Improvements in European mapping of NO2 and BaP concentrations and exposure Jan Horálek (CHMI) Peter de Smet, Frank de Leeuw (RIVM), Pavel Kurfürst, Nina Benešová (CHMI), Philipp Schneider, Cristina Guerreiro (NILU)

1. Basic mapping methodology 2. NO2 improved mapping 3. Benzo(a)pyrene improved mapping 4. Conclusion

Data fusion mapping methodology Regression – Interpolation – Merging Mapping Linear regression model of monitoring data (as a primarily source of information), CTM output and different other supplementary data followed by interpolation of its residuals by kriging (so-called residual kriging). Rural and urban background map layers created separately (based on rural resp. urban/suburban background stations) and merged into the final maps using population density.

Good representation in both 1x1 km and 10x10 km maps. Influence of grid resolution – concentration map PM10, annual average, 2014 simple comparison – rural areas Good representation in both 1x1 km and 10x10 km maps. rural 10x10 final merged 1x1 final, aggr. 10x10

Influence of grid resolution – concentration map PM10, annual average, 2014 simple comparison – urban background areas Good representation in1x1 km map, but not in 10x10 km map (bias, RMSE, R2). rural 10x10 final merged 1x1 final, aggr. 10x10

Annual air quality maps Regular annual product: ETC/ACM Technical Paper „European air quality maps“ Concentration maps (PM10, PM2.5, O3, NO2, NOx), probability of exceedance maps, exposure tables, uncertainty analysis, inter annual difference maps, trends. Most recent: ETC/ACM TP 2016/6, maps and tables for 2014 http://acm.eionet.europa.eu/reports/

Regular mapping concentration maps PM10 – annual average, 2014

Regular mapping concentration maps PM10 – 90.4 percentile of daily means, 2014

Regular mapping concentration maps PM2.5 – annual average, 2014

Regular mapping concentration maps O3 – 93.2 percentile of max. daily 8-hour means, 2014

Regular mapping concentration maps O3 – SOMO35, 2014

Regular mapping concentration maps NO2 – annual average, 2014 (current methodology)

1. Basic mapping methodology 2. Improved NO2 mapping 3. improved benzo(a)pyrene mapping 4. Conclusion

Improved NO2 mapping Motivation for improvement Current methodology gives poorer results compared to the maps of PM. Current methodology concerns rural and urban background areas only not accounting for hot spot (traffic) locations, although traffic is the most important source of NO2.

Improved NO2 mapping Inclusion of land cover and road data in NO2 background mapping. Creation of traffic layer map, based on traffic stations. Subsequently, inclusion of this traffic map layer in NO2 map and exposure estimate. Analysis executed on 2013 data. Detailed analysis presented in ETC/ACM Technical Paper 2016/12 „Inclusion of LC and traffic data in NO2 mapping“

LC and road data inclusion in NO2 background mapping Motivation: Supplementary data often used in land use regression (LUR) modelling examined for inclusion. Requirement: data available at the European scale. No traffic intensity data available accross Europe. Two data sources used: Corine Land Cover (CLC, 100m grids) Global Road Inventory Project (GRIP) data base of PBL 8 general CLC classes, with different aggregation: 1x1 km radius 5 km

LC and road data inclusion in NO2 background mapping Global Road Inventory Project (GRIP) – 5 road classes, four classes used only: Together with other supplementary data – 37 variables (EMEP model, altitude, meteo, population, LC, road data): Input to linear regression model analysis (stepwise regression + backwards elimination).

LRM analysis – 5 variants selected for further analysis. NO2 annual average, 2013 – rural and urban background map layers improvement by LC and road data inclusion LRM analysis – 5 variants selected for further analysis.

NO2 annual average, 2013 – rural and urban background map layers improvement by LC and road data inclusion For selected variants: spatial interpolation of residuals, 1x1 km resolution. The variants compared by cross-validation. Inclusion of land cover and road data in NO2 background mapping brings clear improvement of the uncertainty results.

NO2 annual avg., 2013 – rural and urban background map

NO2 annual average, 2013 – traffic map layer creation Based on 855 urban traffic stations. (Rural traffic stations not considered, due to their small number, i.e. 19). Supplementary data – 37 variables (EMEP model, altitude, meteo, LC, …): input to linear regression model analysis (stepwise regression + backwards elimination). 4 variants selected for further analysis.

NO2 annual average, 2013 – traffic map layer creation Spatial interpolation, 1x1 km resolution Different LRM variants, comparison based on cross-validation. Different variants give quite similar results. Based on RRMSE (24%) and R2 of cross-validation scatterplot (0.51): estimates of urban traffic air quality is reasonable.

NO2 annual average, 2013 – urban traffic map layer Map can be applied for urban traffic areas only.

