Supplementary material
Yüklə 262,86 Kb.
|
- Bu səhifədəki naviqasiya:
- CONTENTS
- 2 Results S15
- References
- 1.2 Surface NO 2 estimates
- Parameterisation Selected Configuration Item
- 1.3 Land-use variables Table S2
- Variable
|
SUPPLEMENTARY MATERIAL A national satellite-based land-use regression model for air pollution exposure assessment in Australia Luke D. Knibbs1* Michael G. Hewson2 Matthew J. Bechle3 Julian D. Marshall3 Adrian G. Barnett4
* Corresponding author Public Health Building The University of Queensland Herston, QLD 4006, Australia p: +61 7 3365 5409 e: l.knibbs@uq.edu.au
The Ozone Monitoring Instrument (OMI) observes atmospheric column density of NO2 on a daily basis. It was launched in July 2004 and is one of the four instruments on board the NASA Earth Observing System Aura satellite designed to measure atmospheric trace gases and aerosol optical properties. The Aura satellite crosses the equator in a sun-synchronous polar orbit at approximately 13:30 hours local time for the daylight ascending orbit (Torres et al., 2007) and it passes over Australia in the mid- to late-afternoon. OMI measures top of atmosphere radiance from 270-500 nm in the ultraviolet and visible regions of the solar spectrum with a spatial resolution at nadir of 13 × 24 km (Levelt et al., 2006). OMI NO2 tropospheric column amount data (cloud screened at 30%) is available via the NASA Giovanni Aura/OMI online visualisation and analysis web site (http://gdata1.sci.gsfc.nasa.gov/daac-bin/G3/gui.cgi?instance_id=omi) (Acker & Leptoukh, 2007). The level 3 NO2 concentration product, OMNO2d.003, is a daily, global dataset gridded at 0.25 × 0.25 degrees spatial resolution with units of 1015molecules/cm2. Briefly, the OMI-derived NO2 algorithm: (1) takes radiance values in ultraviolet and visible bands known to absorb NO2; (2) applies radiometric and de-striping corrections to column concentrations, and then; (3) calculates both tropospheric and stratospheric contributions to the atmospheric column NO2 (for pixels which have less than 30% cloud cover) by incorporating satellite path slant corrections and a range of air mass factors, and; (4) converts the data from satellite swath grid to a consistent grid size (Bucsela et al., 2013). The NO2 concentration at any given grid point is a weighted average of a number of OMI measurements, given the grid points are calculated from OMI swath pixels which are 13 x 24 km at nadir but become larger toward the swath edge. The contribution of individual OMI pixels to NO2 concentrations above a given geographic location is a function of the daily variance in Aura ephemeris. While OMI tropospheric columns from the standard product are prone to seasonal bias (Lamsal et al., 2010), surface NO2 estimates derived from OMI columns are well-correlated with corrected ground level measurements, with bias under 30% and without substantial seasonal variation (Lamsal et al., 2008; Lamsal et al., 2010; Novotny et al., 2011). Notwithstanding the confines of algorithms and a dynamic radiometric row anomaly more evident since January 2009, OMI NO2 data sets offer consistent quality for quantitative investigations of the relationships between pollution, its sources, and populations (Lamsal et al., 2013). 1.2 Surface NO2 estimates Data source We downloaded and subset the global daily average OMI tropospheric NO2 data using NASA Giovanni website functions, for: (1) a spatial domain encompassing the Australian continent; and; (2) temporal aggregation to each calendar month from January 2006 through December 2011. The OMI row anomaly issue is normalised in the present study by the level 3 OMI NO2 concentration product weighted average grid point calculation algorithm, as well as our selection of mean daily NO2 concentration time averaged per month. Modelling surface and tropospheric NO2 We estimated the ground level NO2 concentrations from the OMI observed tropospheric column NO2 concentrations by determining surface-to-column ratios across Australia using modelled ground-level and tropospheric NO2 levels (Novotny et al., 2011). We used a gas phase chemistry and aerosol transport version of the Weather Research and Forecasting (WRF) model; WRF-Chem. WRF-Chem is a multi-scale, “on-line” fully connected, atmospheric chemistry edition of the WRF non-hydrostatic, fully compressible, community meteorological model (Grell et al., 2005). The fully coupled nature of the chemistry transport and meteorology applications means that during computation, the same transport schemes, horizontal and vertical grids, physics schemes and model time steps operate together, which removes the need for inter-model interpolation. Verification trials have shown that WRF-Chem improves on the previous coupled chemistry/meteorological models from which it was developed (Grell et al. 2005). WRF-Chem requires the modeller to select combinations of computational schemes representing aerosol transport, gas-phase chemistry, atmospheric physics, cloud microphysics as well as land use energy-balance parameterisation. Furthermore, WRF-Chem configuration also involves making informed decisions on which meteorological boundary conditions, aerosol emissions (anthropogenic, biogenic and background), model domain sizes and grid spatial resolution to use. These model component choices are made considering any computational resource constraints. In our study, 6 years of Australia-wide daily average NO2 concentrations were required, ordinarily representing an extensive computational commitment. WRF-Chem version 3.5 was structured on The University of Queensland’s Research Computing Centre (RCC) to produce the required output using 2 parallel batch runs of 64 CPUs each. With this constraint, WRF-Chem was configured using a single domain, 60 km spatial resolution, and time steps of 6 minutes to produce daily aggregated NO2 concentrations. The WRF-Chem models contained 27 vertical levels. The WRF-Chem physics scheme configurations used in this study are listed in Table S1. Table S1. WRF-Chem configuration.
