Air Quality Assessment and Management Land use patterns and spatial interpolation of air pollution monitoring data Stijn Janssen1, Clemens Mensink1, Gerwin Dumont2 and Frans Fierens2 (1) VITO, Belgium (2) IRCEL, Belgium 11th EIONET Workshop on Air Quality Assessment and Management La Rochelle, October 26-27, 2006
Introduction Belgium is a rather small country with highly urbanized regions Air quality is sampled by a dense network of monitoring sites e.g. more than 50 stations for NO2 and PM10 in Belgium Air quality is forecasted by statistical models forecasts for monitoring sites only For monitoring and forecast data: pollution levels are only representative for POINT locations
Introduction Real-time measurements and daily forecasts are published one-line by IRCEL Need for reliable maps to inform the public How to interpolate point values to an air quality map? ?
Introduction
RIO-model: Methodology Observation: Sampling values depend on land use in (direct) vicinity of the monitoring site Conclusion: Interpolation schema needs to know this relation between land use and air quality levels Approach in RIO-model: Quantify this at a statistical level (mean and standard deviation) Create land use indicator to express relation
RIO-model: Land use indicator 2 km Land use indicator For each station: Determine buffer (e.g. 2km radius) Characterize land use by CORINE class distribution inside buffer
RIO-model: Land use indicator Land use indicator is based on CORINE class distribution Calibration of coefficients ai: multi-regression to optimize trend for mean and standard dev. of monitoring data <NO2>
RIO-model: Trends Trends in mean and standard dev. of sampling values: sNO2 <O3> <NO2> <PM10> sO3 sPM10
week/weekend variations RIO-model: Trends Hourly and week/weekend variations in trends hourly variations week/weekend variations
RIO-model: Detrending Use relation between land use indicator and AQ statistics to “detrend” monitoring data: Remove local character of sampling values
RIO-model: Interpolation Detrended sampling values are interpolated with ordinary Kriging Correlation function R(r) based on historical time series Much more information than in standard Kriging
RIO-model: Spatial correlation Spatial correlation functions depend on: Pollutant (O3, NO2, PM10,…) Aggregation value (day avg, max1h, max8h, hourly value,…)
RIO-model: Methodology For each grid cell: Interpolate detrended values with Kriging Determine local bCORINE-value Get corresponding trend shift Add trend to interpolation result
RIO-model: Validation Validation: leaving-one-out. Compare with standard IDW O3 NO2 PM10
RIO: Results Real-time NO2 concentrations for 25/03/2003 (max 1h values) IDW RIO
RIO-model: Results Real-time PM10 concentrations for 24/03/2005 (day average values) IDW RIO
RIO-model: Results Year average O3 concentrations for 2002 IDW RIO
RIO-model: Results Year average NO2 concentrations for 2002 IDW RIO
RIO-model: Results Year average PM10 concentrations for 2002 IDW RIO
RIO-model: Results Year average PM10 concentrations for 2005 IDW RIO
MODIS AOD 2003, Koelemeijer et al, Atm. Env. 40, 5304, 2006 RIO-model: Validation Validation: Compare PM10-map with Aerosol Optical Depth satellite measurements (2003) MODIS AOD 2003, Koelemeijer et al, Atm. Env. 40, 5304, 2006
RIO-model: Extensions Relation between land use indicator and air quality statistics can be used for: Assessment of spatial representativeness of monitoring locations Downscaling of model results: redistribution of model concentration inside grid cell Land use indicator can be optimized with satellite data Aerosol Optical Depth for PM10
Conclusion RIO is an interpolation scheme for ambient air pollution (O3, NO2,PM10 ,… ) Applicable for historical, real-time or forecast values Kriging is used as interpolation tool Land use model is applied to incorporate local patterns Detrending is essential step for interpolation of air quality values