Regional air quality forecasting over Europe: the MACC-II

Regional air quality forecasting over Europe: the MACC‐II ensemble system
Virginie Marécal
CNRM‐GAME (Météo‐France/CNRS)
M. Plu, Ph. Moinat, L Rouïl,
and the MACC‐II regional consortium (INERIS, MET.NO, FRIUUK, KNMI, TNO, SMHI, FMI, CERFACS, AEMET, AUTH, LISA, CERFACS)
WWOSC, Montreal, 19 August 2014
MACC‐II: Monitoring Atmospheric Composition and Climate‐
Interim Implementation
A step towards Copernicus European Atmospheric Services
ESA
MACC-III
(FP7)
FP6
2005
2009
GMES
(FP7)
2011
2014
(H2020)
2015
Atmospheric
services
2020….
airText
Air quality over Europe:
a bridge between the global scale
and local applications
Regional air quality in MACC‐II : 2 complementary chains of production
Global model products
Regional emissions
Observations
Annual
production of
air quality
reanalyses
Daily production
of air quality
forecasts and
analyses
USERS
Production based on a regional air quality ensemble of 7 models
Current geometry
Assimilation method
Operations
0.1°, L8, top : 500hpa
Optimal Interpolation
run @ INERIS
met.no
0.25x0.125°, L20, top : 100hpa
Variational 3d‐var
run @ met.no
EURAD
15km, L23, top : 100hpa
Variational, 3d‐var
run @ ECMWF
15km, L4, top : 3.5km
Ensemble Kalman Filter
run @ KNMI
0.2°, L40, top : 100hpa
Variational, 3d‐var
run @ SMHI
0.2°, L47, top : 5hpa
Variational, 3d‐var
run @ MF
0.2°, L46/8, top : 100hpa
Variational, 4d‐var
run @ FMI
CHIMERE
INERIS, CNRS
EMEP
FRIUUK
L‐EUROS
TNO, KNMI
MATCH
SMHI
MOCAGE
MF, CERFACS
SILAM
FMI
State-of-the-art models operated by the institutes also in charge of their development
MACC‐II ensemble based on 7 limited‐area Chemistry‐Transport Models (CTMs)
Transformations
Hom./Het. chemistry
Photolysis
Aerosol microphysics
Forcings
Radiation
Wind
Temperature
Humidity
Precipitation
external
Transport
Advection
Turbulent diffusion
Convection
Particle sedimentation
Source
Anthropogenic emissions
Natural emissions
Loss
Wet deposition
Dry deposition
Atmospheric
concentrations
Daily production of forecasts and analyses Input data
Meteorological forcings from ECMWF
Anthropogenic emission
Fire emissions
Boundary conditions for chemistry from global production Surface measurements of AQ monitoring stations from EEA NRT database
(Satellite data)
Local Production
96h‐Forecasts from
individual models
O3, NO2, CO, SO2, PM2.5, PM10
NO, NH3, NMVOC, PAN
Analyses from
individual models
for previous 24h
O3
(CO, NO2, SO2, PM2.5, PM10)
Central production at Météo‐France
96h‐Median
ensemble forecast
O3, NO2, CO, SO2, PM2.5, PM10
NO, NH3, NMVOC, PAN
Median ensemble analyses for previous 24h
O3
Verification
products
Evaluation of the forecasts
Comparison with the European surface air quality stations
Statistical indicators
Median ensemble+observations Seasonal verifications: 6‐monthly reports
Evolution of the performances from one year to another
 Depend on the model developments, the variability of the meteorological
conditions and on the change in emissions
 Important effort on aerosol representation in the models with impact on PM10 performances
PM10 mean bias (g/m3)
Spring 2013
-9
Spring 2014
-4
-10
-5
Forecast time
Forecast time
Specific evaluation over the East and West Mediterranean
area with high resolution forecasts
Example of the forecast with a 5kmx5km resolution
AEMET forecast
MACC‐II ENS forecast
Website
Ensemble
Analysis &
Forecasts
http://macc‐raq.gmes‐
atmosphere.eu
Individual
24H Analysis
Individual
96H Forecasts
NRT
Observations
Verification
Mediterranean
zooms
Why using an ensemble of 7 models for air quality forecasts
over Europe ?
All air quality models have their golden days
… but also their bad days
Ensemble methods give on
average better performances than
individual models
Ensemble methods also give an
estimate of the model uncertainties
How to get optimal benefit of the ensemble of 7 models ?
 The median: an efficient method that rejects outliers
currently used in MACC‐II
 More refined methods (Pagowski et al, 2006 ; Galmarini et al, 2013) take into account the performance of each model Comparison of various methods to combine optimally the MACC‐II regional models for summer 2013 (1)
– ENS : at each gridpoint, median of the 7 models
– RMS : at each gridpoint
• the rmse of every member is computed with regard to the ensemble analysis on the day before
• the ensemble is the sum of the members weighted by the inverse of rmse.
Map of weights
for one day and one
model
Comparison of various methods to combine optimally the MACC‐II regional models for summer 2013 (2)
– KZF (Galmarini et al, 2013) : at observation station
• the time series of each model over the last 3 months is decomposed
into 4 spectral components : long‐term, synoptic, diurnal, intra‐day
• these spectral components are compared to the spectral components of the time series of observations
• the KZF ensemble is the sum of the 4 best spectral components applied to the forecast time series.
Results: mean bias of ozone
Bias is reduced
for RMS and KZF
compared to ENS
0
But bias for ENS ~0
during daytime
Results: root mean square error for ozone
During the night :
lowest RMSE for KZF
During daytime :
ENS still the best
Results: correlation for ozone
During the night :
Best correlation for KZF
During daytime :
ENS still the best,
RMS is close to ENS
 Conclusion of the comparison:
– During daytime (ozone peaks), the median is still the best method : low bias, best rmse and correlation;
– KZF is the best method during the night.
 Perspectives :
– try other methods : remove model bias (depending on hour
of the day);
– test on other species;
– test on upper vertical levels, but how to validate ?
Other application of the ensemble of the 7 modeling systems
Global model products
Regional emissions
Observations
Annual
production of
air quality
reanalyses
Daily production
of air quality
forecasts and
analyses
USERS
Yearly production of air quality re‐analyses for assessment
reports
Input data
Local production
Centralised
Production
(INERIS)
Yearly re‐analyses from individual
models for O3, NO2, PM2.5, PM10
Experiment with
pollen (FMI)
Median ensemble Yearly re‐analyses
O3, NO2, PM2.5, PM10, Pollen re‐analyses from FMI
Meteorological forcings from ECMWF
Anthropogenic emission
Fire emissions
Boundary conditions for chemistry from global production Same surface measurements from AIRBASE
(split in 2 sets for data assimilation and data for validation)
(+ Satellite data)
Verification/validation products and
processes Ozone peak values : number of hours when 120 ug/m3
(hourly average) is exceeded
2011
2010
Health indicator for ozone : SOMO35
2011
2010
Summary and perspectives
 MACC‐II air quality forecast/analysis system is based on an ensemble of 7 state‐of‐the‐art models
 Continous improvements of the system thanks to the upstream research
developed in the 7 institutes in charge of the modelling/assimilation
 Research is also done on ensemble methods for an optimal use of the 7 modeling systems
 The number of forecast and analysis/re‐analysis products increase
according users’needs and the number of users increases too
 Towards a fully operational system to be part of the Copernicus
Atmospheric services