Story 3


From late October 2013 to early June 2015, the Citizen Sense research project collaborated with residents of northeastern Pennsylvania to develop a citizen-led air-quality monitoring project. Residents in this area were particularly concerned about the inadequate environmental regulation and monitoring in relation to expanding hydraulic fracturing activities, and had already begun to undertake community activities for monitoring environmental pollutants.

Citizen Sense worked with local residents to develop a monitoring kit that included Speck PM2.5 sensors, BTEX badges (for monitoring benzene toluene, ethylbenzene and xylene, volatile organic compounds commonly associated with petroleum-related activities), a community platform for mapping monitoring locations and viewing real-time and historic data, and a Frackbox, which monitored nitrogen oxides, ozone, volatile organic compounds, temperature, humidity and wind direction. Residents were also provided with a logbook of instructions, which suggested several options for recording observations of environmental conditions and health effects.

The Citizen Sense kit and sensors were distributed in October 2014 during a monitoring workshop and walk. In total, 30 monitors and kits were distributed to residents, and 3 Frackboxes were placed near infrastructure. The monitoring period ran for over 7 months, until June 2015. During peak monitoring activity, there were 23 active monitoring sites, and there was consistent monitoring taking place at up to 16 sites over a period of 7 months.

Sensing Practices video, Catherine Pancake

The data stories are generated using the Citizen Sense Airsift Data Analysis Toolkits, which were developed to allow for citizen-led interpretation of datasets. The core data available for interpretation is the PM2.5 sensor data using the Airsift PM2.5 Data Analysis toolkit. The Frackbox data and Airsift Frackbox Data Toolkit are also available as part of the resources section of our website. The Airsift toolkits use and adapt the open source software, openair, developed by atmospheric scientists for the analysis of air pollution data. In order to blur the exact monitoring locations, the monitoring locations have been labeled with township locations, which can be found in relation to Pennsylvania counties.

The 5 data stories presented on this site demonstrate the different patterns that have emerged from the data. Our hope is that the stories and the Airsift toolkits will provide a method and guide for others to undertake their own analysis of this citizen-gathered dataset, and to contribute to the wider development of citizen-led environmental monitoring and data analysis tools and practices.

Paul Karpich lives in Dimock Township in Susquehanna County, Pennsylvania, an area heavily populated with infrastructure from the unconventional natural-gas industry. The area has no zoning ordinances to prevent gas exploration and development, and Paul’s house is surrounded on 3 sides by compressor stations, a situation that Paul describes as living in a “death triangle.” After experiencing conditions that Paul felt had impacted on his health and well-being, he looked for ways to monitor the emissions and noise from the surrounding compressor stations. Along with other local residents, Paul collaborated with Citizen Sense to design a sensing kit in response to his concerns about air quality. Paul used this kit to monitor the air quality at his home, collecting over 7 months of data using the PM2.5 monitor. Paul located the PM2.5 monitor on the west-facing wall of his home, in an area with meadows and trees where there are no visible houses nearby.

Local sources of particulate pollution
As can be seen in Figure 1, which includes geo-location data on hydraulic fracturing infrastructure documented by resident and participant Meryl Solar, the Dimock monitoring location is close to 3 natural-gas compressor stations. In relation to the monitoring site, the first compressor station is located 3000 feet to the southeast, the second compressor station is located 5000 feet to the southwest (and is visible from the monitoring location), and the third compressor station is located 7500 feet to the north. (Note that this map is historic data to reflect conditions at the time of monitoring. The landscape is continually changing, and to see a more up-to-date map of current conditions see

The monitoring location is setback from any major roads. The monitor was positioned approximately 90 feet east of the dirt track access road. There are also approximately 10 operating well pads within 1 mile of this site. There are 4 operating well pads within 5000 feet of the property; the first well pad is located to the southeast, the second and third well pads are located to the west, and the fourth well pad is located to the north. Additionally, within a radius of 7 miles there are a number of large compressor stations in all directions. There are over 30 compressor stations within a 20-mile radius of the monitoring location.

There are 2 small working farms nearby the monitoring location: the first is located around 5000 feet to the west adjacent to the Lathrop compressor station; the second is approximately 7000 feet to the south of the monitor. Both local farms are less than 200 acres in size and are non-industrial, with a dairy production of approximately 25 cows per farm. During the monitoring period, no fields close to the location were worked for planting. Hay harvesting took place in June, July and August 2015 for 6 days lasting no longer than 3 hours per day. The participant also mowed his lawn each week for an hour during May until June 2015. The Procter & Gamble factory is 11 miles southwest of the monitoring site. There are also a number of operating quarries 15 miles north of the site.

At the end of May and June 2015, there were high pollen counts in the local area.

Regional sources of particulate pollution
25 miles to the north is the city of Binghamton, New York, and 25 miles to the south to southeast is the city of Scranton. The cities of Pittsburgh, Pennsylvania, and Harrisburg, Pennsylvania are located over 200 miles southwest from the site.

