This project aims at utilizing BAM－1020 PM2.5 values to calibrate the PM2.5 sensors. The calibration results were then applied to analyze the measured PM2.5 data of the randomly selected sensors deployed at sixteen counties or cities to evaluate their performance. Finally, the main goal of this project is to promote the application of the smart IoT sensor network through public education and training workshops to disseminate its applicable range and discuss the factors influencing the sensor reading.The results from the project indicated that the relative humidity （RH） and PM2.5 concentration have a significant effect on the PM2.5 sensor readings which increase with increasing RH. Besides, original manufacture’s data always overestimate PM2.5 concentrations. Therefore, calibration of the sensors data is needed before use. An increase in wind speed and internal leakage will decrease the measured values because of the decrease in sampling flowrate. Currently, the flowrate of PM2.5 sensor is too low so that the detected number concentrations are also very low which lead to very small calculated PM2.5 mass concentrations. Taking PMS5003 as an example, calculated PM2.5 mass concentration is only 1／2004 times indicated values.Hourly average PM2.5 sensor data of Plantower PMS5003 （manufacturer calibration） were first calibrated by those of BAM－1020 using linear regression and the obtained R－squared （R＾2） values were 0.77, 0.60, 0.77, 0.76 and 0.69 for Keelung, Taoyuan, Zhongming, Tainan and Pingtung, respectively. To improve the data quality further, non－linear regression based on PM2.5 concentrations and RH values were then used to calibrate the sensor data. R2 values for Keelung, Taoyuan, Zhongming, Tainan, and Pingtung were increased to 0.80, 0.64, 0.81, 0.81 and 0.74, respectively, while MNB values were decreased substantially to ±10％ from ＋27.50~＋54.54％ and MNE values were also decreased substantially to 36％ from 49.46~66.64％. The non－linear regression equations were also used to calibrate the Sensirion SPS30 and Honeywell HPMA115S0 data. After calibration, the MNB and MNE of hourly average PM2.5 concentrations were reduced to ＜±10％ and ＜17％ from－21.41~＋39.45％ and 32－78~42.92％, respectively. On the aspect of 24－hour average PM2.5 concentrations, MNB and MNE were reduced to ＜±13％ and ＜19％, respectively, for Plantower PMS5003 Sensirion SPS30 and Honeywell HPMA115S0 sensors after non－linear regression. Therefore, well calibrated sensors can be used for supplementary monitoring of air quality stations. It was found that the PM2.5 data of PMS5003 had no obvious aging and data quality degradation after 24－month deployment. It was also found that the Plantower PMS5003 flowrate is too low and PM10 data are almost same as PM2.5 data （PM10 data are severely underestimated）. In order to resolve this problem, sensor flowrate was increased to 0.1 L／min to sample more coarse particles （PM10－2.5）. As a result, PM10－2.5／PM10 ratio was increased to 47.8％ as compared to 20.7％ of the original sensor data. PM10 data were found to be more accurate and can be used for PM10 monitoring after theoretical calculation and calibration with the PM10 data of air quality station in Hsinchu city. This project audited the performance of PM2.5 sensors deployed at 16 counties or cities. Until now, the auditing of 905 sensors has been finished accounting for 12.2％ of 7,400 deployed sensors. Sensors are less accurate at low PM2.5 concentrations of ＜15 μg／m3 leading to low quality index. Auditing criteria is based on concentration difference of ±5 to ±8 μg／m3 when PM2.5 concentration less than 15 g／m3； or less than 30％ or 50％ error when PM2.5 concentrations ≥15 μg／m3 at station auditing and field auditing, respectively. The best 72－hour quality index is over 80％ in most counties and cities except Hsinchu county. The quality index based on the whole auditing period （168~192 hours） is worse than that of best 72－hour period. Among these sensors, the Plantower and Sensirion showed the best stable performance while the quality criteria was exceeded by Honeywell and Amphenol sensors when PM2.5 concentrations were ＞30 g／m3, ＞ 20 g／m3 , respectively. ITRI sensors were below the quality criteria in the range of 10－30 g／m3. However, the effect of PM2.5 concentration ranges needs to be studied further since measured concentrations are less than 15 g／m3 during the auditing period for ITRI sensors.There were over 250 people attending five public education and expert training workshops, which have improved the understanding of the deployment status of PM2.5 sensors and the limitation of the utilization of sensor data. It is hoped these workshops can enhance the performance of PM2.5 sensors and increase the confidence of the domestic people on sensor data.

PM2.5 sensor, low cost air sensor, Air quality monitoring, Environmental internet of things, Multiple variable regression