Elsevier

Atmospheric Environment

Volume 44, Issue 10, March 2010, Pages 1341-1355
Atmospheric Environment

A new method to estimate air-quality levels using a synoptic-regression approach. Part I: Present-day O3 and PM10 analysis

https://doi.org/10.1016/j.atmosenv.2009.06.029Get rights and content

Abstract

In order to make projections for future air-quality levels, a robust methodology is needed that succeeds in reconstructing present-day air-quality levels. At present, climate projections for meteorological variables are available from Atmospheric-Ocean Coupled Global Climate Models (AOGCMs) but the temporal and spatial resolution is insufficient for air-quality assessment. Therefore, a variety of methods are tested in this paper in their ability to hindcast maximum 8 hourly levels of O3 and daily mean PM10 from observed meteorological data. The methods are based on a multiple linear regression technique combined with the automated Lamb weather classification. Moreover, we studied whether the above-mentioned multiple regression analysis still holds when driven by operational ECMWF (European Center for Medium-Range Weather Forecast) meteorological data. The main results show that a weather type classification prior to the regression analysis is superior to a simple linear regression approach. In contrast to PM10 downscaling, seasonal characteristics should be taken into account during the downscaling of O3 time series. Apart from a lower explained variance due to intrinsic limitations of the regression approach itself, a lower variability of the meteorological predictors (resolution effect) and model deficiencies, this synoptic-regression-based tool is generally able to reproduce the relevant statistical properties of the observed O3 distributions important in terms of European air quality Directives and air quality mitigation strategies. For PM10, the situation is different as the approach using only meteorology data was found to be insufficient to explain the observed PM10 variability using the meteorological variables considered in this study.

Introduction

Since high concentrations of O3 and PM10 affect the public health, much attention is paid to the improvement of the accuracy of short-term deterministic and statistical prediction models and the development of robust long-term air-quality prediction systems. Complex models, including a full description of atmospheric chemistry and meteorological processes, are often used with respect to the former (Giorgi and Meleux, 2007). Although these techniques are shown to be powerful for short-term predictions, the complex climate–air-quality modelling systems, together with their computational/technical characteristics, are at present less useful for long-term (decadal) predictions (Giorgi and Meleux, 2007).

Therefore, statistical downscaling methods were developed to determine predictive relationships between air pollution concentrations and individual meteorological parameters. This was done for different air-quality variables and different locations. Different methods can be distinguished namely multiple linear regression (MLR) analysis (Barrero et al., 2006), nonlinear multiple regressions (Cobourn, 2007), artificial neural networks (ANN) (e.g. Hooyberghs et al., 2005), and generalized additive models and fuzzy-logic-based models (Cobourn et al., 2000). Among these approaches, regression methods are well documented because of their ease of implementation and their low computation requirements (Wilby and Wigley, 1997, Wilby et al., 2004). The technique first detects present-day relationships between local meteorological variables (or “predictors”) and e.g. air-quality variables (or “predictands”). These relations based on (non) linear multiple regressions have been described in literature numerously, as for e.g. in Chaloulakou et al. (2003) and Ainslie and Steyn (2007). Nevertheless, the use of another circulation patterns as a downscaling tool is widespread, although less common in air-quality research. In that respect, this technique is adopted by e.g. Comrie and Yarnal (1992) and Davies et al. (1992) to explain observed ozone variability at measuring sites in the Europe, US and Canada. Some authors have tried to combine the above-mentioned two downscaling techniques. A stratification based on the circulation pattern is adopted to introduce nonlinearity into the model (Huth et al., 2008a, Huth et al., 2008b) under the assumption that the relations between large-scale predictors and predictands may vary depending on the type of the synoptic pattern. This technique is applied for downscaling surface meteorological variables by e.g. Cavazos (1999). Nevertheless, this approach has not been adopted in terms of air quality assessment yet. To the author's best knowledge, there has been only one study so far that used a within-synoptic-type air pollution model to study future air pollution levels for a variety of pollutants (Cheng et al., 2007a, Cheng et al., 2007b).

The aim of our study is to test a simple linear regression method together with a stratification of the dataset by its daily synoptic patterns in their ability to hindcast levels of O3 and PM10. To achieve this goal, regression results from previous research based on station measurements from Cabauw (Demuzere et al., 2009) are used to reconstruct the observed maximum eight hourly mean O3 [μg m−3] (hereafter referred to as m8O3) and daily mean PM10 [μg m−3] levels for the period 2001–2004 (calibration) and 2005–2006 (evaluation). Furthermore, the Lamb weather type classification is used as a synoptic circulation-typing tool to enter nonlinearity into the downscaling model in order to construct a robust method for the improved projections of air-quality levels. These methods are calibrated on observed air quality data from the Cabauw measurement tower and the measurement station of Zegveld-Oude Meije (The Netherlands) for the period 2001–2004.

