–1000 μm in size (known as microplastics, or MPs) emitted directly into the environment (“primary”) or created due to degradation (“secondary”) may be transported through the atmosphere, similar to other coarse-mode particles, such as mineral dust. MPs can thus be advected over great distances, reaching even the most pristine and remote areas of the Earth, and may have significant negative consequences for humans and the environment. The detection and analysis of MPs once airborne, however, remains a challenge because most observational methods are offline and resource-intensive, and, therefore, are not capable of providing continuous quantitative information.
In this study, we present results using an online, in situ airflow cytometer (SwisensPoleno Jupiter; Swisens AG; Emmen, Switzerland) – coupled with machine learning – to detect, analyze, and classify airborne, single-particle MPs in near real time. The performance of the instrument to differentiate single-particle MPs of five common polymer types (including polypropylene, polyethylene, polyamide, poly(methyl methacrylate), and polyethylene terephthalate) was investigated under laboratory conditions using combined information about their size and shape (determined using holographic imaging) and fluorescence measured using three excitation wavelengths and five emission detection windows. The classification capability using these methods was determined alongside other coarse-mode aerosol particles with similar morphology or fluorescence characteristics, such as a mineral dust and several pollen taxa.
The tested MPs exhibit a measurable fluorescence signal that not only allows them to be distinguished from the other fluorescent particles, such as pollen, but can also be differentiated from each other, with high (> 90 %) classification accuracy based on their multispectral fluorescence signatures. The classification accuracies of machine learning models using only holographic images of particles, only the fluorescence response, and combined information from holography and fluorescence to predict particle type are presented and compared. The results provide a foundation towards significantly improving the understanding of the properties and types of MPs present in the atmosphere.
- Preprint
(1481 KB) - Metadata XML
-
Supplement
(1903 KB) - BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2023-2853', Anonymous Referee #1, 17 Feb 2024
The manuscript entitled “Merging holography, fluorescence, and machine learning for in situ, continuous characterization and classification of airborne microplastics” by Beres et al. presents a detailed examination of microplastic (MP) detection, focusing on the deployment of a commercially available online airflow cytometer (SwisensPoleno Jupiter) combined with machine learning to identify, analyze, and classify airborne microplastic particles. Emphasizing the method's efficacy in real-time classification of different microplastic polymer types, the study assesses the technology under controlled conditions, employing metrics of size, shape, and fluorescence. References substances, such as mineral dust and pollen, are used for comparison to assess to what extent particle classes can be discriminated with this approach. The results are thoroughly discussed/integrated in the context of previous studies and existing knowledge.
My overall recommendation is that the manuscript should be published after some relatively minor changes. From a scientific perspective, it clearly fits the scope of AMT. It is a thorough and interesting study, relevant for the field of atmospheric research and beyond. I am convinced that the study will become a useful resource for the emerging field microplastic detection in the atmosphere. It underlines that the combination of size, shape and fluorescence detection might provide sufficient specificity to quantify airborne microplastic (to some extent) also under real ambient conditions. From a formal perspective, the quality of the manuscript is high - it is well-written, the figures and tables are clear, and all arguments and aspects are presented clearly.
Below, the authors will find some general comments that they may want to consider:
Page 3, line 75 on statement “Some commercially available polymers have previously been examined for their autofluorescence”: It might be helpful for the readership to add a few sentences already in the introduction on the molecular motifs that are responsible for the fluorescent emission. Especially for the non-aromatic polymers, which one typically would not consider efficient fluorophores, the observed strong emission reported already in previous studies is an interesting phenomenon.
Page 3, line 82 on statement “No study has used the intrinsic fluorescence of polymers for airborne particle identification and characterization in situ.” I think the preprint by Gratzl et al., 10.26434/chemrxiv-2023-qzhr8 could/should be cited in this context.
Page 3, line 90 on statement “ … assess the fluorescence response of various common microplastics”: I wonder why polystyrol was not included here. It is widely used probably also shows a characteristic fluorescence due to the aromatic structure. Along these lines and in more general terms, according to which criteria were the five polymers selected.
The study by Ornik et al. should be cited (and probably also discussed) somewhere https://doi.org/10.1007/s00340-019-7360-3
Table 1: Did the authors receive any information on the age of the pollen samples? Such commercially available biological reference substances are not necessarily freshly collected, which brings up the question on how atmospherically representative the derived fluorescence signals are.
