Global reactive gases forecasts and reanalysis in the MACC project

Abstract

The EU FP7 projects MACC (Monitoring Atmospheric Composition and Climate, 2009–2011) prepared for the operational Global Monitoring for Environment and Security (GMES) atmospheric core service on greenhouse gases, reactive gases and aerosols which is envisaged to start in 2014. This paper describes the data assimilation and modelling system which has been implemented for global monitoring of reactive gases in the troposphere and stratosphere. The MACC reactive gases system uses a coupling software to integrate the Model for Ozone And Related Tracers, version 3 (MOZART-3) chemistry transport model with the Integrated Forecast System (IFS) of the European Centre for Medium-range Weather Forecasts (ECMWF). The focus is placed on the tropospheric simulations with this MACC-IFS-MOZ model. The MACC reanalysis (2003–2010) and the forecasts performed in near real time (NRT) benefit from the multi-sensor approach for data assimilation of total columns, tropospheric columns and vertically resolved observations of ozone, CO and NO₂. Daily biomass burning emissions are integrated in real time using the global fire assimilation system (GFAS) that was developed within MACC. Other emissions are taken from a state-of-the-art global inventory that was developed across several EU projects. The MACC reanalysis and tracer forecasts are routinely evaluated with ground-based and airborne in-situ observations and independent satellite retrieval products. We present the system set-up for reactive gases, give an overview on the service and technical developments during the project, and indicate how MACC global reactive gases products could provide information on non-CO₂ greenhouse gases.

Keywords:

• GMES
• reactive gases
• monitoring
• forecast
• air quality
• emissions

1. Introduction

While tropospheric ozone is a direct greenhouse gas, other short-lived atmospheric trace gases like carbon monoxide, volatile organic compounds and nitrogen oxides act as indirect greenhouse gases by controlling ozone and the abundance of the OH radical, which itself affects also the atmospheric burden of methane (e.g. Prather and Ehhalt 2001; Rigby et al. 2008; Montzka et al. 2011). In order to better understand this complex chemical interplay, it is essential to monitor the key reactive gas species on a global scale.

Under the framework of the European initiative Global Monitoring for Environment and Security (GMES), an operational atmospheric monitoring and forecast service will be established which is envisaged to start in 2014. The EU research projects GEMS (Global and regional Earth-system Monitoring using Satellite and in-situ data, 2005–2009; Hollingsworth et al. 2008), MACC (Monitoring Atmospheric Composition and Climate, 2009–2011) and MACC-II (2011–2014) have been designed to prepare for this novel and comprehensive service which will comprise monitoring of greenhouse gases as well as reanalysis and forecasts of aerosols and reactive gases, both on the European and on the global scale. The global and regional service lines provided from the MACC project are in a pre-operational state and already provide valuable services to various users. MACC products featured prominently during the eruption of the Eyjafjallajökull volcano in 2010 and during the arctic ozone hole in the winter of 2011. They will emerge into a fully operational GMES atmospheric service for monitoring and forecast of global and regional air quality, aerosols and greenhouse gases during the coming years.

The overall structure of the MACC project is shown in Figure 1. Satellite and in-situ data from various agencies and institutions as well as other research projects and global networks are collected and processed to be fed into a comprehensive data assimilation and modelling framework consisting of the Integrated Forecast System (IFS) at the European Centre for Medium-range Weather Forecast (ECMWF) and various chemistry transport models. Special emphasis is put on the development of an up-to-date inventory of emissions including the near real time (NRT) monitoring and processing of satellite data on open burning of biomass. The main service lines on the global scale constitute daily NRT forecasts (up to 96 h into the future) for stratospheric ozone and a variety of tropospheric trace gases, greenhouse gases and aerosols, and a comprehensive reanalysis covering the period 2003–2010. In the following, we will describe the pre-operational MACC services for tropospheric reactive gases with emphasis on the model system using the Model for Ozone And Related Tracers, version 3 (MOZART-3) chemistry transport model. Further information about the project, data products and links to external validation activities can be found at the MACC website http://www.gmes-atmosphere.eu

Figure 1. MACC project structure and data flow.

