the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 06 Nov 2025
Identification of Dust-Dominated Periods and a PM2.5 Correction Based Solely on Plantower PMS Sensor Observations
Abstract. The Plantower PMS5003/6003 sensor is widely used for low-cost monitoring of particulate matter (PM), but it substantially underestimates PM2.5 and PM10 during dust events. This limitation is especially critical in the arid regions, such as the western United States, where windblown dust frequently degrades air quality, visibility, and public health. Identification of dust episodes typically relies on federal reference or equivalent method (FRM/FEM) monitors, often supplemented with satellite or meteorological data, but these resources have limited spatial resolution. This study examines whether dust episodes can be directly identified from PMS5003 sensor outputs. We analyzed measurements from 109 PMS sensors collocated or nearby 75 U.S. EPA monitoring sites with hourly FEM PM2.5 and/or PM10 between January 2017 to May 2025. Two dust-event cutoff thresholds (threshold1 and threshold2) were developed using the sensor-reported ratio of coarse (2.5–10 µm) to ultrafine (0.3–1 µm) mass concentration, and relative humidity, to identify potential dust events. The thresholds can be used in real time, relying on the preceding 336 hourly measurements (consistent with PurpleAir’s public archive display). To improve PM2.5 estimates from the PMS sensor (pm2.5_alt, a common correction for Plantower PMS measurements reported by PurpleAir), this study used pm2.5_alt measurements identified as potential dust events to develop a correction factor for dust events through non-linear regression. This correction reduced the mean bias error between pm2.5_alt and FEM PM2.5 for 103 of the 109 sensors by 59.8 ± 24.3 %, while ensuring that the mean bias error did not fall below -1 µg/m³. This framework enhances the utility of PMS5003/6003 sensors for detecting and quantifying dust-related pollution episodes, extending monitoring capabilities in regions where regulatory coverage is limited.
- Preprint
(1887 KB) - Metadata XML
-
Supplement
(8140 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- CC1: 'Comment on egusphere-2025-5063', Carl Malings, 19 Nov 2025
-
CC2: 'Comment on egusphere-2025-5063', Carl Malings, 19 Nov 2025
Publisher’s note: this comment is a copy of RC1 and its content was therefore removed on 20 November 2025.
Citation: https://doi.org/10.5194/egusphere-2025-5063-CC2 -
CC3: 'Comment on egusphere-2025-5063', Carl Malings, 19 Nov 2025
Apologies for the duplicate post; the contents of each should be the same.
Citation: https://doi.org/10.5194/egusphere-2025-5063-CC3 -
RC1: 'Comment on egusphere-2025-5063', Carl Malings, 20 Nov 2025
General Comments
This paper investigates an important limitation of existing low-cost sensors for providing reliable data during dust-impacted events, and proposes an approach to both identify these events and compensate for sensor underestimation of PM2.5 during identified events. The proposed approach is potentially generalizable to many currently deployed low-cost sensors using the PMS sensing technology, thereby enhancing the value of the data provided by these sensors.
Currently, the manuscript lacks a thorough investigation of the performance of the proposed methods of dust-impacted measurement detection, instead focusing on the calibration approach applied to these measurements. Prior to this, assessment of the accuracy of the identification approaches should be carried out by means of common classification metrics, such as false positive and false negative rates, precision (ratio of true positives to all positives), and recall (ratio of true positives to all actual events, as determined by the FEM-derived CF). While Lines 326-332 discuss the relative fraction of detected events corresponding to different CF bins, it is not clear from this discussion what fraction of dust-impacted events (as determined by the FEM-derived CF) were correctly identified by the detection algorithm, or what fraction of events identified by the detection algorithms corresponded to high-CF events. This is a critical aspect of the analysis which is currently overlooked.
Besides this major point, and the other specific comments listed below, I believe that the paper can be suitable for publication after revision.
Specific Comments
Lines 58-59: This is a continuation of the point made in lines 48-49; these can be combined.
Line 103: This is a sentence fragment.
Lines 134-136: Please clarify what is meant by “alternate PMS5003” as compared to PMS5003, and why the ratio of these quantities is relevant here.
Line 142: “are subset” should be “are a subset”.
Line 143: “month” should be “months”
Line 145: Clarify why “at beach” disqualified these sensors.
Line 145: If 4 sensors were included in the study, they should not be counted among the non-included sensors.
Line 155: It is mentioned later that T is also included, though it does not seem to be used elsewhere in the analysis.
Line 202: Can you explain why mass concentrations were used here, rather than the number counts directly?
Line 205: Parameter name is accidentally subscripted.
Figure 1: The arrow denoting threshold2 is misleading, as it points in the opposite direction from points exceeding the threshold. If these arrow are meant to help indicate the meaning of the lines, I’d suggest removing them, and just placing the text directly adjacent to the lines.
Line 276: All circles are empty, not solid.
