TRANSPHORM. Deliverable 2.4.1, type R

Transcription

1 TRANSPHORM Transport related Air Pollution and Health impacts Integrated Methodologies for Assessing Particulate Matter Collaborative Project, Large-scale Integrating Project SEVENTH FRAMEWORK PROGRAMME ENV Transport related air pollution and health impacts Deliverable 2.4.1, type R Report on the enhanced resolution and sub-grid downscaling of transport emissions Due date of deliverable: project month 24 Actual submission date: project month 28 Start date of project: 1 January 2010 Duration: 48 months Organisation name of lead contractor for this deliverable: Scientist responsible for this deliverable: Met.no Jan Eiof Jonson Revision: [0]

2 Summary Regional scale models that can cover all of Europe do not resolve concentrations in urban areas (< 5 km) where the majority of the population lives and where most pollutant emissions are higher. This is mainly due to computational restraints but also due to the availability of sufficiently high resolution emission and meteorological data. There are basically two methods available for improving regional scale exposure assessment in urban areas. The first is to run both meteorological and air quality models at enhanced resolution, after the development or adaptation of high resolution emission inventories (usually bottom up). This has the constraint that the enhanced model calculations are limited to a smaller region. The other method is to develop sub-grid downscaling methods for adjusting or adapting gridded regional scale concentrations so that the concentration or exposure in the urban areas is improved. A number of these methods for regional downscaling have been investigated within the TRANSPHORM project, using differing approaches, and this deliverable outlines these methodologies. Two cases are described where enhanced resolution modelling is applied to improve the level of spatial detail, using the EMEP and Enviro-HIRLAM models. In addition three different sub-grid downscaling methods, intended to improve the exposure assessment at the urban scale for all of Europe, are described. These involve the redistribution of concentrations based on fine scale emission data, on the statistical covariance of population and on a parameterised model for determining the urban increment. Enhanced resolution modelling Two examples of enhanced resolution modelling are described. The EMEP model has been run with enhanced resolution of 1/8 x 1/16 degrees latitude longitude resolution (~7 km, the native resolution of the emissions) for a full year. For this report the calculations have been made for the year The fine scale model run shows clear improvements in the level of detail. The integration of a multi-scale meteorological-chemical system for transport modelling at DMI is described. DMI is using both on-line and off-line modelling systems. The systems are also designed for use in forecast mode, but can also be used for long term, multi year, model runs. Shipping routes and harbour activities, aviation related emissions and emissions from major roads are implemented in the system. Sub-grid downscaling of regional scale concentrations As a test case for Oslo the regional EMEP 0.2 x 0.2 degrees (~20 km) model output has been downscaled for the city of Oslo into a 0.05 x 0.05 degrees grid (~4 km). This downscaling method redistributes the EMEP concentrations according to the fine scale emission distribution, based on a bottom up emission inventory for Oslo. Compared to the regional model calculations for Oslo, PM 2.5 concentrations in the most polluted grids are approximately doubled. A covariance downscaling method is applied that converts gridded regional model PM 10 concentrations to a gridded population weighted concentration using finer scale information concerning the covariance of the population density, emission data and altitude data. The method is applied to all of Europe to determine the population weighted concentrations of PM 10 originally calculated by the 2

3 EMEP model with a 50 x 50 km 2 resolution, effectively attaining a resolution of ~5 km. Application of this method results in an increase of 15% in the total population weighted concentrations for all of Europe. The third downscaling method attempts to define functional relationships between local meteorological parameters, city characteristics and urban emissions to derive an urban increment. The relationships are determined from measured urban increments at a number of locations for key cities. The derived functional relationship (parameterisation) can then be used to estimate the urban increment for any arbitrary location. 3

4 Contents Summary... 2 Contents Enhanced resolution modelling: EMEP regional calculations Enhanced resolution modelling: DMI-integration of a multi-scale meteorological-chemical transport modelling system Downscaling Outline Off-line modelling framework On-line modelling framework Urban Scale High Resolution Modelling Modelling on regional / European/ Denmark scale Modelling on urban / country scale Sub-grid downscaling: Redistribution of EMEP regional model results using fine scale emissions Step Step Step Step Preliminary Results Sub-grid downscaling: Covariance downscaling method for exposure assessment of PM 10 when using regional scale models Description Application Example Future application in TRANSPHORM Sub-grid downscaling: An approach for determining urban concentration increments Rationale and Method Overview Methodology and Results Conclusions and future application Conclusions Acknowledgements References

5 1 Enhanced resolution modelling: EMEP regional calculations The EMEP model is highly flexible. Both model domain, ranging from global to local, and model resolution, can be changes by simply changing a few parameters prior to compilation. As a result the main restraint is the availability of model input data as meteorological data, emissions and land-use, and computer resources. The EMEP photochemistry model has been run for a full year in the same resolution as the TRANSPHORM/MEGAPOLI emissions (1/16 1/8 degrees latitude longitude resolution). So far the calculation has been made for the meteorological year The TRANSPHORM year 2005 will follow later when we receive the final emissions for that year. The calculations have been made with the ECMWF IFS meteorological driver. The meteorological data have been interpolated to the model grid from the original T799 grid used in the IFS calculations. Some preliminary results from the 2009 calculations are presented below. The fine scale model run shows clear improvements in the level of detail compared to model calculations with a coarser resolution. Preliminary comparisons with EMEP model calculations with 50 x 50 km 2 resolution show only slight improvements when comparing with EMEP measurements. The EMEP measurement sites are however specifically selected as background sites. We expect that the increase in resolution will result in larger improvements in more urbanized areas. This will be investigated in more detail with the model calculations with 2005 emissions. Figure 1.1a shows PM 2.5 for the regional model domain for year Emissions have been provided by TNO in the same format as for TRANSPHORM/MEGAPOLI, but the emissions have been scaled to the EMEP national totals. High PM 2.5 levels are calculated in and around the large city agglomerations, and in particular in the Po valley in Northern Italy. In Figure 1.1 b we are zooming in on the region 3 degrees west to 10 degrees east and 45 to 55 degrees north. Major cities in the region can be seen as PM 2.5 hotspots. In Figure 1.1 c we zoom in further on the region 3 to 6 degrees east and 50 to 53 degrees north. Major city hotspots in this region include Brussels and Antwerpen in Belgium and Rotterdam in the Netherlands.

