Setting up PM sensor

The Plantower Laser PM dust sensor PMS 3003 is a low-cost (~ $15 USD) commercially-available laser particle sensor. This sensor is used for measuring PM in a network of low-cost devices from Location Aware Sensing Systems. The sensor effective measurement range is between 0 and 500 µg/m, with a resolution of 1 µg/m. The working conditions (temperature and relative humidity) of the sensors are -10 to 60 °C and 0 to 99%, characteristics which are suitable for measuring the PM in ambient air in Ho Chi Minh City. A Plantower Laser PM dust sensor and a LinkIt ONE board were combined to become a low-cost sensor (<100$), as illustrated in Figure 1a. The data from two of these low-cost sensors were compared against those obtained with a reference method, using a FH 62 C14 beta attenuation monitor (BAM) (Thermo Scientific, USA). The BAM was situated in the air monitoring station of the Faculty of Environment, VNU-HCM University of Science, located on the rooftop of a 11-storey building. The BAM is calibrated twice per year by mass foils. The two sensors and the BAM were located physically close to each other so that the two devices could measure the same air (Figure 1b).

Collecting data

The instruments were run for three months from October 2018 to December 2018, and one-minute sensor readings were averaged over one-hour and 24-hour periods, for comparison with BAM and gravimetric data (obtained with an SKC Inc Impactor at 10 L/min over 24 hours on quartz fiber filters). Initially the two low-cost sensors were tested over two days to compare their responses to atmospheric PM concentrations. Subsequently, data from the sensors and BAM were collected at the same time and same location, with a 1-week period of coincident impactor measurements.

Calibrating the low-cost sensor

The laboratory calibration method was provided by the EPA 2014 Air Sensor Guidebook 11. We compared the response of the low-cost sensor with data from the reference device. Then, we created a calibration curve that relates the responses of the low-cost sensor to the reference instrument using a mathematical equation. We used a set of one month of sensor data for plotting a calibration curve and then applied the mathematical equation to the data from the following month. Finally, bias was calculated to assess the performance of low-cost sensor.

The response of two low-cost sensors to PM

As shown in Figure 2, there was a very good correlation between the two low-cost sensors (R = 0.95, slope = 1) which is in agreement with the study by Sayahi, which also found high agreement between the same type of low-cost sensors 12. A statistical t-test was used to determine if there are any significant differences between the two measurements, and the results are shown in Table 1. The means of the two sensors were 33.86 and 34.16 µg/m, with variance of 69.46 and 72.40, respectively (n = 2634). The t stat value was smaller than the t critical, showing that there was no difference between the two sensors. Sensor precision was calculated via the coefficient of variation (CV) of the raw outputs: CV = Cs/Cm (where CV is the precision, Cs is the standard deviation of the measurements, and Cm is the measurement mean), and with acceptable CV values being smaller than 10% 7. The CV was calculated for each minute for the two sensors and varied from 0 to 21.4%, with a mean of 3.1% (with the third quartile of CV being less than 5%). These results suggest that the two low-cost sensors can be treated similarly when monitoring PM concentrations.

Comparison of sensor to BAM

The one-minute data from the sensors showed very similar variations of PM to those measured by the BAM. However, there were many spikes in the sensor data (Figure 3 a) which were not evident in the BAM record. Although the 24-hour averaged data of the sensors give very similar results to the BAM, the spikes in the 24-hour averaged data were higher than those of the BAM. The study of similar sensors in Hanoi showed a higher correlation of hourly data between a Panasonic sensor and a BAM (3.1 km away from Panasonic sensor), with R = 0.85 and slope of 0.99 9. The Panasonic and Plantower sensors are both low-cost sensors which are based on the light scattering principle. However, there are no studies comparing these sensors against each other.

To compare sensor data to the true value of PM obtained from BAM, bias calculations can be performed. Bias is a fixed value that is added or subtracted from the true value due to the response of the sensor 11. A bias calculation is as follows: , where is the bias, is the average of the sensor, and is the true concentration of the pollutant. Zero bias is ideal, but bias values lower than 0.1 can be acceptable. The bias calculated for one-minute, one-hour and 24-hour averaged data were 0.66, 0.43 and 0.11, respectively. The bias of the one-minute data was large; however, the bias of the 24-hour averaged data was acceptable.

Over a seven-day period, both the sensor and BAM data were compared against a gravimetric method and both showed high correlations with the impactor data (R = 0.95 for sensor and R = 0.99 for BAM, (n = 7)). The 24-hour averages of the sensor, BAM and filters were 34.6 ± 10.5, 31.9 ± 9.9, and 30.9 ± 10.6 µg/m, respectively.

Calibration of low-cost sensor

Calibrations of the sensor measurements were performed using 24-hour averaged data from the low-cost sensor because these data gave the highest correlation with the BAM. A calibration curve that relates sensor to BAM data was created by linear regression resulting in the equation Y = 0.74X + 4.41, where Y is the calibrated value and X is the measured sensor value. Calibrated 24-hour averages of PM are shown in Figure 4.

After the calibration, the sensor data were much closer to the true value of PM, with the bias decreasing from 0.11 to 0.01. To test the calibration, the linear equation was applied to 10 more days of sensor data. The mean of the 24-hour averaged data of the sensor, BAM, and sensor were 27.0 ± 6.0, 24.8 ± 4.9, and 24.3 ± 4.5, respectively. The bias also decreased from 0.08 to -0.01. These results suggest that a linear relationship can be used for calibrating 24-hour averaged data obtained from the Plantower Laser PM dust sensor. For comparison with other studies, as Table 2 shows, this sensor and reference method can be used, along with other low-cost sensor and reference methods.

The Plantower sensors give the highest correlation coefficient when comparing 24-hour measurements to the reference method. However, it is not good for measuring higher time resolutions, such as hourly and minute data. The Panasonic and DC1700 could effectively compare hourly and minute data; however, the DC1700 (~ $400 USD) has a much higher cost than the Panasonic or Plantower sensors (~ $15 USD). The DC1700 is a completed sensor system that can measure PM in air without any further buildup.