Line 282: Does the model incorporate FEM PM2.5, or is it only calibrated using these data?
Lines 282-284: This describes that the calibration was determined using identified dust events based on the post-processing and near-real-time approaches. How would the results differ if calibration were based on identified events using the FEM-derived CF instead?
Line 286: Parameters adj_pm2.5_alt and pm2.5_alt are accidentally subscripted.
Lines 301-302: Why is MBE the only metric considered? I would suggest also including a metric like Mean Absolute Error or Root Mean Square Error, including normalized variants of these metrics, as the issue being addressed is not one of constant measurement offset, but of relative underestimation which varies in proportion to the magnitude.
Lines 303-304: Parameters are accidentally subscripted. Furthermore, these definitions are counter-intuitive, e.g., a “positive bias” usually implies that the estimate (pm2.5_alt or adj_pm2.5_alt) is higher than the “true” value (FEM PM2.5).
Lines 335-339: The description implies that only data identified as being associated with dust events are presented in the figure, but my understanding is that all data are shown in the figure, with black circles denoting data associated with identified dust events, via the post-processing approach. This should be clarified. Same comment applies to Lines 373-375.
Line 403: The final statement about the impact of A is not needed; this was discussed when the correction was defined above.
Lines 404-418: I think that the impact of using sensor-specific calibration factors discussed here is an important point to understand, and to further emphasize it, I would suggest adding a third set of points to figures 3 and 5, showing results if the sensor-specific calibration factor were used.
Citation: https://doi.org/10.5194/egusphere-2025-5063-RC1 -
RC2: 'Comment on egusphere-2025-5063', Dimitrios Bousiotis, 28 Nov 2025
Dust events can be a significant threat to public health, as it involves sharp increases of the coarse particles which are associated with several health hazards. The limitations of the reference air pollution measuring networks are known, as due to the high cost of the instruments used they can only be deployed in limited numbers. Low-cost sensors can provide an alternative which can potentially increase the number of monitoring points, though their usability in specific scenarios is not fully explored. This study attempts to reveal the capabilities of the low-cost sensors in measuring the coarse particles during dust events despite their inherent limitations. The study tests previously used (and new?) methodologies with measurements collected from Plantower sensors.
My main objection is the title of the manuscript and the way the outcomes of the analysis are presented. The title mentions “Identification of Dust-Dominated periods”, which when I first read, I was expecting that through the methodologies used dust events would be identified by the low-cost sensor data and then evaluated using the reference data. Instead, the dust events are a group of days which comply to specific rules without having any kind of evaluation (whether they were really dust events or something else, which is a possibility correctly pointed in the Introduction and Methodology parts). Similarly, in the Introduction the phrase “examines whether dust episodes can be directly identified” and in the conclusions the phrase “successfully identified” are used, which point exactly on the same thing (that the LCS provided and the regulatory instruments approved). I suggest that both the title and the parts that mention successful identification of the events are removed or at least toned down to the possible ability of the sensors to identify the events. The calibration (correction) of the sensor parts are useful though and probably should be the focus of the title and conclusions.
Line 203. The assumed density of the particles is too low. Household dust is reported at about 1.6 and mineral dust at about 2.7. This low density would probably lead to great underestimations of the mass of the coarse particles.
Line 313. I don’t understand the numbering scheme in reporting the number of potential dust events. What do 0 – 3785 or 0 – 9% mean? Does it start from zero because they are “potential”?
Minor comments
Please add the six size bins reported by the PMS sensors in chapter 2.1 when the sensors are presented.
In line 146 it is mentioned that the data is downloaded by the company’s site. Are these given as raw or calibrated data?
In line 161 the variable pm2.5_cf_1 is mentioned. What is it (according to the manufacturer)? How is it named and presented in the datasets?
While out of the scope of the specific paper, but were there any discrepancies from the different FEM measurements reported at the same sites (as mentioned in line 181)?
Line 205. CFtoUF is inconsistent with the naming used in other spots (CF_to_UF).
How did the regression perform in the correction at chapter 2.5.3?
In line 365, please use the full names of the states.
In line 442, please use Celsius as well (can be put in parentheses).
Line 451, while I like the optimism, a major shortcoming with the use of low-cost sensors is the variability of their performance in time. A factory “Factor A” would only be true for a specific period (as pointed by figure S5), though it would be a proxy of its general performance. In general, though it would be an improvement from the “tabula rasa” that every sensor comes as. Consider adding a small asterisk about this.
Citation: https://doi.org/10.5194/egusphere-2025-5063-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 178 | 39 | 27 | 244 | 36 | 11 | 11 |
- HTML: 178
- PDF: 39
- XML: 27
- Total: 244
- Supplement: 36
- BibTeX: 11
- EndNote: 11
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Publisher’s note: this comment is a copy of RC1 and its content was therefore removed on 20 November 2025.