6 6 a) b) c) Figure 1.1. PM2.5 concentrations in µgm -3 for a) the full regional domain, b) zooming in on the region 3 degrees west to 10 degrees east and 45 to 55 degrees north and c) further zooming in on the region 3 to 6 degrees east and 50 to 53 degrees north.

7 7 The process of validating the results with measurements has just started. Figure 1.2 shows the annual model versus measurements scatter plots for PM 2.5 and PM10. Both PM 2.5 and PM10 are under-predicted compared to measurements. The model version used in these model runs is with SOA (secondary organic aerosols). Previous model calculations have shown that including SOA result in substantial increases in PM levels (Tsyro, 2009). Figure 1.2 Log-log scatter plot of model versus measurements for PM2.5 (left) and PM10 (right) in µgm -3. PM2.5 measured: 10.69, modelled: 8.37, correlation: PM10 measured: 14.55, modelled: 11.49, correlation: 0.

8 2 Enhanced resolution modelling: DMI-integration of a multiscale meteorological-chemical transport modelling system Alexander Baklanov, Roman Nuterman, Bjarne Amstrup, Alexander Mahura, Ashraf Zakey, Ulrik Korsholm, Iratxe Gonzalez-Aparicio Danish Meteorological Institute, DMI, Copenhagen, Denmark This contribution is focusing on the realisation of the main goal of this Task 2.4.1: to enhance the resolution and develop sub-grid downscaling methods for transport sources, for improved exposure assessment on the European scale, using regional scale models. 2.1 Downscaling Outline Usually, the up-scaled city-scale (sub-meso) or meso-scale models consider parameterisations of urban effects or statistical descriptions of the urban building geometry, whereas the micro-scale (street canyon) models are obstacle-resolved and consider a detailed geometry of the buildings and the urban canopy. Global ECHAM5 MESSy MACC MACC/GEMS regional domain Denmark- scale domain City- scale domain Street- scale selected domain (Jagtvej) Figure 2.1. Downscaling of global and European-scale models for the city and streets in Copenhagen, DK. The first element in the chain is the TRANSPHORM (as well as GEMS/MACC ensemble) forecast and assessments from regional/ meso-scale models as input to DMI s downscaling system (Figure 2.1). After that, two nested domains are used for downscaling from the regional- to meso- and city-scale using integrated (ACTM+NWP) Numerical Weather Prediction Atmospheric Chemistry Transport models (Figure 2.2): (i) off-line CAMx+HIRLAM or (ii) on-line Enviro-HIRLAM (Environment - HIgh Resolution Limited Area Model).

9 D2.4.1 TRANSPHORM Deliverable Figure 2.2. Downscaling of global-regional-urban-street scale forecasts for selected European metropolitan areas (SP2: the right part for WP2.4 and the left part for WP2.3) and supplying input for assessment studies (SP3). For regional scale, urban parameterisations are based on the roughness and flux corrections approach (EMS-FUMAPEX, 2005); for urban-scale they are based on the building effects parameterisation, BEP (Martilli et al., 2002). Then a Micro-scale Model for Urban Environment (M2UE) is used for local- and micro-scale nesting. 2.2 Off-line modelling framework The Atmospheric Chemistry-Aerosol-Transport models/modules depend on applicable meteorological driver as well as selected domain of interest. At present, output from several nested versions of DMI-HIRLAM (original HIRLAM ( model adapted and refined for Denmark) is applied: (Figure 2.3): T km, 40 vertical layers; M km, 40 vertical layers; S km, 40 vertical layers; S km, 40 vertical layers; U01/I km, 40 vertical layers (experimental urbanised version). 9

10 D2.4.1 TRANSPHORM Deliverable Figure 2.3: Examples of operational and research NWP DMI-HIRLAM modelling areas. The current operational DMI forecasting modelling system (Yang et al., 2004) includes the pre-processing, climate file generation (based on CORINE dataset ( data assimilation, initialization, forecasting, postprocessing, and verification. It includes also a digital filtering initialization, semi-lagrangian advection scheme, and a set of physical parameterizations such as Savijaervi radiation, STRACO condensation, CBR turbulence scheme, and ISBA land surface scheme. The lateral boundary conditions are received every 6 hour from ECMWF ( The system runs on the DMI CRAY-XT5 supercomputer and produces output files after forecasts. DMI CRAY-XT5 is based on AMD Opteron Quad-Core Microprocessors. There are two identical systems (one for research and another one for operational weather forecast), each system contains 256 compute nodes (8 cores), in total 4096 cores with peak performance 38 Tflops, memory 8.2 TBytes and disk space 110 TBytes. An interface between the NWP and ACT models was built. Through this interface necessary information is extracted from the HIRLAM output, and then it is used by the ACT model. A comprehensive script system (based on Perl and Linux Shell) was built to couple both models together in order to produce air quality forecasts. The offline ACT model is CAMx (Comprehensive Air Quality Model with extensions). An updated version of Carbon Bond IV (CB-IV) Mechanism (Gery et al., 1989) with improved isoprene chemistry and aerosol chemistry is used. The Tropospheric Ultraviolet and Visible radiation model (TUV) (Madronich, 2002) is applied to calculate photolysis rate coefficients, and emissions from EMEP ( or TNO ( The horizontal and vertical resolutions of the model depend on a resolution of the meteorological and emission data. At present the model is run over a 0.2º 0.2º horizontal grid (Figure 2.4a), and it has a vertical resolution of 25 levels. It is also planned to use 0.02º 0.02º horizontal grids for Denmark with further nesting for Copenhagen (Figure 2.4b) with 25 vertical layers. These vertical levels cover the lowest 3 km of the troposphere. The amount of chemical compounds, which is transported from the free troposphere into the atmospheric boundary layer, is determined by the meteorological information and the concentration of the chemical compounds in the free troposphere. The advection is solved using the Bott scheme. 10

