Summer Virtual Series Session 3

Virtual Summer Series

Event Date

Location
Virtually - Pacific Time

Full session moderated by: R. Subramanian, QEERI & OSU-Efluve & Ethan McMahon, US EPA

Register for Session 3 Here

Source characterisation and emission indices estimation using hyperlocal measurements from a low-cost sensor network - London Heathrow airport

Presented By: Lekan Popoola, University of Cambridge

Description: Low cost sensors are becoming increasingly available for studying urban air quality. By making hyperlocal air quality measurements at high time resolution, this technique can be used in identifying pollution sources. When coupled with CO2 measurements, real world emission indices can be estimated. 

We present results for a study at London Heathrow airport (LHR), where 40 nodes were deployed for a period of one-year measuring CO, NO, Ox (NO2+O3) and CO2 as well as temperature, relative humidity, wind speed and direction at 20s second resolution. We will show how the observations were used in distinguishing long-range transport and airport related emissions, thereby allowing the direct measurement of emission indices for the different airport activities.

In this talk, we show how the observations were used to constraint an air quality model (ADMS-airport) to create a powerful tool for predicting air quality.

Follow Up Questions and Answers:

  • Non-local cannot simply be the shared baseline, because that does not prove that it comes from offsite rather than LHR. Is it derived from other sensors roadside or in central London? - Amy Heidner
    • We have high dense sensor network, measuring at 20s, we quantify the most probable reading across the sensor network for a given window (1 hour), this is characteristic of the most network baseline = non-local signal for that hour. See more detail in the published article. - Lekan Popoola https://www.sciencedirect.com/science/article/pii/S1352231018306241
  • You seem to have been able to split local and other emissions but on the basis of ground based traffic and aircraft movements. At the end you state that the extra runway will not contribute to exceedances but you only include the additional aircraft emissions. Should you not also include additional traffic movements due to increased traffic that results from the additional capacity? - Peter Fleming
    • We didn’t explicitly model the additional airport related traffic due to the third runway as there are a lot of unknown variables involved. Having said that, using Emission Factor Toolkit (FET), we expect 80% project reduction in NOx around the airport and if the additional airport traffic follows the similar vehicle standards, it will results in marginal annual NO2 contribution, and unlikely to increase the traffic related fraction by such amount that will lead to exceed annual NO2 limit in 2030.  - Lekan Popoola
  • Are you able to share what specific sensors you used in your study? Are they generally available for sale? - Victor Aprea
    • The CO2 sensor was the Senseair K series NDIR. - John Saffell
  • how sensitive and accurate of the CO2 sensors?- Zhongren Pen
    • Senseair K30 uses ABC algorithm for self calibration, that means outdoor air is used as reference to set a 400 ppm baseline. - Maryna Lotsman
    • CO2 sensors are accurate to within 10% of reference Cavity CO2 devices as shown in the collocation plots. - Lekan Popoola
    • ** @Lekan/Maryna - I find the NDIR CO2 sensors are affected by ambient humidity/get fogged up. How do you deal with that? - R. Subramanian
  • Which reference did you use for the monitoring with low-cost sensors ¿? Did you design your own protocol or take standardized references? - Gonzalo Rosado
    • we tend to use recommended reference or equivalent instruments and we try and calibrate in the field rather than in lab - Lekan Popoola

Data Management Lessons from the RAMP Sensor Network

Presented By: Carl Malings, NASA

Description: The Real-time Affordable Multi-Pollutant (RAMP) sensor package is a low-cost air quality monitor combining five internal gas sensors, external or built-in optical particle mass sensors, battery power, and cellular data communication. More than 60 RAMPs have been deployed across four continents since 2017, generating more than 50 gigabytes of raw data. Alongside the sensors themselves, a system has been developed for storing, cleaning, and processing their data. Data analytics include the application of traditional regression methods as well as more complicated machine learning schemes to calibrate low-cost sensor data to regulatory-grade instruments. Analyzing data and drawing conclusions presents additional challenges. This presentation will outline our system, as well as the scientific and practical lessons learned during its development. Deployments of the RAMPs have varied from dense networks in traditionally monitored urban areas (Pittsburgh, PA) to individual sensors deployed in previously unmonitored regions (e.g. Niger, Rwanda, Ghana). For example, in Kigali, analysis of daily and weekly variability during the dry and wet seasons allowed a tentative attribution of pollution sources for fine particulate matter. The capabilities of electrochemical gas sensors to respond quickly to concentration changes, e.g. to chart vertical profiles using a balloon-mounted RAMP, has also been assessed. Efforts to integrate RAMP network data with other data sources, including existing regulatory monitoring networks (to track sensor performance over time and automatically correct for drift) and remotely sensed satellite aerosol data will also be summarized. Open areas of system development and perceived future data needs will also be discussed, including the need to better integrate stationary and mobile data. Overall, lessons learned in the deployment and management of the RAMP sensor network provide useful insight to future air quality sensing efforts using low-cost sensors.