2. Modelling and assimilation approach

The meteorological forecast and data assimilation system IFS (Integrated Forecast System, http://www.ecmwf.int/research/ifsdocs) at the ECMWF has been coupled to an updated version of the global chemistry transport model MOZART-3 (Kinnison et al. 2007) in order to build the coupled MACC system MACC-IFS-MOZ (Flemming et al. 2009). For a coupled simulation, both models are running in parallel and exchange several two- and three-dimensional fields every hour using the OASIS4 coupling software developed in the PRISM project (Valcke and Redler 2006). Data assimilation of the MACC species O₃, CO, NO_x, HCHO and SO₂ takes place in IFS, which provides basic meteorological data and species initial concentrations to MOZART. Chemical reaction equations, emissions and deposition are calculated in MOZART. Transport processes are parameterised in both models. MOZART provides updated tendency terms for chemistry, emission and deposition sources and sinks for the MACC species, which are then used to constrain the data assimilation in the next cycle.

MOZART uses the same 60 vertical hybrid layers as in current IFS simulations reaching from the surface to 0.1 hPa. The MOZART system of chemical reactions consists of 115 species, 71 photolysis reactions, 223 gas phase reactions and 21 heterogeneous reactions.

Both the global reanalysis and the NRT forecasts benefit from the multi-sensor approach for data assimilation: total columns, tropospheric columns and vertically resolved observations from different satellite sensors and platforms are assimilated and are pivotal to improve the simulations by removing biases; for example, with respect to polar stratospheric ozone chemistry or during exceptional pollution episodes. Integrated Forecast System (IFS) uses an incremental formulation of 4-dimensional variational (4D-Var) data assimilation. In 4D-Var, a cost function J is minimised to combine the model background and the observations to obtain the best possible forecast by adjusting the initial conditions (Courtier et al. 1994). The background error statistics for ozone currently used in the IFS were also used for the assimilation of the MACC ozone as part of the coupled system. The background error statistics for the other species were calculated with the National Meteorological Center (NMC) method (Parrish and Derber 1992). The assimilation experiments in MACC apply averaging kernels for CO and NO₂ observations and use the variational bias correction scheme for O₃ and CO observations (Dee 2004; McNally et al. 2006; Auligné et al. 2007). For a far more detailed description of the ECMWF data assimilation approach for trace gases the reader is referred to Inness et al. (2012). The satellite data assimilated in the forecasts and in the reanalysis are summarised in Tables 1 and 2, respectively.

Table 1. Reactive gases data assimilation in the MACC NRT analysis. Data labelled active are assimilated in the analysis. Data labelled passive are only monitored, which means that the observations are not fed back into the model system as part of the assimilation. This allows for detection of potential satellite instrument biases or drifts without influencing the quality of the model analysis.

Instrument	Satellite	Provider	Version	Species	Type	Period	Status
MLS	AURA	NASA	V02	O3	Profiles	20090901 -	Active
OMI	AURA	NASA	V883	O3	Total column	20090901 -	Active
SBUV	NOAA	NOAA	V8	O3	Six layer profiles	20090901 -	Active
SCIAMACHY	Envisat	KNMI		O3	Total column	20090916 -	Active
GOME-2	MetOp-A	Eumetsat/DLR	V003	O3	Total column	20090901 -	Passive
SEVIRI	MSG	Eumetsat		O3	Total column	20090901 -	Passive
IASI	MetOp-A	LATMOS		CO	Total column	20090901 -	Active
MOPITT	TERRA	NCAR	V4	CO	Total column	20100330 -	Passive
GOME-2	MetOp-A	Eumetsat/DLR	V003	NO2	Tropospheric column	20100714 -	Passive
OMI	AURA	KNMI	V883	NO2	Tropospheric column	20090913 -	Passive
SCIAMACHY	Envisat	BIRA		SO2	Total column	20090901 -	Passive

3. Global forecasts

Global 96-h daily forecasts for concentrations of ozone and several other reactive gases together with meteorological parameters are generated with MACC-IFS-MOZ using a horizontal resolution of 1.125° for IFS and 1.875° for the chemistry transport model. Currently, these forecasts are running with a time lag of one day to the operational weather forecast simulations. The daily forecasts started in September 2009 and have run with almost no interruption since then. For graphical forecast products, see http://www.gmes-atmosphere.eu/services/gac/nrt

Validation of O₃, CO and NO₂ is carried out regularly using various independent in-situ and satellite observational data sets (see e.g. Elguindi et al. 2010; Flemming et al. 2011). On a dedicated verification web page, the daily simulation results are evaluated with surface station, aircraft and independent satellite data in NRT. Since the start of the MACC-II successor project, a detailed validation report is produced every three months. Observational data that are available in NRT are sparse and currently consist of surface observations for ozone and CO from the global atmospheric watch (GAW) programme as well as NO₂ tropospheric columns from SCIAMACHY. Figure 2 shows a validation example of CO concentrations from the NRT MACC-IFS-MOZ forecasts with observations from the Hohenpeissenberg observatory, Germany. The model captures most of the synoptic features that are present in the measured data. In most cases, the 24-h forecast (black line) is superior to the longer-range forecasts (coloured lines). Altogether, 12 GAW stations provided data to MACC with much faster turnaround time than usual (between one day and three months). Beginning in 2012, NRT validation is also provided for vertical profiles of ozone and carbon monoxide from ascending and descending In-service Aircraft for a Global Observing System (IAGOS) aircraft ( http://www.iagos.fr/macc/).

