How do air quality models perform with different validation datasets and different spatial resolution setups?
0 likes106 views
The document discusses the validation of air quality models, specifically Land Use Regression (LUR) and Atmospheric Dispersion (AD) models, against multiple datasets to assess their performance and spatial resolution. It emphasizes that validation metrics can vary significantly based on the model and dataset used, and highlights the importance of high spatial resolution for accurate air pollution exposure estimates. The findings have implications for epidemiological studies and health impact assessments by suggesting that location proxies may not represent actual exposure accurately.
How do air quality models perform with different validation datasets and different spatial resolution setups?
- 1. 1 Center for Advancing Research in Transportation Emissions, Energy and Health U.S. Department of Transportation University Transportation Centers Program HaneenKhreis, PhD www.carteeh.org
- 2. 2 “The Impact of Different Validation Datasets on Air Quality Modeling Performance”
- 3. 3 •Not possible to make sufficient air pollution exposure measurements for epidemiological and health impact assessments 143,472 children •Many studies rely on air pollution modeling •Commonly used models include: • Land use regression (LUR) • Atmospheric dispersion (AD) Background
- 4. 4 LUR modeling • Repeated measurements (passive samplers) at N sites • Average measurements over longer term and adjust for temporal variations • Regression model to combine measurements with GIS-based predictors • Apply model to non- measured locations
- 5. 5 LUR modeling • BUILDINGS Local land use, Area/number of buildings, m2/N(umber) • TRAFLOAD Local road network, Total traffic load of all roads in a buffer (sum of (traffic intensity*length of all segments)), Veh. Day-1 m • NATURAL Semi-natural and forested areas, m2 • HEAVYTRAFMAJOR Heavy-duty traffic intensity on nearest major road, Veh. Day-1
- 7. 7 •Models are only validated using one validation dataset •Their estimates at select receptor points are sometimes generalized to larger areas •This may lead to unsatisfactory validation and/or inaccurate insights about the models’ performance and suitability for large-scale application Background
- 8. 8 •Objective 1 explore the effect of different validation datasets on the validation results of two commonly used air quality models •Objective 2 explore the effect of the model estimates’ spatial resolution on the models’ validity at different locations Objectives
- 9. 9 Study area
- 10. 10 Annual (2009) NO2 and NOx LUR model Validation against 4 different datasets Spatial resolution analysis Estimate at exact location of validation point Estimates at centroid of 100x100m grid in which validation point fell AD model Validation against 4 different datasets Spatial resolution analysis Estimates at exact locations of validation point Estimates at centroid of 100x100m grid in which validation point fell Methods
- 11. 11 Measurement campaign and dataset (n = 126) Pollutants measured Measurement device Year and time interval for final dataset Locations and purpose of measurements ESCAPE diffusion tubes (n=41) NO2 and NOx Ogawa badges 2009 (annualized) At the façade of homes of study subjects as the primary objective of the ESCAPE project was to characterize residential exposures and associated health CBMDC diffusion tubes (n=29) NO2 “Diffusion tubes” 2009 (annualized) Three sites were not close to main road whilst the rest were kerbside sites at 0.5-5m from the nearest road, monitoring undertaken to review and assess air quality progress de Hoogh diffusion tubes (n=48) NO2 Palmes tubes Four 2-week periods during 2007-2008 Close to the front door of 48 homes of study subjects from the Born in Bradford cohort to characterize their residential exposures and compare with future ESCAPE work CBMDC fixed-site monitoring (n=8) NO2 Automatic urban network chemiluminescence 2009 (annualized) Two sites were classified as urban background whilst the rest were kerbside sites at 1.5-2 m from the nearest road, monitoring undertaken to review and assess air quality progress Methods
- 12. 12 Annual (2009) NO2 and NOx LUR model Validation against 4 different datasets Spatial resolution analysis Estimate at exact location of validation point Estimates at centroid of 100x100m grid in which validation point fell AD model Validation against 4 different datasets Spatial resolution analysis Estimates at exact locations of validation point Estimates at centroid of 100x100m grid in which validation point fell Methods