According to the Intergovernmental of Climate Change, the magnitude of the effect of climate change on O3 is still uncertain (IPCC, 2001). Hence, before the above-mentioned downscaling approach is used to bridge the gap between what is produced by AOGCMs (Atmospheric–Ocean Coupled Global Climate Models) and what is needed in climate impact research (Part II of this analysis), the observed local relations between meteorological and air-quality variables need to be tested on the larger scale. After all, AOGCM-output is at present only available typically scales of 300 × 300 km2. Therefore, the validity of these station-based synoptic-regression configurations as an air quality downscaling tool is tested using low-resolution operational ECMWF (European Center for Medium-Range Weather Forecast) data for the period 2005–2006. In this way, a regression-based model forced with meteorological ECMWF data is used as a prototype for a modelling system in which AOGCM-output is downscaled for the purpose of obtaining projections for future air-quality levels.

Section snippets

Data

High temporal resolution meteorological data for the period 2001–2006 from the rural measurement station of Cabauw (The Netherlands), partly operated by the KNMI (Royal Dutch Meteorological Institute were used. Ten-minute measurements are averaged to daily values. More details on the measurement site characteristics and quality control are provided in Demuzere et al. (2009). The calibration of the multiple linear regression equation is based on local observations of 2 m air and dew point

Methods

Multiple linear regression models have widely been adopted to reconstruct observed time series of O3 and PM10 for various heterogeneous regions based on measurements (Barrero et al., 2006). In addition, Huth et al., 2008a, Huth et al., 2008b points out the power of a pointwise linear regression method (using grid point values instead of principal components as predictors) in comparison with nonlinear methods such as neural networks. And although Huth et al., 2008a, Huth et al., 2008b do not

Results and discussion

First, the various multiple linear regression equations described in Section 3 are calibrated for the period 2001–2004, using observational data (Section 4.1). Secondly, it is tested which of the methods is most suitable for hindcasting m8O3 and daily mean PM10 concentrations for the independent evaluation time period 2005–2006, using observed meteorological data as input to the regression-based model. Thirdly, it is tested whether the MLR equations derived from observed meteorological data

Discussion

Deficiencies in the regression-based models can be due to 1) an inadequate representation of the predictor variables and 2) the fact that most downscaling methods tend to resolve only part of the total variance (Barrero et al., 2006). In the previous section we found substantial differences between ECMWF data and observed meteorological variables. This difference can be due to 1) deficiencies in the ECMWF model or 2) the low spatial resolution used in our analysis. After all, in order to test

Conclusion

The primary aim of this paper is to evaluate a variety of regression-based methodologies to hindcast levels of m8O3 and PM10 from meteorological predictors. In order to quantify the performance of the regression-based methods, several statistical indices are used besides the common first and second order moments: fit between modelled and observed series using the explained variance, shape of the distribution in terms of skewness and kurtosis and the performance against a persistence reference

Acknowledgments

This research is funded by a PhD grant of the Institute for the Promotion of Innovation through Science and Technology Flanders (IWT-Flanders). Furthermore, it was conducted in the framework of the CLIMAQS project, with financial support of the Institute for the Promotion of Innovation by Science and Technology in Flanders (IWT-Flanders). ECMWF is acknowledged for providing operational ECMWF data. Radan Huth and Ricardo Trigo are acknowledged for their useful comments on this manuscript.

References (31)

  • C.S.Q. Cheng et al.

    A synoptic climatological approach to assess climatic impact on air quality in south-central Canada. Part I: historical analysis

    Water Air and Soil Pollution

    (2007)
  • C.S.Q. Cheng et al.

    A synoptic climatological approach to assess climatic impact on air quality in south-central Canada. Part II: future estimates

    Water Air and Soil Pollution

    (2007)
  • W.G. Cobourn et al.

    A comparison of nonlinear regression and neural network models for ground-level ozone forecasting

    Journal of the Air & Waste Management Association

    (2000)
  • T.D. Davies et al.

    Surface ozone concentrations in Europe – links with the regional-scale atmospheric circulation

    Journal of Geophysical Research – Atmospheres

    (1992)
  • M. Demuzere et al.

    An analysis of present and future ECHAM5 pressure fields using a classification of circulation patterns

    International Journal of Climatology

    (2008)
  • Cited by (0)

    View full text