11 11 (a) (b) Figure 2.4. (a) Operational modelling areas and (b) nested modelling areas for MACC and TRANSPHORM projects. The produced model output (surface meteorological and chemical fields) is extracted from binary files (with time resolution of one hour) and is written in ASCII files for subsequent usage as input in nested Computational Fluid Dynamics model M2UE. 2.3 On-line modelling framework The Enviro-HIRLAM is an on-line coupled NWP (HIRLAM) and ACT model for research and forecasting of meteorological as well as chemical weather (Baklanov et al., 2008). The integrated modelling system is developed by DMI and is included by the European HIRLAM consortium as the baseline system in the HIRLAM Chemical Branch ( The model description was done in (Baklanov et al., 2008; Korsholm, 2009; Korsholm et al., 2009). Enviro-HIRLAM contains parameterisations of the direct, semi-direct, first and second indirect effects of aerosols. Direct and semi-direct effects are realised by modification of Savijarvi radiation scheme (Savijärvi, 1990) with implementation of a new fast analytical SW and LW (2-stream approximation) transmittances, reflectances and absorptances (Nielsen et al., 2011). Simplified analytical parameterization for inclusion of direct aerosol effect on short-wave radiation was developed based on Koepke et al. (1997) using the DISORT model and considering the full spectral radiance field. The species include BC (soot), minerals (nucleus, accumulation, coarse and transported modes), sulphuric acid, sea salt (accumulation and coarse modes), "water soluble", and "water insoluble. Condensation, evaporation and autoconversion in warm clouds are considered to be fast relative to the model time step and are not treated prognostically. The bulk convection and cloud microphysics scheme STRACO (Sass, 2002) and the autoconversion scheme by Rasch and Kristjansson (1998) form the basis of the parameterisation of the second aerosol indirect effect. As aerosols are convected they may

12 12 activate and contribute to the cloud droplet number concentration, thereby, decreasing the cloud droplet effective radius affecting autoconversion of warm cloud droplets into rain drops. Cloud radiation interactions are based on the cloud droplet effective radius (Wyser et al., 1998). As it decreases warm cloud droplet size and reflects more incoming short wave radiation, thereby, we parameterise the first aerosol indirect effect. A clean background cloud droplet number concentration is assumed and the anthropogenic contribution is calculated via the aerosol scheme. The model setup is quite similar to the HIRLAM system (described in the previous section) except the chemistry initialization. So the following data should be specified in addition to that mentioned above: initial concentrations, boundary concentrations (obtained from either measured ambient data or from large/ regional-scale models) and gridded emission sources. The anthropogenic emissions are provided by TNO. The resolution of emission inventory is 6 6 km. The Enviro-HIRLAM output (meteorological and chemical fields) is stored in GRIB1 format (with time resolution of one hour). To extract gridded data, the ECMWF library ( is used in conjunction with Perl script system. After extraction the output is written in ASCII files for subsequent usage as input in nested M2UE model. Long-term version EnvClimA The Enviro-HIRLAM Integrated Online-Coupled Multi-Scale Meteorological-Chemical Transport Modelling System involves several models. The Enviro-HIRLAM model is an online coupled NWP (HIRLAM) and ACT model for research and short-term forecasting of meteorological as well as chemical weather (Baklanov et al., 2008; Korsholm et al., 2008; Korsholm, 2009). For long-term multi-year runs and climate studies a more economical version EnvClimA is used. The climate component of the EnvClimA version is the Regional Climate Model (version RegCM4, Shalapy et al., 2012), and the Environment component is the same as in the Enviro-Hirlam model. The RegCM model, developed at the Abdus Salam International Centre for Theoretical Physics (ICTP), is a hydrostatic, sigma coordinate model (Pal et al. 2007). The chemistry component in EnvClimA includes the condensed gas-phase chemistry which is based on CBM-Z (Zaveri and Peters, 1999) and uses lumped species that represent broad categories of organics based on carbon bond structure. The computationally rapid radical balance method (RBM) of (Sillman et al., 1991) and (Barth et al., 2002) is coupled as a chemical solver to the gas-phase mechanism to provide a solution to the tendency equation for photochemical production and loss. Photolysis rates are determined as a function of various meteorological and chemical inputs and interpolated from an array of pre-determined values based on the Tropospheric Ultraviolet-Visible Model (Madronich and Flocke, 1999) with cloud cover corrections by (Chang et al., 1987). Cloud optical depths and cloud altitudes from EnvClimA are used in the photolysis calculations, thereby directly coupling the photolysis rates and chemical reactions to meteorological conditions at each model time step. The effects of dry deposition are included as a flux boundary condition in the vertical diffusion equation. Dry deposition velocities for 31 gaseous species are calculated from a

