Calibr alibration ion and and Data a Fus Fusion ion of of PM2.5/ 2.5/PM10 10 sens ensor ors in in the he Net Nether herlands lands Joos oost Wes esseling eling
Us Using ing AQ Q sens ensor ors • What are we using the AQ sensors for? • When used in well organized projects, run by professionals, using well known hardware, QA/ QC and calibration is usually addressed. • When you have hundreds of sensors producing data every 5 minutes, some well managed, some not, all kinds of housings at different kinds of locations, QA/QC and calibration can be a challenge. Here: on-the-fly “fusion” calibration of up to 750 PM sensors in the Netherlands of the type Nova SDS-011. RIVM Sensors | FAIRMODE | Oct 2019 2
Calibr alibration: ion: humidit humidity All low-cost optical PM sensors are • very sensitive to relative humidity. The effects of humidity are quite • similar for PM10 and PM2.5. There is a large variation in the • effect of the humidity à more issues, effect location, housing, ... Fit to the data: RIVM Sensors | FAIRMODE | Oct 2019 3
Calibr alibration: ion: Fus Fusion ion In tests with groups of measurements in a number of • Dutch cities, we observed that the differences between hourly results of sensors in small areas are quite small, and the average PM gradients are small. RIVM Sensors | FAIRMODE | Oct 2019 4
Calibr alibration: ion: Fus Fusion ion In tests with groups of measurements in a number of • Dutch cities, we observed that the differences between hourly results of sensors in small areas are quite small, and the average PM gradients are small. Option: compare the average of all sensors located in a • short distance around an official measuring station to the official hourly results à local correction factor. Interpolate the correction factors over the country. • Combine the calibrated sensor data with existing AQ • maps using the uncertainties as weights for the whole country to provide more details. So, we do not work with individual pre-calibrated • sensors, but with the available overall combined data. RIVM Sensors | FAIRMODE | Oct 2019 5
Res esult ults of of the he fus usion-calibr ion-calibration ion PM2.5 The fusion scheme was applied to all hourly sensor • C avg = 11.8 ug/m3 C avg = 10.8 ug/m3 data (Nova SDS-011) in the period Jan-Aug 2019. Raw Calibrated Compare the results for sensors that are co-located • or are very close (<250 meter) to official stations. For a clean test, the nearest official measurement • was not used in calibrating each sensor. The 8-month average raw and fusion-calibrated • C avg = 10.7 ug/m3 C avg = 10.7 ug/m3 PM2.5 concentrations are quite similar. Small/blue markers represent data from co-located sensors and larger orange markers represent sensors that are up to 250 meter from the official location. RIVM Sensors | FAIRMODE | Oct 2019 6
Res esult ults of of the he fus usion-calibr ion-calibration ion PM2.5 The hourly scatter of the calibrated • sensor data around the nearest official is much smaller than that of the raw data. During hours with high relative • humidity the effects of the calibration are very prominent. The raw data severely over-estimate • the nearby official data. After the calibration, the data scatters • around official data, although with a relatively large uncertainty. Small/blue markers represent data from co-located sensors and larger orange markers represent sensors that are up to 1000 meter from the official location. RIVM Sensors | FAIRMODE | Oct 2019 7
RIVM da data a por portal al Presently up to 750 sensors every hour in the data portal. https://samenmeten.rivm.nl/ Development and Implementation of a Platform for Public Information on Air Quality, 8 RIVM Sensors | FAIRMODE | Oct 2019 Sensor Measurements, and Citizen Science, Atmosphere 2019, 10(8), 445
https://samenmeten.rivm.nl/ 9 RIVM Sensors | FAIRMODE | Oct 2019
Fusion-calibr Fus ion-calibration ion in in pr pract actice ice • RIVM is currently involved in several projects with/for/by citizens. • It is important to present data that are properly calibrated. • One project is near a large steel plant that is quite close to the coast. • In practice the PM2.5/PM10 ratios are more complicated than in an average urban environment. • Using the available local official data in the calibration of the sensors is important. Blue: raw sensor data, Purple: calibrated sensor data, 10 RIVM Sensors | FAIRMODE | Oct 2019 Red: official result (@160 meter)
Data a Fus Fusion ion of of sens ensor ors and and ma maps ps PM2.5 • Look at the data of September ug/m3 26, 2019, 09:00 (UTC). • There were significant effects of the high humidity, especially at the coast and in the middle of the country. • 90% ≤ RH ≤ 97% • After the calibration there is much less variation in sensor Raw sensor data Calibrated sensor data values. Correction field RIVM Sensors | FAIRMODE | Oct 2019 11
Data a Fus Fusion ion of of sens ensor ors and and ma maps ps Official data • The structure of the PM2.5 PM2.5 Official map + ug/m3 concentration map changes uncertainties Calibrated sensors + slightly due to the sensor uncertainties data. • There are more detailed sub structures in the map. Background map Background /calibrated sensor map RIVM Sensors | FAIRMODE | Oct 2019 12
Eas aster er fir ires es Effect of traditional Easter fires in Germany. PM10, from 20 April 23:00 hours 02:00 05:00 2019, 23:00 until 21 April 2019, 14:00, Steps of 3 hours. www.youtube.com/watch?v=dJveRLMRaiA 08:00 hours 11:00 14:00 RIVM Sensors | FAIRMODE | Oct 2019 13
Conc oncluding luding remar emarks ks Data from low-cost PM10/PM2.5 sensors (SDS-011) can be combined with that • of nearby sensors and nearby official measurements to yield a correction field. The calibration algorithm does not need much information about the sensors. • It takes all effects into account, not only humidity. • Tests of the data fusion calibration scheme show it performs at least as good as • using local humidity correction for the sensors. The local correction factors for sensors can be used to calculate a sensor-data • field, including uncertainties, for the whole country. The sensor-data field can be combined with official hourly back ground map in a • data fusion approach, yielding new and more detailed maps. All the information can be used for a QA/QC scheme. • RIVM Sensors | FAIRMODE | Oct 2019 14
Ques Questions ions ? RIVM Sensors | FAIRMODE | Oct 2019 15
Recommend
More recommend