Inclusion in urban map layer: Inclusion of traffic map layer in background map Inclusion in urban map layer: Weight: based on buffer around the streets/roads (GRIP database, source: PBL). Buffer: tentative. 75 meters around roads of class 1 and 2, 50 meters around roads of class 3. (Should be refined.)

NO2 annual average, 2013 – final merged map, 1x1 km

NO2 annual average, 2013 – difference „final – background“

Bad representation of traffic areas in urban background layer. Improved NO2 mapping – continuation NO2, annual average, 2013 – urban traffic areas simple comparison – map layers vs. traffic stations Bad representation of traffic areas in urban background layer. traffic map layer background map layer

Improved NO2 mapping – continuation Inclusion of traffic map layer in exposure estimate Weight: based on a buffer around the roads (GRIP database, source: PBL), similar as for map creation. However: We go inside 1x1 km grid in the population exposure estimate.

NO2 annual average 2013 – exposure estimate, comparison of estimate based on background (top) and final merged map (i.e. including traffic layer) Difference of population in exceedance: cc. 1.7% of European population. It is recommended to include the traffic map layer in the routine NO2 maps and exposure estimates

1. Basic mapping methodology 2. NO2 improved mapping 3. Benzo(a)pyrene improved mapping 4. Conclusion

Improved BaP mapping Motivation for improvement Current methodology gives quite poor cross-validation uncertainty results. Thus, BaP maps not prepared regularly. As the areas with the relative interpolation standard error smaller than 0.6 are presented only, the mapped domain is limited.

First BaP map (ETC/ACM TP 2014/6) BaP – annual average, 2012 (current methodology)

Improved BaP mapping Pseudo stations data approach, based on PM2.5 or PM10 data. Analysis executed on 2013 data. Detailed analysis presented in ETC/ACM Technical Paper 2016/3 „Potential improvements on benzo(a)pyrene (BaP) mapping“

Pseudo station data approach Similar methodology examined like routinely used for PM2.5 mapping: BaP estimates at PM (either PM10 or PM2.5) station locations without BaP measurement are based on actual PM10 resp. PM2.5 measurement data and different supplementary data (like coordinates and meteorology), using multiple linear regression: Prior to linear regression, a logarithmic transformation on measurement data can be applied. In such case, a back-transformation of the estimated values has to be performed.

Pseudo station data analysis For PM10 and PM2.5 variants, both with and without lognormal transformation, MLR applied, including data selection: For selected four variants, uncertainty analysis executed, leading into the choose of variants with logarithmic tranformation:

PM2.5 station data – original and pseudo based on PM10

PM2.5 station data – original and pseudo based on PM2.5

Comparison of current and improved methodology Current method compared with the mapping using pseudo stations in two variants (based on PM10, based omn PM2.5), and additionally with the current method using PM2.5 map layers as a supplementary variable. At first, the variants compared by cross-validation: Inclusion of pseudo data gives better results compared to current method (better for rural areas and similar for urban areas). For urban areas, the best is alternative current method.

Comparison of current and improved methodology Next to the cross-validation, the methods were compared by the means of their relative interpolation standard error. The best results are given by methods using pseudo stations based on . The method using pseudo stations based on PM10 gives better results compared to the method using pseudo stations based on PM2.5. The main reason is in the number of the stations: the use of the pseudo stations in the interpolation decreases the interpolation error.

BaP annual avg., 2013 – current method

BaP annual avg., 2013 – method using pseudo stations based on PM10

BaP annual avg., 2013 – method using pseudo stations based on PM2.5

BaP annual avg., 2013 – alternative current method

Comparison of current and improved methodology The method (ii) using pseudo stations based on PM10 shows the largest area with the acceptable uncertainty, while the method (iv) using PM2.5 rural and urban background maps shows the best cross-validation results. For the time being with the lack of the BaP stations, the most promising it seems to be the method using pseudo stations based on PM2.5. Furthermore, in order to increase spatial coverage, these could be complemented with pseudo stations based on PM10 at locations with PM10 measurements only.

1. Basic mapping methodology 2. NO2 improved mapping 3. Benzo(a)pyrene improved mapping 4. Conclusion

Conclusions Inclusion of land cover and road data in NO2 background mapping brings clear improvement of the uncertainty results. Based on this, it is recommended to implement these supplementary data sources in the routine methodology. Next to this, it is recommended to include the traffic map layer into the routine NO2 maps and exposure estimates. BaP maps created with the pseudo stations (either based on PM10 or PM2.5 data) give better uncertainty results compared to the current method (both in the means of cross-validation and their relative interpolation standard error). Most promising – method using pseudo stations based on PM2.5, potentially complemented with pseudo stations based on PM10 at locations with PM10 measurements only.

Thank you for your attention.