WRF-Chem meteorological boundary conditions were updated every six hours by data sourced from the National Centre for Environmental Prediction (NCEP) global 1-degree database of tropospheric analyses, which is a collection of meteorological observations used in the US Global Forecast System (GFS). Each model run calculated a month of NO2 concentrations and included a model spin-up time of five days to ensure stable meteorological physics operation during the model run. The WRF-Chem aerosol and gas phase molecule emissions were provided by two global data sets; GOCART global anthropogenic and background emissions as of 2006 with a spatial resolution of 1 degree (Chin et al., 2002), and the EDGAR (Emission Database for Global Atmospheric Research) global anthropogenic emissions for 2005 with spatial resolution of 0.1 degree (Olivier et al., 2005). WRF-Chem was configured with the MADE-SORGAM (Modal Aerosol Dynamics Model for Europe - Secondary Organic Aerosol Model) aerosol transport scheme and RADM2 (Regional Acid Deposition Model version 2) gas phase chemistry scheme because they offered timely model output. MADE was developed in Europe in 1998 (Ackermann et al., 1998), with the capacity to model secondary organic aerosol (SORGAM) added in 2001 as described by Schell et al. (2001). Fast et al. (2011) concluded that MADE-SORGAM performed as well as a more advanced 8-bin sectional aerosol parameterization while being computationally cheaper. Since we used WRF-Chem to calculate an average daily ratio of surface-to-column NO2 concentration aggregated to a monthly time scale Australia-wide, the model configuration trade-offs were considered suitable. In a study comparing a similarly configured WRF-Chem that measured anthropogenic emissions for the whole of Europe (albeit with a different anthropogenic emission source database), Tuccella et al. (2012) noted that WRF-Chem NO2 replicated measured time-series NO2 within ±15%. The atmospheric chemistry model was used to create a ratio of near ground to tropospheric NO2 concentrations for each month of the study. After each WRF-Chem run, NO2 concentration at the 60 km spatial resolution was extracted and aggregated to the daily average for each calendar month of 2006 to 2011 using ‘netCDF kitchen sink’ LINUX file manipulation utilities. At each model grid-point the extraction created a netCDF file of latitude, longitude and NO2 concentration in ppmv for each model level from near ground level (i.e. surface) to the average height of the tropopause, which was set to 14 km and meant that the first 22 WRF-Chem model level NO2 concentrations were extracted to represent the troposphere (i.e. column). The tropopause height varies latitudinally, seasonally and daily due to the heterogeneous nature of heat sources over time and space – according to figures presented by Sturman and Tapper (2006), a 14 km median tropopause height over Australia is reasonable. The NO2 emission data sets were created in netCDF format because the intrinsic WRF-Chem model output file format can be read by ArcGIS (ESRI Inc., Redlands, USA), in which further OMI and model NO2 comparison computation and emissions mapping was undertaken. The OMI tropospheric NO2 data was sourced in text file format so that that ArcGIS point class shape-files could be readily produced for further processing. Estimating surface NO2 from OMI columns We used ArcGIS to determine the ratio of WRF-Chem surface-to-column NO2 at each grid point for each month of the study. We then applied this ratio to OMI tropospheric NO2 columns to elicit a daily surface average NO2 concentration at 0.25 x 0.25 degrees spatial resolution for each month from January 2006 to December 2011. Since the OMI and WRF-Chem NO2 concentration had different spatial resolutions and therefore different grid structures, the ArcGIS model firstly calculated grid-point pairs using a GIS “one to one” spatial join and intersect procedure. The maps of monthly and annually aggregated daily average OMI surface NO2 concentrations were created using kriging interpolation in ArcGIS. Of the various techniques for treating trend common in atmospheric data geostatistical analysis, we chose universal kriging with linear drift, as described by Webster and Oliver (2007). Since universal kriging utilises localised parameter mean values a logarithmic transformation of the OMI and WRF-Chem data sets was not warranted. This determination also acknowledges that the ArcGIS kriging semivariogram spatial correlation calculations using 12 points was only 0.04% of the grid points in the size of the domain – the data set mean has less influence and local clustering influences dominate the interpolation. 1.3 Land-use variables Table S2. The type and source of independent land-use variables considered in the model.
Yüklə 262,86 Kb. Dostları ilə paylaş:
|