REGIONAL AIR QUALITY FORECASTING AT THE (MIDI-PYRENEESAIR QUALITY OBSERVATORY) 2nd CHIMERE workshop LMD Palaiseau Adrien Royer - Vincent Crassier
air pollution monitoring stations 19 permanent stations (without industrial) 1 Carte
2 Regional forecast systems PREVAIR by SYRSO + PREVAIR gives an additional forecasting for the french territory
3 Why forecasting / Context To cover territories where there is no monitoring. To inform population if a pollution event occcured or a pollution event is coming with a good probability (forecasting is now a regulatory element for trigger level ). To create communication supports to explain and illustrate phenomenons with medias and public To prospect territories and look for places where to perform mobile campaign in order to get better knowledge.
4 System Automatic tasks with crontab on Ubuntu 64 bits (start at 22h UTC with 18h GFS data) Shell and Python scripts to follow steps and for the post-processing Now, Virtualization with VirtualBox to put a forecasting system quickly on a rescue computer with any OS. Architecture WRF Version 3.4 (april 2012) NCAR Resolution : 3 km CHIMERE Version 2011a (sept. 2011) LMD Resolution : 4 km Post-processing duration : 2h00 duration : 5h00 duration : 0h30
5 1st Step : Weather Forecasting with WRF 2 nested domains with GFS Input data Update : static geographical data in WPS: NVDI (with A. Rosso AirParif) Corine Land Cover Aster DEM for elevation To increase accuracy, Météo-France s data assimilation : For D-1: Météo-France Radome Data For D-Day and D+1: Arome/Arpege Forecasting with virtual stations
6 2nd Step : Regional Emissions Inventory to Chimere Convert total annual emissions (once for every update of the emissions inventory) from spot, linear and surface emissions sources to emissions grid. Convert the emissions grids to AEMISSION.nc (currently in test, daily) Consider holidays, public holidays For SNAP 2 : emissions from heating Evaluate from difference between weather (Windchill calculate from WRF forecasting) and a set point of heating temperature for building. For SNAP 7 : emissions from traffic Evaluate from monthly cars/trucks count and daily profiles for urban and rural areas. Spatial distribution over region (to split touristic area and jobs area). Keep temporal keys from Chimere emi2011 for the others emissions sources
7 3rd Step : Air Pollution Forecasting with CHIMERE 2 nested domains 4km & 20km 5 hours with 8 cpu (3GHz) (test with 24 cpu and future test on Calmip supercomputer, Toulouse P. Sabatier University) Default parameters (chimere.par Heat wave test case)
7 3rd Step : Air Pollution Forecasting with CHIMERE 2 nested domains 4km & 20km 5 hours with 8 cpu (3GHz) (test with 24 cpu and future test on Calmip supercomputer, Toulouse P. Sabatier University) Default parameters (chimere.par Heat wave test case)
8 4th Step : Statistical adjustment Look for predictor variables with ACP and stepwise regression (student in statistics M. Zhang) Calculate a multivariate statistical model from predictor variables O3 model with 2 years time series Example of graphic for statistical model quality control : (daily made for all PM stations) PM10 model: everyday, moving model with 1 month of data
10 4th Step : Statistical adjustment Look for predictor variables with ACP and stepwise regression (student in statistics M. Zhang) Calculate a multivariate statistical model from predictor variables O3 model with 2 years time series => Pb with peaks prediction need to improve method with long-term time series and use selective data sampling PM10 model: everyday, moving model with 1 month of data => Good results Spatial distribution with Inverse Distance Weighted Currently, kriging with external drift (pb a few number of stations don t allow to obtain a good result => create virtual stations) Future work and further tests : Using the mobile campaigns to reconstruct time series for statistical model, Seeking for predictor variables for co-kriging, Virtual stations for kriging, Peaks model for O3 Kalman filter (Try to adjust emissions with historical data series (instead of inverse modeling))
9 4th Step : Statistical adjustment
10 4th Step : Statistical adjustment 80 70 Scores PM10 in 2013 for Toulouse Urban Station (Berthelot) Measures Chimere without adjustment 60 50 40 30 20 10 0 01/01/2013 20/02/2013 11/04/2013 31/05/2013 20/07/2013 08/09/2013 28/10/2013 Scores with statistical adjustment for 2012
11 5th Step : Post-processing Compute daily statistics : daily max, daily mean, 24h moving average, Estimate population exposure and area exceeding regulatory levels. Create maps for region and departments for daily statistics and hourly data. Scores to compare forecasting with measurements. ATMO index for the 3020 towns of Midi-Pyrénées region Export to web server powered with Python and a lot of libraries (matplotlib, numpy, PyNGL, netcdf4, wxpython, pygrib, shapely, simplekml, osgeo, gdal, rpy, MySQLdb, cx_oracle,...)
12 Estimate population exposure + area exceeding regulatory limit Compute a shape area from PM10, O3, NO2 regulatory pollutants for D, D+1, D+2 If a level is exceeded : Calculate both area and population (intersect department boundaries and population in buildings). send an email with a picture (map) to mailing list make GIS MapInfo and KMZ files
12 Estimate population exposure + area exceeding regulatory limit Compute a shape area from PM10, O3, NO2 regulatory pollutants for D, D+1, D+2 If a level is exceeded : Calculate both area and population (intersect department boundaries and population in buildings). send an email with a picture (map) to mailing list make GIS MapInfo and KMZ files
13 Convert Chimere forecasting to ATMO index Intersect concentrations levels with the 3020 town s boundaries to get ATMO indexes (national scale) Export a csv file to our website Make maps and KMZ file.
14 Examples of automaticaly created maps/graphics Daily max for NO2 12/11/2013
14 Examples of automaticaly created maps/graphics Daily max for NO2 12/11/2013
14 Examples of automaticaly created maps/graphics Daily max for NO2 12/11/2013
15 GUI to manually extract data from netcdf files To easily implement : storage storage extracting maps on demand scores on demand customise display time saving
15 GUI to manually extract data from netcdf files To easily implement : storage storage extracting maps on demand scores on demand customise display time saving
15 GUI to manually extract data from netcdf files To easily implement : storage storage extracting maps on demand scores on demand customise display time saving
16 Feedback Problem with low O3 peak concentrations during summer (source: WRF mixing, PBL, temperature, anthropogenic or biogenic VOC emissions?) Sometimes, wind power affects PM concentrations too much. Switch species unit gives rise to a problem with concentrations forecasts (ppb => μg/m 3 )
17 Feedback Wind effect on PM10 forecasting - october 3rd, 2013 without wind (Vent d Autan) with wind (Vent d Autan)
2nd CHIMERE workshop 13/11/2013 17 Feedback Wind effect on PM10 forecasting without wind (Vent d Autan) - october 3rd, 2013 with wind (Vent d Autan)
18 Feedback Wind effect on PM10 forecasting Graph october 3rd, 2013 50 45 Toulouse Measure / Chimere / WRF BRTLOT Chimere BRTLOT Mesures 90 BRTLOT WRF Wind speed (m/s) 80 Concentration (µg/m 3 ) 40 35 30 25 Graph biais mesures modele en moyenne journaliere. 20 15 10 5 70 60 50 40 30 20 10 Wind speed (km/h) 0 0 30/09/2013 00:00 01/10/2013 00:00 02/10/2013 00:00 03/10/2013 00:00 04/10/2013 00:00 05/10/2013 00:00 06/10/2013 00:00 Need to test in switching off wind erosion and resuspension for dusts
19 Extra model animation for communication