Follow Up Questions and Answers:

  • What are the main contaminants detected and sources in Sub-Saharan Africa? - Laura Rosales
    • Major local sources are vehicle emissions (vehicles in use tend to be older, and so have higher emissions than those in common use in US/Europe) and wood, charcoal or liquified petroleum gas burning for domestic cooking. Diesel burning by backup generators can also be a major local source. Contaminants of interest here are CO and NOx, and also fine PM. Common regional sources are biomass burning either due to wildfires or to burning of agricultural wastes, and also dust (especially closer to the Sahara Desert). These produce mainly fine or coarse PM which can be transported very far from the source. - Carl Malings
  • Do the RAMPs use the cellular network?  If so, how is that funded? - Amy Heinder
    • Yes, RAMPs communicate their data over the cellular network. You can set the frequencies of data collection and communication (which affects battery usage and data transmission rates). The RAMPs have SIM cards inside; in some cases we’ve purchased pre-paid SIM cards similar to what you would buy for a mobile phone. We have mostly transitioned to using the Things Mobile SIM card (www.thingsmobile.com), which is specifically designed to work with “internet of things” devices which transmit data only. Funding for this is provided for by  the grants which fund this research. Major funding providers for this research were the US EPA, Heinz Endowments, and French ANR. - Carl Malings/ R. Subramanian
  • What kind of RAMP sensor errors do you contend with? - Sridhar Rajagopal 
    • I would split the sensor errors into three basic categories. First is a hardware failure of the sensor itself; sometimes an individual electrochemical cell is defective or breaks for some reason, and all cells lose their sensitivity and responsiveness over time. We have tried to implement automatic codes which track sensor responses and flag anomalous behaviour (e.g. signals that are too high or too low or vary much more quickly or some slowly than is typical). Second is sensor noise, and this is mainly due to the sensor being sensitive to other factors besides what it is meant to detect. We adjust for these as much as possible through field calibration using algorithms which “learn” the relationships between the variations of each sensor’s responses and the variations in the concentrations of the target pollutant (as measured by trusted reference instruments), and therefore try to tune out spurious signals as much as possible, but this is not perfect. We can also average data over time to minimize the impact of transient responses. Third is sensor bias, and this is due to the slightly different behaviors of sensors as they age and are exposed to new environments. Again, this is best accounted for by calibration, and especially through calibration in an environment which is as similar as possible to the environment where the sensors will be deployed (in terms of ambient temperature and humidity and the mix of atmospheric pollutants). These issues and our solution approaches are discussed in more depth in our publications, which can be found here (most of them open access): https://www.cmu.edu/epp/afriqair/pages/resources.html - Carl Malings/R. Subramanian
  • Where can I learn more about the Hygroscopic Growth Correction for PM? - Victor Aprea

Reliable NO2 measurements from low-cost air quality sensors deployed in large networks

Presented By: Geoff Henshaw, Aeroqual

Description: Dense low-cost sensor networks are becoming increasingly popular and offer novel opportunities to measure air quality at the neighbourhood scale in near real-time. However, the success of these networks strongly depends on the reliability of the data. We extend a management and data correction framework previously developed for O3 to NO2 measurements from electrochemical sensors deployed in a dense network in Southern California (~100 sensors). The framework is based on the idea that data from low-cost sensors can be remotely corrected using data from a reliable proxy site (e.g. a regulatory monitoring site). A proxy site is a site where the probability distribution of the measurements of interest is similar to that at the sensor site. A suitable proxy site for NO2 proved to be a site with similar land use (e.g. distance to motorway) characteristics to the sensor site of interest. We show that the three NO2 sensor response parameters, offset, O3 response slope and NO2 response slope can be estimated by minimising the Kullback-Leibler divergence between sensor and proxy NO2 distributions. Using this approach, we were able to remotely detect and correct for sensor drift and improve the accuracy of the data from the electrochemical NO2 sensors. Using the corrected data we were able to detect local-scale effects on air quality that were not captured by the more sparsely distributed stations.

