Providing Fine Temporal and Spatial Resolution Analyses of Airborne Particulate Matter Utilizing Complimentary In Situ IoT Sensor Network and Remote Sensing Approaches

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study analysed the airborne particulate matter using a two-pronged approach that combines data from in situ Internet of Things (IoT) sensor networks with remotely sensed aerosol optical depth (AOD). The authors involved setting up a network of custom-designed PM sensors and expanded their analysis to a national scale by developing machine learning models that estimate hourly PM2.5 levels throughout the continental United States.

 They demonstrate that the combination of AOD data with meteorological analyses and additional data sets can effectively model PM2.5 concentrations. The reconstructed PM2.5 surfaces created in this study are invaluable for monitoring pollution events and performing detailed PM2.5 analyzes.

 Generally, the approach and methods looks scientifically sound and the results seems to be reasonable.

 

Some comments are:

 

1. When compare AOD and PM2.5, the relative humidity is important because PM2.5 usually control RH. The authors should explain more specifically how control RH when compare AOD and PM2.5.

2. Explain more detail regarding reconstructed method

Author Response

We sincerely appreciate the time and effort that you dedicated to reviewing our manuscript and are grateful for your insightful comments and suggested improvements. Thank you!

 

Please find a PDF of or responses to your points. We appreciate you!

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Good work.

References are numerous, up-to-date and appropriate for the subject. No unnecessary self-citations are reported.

Figures and tables are sufficiently clear and explanatory.

The paper is based on different methodologies and experimental approaches.

A homogenization in citations is required. At line 72: "3, Lary et al., 2014 developed a machine learning model to provide daily distributions...". At line 78: "In a separate study, 10, Yu et al. , 2022 enhanced the modeling of PM2.5 concentrations..."

Author Response

We sincerely appreciate the time and effort that you dedicated to reviewing our manuscript and are grateful for your insightful comments and suggested improvements. Thank you!

 

Please find a PDF of or responses to your points. We appreciate you!

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Comments on “Providing Fine Temporal and Spatial Resolution Analyses of Airborne Particulate Matter Utilizing Complimentary In-Situ IoT Sensor Network and Remote Sensing Approaches”

The authors used ground PM measurements from EPA, OpenAQ, and sensors in conjunction with reanalysis and remote sensing datasets to train a machine learning model to predict spatio-temporal PM across the USA. I have the following major comments:

1. One of my major comments is on the length of the article. The manuscript is too long, and several sections or sub-sections can be shortened. The quantity is dominating the quality. For example, the second paragraph of Section 2.2 (related to AOD). The definition and the given information on the AOD are well understood and known across the scientific community.
2. In relation to the PM sensor, nowhere in the text is the accuracy of the sensor measurements discussed. But the authors used the sensor data alongside the EPA and OpenAQ data for model building.
3. All the low-cost PM sensors work on the light scattering principle, and the measurements of particles less than 0.3 microns contain huge uncertainty. In addition, the PM0.1 data provided by the IPS7100 is based on extrapolation (not measurement). The authors have provided their modeling results for PM0.1 in figures 3 and 5. Needs justification.
4. Among the objectives set for the analysis, the first one (importance of the spatial and temporal synchronization of measurements) seems vague. To achieve that objective, the authors conducted all PM size fraction modeling using MINTS data. First, the mass concentration values of PM0.1 seem to be too low (as low as 0.001, as per figure 3); instead, the authors might have used particle count to model rather than the mass concentration. Second, the train ‘R’ values are always one, and the huge increase in test RMSE (20x) compared to train RMSE indicates some sort of over-fitting (I understand that the authors have taken care of potential over-fitting in their ML model configuration). Provide reasons. Third, I didn’t see any application of this exercise. We can’t use this modeling exercise for any sort of prediction. The authors need to provide the rationale for this objective and the corresponding exercise.
5. Provide units for the RMSE and PM statistics, where applicable.
6. PM2.5 modeling. The inclusion of MINTS data has improved the model performance by 0.015 in terms of Test-R and by 0.16 in terms of Test-RMSE (as per Table 6). Is the whole exercise of incorporating MINTS data (in PM2.5 modeling) worth the effort?
7. Section 2.5: Generally, AODs are path-length normalized (I am not an expert on GOES AOD data). The authors need to verify their arguments related to the AOD dependence of zenith angles and other parameters.
8. Table 1. The authors can provide more information related to the variables provided here, like their spatial and temporal resolutions, etc.
9. Section 3.2.1. Lines 422 to 424. Are the authors using the instantaneous (5-minute) AOD against the hourly averaged PM2.5 for the model training?
10. Check for inconsistencies. At one place, the train and test split was mentioned as 80-20; in other places, it was 90-10. Related to MINTS temporal resolution, it’s mentioned as 3s in one place and 6s in another place.
11.  In Figure 12, the authors are comparing the hourly PM2.5 against the annual threshold. It is advised to compare the annual mean PM2.5 in each grid against the annual threshold. Similar exercise on daily mean values against the daily PM2.5 threshold can also be performed.

Author Response

We sincerely appreciate the time and effort that you dedicated to reviewing our manuscript and are grateful for your insightful comments and suggested improvements. Thank you!

 

Please find a PDF of or responses to your points. We appreciate you!

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

I am satisfied with the author's responses.