BAM-Based Accuracy Comparison

We compare the PM2.5 measurements to show the reliability of our PM sensor. We use 6 different sensors and compare them with each other along with our BAM sensor. This comparison has been done by comparing the data recordings taken in the IIT Delhi Campus for 11 days.

1. Time series plot with our sensors

   We start out by plotting a time series-based plot to show how well our sensors are calibrated with each other. The values being recorded are extremely similar by our sensors and vary very slightly from each other.

   

2. Time series plot with BAM sensor

   we have plotted a time series plot using all our sensors and the reference grade BAM sensor. We observe that the values from our sensor are extremely similar in terms of the trend followed but are higher than the values recorded by the BAM sensors. This discrepancy between these measurements has been studied in several other papers [ 1, 2 ] and is prevalent in low cost sensors.

   

3. Correlation between sensors

   1. Here we plot the pair plot to show how the data from our sensors is correlated to the BAM PM data. The trend followed is extremely similar and thus, the correlation is high. This is fwhiteurther corroborated with the below plot which shows the correlation values between all the sensors.

      

   2. Correlation Matrix

      

Thus, as a conclusion for the BAM-based accuracy analysis, we observe that though there is excellent correlation, the absolute values might be over-reported by PMS7003 when PM values are higher. However, this is the best available low-cost sensor for now and so, our further work and challenges include better mapping of PM values at higher ranges from low-cost sensors to the values recorded by reference machines.

Static Sensor-Based Accuracy Comparison

We additionally compare the quality of our mobile PM measurements with static sensor measurements in nearby locations. To compare our measurements, we first locate the mobile sensors that were close (≤ 150m), to any static sensor. We found three static sensors satisfying this criteria, which were installed at CRRI Mathura Road, Delhi, Jawaharlal Nehru Stadium, Delhi and ITO, Delhi respectively (referred to as CRRI, JNS and ITO respectively in the below correlation based time-series plot).

We compute the correlation between the hourly mean of all PM2.5 values recorded by a static sensor and its nearby mobile sensors. We expect to see a high positive correlation between both the hourly averages. Below is shown the daily correlation values of all three locations during our deployment. We observe a high correlation across most days. Differences in raw PM2.5 values between two types of sensors are caused by a variety of reasons including the difference in heights they have been installed at, the difference in the amount of exposure of the sensor to direct smoke and dirt, the difference in measurement technique and the averaging procedure introduced by us to compare the sensors.

We also found 15 instances where the correlation was found to be negative. These were cases where PM values were extremely close in magnitude and thus, their trend became trivial. The mean correlation value of all the significant correlations was 0.85. Only one significant negative correlation was found.
We observe around 17-18 static sensors that are close to our trajectories, the exact number depending on the day. However, we could only manage to compare our mobile sensors with the data from 3 static sensors. The reason was because of the constraints imposed to perform a meaningful and reliable comparison between the sensors along with the irregularity of the static sensor recordings. Finally, we show below the proportion of mobile sensors whose values were included in the comparison. This proportion is consistently high making our results even more reliable.

References

1. Stavroulas, Iasonas, Georgios Grivas, Panagiotis Michalopoulos, Eleni Liakakou, Aikaterini Bougiatioti, Panayiotis Kalkavouras, Kyriaki M. Fameli, Nikolaos Hatzianastassiou, Nikolaos Mihalopoulos, and Evangelos Gerasopoulos. 2020. "Field Evaluation of Low-Cost PM Sensors (Purple Air PA-II) Under Variable Urban Air Quality Conditions, in Greece" Atmosphere 11, no. 9: 926. https://doi.org/10.3390/atmos11090926

2. Bulot, F.M.J., Johnston, S.J., Basford, P.J. et al.Long-term field comparison of multiple low-cost particulate matter sensors in an outdoor urban environment.Sci Rep 9, 7497(2019). https://doi.org/10.1038/s41598-019-43716-3