13 13 big leaf multiple resistance model (Wesely, 1989; Zhang et al., 2002, 2003) with aerodynamic, quasi-laminar layer, and surface resistance acting in series. The processes assume 20 land-use types and make a distinction between uptake resistance for vegetation, soil, water, snow and ice. In the dry deposition scheme we considered both of stomata and non-stomata resistances, which are necessary as the stomata uptake occurs only during the daytime for most chemical species. This leads to a more accurate representation of diurnal variations of dry deposition, a process crucial for climate-chemistry interaction. The aerodynamic resistance is calculated from the model boundary layer stability, wind speed and surface roughness, where a quasi-laminar surface layer is incorporated. The modelling system is also coupled with aerosol modules and includes direct and indirect aerosol effects. For multi-year simulations the EnvClimA version is used, short-term episodes and feedback mechanisms are analysed using the original Enviro-HIRLAM model. 2.4 Urban Scale High Resolution Modelling The boundary layer in the urban areas has a complex structure due to multiple contributions of different urban features, including variability in roughness, heat fluxes, etc. All these effects can be included to some extend into NWP HIRLAM or on-line Enviro-HIRLAM models. The DMI HIRLAM/Enviro-HIRLAM U01/I01 models (resolution of 1.4 km) with domains shown in Figure 2.2 are employed for high resolution urbanized modelling. Three levels of urban parameterisations were realised in the model: The simple urbanization of NWP includes modifications (anthropogenic heat flux, roughness, and albedo) of the land surface scheme, so-called the Interaction Soil Biosphere Atmosphere (ISBA) scheme originally proposed by Noilhan & Planton (1989). The Building Effect Parameterization (BEP) module. It includes the urban sub-layer parameterization suggested by Martilli et al. (2002) and is used to simulate the effect of buildings on a meso-scale atmospheric flow. For local- and micro-scale nesting, the Micro-scale Model for Urban Environment (M2UE) model is used. This is a comprehensive CFD-type obstacle-resolved urban wind-flow and dispersion model (based on the Reynolds averaged Navier-Stokes approach and twoequation k-e turbulence closure). Boundary and initial conditions for the nested M2UE model are used from the HIRLAM+CAMx or Enviro-HIRLAM models. The urban- and street-scale models used in the TRANSPHORM project are described in details in the Deliverable (Urban scale modelling). So, we will not discuss these issues here and focus on the improvements of regional and meso-scale models. 2.5 Modelling on regional / European/ Denmark scale For the downscaling, the regional scale runs (HIRLAM+CAMx HIgh Resolution Limited Area Model + Comprehensive Air quality Model with extensions - modelling system) are done based on the regional DMI HIRLAM+CAMx output (on the GEMS domain, see Figure 2.3). The model runs are performed for both the GEMS (European scale) and DK (Denmark) area domains 4 times a day employing the DMI supercomputing system (CRAY-XT5).

14 14 Run based on MOZART-GEMS output (from HIRLAM+CAMx) For the GEMS area, the forecast length is equal to 72 hr with forecasts produced at 00 UTC; and runs are restarted at 06, 12 and 18 UTCs. For this area, the run is performed using 3D meteorology from the DMI-HIRLAM-T15 model (0.15 deg. horizontal resolution, 40 vertical levels, 400 sec time step) output. The chemical boundary conditions (if available) are taken from 00 UTC run of the global IFS/MOZART modelling system from ECMWF or the 00 UTC run from the previous day. Meteorological data are recalculated from the lowest 25 levels as used in HIRLAM to the same 25 vertical levels used by CAMx. At the present setup, for the GEMS area the individual run takes approximately 1.5 h of CPU time (on CRAY-XT5), and is finished around 04:20 UTC after nominal time (i.e. after 00 UTC). For the DK area, the run is performed 4 times daily (with a forecast length of 36 hrs) using 3D medio meteorology from the DMI-HIRLAM-S03 model (0.03 deg horizontal resolution, 65 (40 prior November 2010) vertical levels, 90 s time step) output. This run uses initial and boundary conditions from the early corresponding run for the GEMS domain. At the present setup, for the DK area the individual run takes approximately 4 hrs (up to 5 hrs, on occasion) of CPU time and until, at least, 08:30 UTC after the nominal time. Hence, the meteorological and chemical boundary conditions for the urban (as well as street scale) runs become available after that time. Therefore, evaluation of optimal setup for the Copenhagen downstream chain was carried out because of the current (operational duties of DMI in daily numerical weather prediction) load of the CRAY-XT5 computer system. For that the existing DK setup was revised for the same domain size but with reduced (for example, 0.04 deg.) horizontal resolution and increased time step (variable adjustment within CAMx and it does vary a lot). (a) (b) Figure 2.5. HIRLAM+CAMx simulation results for (a) NO x and (b) O 3 concentration [µg/m 3 ] within the European scale modelling domain (horizontal resolution of 0.2 x 0.2 deg) on 5 Sep 2011, 10 UTC.

15 15 Figure 2.6. Monthly mean ozone concentration (µg m-3) at eight non-mountain European measurement stations along with modelled values from the baseline and emission simulations (Korsholm et al., 2012). An example of the HIRLAM+CAMx simulation results is shown in Figure 2.5. The spatial distribution of the NO x (Figure 2.5a) and O 3 (Figure 2.5b) concentrations within the European scale modelling domain (horizontal resolution of 0.2 x 0.2 deg) is given at 10 UTC on 5 Sep As seen in Figure 2.5a, the elevated NO x concentrations (up to 50 µg/m 3 ) are observed over the urban areas of Denmark. Almost for the entire territory of Denmark the O 3 concentration varied between µg/m 3, although south of the Jutland Peninsula ozone level are lower (i.e. less than 50 µg/m 3 ) (see Figure 2.5b).

16 16 Figure 2.8 shows observations at the building roof level vs. HIRLAM+CAMx forecasts up to 48 hours. Concentrations of NO x and O 3 based on the European (green line) and Denmark (read line) scales model runs vs. observations (blue line) are given for 4-5 September Note that observations of these chemical species were from An increase in NO x concentration observed during morning hours on 5 Sep 2011 (i.e. forecast length 34 hour) was not fully captured on the DK scale, and even less captured on EU scale by the GEMS forecast. The reason of the peak on this date is related to complex meteorological conditions leading development of local vortex over the Copenhagen metropolitan area, which existed during several hours and forced accumulation of pollution locally. Additionally to the above described study a long-term evaluation of the HIRLAM+CAMx model for the European regional scale versus annual measurement data from European monitoring stations was done for the year 2009 within the EURIMA project, analysing a possible influence of building insulation on outdoor concentrations of regional air-pollutants (Korsholm et al., 2012). Figure 2.6 shows monthly mean ozone concentration (µg m -3 ) at eight non-mountain European measurement stations along with modelled values from the baseline and emission simulations. As we can see from the figures the model shows a relatively good correspondence with the measured values for most of the stations. 2.6 Modelling on urban / country scale The Enviro-HIRLAM model is run for the Denmark area (modelling domain U01, including Zealand Island and Jutland Peninsula) at horizontal resolution of 1.4 km, 40 (up to 60) vertical levels, time step of 30 seconds (up to 90). The model is run, at least, 2 times a day with a forecast length of 24 hr, and saving generated 3D output (meteorology and chemistry) every one hour interval. Enviro-HIRLAM uses boundary conditions from the operational DMI-HIRLAM-S03 model (with a forecast length of 36 hrs) at given times. Individual runs require (including selected chemistry and aerosol effects; i.e. shortened list of chemical species compared to the CAMx runs) up to 2 hours of CPU time. The run will be initiated for 00 UTC forecast when the boundary conditions will became available from the S03 domain. Note that the runs can be done in two modes: 1) urbanized Enviro-HIRLAM-U01 generates both types of (chemistry + meteorology) output used as input by M2UE; and 2) CAMx generated output in -S03 domain is directly used in M2UE than Enviro-HIRLAM can provide only 3D meteorological fields improved in urban areas for M2UE (and hence, total CPU time will be reduced). The urban effects are taken into account through implementation of the Building Effect Parameterization (BEP) module. In each grid cell a fraction of an urban class is assigned from CORINE (2000) land-use dataset. Moreover, each urban grid cell is attributed to selected urban districts (city centre/ high building district; industrial commercial district; residential district, and rural area) with specific morphological parameters, representing possible influence of urban features on thermal and dynamical characteristics and formation of meteorological fields. Urban land-use data were pre-processed into mentioned above urban districts using GIS (ArcView) technique. For example, for the Copenhagen metropolitan area

17 17 approximately 500 urban grid cells were identified. During the runs the urban module is activated in the ISBA (Interaction Soil-Biosphere-Atmosphere) land surface scheme and only in grid cells where the urban fraction is more than zero. Note that the inclusion of urban module has increased the total computational time by only a few percent. (a) (b) Figure 2.7. HIRLAM+CAMx simulation results for (a) NO x and (b) O 3 concentrations [ug/m 3 ] within the Denmark scale modelling domain (with GEMS area boundaries; horizontal resolution of 0.04 x 0.04 deg) on 5 Sep 2011, 10 UTC. An example of the HIRLAM+CAMx simulation results is shown in Figure 2.7. The spatial distribution of the NO x (Figure 2.7a) and O 3 (Figure 2.7b) concentrations within the Denmark scale modelling domain (horizontal resolution of 0.04 x 0.04 deg) is given at 10 UTC on 5 Sep As seen in Figure 2.7a, the elevated NO x concentrations are observed over large parts of Denmark (compared to Figure 2.5a), reaching levels of more than 250 µg/m 3. Moreover, lower ozone concentrations, having irregular pattern, are observed over the urban areas (see Figure 2.7b). Figure 2.8 (top) shows observations at the building roof level vs. HIRLAM+CAMx forecasts up to 48 hours forecast length. The NO x and O 3 concentrations based on the European and Denmark scales model runs are given for 4-5 Sep Note that the observations of selected chemical species were obtained from As seen in Figure 2.8 (top), the peak of NO x observed at the morning on 5 Sep 2011 (see starting 34 hour on horizontal axis) was not completely captured on a country scale, and even less captured on a large scale based on the GEMS forecast. The possible reason for such a peak could be related to specific meteorological conditions creating the local circulation over the urban area of Copenhagen, which lasted for a few hours and led to a local build-up of air pollution.

18 18 Figure 2.8. Observations at the building roof level vs. HIRLAM+CAMx forecasts for (top) NO x and (bottom) O 3 for the European (horizontal resolution of 0.2 x 0.2 deg) and Denmark (horizontal resolution of 0.04 x 0.04 deg) scale modelling domains during 4-5 Sep Although the Enviro-HIRLAM urban scale runs have been tested and verified for the Copenhagen metropolitan area, due to high computational resources required for part of the online chemical transport modelling (compared with the meteorological modelling) these runs have been omitted from the down-scaling chain to satisfy the chain overall operational requirements and needs.

19 19 3 Sub-grid downscaling: Redistribution of EMEP regional model results using fine scale emissions The regional scale output of model results from the EMEP model for 2005 is on a degree resolution. This resolution is too coarse to be applied in urban areas with high population densities and large variability in pollutant concentration levels over short distances. At present regional models can not be run on the resolution required for city scale applications, mainly because of computational restraints. Furthermore, emissions on such fine scales are only available on a city level, and not for the entire European domain. Downscaling of the regional model computations can give an estimate of pollution levels within the city domain, but is likely to be less accurate than running a city scale model. Downscaling of regional model results may not replace local scale modelling, but can at least provide a screening method for selecting cities that are suitable for a more detailed study involving city scale modelling. Running a city scale model requires, in addition to a fine scale emission inventory, appropriate meteorological data input data, detailed land use data etc. In order to get an indication of the variability in pollutant levels within an urban area, we have started testing methods for downscaling of the regional model results. It should be noted that unlike the model version used in section one, this model version does not include SOA. As an example of the method we have started with the downscaling of the PM2.5 model results for Oslo. The downscaling is performed in 4 steps: 1. The contribution from Oslo is calculated by the regional EMEP model. This contribution is calculated by subtracting a model calculation where PM2.5 emissions from Oslo are removed from the emission input. 2. Splitting the native EMEP grid into 4 4 grid boxes. The resulting resolution of the downscaled grid is degrees. 3. Interpolating the local scale emissions of PM2.5 (provided by NILU) into the same degree grid. 4. Rescaling the concentrations from the EMEP regional model on the grid with the local scale emissions. 3.1 Step 1 The EMEP photochemistry model is run for one full year with a 0.2 times 0.2 degrees resolutions with TNO emissions provided through the TRANSPHORM project. The emissions are interpolated to model grid from the original 7 7 km 2 grid. Figure 3.1, left shows PM2.5 concentrations in and around Oslo as calculated by the EMEP regional model with a 0.2 times 0.2 degree resolution. Figure 3.1, right shows the contribution to PM2.5 when excluding PM emissions in Oslo, represented as the four central grid boxes in the plot, denoted the regional Oslo delta (ROD).

20 Step 2 The annual output from the regional EMEP model calculations is split into a 0.05 times 0.05 degrees grid. Concentrations/levels in the refined grid retain the original concentrations/levels from the original grid. Figure 3.1. Left: PM2.5 (µgm -3 ) concentrations in and around Oslo as calculated by the regional scale EMEP model with a 0.2 times 0.2 degree resolution. Right: The contribution to PM2.5 from emissions within Oslo, the regional Oslo delta (ROD). 3.3 Step 3 PM2.5 emissions for larger Oslo, including Oslo and parts of the surrounding area, has been provided by NILU. The highest emissions are located near the centre of Oslo. A line of relatively high emissions is extending eastward from central Oslo. The emissions, interpolated to the 0.05 times 0.05 degrees grid (E_PM25 loc ) is shown in Figure 3.2, left. These emissions are used for scaling only, and may, if aggregated to the 0.2 x 0.2 degrees grid, differ from the emissions used in the EMEP model calculations. Figure 3.2 Left: local PM2.5 (gs -1 km -2 ) emissions in Oslo interpolated to 0.05 x 0.05 degrees (E_PM25 loc ). Right: PM2.5 (µgm -3 ) concentrations on the same grid scaled with the local emissions.

21 Step 4 The fine scale concentrations of PM2.5 over Oslo is calculated by rescaling the difference in concentrations when excluding the PM2.5 emissions over Oslo as shown in Figure 3.1, right (ROD, Regional Oslo Delta). This calculation of PM2.5 concentrations in Oslo (PM2.5loc) on a 0.05 x 0.05 degrees grid is (so far) performed in a very simplistic way: PM2.5loc = RC + ROD * E_PM25 loc /avge, where RC is the regional contribution to PM2.5 concentrations and avge is the average over all non-zero local PM2.5 emissions. 3.5 Preliminary Results Compared to the regional scale calculation alone, PM2.5 concentrations in the most polluted grid are approximately doubled. For the most polluted grid cells ROD as calculated by the regional model in increased by a factor of three. The most polluted grids are located close to the city centre, where concentrations close to 25 µgm -3 are calculated. Elevated concentrations are also seen east of the city (towards Lillestrøm and the main airport), and to the west. The area covered by the local emissions as provided by NILU is smaller than the area of the 4 grid boxes where emissions are removed from the regional model calculations. However, a large part of the area where regional emissions are removed, but without local scale emissions is covered with forests, where emissions should be small. The downscaled concentrations close to the city centre are likely to be too high. Annual measurements from several traffic sites in central parts of Oslo are all below 15 µgm -3. The scaling of the PM2.5 concentration using the local emissions from NILU is very coarse, and does not allow for dispersion of locally emitted pollutants out of the grid (this is the main reason why we have not used a finer resolution). An improved method will be sought for this scaling.

22 22 4 Sub-grid downscaling: Covariance downscaling method for exposure assessment of PM 10 when using regional scale models A methodology has been developed (Denby et al., 2011) for estimating the impact of sub-grid variability on long term static population exposure estimates when calculated using regional scale models (CTMs). The method parameterises the population-concentration covariance within a CTM grid and corrects for this sub-grid covariance. The method converts a CTM grid concentration to a CTM grid population weighted concentration, which is the exposure indicator required for long term exposure and health impact studies. This approach can be seen as an alternative to the urban increment methodology. The inferred resolution of the sub-grid covariance is limited by the available proxy data used for the parameterisation. In this case this is 3 5 km with the limiting factor being the available resolution of emission data. 4.1 Description The population weighted concentration C pw,j in a CTM grid j can be written as C pw, j cov j(c, p) = C j 1 + C j Pj with the sub-grid covariance in the CTM grid j being given by 1 cov j(c, p) = n 1 n ( ci C j )( pi Pj ) i Here c i and p i are the sub-grid concentrations and population density respectively (indexed with i) within the CTM grid j. C j and P j are the mean CTM grid concentration and mean subgrid population density in the CTM grid j respectively. The covariance correction factor (COV cp ) needed to convert the CTM grid concentration to the CTM grid population weighted concentration is given by COV cp, j cov = C j (c, p) j P j When the sub- grid covariance is 0 then the population- weighted concentration is the same as the CTM grid concentration. For PM 10 in Europe this parameter has been assessed (using observed concentrations) to vary from to 0.5 for 50 km grid sizes. The covariance correction factor for any particular model grid is unknown but it can be parameterised utilising higher resolution proxy information. This proxy information is contained in emissions, population density and altitude and is available at higher resolutions than the CTM grid. The limiting resolution is the available emission data at around 6 km. The parameterisation used is a linear regression fit to the observed population-concentration covariance using the emission-population covariance and the altitude-population covariance.

23 23 The parameterisation is given by: COV cp (reg)= b 0 + n k= 1 b E,k COV ep,k +b Z COV where the subscript E denotes relevant emissions and Z refers to altitude. Note that due to the normalization of the covariance with the mean values the units of the emissions are irrelevant. For altitude it is not appropriate to use the average altitude and so the average European altitude of 340 m is used to normalize the altitude-population covariance (i.e. Z j =340 m everywhere). 4.2 Application The method can be used to downscale gridded CTM concentrations to a population weighted concentration for each grid. The covariance correction factor for each CTM grid need only be calculated once, based on the current population, emission and altitude high resolution data. If the spatial distribution or the variability of either the population or emission data changes in some way, e.g. for future scenario calculations, then the covariance will change in the CTM grid and the covariance correction factor needs to be recalculated. The method thus reflects the fact that decreasing the covariance of population and emissions (moving emission sources away from populated areas) will decrease the exposure, even though total emissions may remain the same. To apply the method to a CTM calculation the following information is required: Gridded CTM concentration fields (e.g. 50 km) High resolution population density data (e.g. 3 6 km) High resolution relevant emission data. For PM this is primary PM and ammonia emissions (e.g. 3-6 km) High resolution altitude data (e.g. 3 6 km) Correction factor parameters (b 0, b E, b Z ) The covariance correction factor is calculated for each CTM grid using the emissionpopulation covariance and the altitude-population covariance, derived from the high resolution data. This is then applied for each grid in the CTM and the population-weighted concentration in the grid is calculated. 4.3 Example The method has been applied to EMEP 50 km calculations for Emission data used was the MEGAPOLI emission data set from TNO at 1/16 and 1/8 degree. For PM 10, at 50 km resolution, the following regression constants (and their uncertainties expressed as standard deviations) were determined. Standard error and correlation of the regression used to determine the coefficients is also given. The standard error indicates the uncertainty in the parameterised covariance correction factor for each individual CTM grid. Aggregation of the grids will reduce this uncertainty. zp

24 24 50 km b 0 b E (PM 10 ) b E (NH 3 ) b Z Standard error Correlation (r 2 ) PM (SD) (0.009) (0.033) (0.046) (0.006) (0.012) (0.08) The calculated area weighted concentration for all of Europe, the modelled population weighted concentration and the corrected population weighted concentration using the parameterised covariance correction factor are shown in the table and Figures 4.1 and 4.2. Application of the correction leads to a 15% increase in the total population weighted concentrations for all of Europe. Compound Area weighted model concentration (C T ) Population weighted model concentration (C pw,m ) Parameterised population weighted concentration (C pw,p ) Total covariance correction factor Europe (COV cp,t ) PM µg/m µg/m ± 0.5 µg/m ± 0.04 Figure 4.1. Modelled concentration (left) and population weighted concentration (right) calculated using the parametrisation of the covariance correction factor for PM 10. Units are in ug/m 3.

25 25 Figure 4.2. Covariance correction factor for PM 10 calculated using the parameterisation. 4.4 Future application in TRANSPHORM The methodology described here for downscaling regional scale modelled concentrations can be implemented as part of the Integrated Assessment tool in TRANSPHORM, enabling a fast downscaling method for addressing European wide exposure that, in addition, has the potential of addressing structural changes in the way that emissions are spatially distributed (i.e. the covariance of population and emissions). The method may also be applied to urban scale and a study is currently underway in TRANSPHORM to assess this. In that case the emission sources and their spatial distribution may be determined based on bottom up emission inventories, e.g. road links and traffic volumes, and these may be used as appropriate proxy data to calculate the covariance correction factor on the local scale. A number of the target cities in TRANSPHORM, and high resolution model concentration fields, will be used to assess the methodology when applied on the local to urban scale.

26 26 5 Sub-grid downscaling: An approach for determining urban concentration increments A simple approach for accurately estimating an urban concentration increment on top of the regional background for urban areas in Europe was developed. The method operates by establishing a functional relationship between the concentration increment and the local meteorological situation, the city characteristics, the urban emissions and background concentrations. The application of the method was carried out for PM 10 and NO 2. Pollutant concentrations needed for the multiple regression process can be derived both from measurements and high resolution urban scale model results. The results demonstrate the capability of this simple approach to reproduce the urban increment with satisfactory accuracy, thus providing a tool for fast but still reliable quantitative assessments of urban air quality that can subsequently be used in calculations of exposure and health impact assessment. Besides, scenario calculations for the urban increments can be based on scenario emissions and respective modelled regional background concentrations. 5.1 Rationale and Method Overview The simple methodology presented aims at the determination of an urban concentration increment on top of the regional scale background, for urban areas in the entire European region. This is especially relevant for the purposes of health impact assessment of various pollutants, but will be demonstrated here for PM 10 and NO 2. The methodology attempts to define a functional relationship between local meteorological parameters, city characteristics and urban emissions on the basis of measured increments on sample locations. The functional relationship can then be applied on arbitrary locations (urban areas) throughout Europe. Special care was taken during the selection of sampling locations so as to ensure geographical representativeness and adequate data availability. Part of the development of this methodology has taken place within the MEGAPOLI project. 5.2 Methodology and Results The overall procedure for determining the urban increment can be described as a sequence of three main processing stages: 1. Spatial sampling: selection of representative rural-urban measurement station pairs across Europe. 2. Multiple regression analysis for determining a functional relationship. 3. Generalisation: estimation of urban increments for other European cities. In order to extract concentration increments, several station pairs consisting of one rural and one urban background station were selected. The search initially focused on 20 cities that have been studied before in the framework of various projects (e.g. the MERLIN project, URL1), namely Antwerp, Athens, Barcelona, Berlin, Brussels, Budapest, Copenhagen, Gdansk, Graz, Helsinki, Katowice, Lisbon, London, Marseille, Milan, Paris, Prague, Rome, Stuttgart and Thessaloniki.

27 27 Data associated with the main meteorological parameters were available for the year 2003 from the regional model PARLAM-PS (Bjørge and Skålin, 1995). Therefore, it was decided to use the AirBase (Airbase, 2008) datasets for this particular year throughout the calculations procedure. The next step was the selection of appropriate station pairs, satisfying the following criteria: The two stations of the pair should be categorised explicitly as rural (or suburban) background and urban background, respectively. Station categorisation was based on the AirBase metadata entries. Both stations should have better than 90% (for NO 2 ) or 75% (for PM 10 ) data completeness for the reference year (as reported in AirBase). The two stations should be located close enough, preferably within the same km 2 PARLAM-PS cell. Applying this set of criteria, 51 stations for NO 2 and 26 stations for PM 10 were selected. The spatial distribution of the cities corresponding to these station pairs is presented in Figure 5.1. As it can be deduced from this Figure, in many cases, different station pairs had to be selected for NO 2 and PM 10 measurements. Having established a sufficient sample of station pairs, time series of concentration differences were obtained from each station pair, meant to provide an estimate of the urban increment. Figure 5.1. Locations of the cities with sufficient measurement data for the estimation of NO 2 (left) and PM 10 (right) urban increments At the second stage of the methodology, the urban background increments were determined through a piecewise functional relationship which is established on the basis of certain variables that are known to be important (Amann et al., 2007). These variables include the level of emissions within the urban area, the size/area of the urban entity, the urban and regional background concentrations, as well as the wind speed and the atmospheric stability. A novel element of the current approach, as compared to previous similar attempts (Ortiz, 2010), consists of several improvements with respect to the part of the meteorological input, firstly with the use of an advanced interpolator for meteorological parameters at the sub-grid level and, secondly, with the incorporation of atmospheric stability as an important factor

