Research Article: Multi-pollutant surface objective analyses and mapping of air quality health index over North America

Date Published: January 7, 2016

Publisher: Springer Netherlands

Author(s): Alain Robichaud, Richard Ménard, Yulia Zaïtseva, David Anselmo.

http://doi.org/10.1007/s11869-015-0385-9

Abstract

Air quality, like weather, can affect everyone, but responses differ depending on the sensitivity and health condition of a given individual. To help protect exposed populations, many countries have put in place real-time air quality nowcasting and forecasting capabilities. We present in this paper an optimal combination of air quality measurements and model outputs and show that it leads to significant improvements in the spatial representativeness of air quality. The product is referred to as multi-pollutant surface objective analyses (MPSOAs). Moreover, based on MPSOA, a geographical mapping of the Canadian Air Quality Health Index (AQHI) is also presented which provides users (policy makers, public, air quality forecasters, and epidemiologists) with a more accurate picture of the health risk anytime and anywhere in Canada and the USA. Since pollutants can also behave as passive atmospheric tracers, they provide information about transport and dispersion and, hence, reveal synoptic and regional meteorological phenomena. MPSOA could also be used to build air pollution climatology, compute local and national trends in air quality, and detect systematic biases in numerical air quality (AQ) models. Finally, initializing AQ models at regular time intervals with MPSOA can produce more accurate air quality forecasts. It is for these reasons that the Canadian Meteorological Centre (CMC) in collaboration with the Air Quality Research Division (AQRD) of Environment Canada has recently implemented MPSOA in their daily operations.

Partial Text

Air quality (AQ), like weather, can affect everyone, but responses differ depending on the sensitivity and health condition of a given individual. Breathing clean air is an important aspect of quality of life (European Environment Agency–World Health Organization (EEA-WHO) 2002). Policy makers require more and more detailed AQ information to take measures to improve or mitigate the impacts of AQ, while epidemiologists seek more accurate exposure estimates to evaluate the health risk (Van de Kassteele 2006). A large number of studies have been published describing the role of air pollution in inducing or exacerbating disease. Health effects related to air pollution include eye irritation, asthma, chronic obstructive pulmonary disease (COPD) heart attacks, lung cancer, diabetes, premature death and damage to the body’s immune, neurological, and reproductive systems (Pope et al. 2002; EEA-WHO 2002; WHO 2003; Sun et al. 2005; Ebtekar 2006; Pope and Dockery 2006; Georgopoulos and Lioy 2006; Institute for Risk Research 2007; Reeves 2011; Crouse et al. 2015). It has recently been estimated, using coupled climate-chemistry global models with concentration-response functions, that up to 3.7 million premature deaths occur annually worldwide due to outdoor air pollution as compared to a reference year before widespread industrialization, i.e. year 1850 (Silva et al. 2013). Similar results have been found by the Global Burden Study 2010 who places outdoor air pollution among the top 10 risks worldwide (Lim et al. 2012). According to a study by the Canadian Medical Association (CMA 2008), 21,000 premature deaths due to air pollution occurred in the year 2008 with 11,000 hospital admissions, 620,000 doctor visits at a cost to the Canadian society evaluated at more than 8 billion dollars. Therefore, it is important to provide effective tools for assessing the quality of the air at any given time and everywhere in Canada. However, it is impossible to get a comprehensive overview of pollutant concentrations over large territories based only on ground-based measurements (Van de Kassteele 2006). To achieve this task, data fusion of observations and models are required. Knowledge of multi-pollutant concentrations in near real time is the first step towards a total environmental risk monitoring system (Georgopoulos and Lioy 2006; Institute for Risk Research 2007). Since pollutants can also behave as passive atmospheric tracers, they can give information about their dispersion and provide links to synoptic and regional meteorological phenomena. Multi-pollutant surface objective analyses (MPSOAs) could also be used to build air pollution climatology, compute local and national trends in air pollutants, and detect AQ model systematic biases (see Robichaud and Ménard 2014a, thereafter RM14a). Finally, initializing numerical AQ models at regular time intervals with MPSOA can produce more accurate air quality forecasts (Blond et al. 2004; Wu et al. 2008; Tombette et al. 2009; Sandu and Chai 2011; Silibello et al. 2014; Robichaud and Ménard 2014b, thereafter RM14b). This is the motivation behind the implementation at Canadian Meteorological Centre (CMC) of the MPSOAs described here. Moreover, MPSOAs are considered an important tool for environmental surveillance since they provide users (e.g. public, air quality forecasters, and epidemiologists) with a more accurate picture of the true state of air quality in the form of geographical maps of chemical species. The pollutants under study here are ozone, particulate matter (PM), and nitrogen dioxide (NO2). Note that PM2.5 (particulate matter with aerodynamic diameter less than 2.5 μm) ozone and nitrogen dioxide (NO2) are also inputs to the Canadian Air Quality Health Index (AQHI) (Stieb et al. 2008).