- 13. 13 Results: validation against different datasets Models combination Validation dataset ESCAPE NOx diffusion tubes (n=41) ESCAPE NO2 diffusion tubes (n=41) CBMDC NO2 diffusion tubes (n=29) De Hoogh NO2 diffusion tubes (n=48) CBMDC NO2 fixed-site monitoring (n=8) ADmodel COPERT dispersion model NOx at points (varying background) R2 = 0.30 COPERT dispersion model NO2 at points (varying background) R2 = 0.33 R2 = 0.20 R2 = 0.59 R2 = 0.24 LURmodel NOx LUR estimates at points R2 = 0.58 NO2 LUR estimates at points R2 = 0.54 R2 = 0.21 R2 = 0.61 R2 = 0.38 (r= 0.62)
- 14. 14 AD vs. LUR annual average NO2/NOx Estimates (µg/m3) at 46,452 specified output points centering each 100m x 100m grid across 40 * 33 km Results: spatial resolution of estimates
- 15. 15 Results: spatial resolution of estimates Validation dataset ESCAPE NOx diffusion tubes (n=41) ESCAPE NO2 diffusion tubes (n=41) CBMDC NO2 diffusion tubes (n=29) de Hoogh NO2 diffusion tubes (n=48) CBMDC NO2 fixed-site monitorin g (n=8) LURmodels NOx LUR estimates at points R2= 0.58 NOx LUR estimates at raster R2= 0.35 NO2 LUR estimates at points R2= 0.54 R2= 0.21 R2= 0.61 R2= 0.38 (r= 0.62) NO2 LUR estimates at raster R2= 0.31 R2= 0.06 R2= 0.32 R2= 0.38 (r=- 0.61) -23%
- 16. 16 Results: spatial resolution of estimates Validation dataset ESCAPE NOx diffusion tubes (n=41) ESCAPE NO2 diffusion tubes (n=41) CBMDC NO2 diffusion tubes (n=29) de Hoogh NO2 diffusion tubes (n=48) CBMDC NO2 fixed-site monitorin g (n=8) LURmodels NOx LUR estimates at points R2= 0.58 NOx LUR estimates at raster R2= 0.35 NO2 LUR estimates at points R2= 0.54 R2= 0.21 R2= 0.61 R2= 0.38 (r= 0.62) NO2 LUR estimates at raster R2= 0.31 R2= 0.06 R2= 0.32 R2= 0.38 (r=- 0.61) -23% -15% -29%
- 17. 17 • LUR and AD models validated against four different datasets Summary and discussion
- 18. 18 • LUR and AD models validated against four different datasets • LUR and AD model estimates made at different spatial resolution and validated against four different datasets Summary and discussion
- 19. 19 • LUR and AD models validated against four different datasets • LUR and AD model estimates made at different spatial resolution and validated against four different datasets • The validation metrics varied substantially (R2 0.20 – 0.61) based on: • which model was used • which validation dataset was used • whether exposure estimates were made at exact validation point or at centroid of containing grid Summary and discussion
- 20. 20 • LUR and AD models validated against four different datasets • LUR and AD model estimates made at different spatial resolution and validated against four different datasets • The validation metrics varied substantially (R2 0.20 – 0.61) based on: • which model was used • which validation dataset was used • whether exposure estimates were made at exact validation point or at centroid of containing grid • Validation results based on actual points’ locations were generally much better than at a grid level Summary and discussion
- 21. 21 • There is value of validating modeled air quality data against various datasets Conclusions
- 22. 22 • There is value of validating modeled air quality data against various datasets • The spatial resolution of the models’ estimates has a significant influence on the validity at the application point (even at 100m level) Conclusions
- 23. 23 • There is value of validating modeled air quality data against various datasets • The spatial resolution of the models’ estimates has a significant influence on the validity at the application point (even at 100m level) • Have implications for epidemiological studies disregarding time-activity patterns or using location proxies Conclusions
- 24. 24 • There is value of validating modeled air quality data against various datasets • The spatial resolution of the models’ estimates has a significant influence on the validity at the application point (even at 100m level) • Have implications for epidemiological studies disregarding time-activity patterns or using location proxies • Have implications for health impact assessments where estimates of air quality models at select points are extrapolated to larger areas/populations Conclusions
- 25. 25 • There is value of validating modeled air quality data against various datasets • The spatial resolution of the models’ estimates has a significant influence on the validity at the application point (even at 100m level) • Have implications for epidemiological studies disregarding time-activity patterns or using location proxies • Have implications for health impact assessments where estimates of air quality models at select points are extrapolated to larger areas/populations • Can improve understanding of the most influential uncertainties/errors across full-chain health impact assessment Conclusions
- 26. 26