Page 8, line 235 on statement “Further details about the SwisensPoleno fluorescence measurement system can be found in (Graf et al., 2023).”: It is a pity that the Graf et al. reference is not available yet. It is cited few times and could be quite useful for a better understanding of the fluorescence response of the instrument. If the publication of this study still needs some time, it might be worth to put some relevant information still in this study to provide the reader a more comprehensive understanding of the technique.
Page 11, line 296 on statement “Here, the water dataset is used as a proxy for the baseline fluorescence response of the instrument”: Is this the standard procedure to determine the background or is there a force trigger function as commonly used for the WIBS?
Page 14, line 345 on statement “The relative fluorescence spectra for MPs exhibit a noticeably higher response in the λex/λem=280/357 nm channel compared to other particles tested”: What is the molecular explanation for this spectral feature?
Citation: https://doi.org/10.5194/egusphere-2023-2853-RC1 -
RC2: 'Reply on RC1', Anonymous Referee #2, 21 Mar 2024
Comments
The manuscript presents a laboratory study of the commercially available SwisensPoleno system, advancing its use for MP identification with a optimised CNN using holography and fluorescence data for the highest classification accuracy. Overall I am very pleased to have reviewed the manuscript, however, I am not clear if the model trained can be used for live measurements and why there were no measurements using a combination of particles to test the system. From what I understand 15 different airborne particle groups have been measured for holographic and fluorescence data, which was later trained and test in 3 models in an offline process.
Expanding on this if the study had an example dataset to finish that had resuspended already collected environmental MPs (or mixture of the 15 tested particles) this would provide a more convincing argument for its use and is shown in examples of other novel sampling or analysis methods. It would further reinforce the later results discussion and conclusions on the SwisensPoleno system and CNN use and impact of environmental samples and their varying fl response which would affect classification. I do appreciate that this is and is stated as a laboratory study only.
Given the manuscripts main development are the models and datasets created it would be good to see this hosted in an opensource platform, referenced in the text, so that other researchers could access it and build on it.
To fix/address:
Lines 3 – 33: Abstract, please include the models you used, ML is fairly ambiguous, even if you just state the most optimal model that is fine.
Lines 30-33: Abstract, stating that the holo+fl model was the most optimal, this is a key result and adaption of the existing MeteoSwiss model, but is not mentioned.
Line 46: MPs are in the majority of publications classified into the 1-5,000um scale, to fit your classification us more I would also add in the newer ISO ruling on size as well as the publication you have referenced. This is a better justification in my opinion (https://www.iso.org/files/live/sites/isoorg/files/store/en/PUB100472.pdf)
Line 55-65: All references are fine, but none are particularly recent, there are many reviews and also studies that show the detection of MP and NP that could be used. Just an observation that doesn’t have to be addressed but was surprising to see.
Line 117-127: Fl measurement regions, further explanation of why these regions would be appreciated (not only that this is what is available through the SwissPoleno), in MP research there is some debate on what wavelength to use and filter at. This could be an extra comment added to the SI given to support researchers more deeply involved with this aspect.
Line 128-133: Clarity on why MeteoSwiss not used, also can a custom python script be loaded to the system? MeteoSwiss can be trained and updated but is its current supervised learning method not useful for MP analysis?
Line 143: Materials section: I would like to confirm why PS was not considered for analysis as it is one of the major MP polymers detected in aerosols, it is also available from Cospheric and Goodfellows. A comment later in the discussion could cover these questions as other MP researchers would have. Is the statement in line 459 related to this?
Line 212: Please state the water quality value.
Line 223: Added details on the flow rate limits and particle size limits of the system would be good to add here, or in section 2.1. The questions I am thinking while read are, is there a reason no PM2.5 particle sizes were assessed (PS would be the only standard easily availably in this size range), is it because the SwisensPoleno has a detection limit. Secondly is there a flow rate limit as well. A web search of the system does not give much results unless in it’s the Mars of Jupiter variant.
Line 227-233: Dataset creation. Will you make this dataset available as an open data source, same goes with the model. Since this is a large aspect of the work I would expect to see this hosted somewhere, University server, Github or other so that the scientific community and others can make use of the data/model and build on it.
Line 261: Can you explicitly confirm here if CNN is used for the holo+fl or not.
Line 375-377: Could you expand your discussion to PA, PMMA and PA as they seem less distinct from the UMAP. Is there any further discussion that can be given to the important of dimensions 1 and 2 in the UMAP plot, are these evenly weight in terms of dimensionality? I am thinking along the lines of k-means plot dimension value assignment.
Line 387 – 500: Need to check classification values given and how stated, several read as incorrect or misleading, detailed in next comments.
Line 402: Should this be 78%?
Line 417: 93%?