In this study, Environment Canada’s air quality model, Global Environmental Multi-scale coupled with Model of Air quality and Chemistry (GEM-MACH) version 1.3.8.2, has been used to produce the “first-guess” forecast. The output of this forecast is blended with surface observations to produce the MPSOA. This air quality model is part of the Canadian Air Quality Regional Deterministic Prediction System (AQRDPS) with a spatial resolution of 10 km (Moran et al. 2012). The objective analysis exploits air quality surface observations from the US Aerometric Information Retrieval Now (AIRNow) program, as well as Canadian observations measured in real time by the provinces and territories (and some municipalities). Figure S1 (supplementary material S1a) depicts the flow chart of the production of the surface Regional Deterministic Air Quality Analysis (RDAQA) in an operational environment at CMC. The observations are acquired in real time (get_obs) and are combined with a first-guess model forecast (get_fcst). The observations are passed through a series of quality controls to check for (a) exceedances of maximum and minimum concentration values, (b) dubious hourly jump detection, and (c) background check of observed-minus-forecast increments (module background check and get_bgcksfc). Details of the quality control algorithm are given in supplementary material S1b. Optimum interpolation uses an exponential decay function over distance (see below), and a first estimate of the error statistics (error variance matrices) is obtained from the Hollingsworth and Lönnberg’s method (Hollingsworth and Lönnberg 1986, thereafter HL86; Lönnberg and Hollingsworth 1986). In RM14a, it was found that a scaling of both the correlation length (mostly deflation) and the background error variance (mostly inflation) had to be done in order to improve the performance of both the bias and error variance of the analyses whenever HL86 method was used. Furthermore, whenever HL86 method is inapplicable (whenever the data is too noisy or too many observations are missing), we have followed Silibello et al. (2014) (thereafter S14) with some modification to deduce background error variance (see below for more details). Although the error correlations are modeled as homogeneous and isotropic, the spatiotemporal variability of the background error variance is taken into account which reflects the intrinsic variability of the surface pollutant concentrations at a given station. A regional bias correction could also be applied for any pollutants depending on the situation (see below). The production of objective analysis is done in the module analsfc and is output as a four-panel product (submodule Four_Panel_Images in Fig. S1a). One question which arises at this point is how to interpolate spatial AQHI values to produce maps. The module AQHI computes the air quality index according to the following formula (Stieb et al. 2008):1documentclass[12pt]{minimal}
usepackage{amsmath}
usepackage{wasysym}
usepackage{amsfonts}
usepackage{amssymb}
usepackage{amsbsy}
usepackage{mathrsfs}
usepackage{upgreek}
setlength{oddsidemargin}{-69pt}
begin{document}$$ mathrm{AQHI}=frac{10}{10.4}*left[ 100*right(Big( exp left(0.000871*{mathrm{NO}}_2Big)-1right) + left( exp left(0.000537*{mathrm{O}}_3Big)-1right) + left( exp left(0.000487*{mathrm{PM}}_{2.5} Big)-1right) right)right] $$end{document}AQHI=1010.4*[100*((exp0.000871*NO2)−1+exp0.000537*O3)−1+exp0.000487*PM2.5)−1

The new objective analyses are presented as a four-panel image. Figure 3a–d shows the results of the proposed (RDAQA system) for ozone, PM2.5, PM10, and NO2, respectively. For each pollutant, the four-panel image provides the model trial field in the top left panel, OA in the top right (fusion of model and observations), and analysis increments in the bottom left (or correction to the model computed by the second term on the right-hand side of Eq. 3). Finally, in the bottom right panel, observations used in the analysis are presented. The units are in ppbv for ozone and NO2 and in units of μg/m3 (micrograms per cubic meter) for PM2.5 and PM10. It is important to note that at the resolved scale (approximately four times the numerical resolution of 10 km, i.e., effective resolution of about 40 km for OA and model grid), certain local conditions such as titration of ozone by NO2 (due to local traffic), individual point sources such as pollution plumes originating from chimneys (in the case of PM2.5), or from other point source are neither correctly resolved by the model nor the analysis.

Air quality, like weather, affects everyone but quite differently depending on the sensitivity and health condition of a given individual. High-resolution MPSOAs are important since they provide users (e.g., policy makers, public, epidemiologists, and air quality forecasters) with a more accurate and detailed picture of the true state of a given chemical species as compared to mapping based on observations or model output alone. Knowledge of multi-pollutant concentrations in near real time is a step towards a total environmental risk monitoring system. Models are generally characterized by known deficiencies for prediction of many pollutants, whereas measurement systems suffer from representativeness problems and lack of sufficient coverage and, thus, are often best suited to providing local air quality information. However, the OI technique, used in operational meteorology for decades, provides an optimal framework to extract the maximum information of both model predictions and observations (Rutherford 1972; Daley 1991; Kalnay 2003; Brasnett 2008). The OA used in this study combines model outputs from the Canadian air quality forecast suite with the US and Canadian observations from various air quality surface monitoring networks. The analyses are based on an OI with an explicit bias correction for pollutants (ozone, PM2.5, PM10, and NO2). The estimation of error statistics has been computed using a modified version of the HL86 and the use of the work of Silibello et al. (2014) to compute observation error variance whenever HL86 was not applicable (too noisy or sparse density of stations). Based on the results obtained in RM14a, using a χ2 (chi-squared) diagnostic (Ménard and Chang 2000), the correlation length obtained by HL86 method was found to be too long and needs to be deflated. Better results were found using a prescribed correlation length similar to that prescribed by S14 which is also consistent with Sandu and Chai (2011) who uses a fixed value of 60 km for correlation length for ozone. Successful cross-validation experiments were performed with an OA setup using a subset of observations to build the objective analyses and with the remainder left out as an independent set of data for verification purposes. A new operational product (called RDAQA) has been implemented at CMC. These analyses fill a gap in the operational suite at CMC.

 

Source:

http://doi.org/10.1007/s11869-015-0385-9