ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
A low-cost sensor network to monitor the CO2 emissions
of the city of Zurich
Dominik Brunner1, Michael Mueller1, Michael Jaehn1, Peter Graf1,
Jonas Meyer2, Christoph Hueglin1, Anastasia Pentina3,
Fernando Perez-Cruz3 and Lukas Emmenegger1
1Empa, Dübendorf, Switzerland
2Decentlab GmbH, Dübendorf, Switzerland
3Swiss Data Science Center, ETH Zurich, Switzerland
© Jörg Sintermann
2ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
Project Carbosense4D (www.carbosense.ch)
Dense CO2 sensor network Atmospheric modelling
Machine Learning
Partners
3ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
Project Carbosense4D (www.carbosense.ch)
Goals:
 Determine CO2 emissions of the city of Zurich over multiple years
 Enhance understanding of biospheric CO2 fluxes over Switzerland
 Describe accurately the 4-D evolution of CO2 over Switzerland
 Learn about sensor networks
 sensor integration and characterization
 network setup and operation
 communication and data processing
 Quality Assurance & Quality Control
low-cost
sensors
low-cost
network=
4ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
Carbosense network, status 27 Aug 2018
MeteoSwiss Zürich
NABEL
Picarro CRDS
(7 units)
SenseAir HPP
(20 units)
SenseAir LP8
(300 units)
207 sensors
deployed
Zurich
5ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
Sensor calibration and characterization
 CO2 sensors respond to temperature, pressure, humidity
 Multi-factor model using Beer-Lambert’s law applied to IR detector output
CO2 (350 – 1000 ppm)
T (-5°C – 50°C)
CO2 (400 – 900 ppm)
p (770 – 1050 hPa), T
Climate chamber Pressure chamber
CO2, T, RH
Ambient measurements
6ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
First year of operation of low-cost LP8 sensors in Zurich
10-17 Jul 2017
 Sensors not stable enough for accurate long-term measurements
 Frequent recalibration too demanding
13-20 Aug 2018
Can we correct for drifts while sensors are in operation?
7ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
Operational sensor data processing
Outlier removal
specifying valid operating range
accounting for sensor drift and ageing
Including
outliers
Outliers
removed
T (°C) T (°C)
-log(IR)
-log(IR)
Sensor drift adjustment
using periods of strong winds
(> 2 m s-1 during at least 1 h)
July 2017 Aug 2018
CO2adjustment[ppm]
8ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
Before drift correction
Sensor drift adjustment
July 2017 Aug 2018
Offset relative to nearby accurate sensor
CO2adjustment[ppm]Operational sensor data processing
Periods of strong winds
After drift correction
9ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
Amplitude of diurnal variation
July 2018
Jan 2018
10ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
CO2 modelling with COSMO-GHG
 Weather prediction model COSMO with GHG tracer extension
 Domain centered over Switzerland, 1 km x 1 km resolution, 60 levels
 Highly efficient code, fully ported to GPUs
CO2-boundary conditions
Global CO2 model CAMS
(ECMWF, experiment ghqy)
Emissions
TNO/MACC-3 (Europe) +
CarboCount (Switzerland)
Biosphere fluxes
VPRM (MPI Jena)
11ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
Animation of anthropogenic CO2 in last week of Oct 2017
12ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
Comparison with observations, Oct 2017
Beromünster, rural, 212 m a.g.l. Dübendorf, suburban, 4 m a.g.l
CO2CO2
anthrop.
biospheric
background
RESP x 4, GPP x 2
T T
Wind speedWind speed
Too strong vertical mixing,
surface not sufficiently
decoupled from higher levels
model
observations
13ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch
Conclusions and outlook
Conclusions
 First year of data from ~200 LP8 sensors, HPP just starting
 Approx. accuracy: Picarro 0.1 ppm, HPP 1.0 ppm, LP8 10 ppm
 Strong contribution from biosphere even in Zurich
 First model results encouraging, areas of improvement identified
Outlook
 Sensors: Deployment of HPPs, further improvement of LP8 data
 Model: Meteo data assimilation, PBL mixing, online emissions & VPRM
 Integration of model and sensor data:
 Use model as transfer standard between Picarro/HPP and LP8
 Geostatistical modeling of differences COSMO-GHG and sensor data
 Further develop city-scale model with final goal of estimating emissions
14ICOS Science Conference 2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch O. Wehrli 2013
With a special thanks to
Swisscom, MeteoSwiss, UGZ Zurich, NABEL and others
for generous support of our sensor network
Markus Leuenberger (University of Bern) for Beromünster CO2 observations
Christoph Gerbig (MPI Jena)
for VPRM data
Copernicus Atmospheric Monitoring Service (CAMS)
for global CO2 model data
Funding through Swiss Data Science Center (SDSC) and EU / Eurostars

A low-cost sensor network to monitor the CO2 emissions of the city of Zurich

  • 1.
    ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch A low-cost sensor network to monitor the CO2 emissions of the city of Zurich Dominik Brunner1, Michael Mueller1, Michael Jaehn1, Peter Graf1, Jonas Meyer2, Christoph Hueglin1, Anastasia Pentina3, Fernando Perez-Cruz3 and Lukas Emmenegger1 1Empa, Dübendorf, Switzerland 2Decentlab GmbH, Dübendorf, Switzerland 3Swiss Data Science Center, ETH Zurich, Switzerland © Jörg Sintermann
  • 2.
    2ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch Project Carbosense4D (www.carbosense.ch) Dense CO2 sensor network Atmospheric modelling Machine Learning Partners
  • 3.
    3ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch Project Carbosense4D (www.carbosense.ch) Goals:  Determine CO2 emissions of the city of Zurich over multiple years  Enhance understanding of biospheric CO2 fluxes over Switzerland  Describe accurately the 4-D evolution of CO2 over Switzerland  Learn about sensor networks  sensor integration and characterization  network setup and operation  communication and data processing  Quality Assurance & Quality Control low-cost sensors low-cost network=
  • 4.
    4ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch Carbosense network, status 27 Aug 2018 MeteoSwiss Zürich NABEL Picarro CRDS (7 units) SenseAir HPP (20 units) SenseAir LP8 (300 units) 207 sensors deployed Zurich
  • 5.
    5ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch Sensor calibration and characterization  CO2 sensors respond to temperature, pressure, humidity  Multi-factor model using Beer-Lambert’s law applied to IR detector output CO2 (350 – 1000 ppm) T (-5°C – 50°C) CO2 (400 – 900 ppm) p (770 – 1050 hPa), T Climate chamber Pressure chamber CO2, T, RH Ambient measurements
  • 6.
    6ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch First year of operation of low-cost LP8 sensors in Zurich 10-17 Jul 2017  Sensors not stable enough for accurate long-term measurements  Frequent recalibration too demanding 13-20 Aug 2018 Can we correct for drifts while sensors are in operation?
  • 7.
    7ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch Operational sensor data processing Outlier removal specifying valid operating range accounting for sensor drift and ageing Including outliers Outliers removed T (°C) T (°C) -log(IR) -log(IR) Sensor drift adjustment using periods of strong winds (> 2 m s-1 during at least 1 h) July 2017 Aug 2018 CO2adjustment[ppm]
  • 8.
    8ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch Before drift correction Sensor drift adjustment July 2017 Aug 2018 Offset relative to nearby accurate sensor CO2adjustment[ppm]Operational sensor data processing Periods of strong winds After drift correction
  • 9.
    9ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch Amplitude of diurnal variation July 2018 Jan 2018
  • 10.
    10ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch CO2 modelling with COSMO-GHG  Weather prediction model COSMO with GHG tracer extension  Domain centered over Switzerland, 1 km x 1 km resolution, 60 levels  Highly efficient code, fully ported to GPUs CO2-boundary conditions Global CO2 model CAMS (ECMWF, experiment ghqy) Emissions TNO/MACC-3 (Europe) + CarboCount (Switzerland) Biosphere fluxes VPRM (MPI Jena)
  • 11.
    11ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch Animation of anthropogenic CO2 in last week of Oct 2017
  • 12.
    12ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch Comparison with observations, Oct 2017 Beromünster, rural, 212 m a.g.l. Dübendorf, suburban, 4 m a.g.l CO2CO2 anthrop. biospheric background RESP x 4, GPP x 2 T T Wind speedWind speed Too strong vertical mixing, surface not sufficiently decoupled from higher levels model observations
  • 13.
    13ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch Conclusions and outlook Conclusions  First year of data from ~200 LP8 sensors, HPP just starting  Approx. accuracy: Picarro 0.1 ppm, HPP 1.0 ppm, LP8 10 ppm  Strong contribution from biosphere even in Zurich  First model results encouraging, areas of improvement identified Outlook  Sensors: Deployment of HPPs, further improvement of LP8 data  Model: Meteo data assimilation, PBL mixing, online emissions & VPRM  Integration of model and sensor data:  Use model as transfer standard between Picarro/HPP and LP8  Geostatistical modeling of differences COSMO-GHG and sensor data  Further develop city-scale model with final goal of estimating emissions
  • 14.
    14ICOS Science Conference2018, Prague, 11-13 Sep 2018 | dominik.brunner@empa.ch O. Wehrli 2013 With a special thanks to Swisscom, MeteoSwiss, UGZ Zurich, NABEL and others for generous support of our sensor network Markus Leuenberger (University of Bern) for Beromünster CO2 observations Christoph Gerbig (MPI Jena) for VPRM data Copernicus Atmospheric Monitoring Service (CAMS) for global CO2 model data Funding through Swiss Data Science Center (SDSC) and EU / Eurostars