Figure 2. Near real time validation with CO data from GAW station Hohenpeißenberg, Germany for August 2011. OBS: Observational data; black line: NRT analysis; coloured lines depict the concentrations forecasted from simulations started one, two and three days before the analysis. Please note that station data are of preliminary status as they have not been quality checked extensively.

While CO is generally reproduced well during the summer the MACC forecasts underestimate observed CO concentrations for European and Asian GAW stations in winter with biases of up to −50%, consistent with IAGOS CO profile validation (not shown). This is a feature that is found in many current chemistry transport models. It is presently unclear if this low bias results from underestimated anthropogenic emissions of CO or volatile organic compounds or is due to other factors associated with biomass burning or natural emissions. For ozone, the model results are mostly in good agreement with measured surface concentrations for European and Asian GAW stations in spring, with relative biases of around 10% and a mean RMSE of less than 10 ppb. Only for the northern summer period, O₃ concentrations are predominantly overestimated by all model runs with mean relative biases of up to 30%. Tropospheric NO₂ columns agree well with SCIAMACHY retrievals, except for an underestimation over East Asia during autumn and winter, which is most likely originating from the use of too low emissions for China.

In order to ease access of the daily MACC reactive gases (and aerosol) data products for regional modelling teams worldwide, a fully interoperable, Open Geospatial Consortium-compliant data server (“MACC boundary condition service”) has been established at Forschungszentrum Jülich. A graphical web interface ( http://macc.icg.kfa-juelich.de:50080/) allows users to select the data sets, region, time period and variables they require and then provides data as NetCDF files for download or allows for interactive visualisation. Due to the concept of interoperability, the MACC data can be directly compared to other data available on web coverage servers anywhere in the world. The web interface is still in construction. Presently, the model results can be compared to surface observations from more than 400 stations of the German Umweltbundesamt (UBA), which delivers air quality data in NRT. An example is shown in Figure 3, where MACC 96-h ozone forecasts for four consecutive days in April 2012 are compared to the observations at the station Hamburg Flughafen Nord. The general level of compliance is mostly dependent on the meteorological situation and on the representativeness of the station.

Figure 3. Screenshot from the graphical web interface of the interoperable MACC boundary condition server. Comparison of MACC 24-h ozone forecasts to UBA observational data for station Hamburg Airport Nord, Germany for four consecutive days in April 2012.

4. Reanalysis

A retrospective reanalysis for the years 2003–2010 was carried out in MACC using a similar configuration as for the NRT forecasts but with enhanced horizontal resolution of 0.7° (IFS) and 1.125° (MOZART). This reanalysis assimilates more satellite data than the NRT run, because it allows for inclusion of data sets not delivered in NRT (Table 2). In contrast to the NRT analysis, SCIAMACHY NO₂ tropospheric columns are also actively assimilated. Graphical reanalysis products are available at http://www.gmes-atmosphere.eu/d/services/gac/reanalysis/macc, and monthly mean data can be downloaded from the Jülich boundary condition server. Verification is done against surface station data from global networks (EMEP, GAW, NOAA GMDL), vertical profiles from ascents/descents of MOZAIC aircraft at international airports (Elguindi et al. 2010), ozone sondes and independent satellite data. Figure 4 shows as example how ozone from the MACC reanalysis compares to MOZAIC profile observations over Frankfurt for the period January 2003–April 2005. For this period, the tropospheric bias is reduced by about 25% compared to the previous reanalysis from the GEMS project, which is mostly due to the improvements in the data assimilation system and the increased amount of vertically resolved satellite data used in MACC. The remaining tropospheric biases are found in the surface boundary layer and around the tropopause, where vertical ozone gradients are large and are not always well captured by both the Chemistry Transport Model (CTM) and the satellite observations. The MACC reanalysis and its validation with independent observational data are described in detail in Inness et al. (2012).