Paul described how the compressors could often be heard at the site, creating a noise disturbance. It was also common to smell unfamiliar odors at the monitoring location, especially at night.

II. Is there evidence of a problem?

The Speck device used to monitor PM2.5 particles is an “indicative” monitor. This means that measurements can give an indication of pollutant concentrations, but cannot be directly compared with national and international guidelines and standards in an “official” or regulatory sense. Despite this, indicative monitors are a well-established method within atmospheric science for carrying out initial surveys of an area to establish whether a potential problem merits further investigation. Indicative monitors are also becoming increasingly available for citizen-based air-quality monitoring, similar to this study.

Indicative daily mean concentrations of PM2.5 are shown as a time-series chart in Figure 2. The World Health Organisation (WHO) guideline of 25 µg m-3 for 24-hour daily mean concentration of PM2.5 is breached on a number of occasions, suggesting that further investigation may be merited. However, it is important to determine whether these breaches were caused by “local” sources of pollution close to the sensor (i.e., within 1000 feet (300 meters)), by sources within a few miles (or kilometers), or by regional sources affecting the whole area. Regional sources of PM2.5 typically include large urban areas, major industrial or agricultural processes, or main transport routes.

Figure 2: Time series chart of 24-hour mean concentrations of PM2.5 at Dimock S 21.

Figure 3 shows an extract of the monitoring at Dimock S 21 presented as 1-minute mean concentrations of PM2.5 from January 12 to 19, 2015. Measurements from the Liberty E 111 site are shown for comparison. Regional sources of pollution appear as broad “humps” of elevated pollution affecting both sites. For example, during the period January 14 to 16, 2015, this can be seen quite clearly as visualized in Figure 3. These regional broad humps were also demonstrated in the data at Bridgewater N 56 and Brooklyn N 64_323. Local sources of pollution appear as short “spikes” affecting only one or the other site, for example, January 14 to 15, 2015, at Dimock S 21 (but not at Liberty E 111). This pattern occurs because pollution mixes in the atmosphere as it travels away from a source, smoothing the speed of changes in concentrations. Local sources often augment regional sources, revealed as a spike on top of a hump (e.g., January 15, 2015, at Dimock S 21). When significant local sources of particulate pollution are evident, they elevate ambient concentrations well above those caused by regional sources.

There are many possible sources of pollution in the area and we have to look at the measurements more closely to see if we can deduce which activities cause these spikes. Knowing the source of pollution is important as some activities produce more toxic particulate matter than others, and actions to mitigate sources should be targeted to the cause of the problem.

Figure 3: Comparison of 1-minute mean concentrations of PM2.5 at Dimock S 21 and Liberty E 111 during January 2015, highlighting local “spikes” and regional “humps.”

III. Characterizing the problem

When is the source most evident?
We are most interested in these “spikes” in pollution relating to the possible local source. We can characterize this source by plotting a number of charts, filtering out concentrations above the regional baseline. We refer to Figure 3 in Data Story 1 to set the regional baseline of 15 µg m-3, which is the approximate concentration at which regional “humps” in pollution peak. Data Story 1 provides the most comprehensive and extensive monitoring data, and is a good point of comparison across all data stories.

Figure 4: Mean PM2.5 concentrations at the Dimock S 21 monitoring site, grouped by hour, month and weekday.

Figure 4 investigates when these “spikes” in pollution occur by grouping concentrations by hour, month and day of the week. Sources of pollution related to commuter or transit traffic typically show peaks in concentrations coincidental with peaks in traffic flow, i.e., morning and evening rush hour with notably lower levels at night and on Sundays. However, on average, the local source at the Dimock S 21 monitoring site is highest early in the morning, before rush hour, and on Sundays. These charts can be used to match patterns in the occurrence of spikes with particulate-generating activities in the area. Data Stories 1 and 2 also show elevated levels in the morning. Dimock S 21 shows these elevated levels between midnight to 3 am, and Bridgewater N 56 and Brooklyn N 64_323 after 5 am.

It should be noted that the weather plays a large role in particulate levels. For example, dust tends to be dispersed more slowly during the hours of darkness, as vertical and horizontal wind speeds are generally lower. This phenomenon may skew charts somewhat.

Which direction is PM2.5 coming from?
Wind direction has a considerable influence on pollution measurements. A sensor will only record emissions from a particular source or activity if the wind blows it from the source towards the sensor. Therefore, we can investigate where a source of pollution is likely to be located by plotting wind direction against pollution concentrations.

Figure 5: A scatter plot showing the relationship between PM2.5 concentrations and wind direction (in degrees).

Figure 5, which documents PM2.5 across the monitoring period and includes short-lived spikes with very high readings that in some cases are related to pollen events, shows how pollutant concentrations at Dimock S 21 are influenced by wind direction. It is clear that the highest peaks are recorded when the wind blows from the north (340O to 0O), southwest (210O to 230O) and elevated levels from the east to southeast (110O to 160O). In contrast to Data Stories 1 and 2, Figure 5 shows there are somewhat elevated PM2.5 concentrations from most wind directions.