Line 426 – 427: Please confirm the model used here.
Line 430: is this not less than 98% accuracy?
Line 449: related to line 402 accuracy reporting
Line 502-506: Should state for singular MP polymer groups? The sentences make it sound like you have measured all the particle types together which from my understanding is not the case.
Line 505: Can you claim that it is in real time classification, not an in real time measurement? My understanding is that individual particle types where measured. Later the analysis was made.
Citation: https://doi.org/10.5194/egusphere-2023-2853-RC2 - AC2: 'Reply on RC2', Nicholas D. Beres, 08 Jun 2024
- AC1: 'Reply on RC1', Nicholas D. Beres, 08 Jun 2024
-
RC2: 'Reply on RC1', Anonymous Referee #2, 21 Mar 2024
Status: closed
-
RC1: 'Comment on egusphere-2023-2853', Anonymous Referee #1, 17 Feb 2024
The manuscript entitled “Merging holography, fluorescence, and machine learning for in situ, continuous characterization and classification of airborne microplastics” by Beres et al. presents a detailed examination of microplastic (MP) detection, focusing on the deployment of a commercially available online airflow cytometer (SwisensPoleno Jupiter) combined with machine learning to identify, analyze, and classify airborne microplastic particles. Emphasizing the method's efficacy in real-time classification of different microplastic polymer types, the study assesses the technology under controlled conditions, employing metrics of size, shape, and fluorescence. References substances, such as mineral dust and pollen, are used for comparison to assess to what extent particle classes can be discriminated with this approach. The results are thoroughly discussed/integrated in the context of previous studies and existing knowledge.
My overall recommendation is that the manuscript should be published after some relatively minor changes. From a scientific perspective, it clearly fits the scope of AMT. It is a thorough and interesting study, relevant for the field of atmospheric research and beyond. I am convinced that the study will become a useful resource for the emerging field microplastic detection in the atmosphere. It underlines that the combination of size, shape and fluorescence detection might provide sufficient specificity to quantify airborne microplastic (to some extent) also under real ambient conditions. From a formal perspective, the quality of the manuscript is high - it is well-written, the figures and tables are clear, and all arguments and aspects are presented clearly.
Below, the authors will find some general comments that they may want to consider:
Page 3, line 75 on statement “Some commercially available polymers have previously been examined for their autofluorescence”: It might be helpful for the readership to add a few sentences already in the introduction on the molecular motifs that are responsible for the fluorescent emission. Especially for the non-aromatic polymers, which one typically would not consider efficient fluorophores, the observed strong emission reported already in previous studies is an interesting phenomenon.
Page 3, line 82 on statement “No study has used the intrinsic fluorescence of polymers for airborne particle identification and characterization in situ.” I think the preprint by Gratzl et al., 10.26434/chemrxiv-2023-qzhr8 could/should be cited in this context.
Page 3, line 90 on statement “ … assess the fluorescence response of various common microplastics”: I wonder why polystyrol was not included here. It is widely used probably also shows a characteristic fluorescence due to the aromatic structure. Along these lines and in more general terms, according to which criteria were the five polymers selected.
The study by Ornik et al. should be cited (and probably also discussed) somewhere https://doi.org/10.1007/s00340-019-7360-3
Table 1: Did the authors receive any information on the age of the pollen samples? Such commercially available biological reference substances are not necessarily freshly collected, which brings up the question on how atmospherically representative the derived fluorescence signals are.
Page 8, line 235 on statement “Further details about the SwisensPoleno fluorescence measurement system can be found in (Graf et al., 2023).”: It is a pity that the Graf et al. reference is not available yet. It is cited few times and could be quite useful for a better understanding of the fluorescence response of the instrument. If the publication of this study still needs some time, it might be worth to put some relevant information still in this study to provide the reader a more comprehensive understanding of the technique.
Page 11, line 296 on statement “Here, the water dataset is used as a proxy for the baseline fluorescence response of the instrument”: Is this the standard procedure to determine the background or is there a force trigger function as commonly used for the WIBS?
Page 14, line 345 on statement “The relative fluorescence spectra for MPs exhibit a noticeably higher response in the λex/λem=280/357 nm channel compared to other particles tested”: What is the molecular explanation for this spectral feature?
Citation: https://doi.org/10.5194/egusphere-2023-2853-RC1 -
RC2: 'Reply on RC1', Anonymous Referee #2, 21 Mar 2024
Comments
The manuscript presents a laboratory study of the commercially available SwisensPoleno system, advancing its use for MP identification with a optimised CNN using holography and fluorescence data for the highest classification accuracy. Overall I am very pleased to have reviewed the manuscript, however, I am not clear if the model trained can be used for live measurements and why there were no measurements using a combination of particles to test the system. From what I understand 15 different airborne particle groups have been measured for holographic and fluorescence data, which was later trained and test in 3 models in an offline process.