Follow Up Questions and Answers:

  • Why use Dew Point rather than RH as an independent variable? - Victor Aprea
    • RH is derived from Dew Point. - Amy Heinder
    • Hi Victor DP is directly proportional to water vapor concentration and is independent of T while RH is related to T in the atmosphere - Geoff Henshaw
    • Absent additional moisture, RH falls as T rises, while DP is constant. - Amy Heinder
    • So T + RH is effectively equivalent to T + DP as model input? - Victor Aprea
  • Why does the 3-pin NO2 sensor match the reference analyzer better than the 4-pin with the aux. electrode? - Dereck Dasrath
    • 4 pin electrchemical sensors can be more noisy than 3 pin because to use 4 pin you  subtract the aux elecrode current from the sensing electrode current which can double the noise - Geoff Henshaw

Part 1 Group Question & Answer

Follow Up Questions and Answers:

  • How do you calibrate / validate data in areas where there are no reference monitors? How have you overcome the issue of O3 cross response on EC NO2 sensors? - Peter Fleming
    • An ideal situation would be to collocate one or more “gold standard” low-cost monitors with reference instruments at a central site, and correct all sensors to that site (similar to the MOMA method as discussed in Geoff’s presentation). If it isn’t feasible to have such a station full-time, it might still be possible to use mobile instruments for part of the year to develop and/or verify calibrations whenever possible (see the example of the Ambilabs Airpointer being used successfully by one government agency in California, as suggested by Andrew Tolley: https://www.youtube.com/watch?v=ExJGk9NLqB8). - Carl Malings
    • In the future it might be possible to use remote sensing estimates from satellites to estimate baseline regional  concentrations for an area, and calibrate low-cost sensors in the region to match that baseline. Unfortunately, satellite data products still aren’t well validated all over the world, and you still need a reliable ground data source in the region to verify the estimates they are providing.  - Carl Malings
    • If it is impossible to have a calibration against a reference monitor, it is still possible to do qualitative comparisons between the low-cost sensors themselves so long as you have a good mutual consistency between the sensors during a collocation. - Lekan Popoola
    • Cross-sensitivities (such as between the NO2 and O3 electrochemical sensors) can be accounted for by combining the responses from multiple electrochemical sensors into an algorithm for estimating the pollutant concentrations. Ideally, during the calibration of this algorithm, it will “learn” which combination of responses from which sensors correspond with an increase in a certain pollutant, and implicitly account for any cross-sensitivities accordingly.  For example, when calibrating an algorithm for O3, we always included readings from both the O3 and NO2 electrochemical sensors as possible inputs, since this was a known cross-sensitivity. See this paper for more details: https://amt.copernicus.org/articles/12/903/2019/  - Carl Malings

Why Particle Size Matters

Presented By: Tom Grillo

Follow Up Questions and Answers:

  • Measuring particle numbers and sizes down to 20nm with a low cost sensor is a long way from production in my opinion! - Brain Stacey
    • The POPS sensor made by Handix claims to go down to less than 150 nm (but costs $10K) - Jim McQuaid
    • In Australia, they have implemented a "visibility" standard in combination with the PM standard. This creates far greater ability to monitor particulate impacts overall. Contact me for more info and to be connected to the Australian govt personnel if interested to learn more about their visibility standard;  atolley@ambilabs.com
    • The POPS is the first OPC that goes down that far, so this suggests potential to push the size capabilities - Jim McQuaid
    • It's closer than you think.  Please keep an eye on our company and hopefully by 2023 this will be something to discuss further.  - Tom Grillo
    • We can optically detect particles down to ~55 nm - see the DMT UHSAS (I used to work for the company). But it costs about $65k :) https://www.dropletmeasurement.com/product/ultra-high-sensitivity-aerosol-spectrometer/ - Subu
  • Are there any resources showing the performance of your device compared to conventional methods? Especially for outdoor applications? - Karoline Johnson
    • The study that was provided by third party research and presented at the fall san jose event in 2018 showed the accuracy and repeatability of our sensors vs. FEM models as well as other low quality and high quality photometers.  The results had us on par with the Grimm,  and exceeding it with resolution.    There is currently a study going on with our outdoor and indoor sensors witht universities throughout S.E. Asia.  Once that is completed the lead researchers will be publishing their work. This will certainly be shared at that time.  The study has been ongoing since January this year, and the data so far that has been shared shows amazing correlation with co-located monitors like the BAMs. - Tom Grillo
  • When you say low cost, how much is the cost? - Zhongren Peng
    • The 9301P-OEM is pricing can be obtained by contacting us directly from the contact links on particlesplus.com - Tom Grillo

Assessing Urban air quality project

Presented By: Monika Vadali, Minnesota Pollution Control

DescriptionMinnesota Pollution Control Agency has deployed a network of 45 multi pollutant air quality monitoring sensor pods across every zip code in the cities of Minneapolis and St Paul. Each pod measures CO, SO2, NO, NO2, O3, PM2.5, PM10, in addition to RH and temperature, for a total of 225 gas sensors and 45 PM monitors. There is one federal monitor within this study area that measures all these pollutants. In the 12 months of monitoring, the sensor data clearly shows that there are differences in air pollution from one zip code to the next. Even within a zip code with multiple sensors, we have observed that there are differences. This is a classic example of how a denser distributed network can help communities identify and address local issues of pollution and have more data to take action if required. We know that even small differences in pollutant levels can cause serious health effects for vulnerable and sensitive populations, especially small children and the elderly. Some areas within the Minneapolis – St Paul study area also have environmental justice areas with populations of color and people of lower income groups. 

The deployment of the sensors, especially selection of locations, has been entirely driven by community input. In Minneapolis, all the sensors are on wooded street light poles in residential neighborhoods and in St Paul, the poles are located on light poles in public school parking lots. All of the data is also being displayed on a public web page, updated on a weekly basis. Communities are looking at this data and we strive to display quality assured data but it has been a challenge. 

This presentation will discuss the one year summary results from neighborhood sensors, special event analysis ( July 4th and Minneapolis riots), sensor and federal monitor comparisons, sensor to sensor comparisons, data quality issues with sensor measurements and application of normalization methodology to improve accuracy.

Follow Up Questions and Answers:

  • Did you inform the casino owners that you were taking measurements? How did they react to the resulting data? - Ethan McMahon
    • No, we did not inform the casino owners about the measurements. Laura Rosales, co-author from the Las Vegas study

Clean Air Tourism: Assessment of Individual Exposure to Indoor and Outdoor Air Pollution on the Las Vegas Strip Using Portable Sensors, Las Vegas, Nevada, USA.

Presented By: Carlos Arambula, Nevada State College & Marina Vanderberg, Nevada State College

Description: Poor air quality has been shown to increase the rates of mortality and morbidity in urban environments, where most of the world’s population lives. Urban dwellers increasingly tend to spend more time indoors. Indoor air quality is assumed to be better than outdoors’ due to extensive sources of contaminants in the later. Exposure to outdoor pollutants varies due to local meteorological conditions. In contrast, indoor air quality is determined by indoor pollution sources and the ventilation systems. Non-smoking indoor areas should have better air quality than smoking areas if ventilation systems are able to sufficiently contain drifting tobacco smoke to avoid the risk to secondhand smoke exposure for non-smokers. 

This study presents the measurement of outdoor air quality using portable sensors along the Las Vegas Strip, a year-long touristic destination in Nevada, USA with ~70,000 daily traffic counts at very low speed and more than 55,000 pedestrians. Additionally, risk to secondhand smoke at smoking- and non-smoking areas inside six casinos was assessed. Monitoring air quality will help to assess the health risks at these environments. Measurements were taken twice each season for one year to evaluate air quality seasonal variability. Measured parameters included particulate matter (PM2.5 and PM10 in µg/m3) by two Nova SDL 607 wearable air quality monitors and carbon dioxide/carbon monoxide (in ppm) by a pSense portable CO2 meter. Air temperature (°C) and relative humidity (%) were also recorded simultaneously. 

Smokers were identified as the major source of particulate matter in both, outdoor and indoor environments. Proximity to smokers resulted in PM2.5 concentration increases up to 343 and 854 µg/m3 in the outdoor and the indoor smoking-areas microenvironments, respectively. We will further discuss individual exposure, air quality seasonal changes and the relationships between PM2.5, PM10, CO2 concentrations and environmental conditions at each environment.


Follow Up Questions and Answers:

  • Did you inform the casino owners that you were taking measurements? How did they react to the resulting data? - Ethan McMahon
    • No, we did not inform the casino owners about the measurements. Laura Rosales, co-author from the Las Vegas study

Part 2 Group Question & Answer