A polar plot is a more intuitive way of looking at this relationship. As indicated in Figures 6 and 7, this plot shows color contours of pollutant concentrations in relation to wind direction and wind speed (zero wind in the center, increasing to 35 ms-1 at the outer ring). The highest mean concentrations are shown in red, lowest in blue.

Figure 6: A polar plot of PM2.5 concentrations higher than 15 ug/m3 at Dimock S 21 (left panel) and Liberty E 111 (right panel). The color scale shows mean PM2.5 concentration, wind direction around the polar axis, and wind speed from the center outward. Note that the mean concentrations shown are relative (e.g., highest mean concentration is 50 µg m-3 at Dimock S 21, and 100 µg m-3 at Liberty E 111).

In common with Figure 5, Figure 6 for Dimock S 21 shows high concentrations when the wind is from the north, northeast and southeast and southwest. There also elevated levels in the northwest. In contrast, the highest levels of pollution are only to the south and northeast of Liberty E 111.

Figure 7 shows the NO and NO2 levels at Dimock S 21 recorded with a Frackbox monitor at the same site. Somewhat similar to Figure 6, NOx levels are highest when the wind is blowing from the southeast. NO2 levels are also high when the wind is blowing from the north and northeast. Both Figures 6 and 7 suggest that the cleanest air is coming from the northwest at higher wind speeds.

Figure 7: A polar plot of NO (left panel) and NO2 concentrations (right panel) at Dimock S 21 (left panel).

Under which weather conditions are PM2.5 levels most evident?
Different sources of pollution will behave in distinct ways according to the weather. For example, windblown dust will primarily occur during dry, windy conditions. Sometimes, you can learn about a source by characterizing this weather-related behavior.

Figure 8: Relationship between PM2.5 concentrations and wind speed, indicating that higher PM2.5 concentrations tend to occur at lower wind speeds.

The relationship between PM2.5 pollution concentrations and wind speed is shown in Figure 8. This suggests that the main source is not wind-blown dust, as peak concentrations are recorded during lower wind speeds. This conclusion is supported by Figure 9, which shows that the highest 1-minute mean concentrations of PM2.5 were recorded during relatively humid conditions (60% to 90% humidity), and during high humidity there would be fewer occurrences of wind-blown dust.

Figure 9: Relationship between PM2.5 and humidity at Dimock S 21.

IV. Drawing the evidence together

Using the tools provided on the Citizen Sense Airsift Data Analysis Toolkits, we have characterized sources of particulate pollution detected by the Dimock S 21 sensor as follows:

  • While regional sources of pollution were detected, there was some evidence of additional local source or sources.
  • The local source(s) appear to be to the north, southwest, and southeast of the sensor.
  • The local sources from the southeast seem to be at higher concentrations at lower wind speeds. Therefore this would suggest PM2.5 levels are not related to the Highway I-81 to the east.
  • Looking at all other local sites and industries capable of generating PM2.5 as documented in Figure 1, the local source(s) to the north could include emissions from the third compressor station, located 7500 feet north of the monitoring site. The local source(s) in the southeast could include the first compressor station (3000 feet to the southeast), and the local source(s) to the southwest could include the second compressor station (5000 feet southwest).
  • The regional source to the southwest could include the Procter & Gamble factory, together with emissions from the cities located in the southwest.
  • Elevated PM2.5 levels are unlikely to be related to re-suspended or wind blown dust, as peak concentrations are recorded during lower wind speeds.
Gas flare documentation, Barbara Scott


The Citizen Sense project is led by Dr Jennifer Gabrys, working in collaboration with researcher Helen Pritchard.

Dr Benjamin Barratt at the Environmental Research Group, King’s College, University of London contributed to the co-authoring and analysis of the data stories.

Lau Thiam Kok contributed to the co-development of the Citizen Sense Airsift Data Analysis Toolkits, using and adapting openair software developed by Dr David Carslaw.

Raphael Faeh contributed to the digital design and layout of the data stories.

Special thanks are due to the participants and residents in northeastern Pennsylvania who contributed to the design, development, and testing of the monitoring kit, as well as to the collection and analysis of data, and communication of results to wider publics and regulators. The Dimock data story received additional contributions from Paul Karpich. For more information on project contributors, see Citizen Sense People.

This data story is prepared under the assumption that all pollutant, cartographic and meteorological measurements are valid and not sufficiently biased to cause misrepresentation of results. Please refer to the Airsift Data Analysis Toolkits and Terms of Use for further information.

The research leading to these results has received funding from the European Research Council under the European Union’s Seventh Framework Programme (FP/2007-2013) / ERC Grant Agreement n. 313347, “Citizen Sensing and Environmental Practice: Assessing Participatory Engagements with Environments through Sensor Technologies.”

read more: