D EVELOPMENT OF A NATIONAL METHODOLOGY FOR ESTIMATING POPULATION EXPOSURE Maxime Beauchamp, Laurent Ltinois, Laure Malherbe*, Frdric Tognet 20th EIONET meeting, Ljubljana, Slovenia, 5-6 October 2015 Context Legal requirement: When concentrations exceed environmental objectives in a zone, Member States have to report the exceedance situation. This includes in particular: the geographic delimitation of the exceedance area; the number of inhabitants within this area ( resident population ). National context: The exceedance area and exposed population in the concerned zones are estimated by the local associations responsible for air quality monitoring (the AASQAs). This characterization is usually based on regional or urban scale modelling depending on the pollutant and the location of exceedances. Application of a common approach is essential to improve comparability of results between zones. 20th EIONET meeting 2 Context 20th EIONET meeting 3 To comply with the European and national legislation and standards and promote good practices, national methodologies and guidelines are developed within technical committees and related working groups involving: the Ministry in charge of the environment the National Reference Laboratory the AASQAs. Automatic measurement of gases Measurement of PM Emissions, modelling and data processing IT systems Communication Rural national sites Technical committees (with long-term assignment) Current or recent working groups (time limited assignment on specific issues) Data validation Station siting criteria and classification PM chemical characterisation Acquisition systems Context 20th EIONET meeting 4 The technical committee on emissions, modelling and data processing Its objectives are to: share experience and knowledge, find collective answers to questions raised by members, identify the best methodologies and practices and contribute to their dissemination. Roadmap for 2015: -Development of quality assurance for modelling -Harmonization of practices to map exceedances and estimate exposure -Drafting of a guidance document on modelling Outline 1.Development of a quality assurance framework for modelling 2.Development of a methodology for estimating population exposure Focus on exceedances of PM 10 and NO 2 limit values in urban areas 5 20th EIONET meeting QA/QC for modelling 20th EIONET meeting 6 Diffusion of FAIRMODE recommendations. Access to the Atmosys on-line version of the Delta tool through the LCSQA website: Organisation of model intercomparison exercises 1st exercise ( ) : modelling of a small domain around a canyon street in the city of Reims. 6 5 participants. 2 models: ADMS-Urban and SIRANE. Data provided by Atmo Champagne-Ardenne. Model results analysed by the LCSQA and compared to measurements at a mobile monitoring site and passive sampling sites. QA/QC for modelling 20th EIONET meeting 7 Inter-comparison exercise. Example of results. 7 NO 2 hourly values, Validation at a traffic mobile monitoring site. Target plot (JRC Delta tool) Exposure assessment: main issues Assessment of exposure to NO 2 and PM 10 exceedances in urban areas is mostly based on urban scale modelling. It involves three main steps: Concentration mapping: how to process modelling results in order to map concentrations and delimit the exceedance area? Spatial distribution of population: how to get a fine spatial distribution of the resident population? Estimation of population exposure: how to intersect air quality and population data to estimate the exposed population? 20th EIONET meeting 8 Concentration mapping 20th EIONET meeting 9 Given gridded concentration data, the exceedance area can be defined as the set of cells where the concentration is strictly above the limit value. On the urban scale, because of concentration gradients, the estimation of population exposure can be very sensitive to the grid resolution. Issue: Due to computation times, concentrations are often modelled on irregular grids (lower resolution in background areas and increased resolution close to the roads). 9 Which method should be applied to produce a high resolution concentration map from irregular modelling data ? What should be the minimum density and spatial distribution of modelling receptor points? Which method should be used to interpolate concentrations? Example of irregular modelling grid Concentration mapping 20th EIONET meeting 10 Interpolation from an irregular to a high resolution grid Modelling datasets provided by the AASQAs for six cities (NO 2 annual average): Nantes, Niort, Orlans, Reims, Tours: 10 one simulation dataset on an irregular grid one simulation dataset on a reference high resolution grid (1 or 2 m) in a smaller domain Bourges: six simulation datasets on various irregular grids one simulation dataset on a reference high resolution grid (1 m) in a smaller domain (data from Previsionair project) 806 points1044 points points Different irregular modelling grids Reference grid 1035 points Concentration mapping 20th EIONET meeting 11 Interpolation from an irregular to a high resolution grid Different interpolation methods compared: Linear interpolation Delaunay interpolation (based on Delaunay triangulation) Point and block kriging with different estimation neighbourhoods Evaluation of the interpolation performed by: Leave-one-out cross validation Comparison of maps Validation on the high resolution reference grid 11 Concentration mapping 20th EIONET meeting 12 Evaluation of the interpolation by leave-one-out cross-validation In cross-validation, the linear and Delaunay interpolation methods have the best scores. Kriging in its usual configuration is not entirely adapted yet for interpolating concentrations from data of contrasting typologies (background/traffic). The geostatistical model needs to be adapted (on-going research). 12 Linear and Delaunay interpolation scores: lower RMSE, higher correlation, variability (std dev) of original data correctly reproduced Similar behaviour for all the cities Concentration mapping Orlans agglomeration. Maps obtained after interpolation on a 2m-resolution grid. 13 Linear interpolation Kriging with 2 neighboursKriging with 10 neighbours Delaunay 20th EIONET meeting Evaluation by map comparison Concentration mapping 20th EIONET meeting 14 Evaluation of the interpolation by validation against the reference simulation grid Concentrations are interpolated from the irregular grid on the high resolution grid and compared to the reference values. Similar conclusions regarding the relative performance of the different interpolation methods.. 14 Linear and Delaunay interpolation: lower RMSE, higher correlation, variability (std dev) of original data correctly reproduced Validation against the reference grid Leave-one-out cross-validation Depending on the pattern of the modelling grid, the highest concentrations tend to be underestimated. Concentration mapping 20th EIONET meeting 15 Sensitivity to the modelling grid In addition to the interpolation method, the spatial organisation of the simulation points significantly contributes to the quality of the interpolated map. The different grids available for the city of Bourges were used to provide recommendations on this issue. 15 City of Bourges. Validation against the reference grid for two different modelling grids. Recommended location of modelling points across the roads to reproduce the non- linear concentration decrease on each site. Spatial distribution of population 20th EIONET meeting 16 Population considered for exposure calculation is the resident population. The spatial resolution of population data has to match the spatial resolution of concentration data. A national methodology to produce spatial population data at a fine detail level was developed under GIS (LCSQA - Ltinois, 2013, 2015); It is based on: - an exhaustive review of national practices; - combination of information from different databases : o National official census data (INSEE IRIS database) o Cadastral data (BD Parcellaire) o Information from property taxes (MAJIC database) 16 Cadastral data Spatial distribution of population 20th EIONET meeting 17 Output data: Spatial distribution of population in residential buildings Population aggregated on different mesh grids (e.g. 200m and 1 km) Local extractions of the data have been distributed to the AASQAs for use in their respective regions (local and regional applications). 17 Population distributed in the residential buidings.Gridded population data Estimation of the exposed population 20th EIONET meeting 18 What is the number of exposed inhabitants ? 18 Example of exceedance zone Spatial distribution of population Practices were not uniform according to whether the residential buildings intersecting the exceedance zone were totally or partially taken into account. Sensitivity of the exposed population to the modelling grid, the interpolation grid and the concentration-population intersection method were performed on Bourges dataset. Estimation of the exposed population 20th EIONET meeting 19 Though the true value of the exposed population is not known, some recommendations could be made: Simulation points (before interpolation): should allow good representation of concentration decrease away from the roads. Resolution of the interpolation grid: should be fine enough to avoid too pronounced smoothing near the roads. 1, m, 2 m and 10 m gave comparable results in the tests. Method to intersect concentration and population data: Distribute population on the same interpolation grid and count the number of inhabitants in each cell above the limit value. or intersect directly with residential buildings. For buildings partially included in the exceedance zone, weight their population according to the intersected area. Conclusions and future works 20th EIONET meeting 20 The AASQAs will be invited to apply the whole methodology for the 2016 reporting exercise. Assistance will be provided by the LCSQA. The developments to adapt kriging interpolation to the specificities of urban pollution are promising. Further tests will be carried out. On the regional scale, the methodology to estimate population and vegetation exposed to exceedances of O 3 target values will be harmonized as well. O 3 data : in-situ observations combined with CHIMERE model Implementation of QA/QC for modelling will continue, in relation to FAIRMODE and new CEN working group guidelines.