Figure 4. Comparison of ozone from MACC and GEMS reanalysis to MOZAIC profiles over Frankfurt for the period January 2003–April 2005. Top: monthly mean reanalysis profiles. Bottom: relative bias (observations – model).

Table 2. Reactive gases data assimilation in the MACC reanalysis.

Instrument	Satellite	Provider	Version	Species	Type	Period	Status
GOME	ERS-2	RAL		O3	Profiles	20030101–20030531	Active
MIPAS	Envisat	ESA		O3	Profiles	20030127–20040326	Active
MLS	AURA	NASA	V02	O3	Profiles	20040808–20101231	Active
OMI	AURA	NASA	V003	O3	Total column	20041001–20101231	Active
SBUV	NOAA-16	NOAA	V8	O3	Six layer profiles	20040101–20101231	Active
SBUV	NOAA-17	NOAA	V8	O3	Six layer profiles	20030101–20101231	Active
SBUV	NOAA-18	NOAA	V8	O3	Six layer profiles	20050604–20101231	Active
SCIAMACHY	Envisat	KNMI		O3	Total column	20030101–20101231	Active
MOPITT	TERRA	NCAR	V4	CO	Total column	20030101–20101231	Active
IASI	METOP-A	LATMOS/ULB		CO	Total column	20080401-20101231	Active
OMI	AURA	KNMI	Col. 3	NO2	Tropospheric column	20041001–20101231	Passive
SCIAMACHY	Envisat	KNMI	V1.1	NO2	Tropospheric column	20030101–20101231	Active
OMI	AURA	NASA	V003	SO2	Total column	20040817–20101231	Passive
SCIAMACHY	Envisat	BIRA		SO2	Total column	20040104–20101231	Passive
OMI	AURA	NASA	V003	HCHO	Total column	20040827–20101231	Passive
SCIAMACHY	Envisat	BIRA	V2	HCHO	Total column	20030101–20101231	Passive

5. Surface boundary conditions

The accuracy of the IFS-MOZART simulations depends on the quality of the surface boundary conditions, which control the tracer surface fluxes, notably emission and dry deposition fluxes as the major source and sink terms. With the MACC reanalysis, an updated anthropogenic and natural emission inventory has been introduced. Based upon a review of existing historical and actual emission inventories, a consistent inventory for anthropogenic emissions has been set up for the base year 2000 (Lamarque et al. 2010; Granier et al. 2011). It was originally designed for the IPCC AR5 assessment (ACCMIP emissions). These emissions are extrapolated for years after 2000 with the Representative Concentration Pathway RCP8.5 scenario (Moss et al. 2010; van Vuuren et al. 2011) and are called MACCity emission inventory. Biogenic emissions are from MEGAN-v2 (Guenther et al. 2006) and other natural emissions are from the POET project (Granier et al. 2005) and the Global Emissions Initiative (GEIA).

For the biomass burning emission inventory, a novel method has been developed using a bottom-up approach: within MACC, a global fire assimilation system (GFAS) for wildfires observed from space has been developed (Kaiser et al. 2012). Global daily fields of observed fire radiative power (FRP) based on MODIS observations (Figure 5) are converted into fire emissions, which are available in NRT and as a retrospective data set since 2003. The two consistent products are available with horizontal resolution of 0.5° lat/lon (GFASv1.0) and 0.1° lat/lon (GFASv1.1). They represent for the first time the large temporal variability of biomass burning emissions. A global wildfire data description and graphical products can be found at http://www.gmes-atmosphere.eu/services/gac/fire, the wildfire emission data GFASv1.0 can be downloaded from the Jülich boundary condition server. They are used for the years 2009–2010 of the reanalysis and will soon also be used for the global NRT forecasts of MACC. For the earlier years of the reanalysis, a combination of GFAS FRP and a preliminary version of Global Fire Emission Database, version 3 (GFEDv3) emissions (van der Werf et al. 2010) is being used.

Figure 5. MACC daily fire product for 30 August 2011. Emissions fluxes are derived from these fire radiative power observations.