Expanding on this if the study had an example dataset to finish that had resuspended already collected environmental MPs (or mixture of the 15 tested particles) this would provide a more convincing argument for its use and is shown in examples of other novel sampling or analysis methods. It would further reinforce the later results discussion and conclusions on the SwisensPoleno system and CNN use and impact of environmental samples and their varying fl response which would affect classification. I do appreciate that this is and is stated as a laboratory study only.
Given the manuscripts main development are the models and datasets created it would be good to see this hosted in an opensource platform, referenced in the text, so that other researchers could access it and build on it.
To fix/address:
Lines 3 – 33: Abstract, please include the models you used, ML is fairly ambiguous, even if you just state the most optimal model that is fine.
Lines 30-33: Abstract, stating that the holo+fl model was the most optimal, this is a key result and adaption of the existing MeteoSwiss model, but is not mentioned.
Line 46: MPs are in the majority of publications classified into the 1-5,000um scale, to fit your classification us more I would also add in the newer ISO ruling on size as well as the publication you have referenced. This is a better justification in my opinion (https://www.iso.org/files/live/sites/isoorg/files/store/en/PUB100472.pdf)
Line 55-65: All references are fine, but none are particularly recent, there are many reviews and also studies that show the detection of MP and NP that could be used. Just an observation that doesn’t have to be addressed but was surprising to see.
Line 117-127: Fl measurement regions, further explanation of why these regions would be appreciated (not only that this is what is available through the SwissPoleno), in MP research there is some debate on what wavelength to use and filter at. This could be an extra comment added to the SI given to support researchers more deeply involved with this aspect.
Line 128-133: Clarity on why MeteoSwiss not used, also can a custom python script be loaded to the system? MeteoSwiss can be trained and updated but is its current supervised learning method not useful for MP analysis?
Line 143: Materials section: I would like to confirm why PS was not considered for analysis as it is one of the major MP polymers detected in aerosols, it is also available from Cospheric and Goodfellows. A comment later in the discussion could cover these questions as other MP researchers would have. Is the statement in line 459 related to this?
Line 212: Please state the water quality value.
Line 223: Added details on the flow rate limits and particle size limits of the system would be good to add here, or in section 2.1. The questions I am thinking while read are, is there a reason no PM2.5 particle sizes were assessed (PS would be the only standard easily availably in this size range), is it because the SwisensPoleno has a detection limit. Secondly is there a flow rate limit as well. A web search of the system does not give much results unless in it’s the Mars of Jupiter variant.
Line 227-233: Dataset creation. Will you make this dataset available as an open data source, same goes with the model. Since this is a large aspect of the work I would expect to see this hosted somewhere, University server, Github or other so that the scientific community and others can make use of the data/model and build on it.
Line 261: Can you explicitly confirm here if CNN is used for the holo+fl or not.
Line 375-377: Could you expand your discussion to PA, PMMA and PA as they seem less distinct from the UMAP. Is there any further discussion that can be given to the important of dimensions 1 and 2 in the UMAP plot, are these evenly weight in terms of dimensionality? I am thinking along the lines of k-means plot dimension value assignment.
Line 387 – 500: Need to check classification values given and how stated, several read as incorrect or misleading, detailed in next comments.
Line 402: Should this be 78%?
Line 417: 93%?
Line 426 – 427: Please confirm the model used here.
Line 430: is this not less than 98% accuracy?
Line 449: related to line 402 accuracy reporting
Line 502-506: Should state for singular MP polymer groups? The sentences make it sound like you have measured all the particle types together which from my understanding is not the case.
Line 505: Can you claim that it is in real time classification, not an in real time measurement? My understanding is that individual particle types where measured. Later the analysis was made.
Citation: https://doi.org/10.5194/egusphere-2023-2853-RC2 - AC2: 'Reply on RC2', Nicholas D. Beres, 08 Jun 2024
- AC1: 'Reply on RC1', Nicholas D. Beres, 08 Jun 2024
-
RC2: 'Reply on RC1', Anonymous Referee #2, 21 Mar 2024
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 677 | 183 | 35 | 895 | 65 | 24 | 20 |
- HTML: 677
- PDF: 183
- XML: 35
- Total: 895
- Supplement: 65
- BibTeX: 24
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1