28 28 contributing to the profile of the urban increments. In order to obtain the relationship between the urban increment and the variables mentioned above, a multiple regression analysis was carried out, using the following formulation: Where: C Sc Sc Sc Sc iue Sc Sc iurban = ωι + φi + γ Sc i Cirural AUE uavg C i urban = Urban increment of pollutant i. E iue = Total emission of pollutant i within an urban entity in tons. A UE = Urban area in km 2. u avg = Urban entity average wind speed in m/s. C i rural = Rural background concentration of pollutant i in µg/m 3. ω i, f i, and γ i = Multiple-regression parameters for pollutant i. S C = Pasquill-Gifford stability class E The regression analysis is performed on the cities identified during the station pair selection process. A separate version of the above formulation is extracted for each of the six Pasquill- Gifford stability classes (S C = 1...6), thus taking into account the dependence of concentrations on atmospheric stability. Values of all the variables were averaged over the periods identified through the stability categorisation. Regarding the data that were used in the multiple regression analysis, the urban areas were defined on the basis of land use as provided by the European 1 km resolution land use map of the CORINE Land Cover 2000 (CLC2000) project (Büttner et al., 2004). Besides, it was also assumed that only primary emissions released from low sources increase concentrations within the cities (Amann et al., 2007). Estimates of the urban emissions for each city were based on the MEGAPOLI European Gridded Emission Inventory (Kuenen et al., 2010), which is a version of the TNO emissions data-set, available for the whole of Europe in a resolution of (latitude longitude) (Figure 5.2).

29 29 Figure 5.2. European yearly emissions map in thousands of tons (CO) based on the TNO emissions dataset (left) and urban emission estimates in tons for NO x based on the TNO emissions dataset and CORINE land use (right) The average 10-metres wind speed was estimated using the Meteorology Generator tool and the 2003 PARLAM-PS dataset. The Meteorology Generator tool utilizes the CONDOR diagnostic model (Moussiopoulos, Flassak and Knittel, 1988; Flassak and Moussiopoulos, 1989) for the calculation of local wind fields ensuring that the effects of local topography are taken into account during the spatial interpolation. Wind speeds for the 20 preselected cities are shown in Figure 5.3. The final stage of the calculations involved the application of the piecewise functional relationship extracted in the previous step to an additional set of urban areas around Europe in order to calculate urban increments. Figures 5.4 and 5.5 depict the urban concentration increments as they were calculated on the basis of measurements (bars) but also using the functional relationships (diamonds) of the proposed methodology m/s ANTWERPEN ATHENS BARCELONA BERLIN ROME BRUSSELS BUDAPEST COPENHAGEN GDANSK GRAZ HELSINKI KATOWICE LISBON LONDON MARSEILLE MILAN PARIS PRAGUE STUTTGART THESSALONIKI Figure 5.3. Mean annual 10-m wind speed for the 20 preselected cities as calculated with the aid of the Meteorology Generator

30 Urban increment Rural background Prague Helsinki Copenhagen Budapest Lisbon Berlin Athens Milan Graz Marseille London Stuttgart Gent Genova Turin Madrid Munich Vienna Figure 5.4. Urban increments in µg/m 3 for NO 2 in calibration and validation urban areas (to the left and right of the dashed line, respectively). Diamonds indicate urban concentrations predicted by the UI methodology Urban increment Rural background Prague Berlin Helsinki Athens Lisbon Katowice London Madrid Stockholm Figure 5.5. Urban increments in µg/m 3 for PM 10 in calibration and validation urban areas (to the left and right of the dashed line, respectively). Diamonds indicate urban concentrations predicted by the UI methodology Urban concentrations needed to extract the increment described in the first step of the methodology can as well be derived from high resolution urban scale model results. Exploring this idea, a pilot application of the methodology was carried out using modelled concentrations originating from CityDelta (Cuvelier et al., 2007) results, for 8 European cities for the year In order to extract the sample increments and the functional relationship, model results for a central point and averaged concentrations at eight peripheral points were used for each city. Figure 5.6 depicts the nine points used for the calculation of the urban increment in the city of Paris, as well as the yearly mean PM 10 concentration map produced by the CHIMERE model.

31 D2.4.1 TRANSPHORM Deliverable Figure 5.6. Locations of the central point (yellow) and the eight peripheral points (blue) used for the city of Paris (left) and spatial distribution of the PM10 mean annual concentration calculated by the CHIMERE model. The calculated concentrations using the model simulations data were then evaluated against measurement data and the station pairs approach results. This comparison relies upon the essential assumption that although the CityDelta model applications refer to the year 1999, they nevertheless represent an adequate approximation of the concentration differences between the periphery and the urban background of each city. This validation indicates that the results of the methodology using data originating from urban models (CityDelta) are in fairly good agreement with those calculated using observed data. Nevertheless, additional research has to take place for verifying that model results can replace measurement data in urban increment calculations without any quality loss in the calculated results. Figure 5.7 presents the validation of the approach for PM10 and NO2 as regards six European cities. 5.61*7/*3.)8)*+ B/+CD):+1761A)> '! &! 4*5(67+6/*83869:*(275(/0;*.:6),-(-+.67<(+*= &" &! %" %! %! $" $! $! #! #" #! "! "!! ()*+,-./* 0/)**1 2-*/34 '()*+),-./01.23 Figure 5.7. Evaluation of the CityDelta based results, against results based on station pairs and measured concentration increments of NO2 (left) and PM10 (right) 5.3 Conclusions and future application An urban increment was determined based on a functional relationship scheme, allowing for a correction of regional background concentrations inside areas with significant urban density by establishing an operational correlation between the concentration increment originating from measurement data and the local meteorological situation, the city characteristics, the 31