6. Use of MACC reactive gases products to assess concentrations and trends of non-CO₂ greenhouse gases

Short-lived atmospheric gases are closely linked to the overall abundance and the variablility of non-CO₂ greenhouse gases, most notably via the hydroxyl radical. Reaction with OH acts as the major sink for methane and other greenhouse gases. On the other hand, the atmospheric lifetime of methane is a measure for the tropospheric burden of OH and thus for the oxidising capacity of the atmosphere. Interestingly, a large uncertainty in the estimate of this important parameter still exists between global chemistry models (Shindell et al. 2006; Duncan et al. 2007, Hoor et al. 2009). Moreover, the role of OH for the interannual variablility of methane is not completely understood yet (Rigby et al. 2008; Dlugokencky et al. 2009; Montzka et al. 2011). As a consequence, there is now rising interest from the greenhouse gas community in the OH concentrations derived from the MACC reanalysis and its interannual variability. The mean chemical methane lifetime in the reanalysis is 10.0 years which is close to the IPCC TAR recommendation of 9.6±1.4 years (Prather and Ehhalt 2001). However, the varying amount of trace gas data assimilated during the reanalysis period induces spurious variability in the data record which limits the usefulness of these data to assess the evolution of the year-to-year variability of the methane lifetime.

As an alternative product, MACC provides OH concentrations from a pure MOZART simulation run in parallel to the reanalysis. This simulation has the same set-up for meteorology and emissions and uses the high CTM resolution as in the reanalysis but does not assimilate any reactive gases. A preliminary analysis shows that the absolute methane lifetime (10.8 year) is somewhat higher than in the reanalysis but the interannual variability 2003–2010 is well captured by this simulation when compared to the study of Montzka et al. (2011). The influence of the potentially low biased wintertime CO emissions in the current inventories on the OH concentrations need to be further evaluated but first results from sensitivity simulations with increased CO emissions show only small reductions in OH, resulting in an increase of methane lifetime of less than 0.2 years.

7. Summary and outlook

The data assimilation and modelling of global reactive gases in the MACC project has achieved a mature state which is ready to be transformed into a fully operational service under the European GMES initiative. This paper provides an overview about the main product lines and services of the reactive gases component in MACC including the efforts to validate such products. The focus here has been placed on the pre-operational MACC-IFS-MOZ system, but we would like to point out that a second forecast stream with a different model set-up – using the TM5 chemistry transport model (Huijnen et al. 2010) instead of MOZART – was run in a research mode almost throughout the entire project duration (Huijnen et al. 2011a). Furthermore, the reactive gases team started to integrate the codes of the chemistry modules of MOZART, TM5 and the French MOCAGE model (Peuch et al. 1999) directly into the weather model IFS in order to avoid inconsistencies that are due to the duplication of transport processes in IFS and the chemistry transport model occurring in the present set-up that uses a coupling software (Huijnen et al. 2011b). This C-IFS model also makes more efficient use of computing resources and will facilitate the coupling of reactive gases, aerosol and greenhouse gases processes, which will improve the overall consistency of the MACC data assimilation and forecasting system.

The main output from the reactive gases project in MACC are daily NRT forecasts and a comprehensive, validated reanalysis for the period 2003–2010. Ozone, carbon monoxide and to some extent also NO_x in these runs are constrained by 4D-Var data assimilation. In the context of non-CO₂ greenhouse gas emission and concentration trends, the MACC reanalysis is only of limited value, because it spans a relatively short period and suffers from some inconsistencies in the time series due to changes in the assimilated observations. The project can offer results from a standalone MOZART simulation, which appears to capture the interannual variability of the methane lifetime rather well. Data sets and graphical products from the reanalysis and the NRT forecasts are made available through the MACC project web pages and via a fully interoperable data service with interactive plotting capabilities at the Forschungszentrum Jülich. Validation of the MACC-IFS-MOZ system has been performed using various ground-based, aircraft and satellite data sets. The validation of the NRT system remains challenging, because only very few data that are suitable for evaluating model systems with grid resolutions of several 10 kms are transmitted in NRT, and such data cannot often be freely accessed.

Due to the relatively poor information content that current satellite data products can provide on tropospheric reactive gases concentrations and their vertical structure, the quality of the MACC system is also strongly dependant on the availability of accurate emissions inventories. The MACC project contributed to the development of new inventories on emissions of short-lived gases from anthropogenic activities and developed a new system to monitor emissions from open burning of biomass (GFAS). The interplay between data assimilation, emissions and the model system remains an active area of research in the successor project MACC-II.

Acknowledgements

We acknowledge all colleagues from the MACC project for their contributions to building, running and validating the MACC data assimilation and modelling system. This study was funded by the European Commission under the framework programme 7 (contract number 218793). ECMWF and Forschungszentrum Jülich are gratefully acknowledged for the provision of computer hardware and computer time. The ground-based data presented in this paper were obtained from the Global Atmospheric Watch (GAW) network and from Umweltbundesamt (UBA), Germany. Aircraft data are from the European MOZAIC project.