Forecasts using neural network versus Box-Jenkins methodology for ambient air quality monitoring data

This study explores ambient air quality forecasts using the conventional time-series approach and a neural network. Sulfur dioxide and ozone monitoring data collected from two background stations and an industrial station are used. Various learning methods and varied numbers of hidden layer processing units of the neural network model are tested. Results obtained from the time-series and neural network models are discussed and compared on the basis of their performance for 1-step-ahead and 24-step-ahead forecasts. Although both models perform well for 1-step-ahead prediction, some neural network results reveal a slightly better forecast without manually adjusting model parameters, according to the results. For a 24-step-ahead forecast, most neural network results are as good as or superior to those of the time-series model. With the advantages of self-learning, self-adaptation, and parallel processing, the neural network approach is a promising technique for developing an automated short-term ambient air quality forecast system.     

A Survey Technique for Determining The Representativeness of Urban Air Monitoring Stations With Respect to Carbon Monoxide

An air quality survey technique for measuring the horizontal spatial variation of carbon monoxide concentrations in urban areas is described; it was used to determine how representative an urban air monitoring station is of concentrations throughout the city.

The survey technique was applied in San Jose, Calif., where 1128 samples were collected over a six-month period and were compared with the values recorded simultaneously at the urban air monitoring station. All samples were collected at “breathing height” within a 13-square-mile grid which included the downtown area as well as surrounding residential and industrial locations. Three basic sampling strategies were employed to answer specific questions about the distribution of carbon monoxide concentration: (7) walking sampling, in which samples were obtained while walking along the sidewalks of congested downtown streets, (2) random spatial sampling, in which samples were collected at randomly selected points in the urban grid, and (3) specialized sampling in the immediate vicinity of the air monitoring station.

The results indicate that pedestrians on downtown streets in San Jose can be exposed to concentrations above the federal air quality standards without these values being observed at the air monitoring station. There also is evidence that, at any instant of time, similar values of carbon monoxide exist throughout this city (within a 13-square mile area), provided that measurements are not made in close proximity to streets. Furthermore, the higher concentrations observed in the immediate vicinity of streets decrease quite rapidly with increasing horizontal distance from these streets.

These findings, in the view of the authors, raise serious doubts as to whether it is possible to determine if air quality standards as currently defined are actually being met in urban areas using data from present-day air monitoring stations.     

Design methodology for optimum dosage air monitoring site selection

An air monitoring site selection procedure has been developed that ranks potential air monitoring sites according to their ability to represent the ambient dosage (i.e. the product of the concentration by exposure time) pattern in a monitoring network. The Dosage Monitoring Survey Design (DMSS) analyzes the dosage impact at grid receptors by dispersion modeling. High-dosage grids become potential monitoring sites. The uniqueness of DMSS is the introduction of a cluster of contiguous grid receptors that exceed a threshold value. One station is assigned to the cluster, thus eliminating redundancies among adjacent high dosage grids. The site selection procedure specifies locations for high-dosage monitoring stations along with the cluster area capabilities of each station. An efficiency term based on the ratio of a station's dosage measuring capabilities to the total dosage in the network provides a method of ranking stations. An analysis of the design procedure shows that as the threshold concentration decreased the distance from the source where the maximum dosage was found increased and there was a slight increase in station efficiency. Also, as averaging time increased, higher efficiencies were achieved for individual stations.     

Development of Criteria for Siting Air Monitoring Stations

A critical problem arises if one attempts to compare data from air monitoring stations in different cities, because there are so many differences in monitoring site locations. Some air monitoring stations are on the 6 th and 8 th floors of tall buildings in downtown areas; some are at ground level beside streets with heavy traffic; some are in residential areas with little traffic or industry; and some are in suburban or nonurban areas. Unfortunately, there is ample evidence that the location of a monitoring station relative to nearby sources (such as highways) affects the values observed at the station, particularly for carbon monoxide and other vehicular pollutants. Thus, a standardized system of site selection, such as the one proposed here, appears essential to improve the comparability and meaningfulness of data obtained from different air monitoring stations throughout the Nation.     

Air Pollution Patterns In New York City

Some preliminary analyses of data selected from three years of smoke shade and sulfur dioxide measurements from the forty air monitoring stations in New York City are presented. The purpose of these analyses is to investigate the spatial-temporal variation in concentration of these pollutants throughout the five boroughs of the city. Air pollution health effects studies in New York City have often used city-wide daily morbidity or mortality statistics and related them to air pollution levels obtained from a single monitoring station. The question of whether readings at one station in New York City can adequately represent the air pollution exposure for the population in the five boroughs is examined in this paper. Some samples of correlation matrices of daily pollution averages obtained from the forty air monitoring stations are presented to illustrate the day-to-day variation in pollution in various sections of New York City. It was found that interstation correlations are not high enough to justify the use of one central pollution measuring station as representative of a large metropolitan area. Sulfur dioxide correlates better between stations than smoke shade; this may reflect the different nature and spatial distribution of sources of the two pollutants. Close proximity of stations, or the fact that they were at similar heights above street or sea level did not necessarily lead to higher correlation coefficients.     

An approach for determining air pollution monitoring sites

EC Directive 2008/50/CE on ambient air quality and cleaner air for Europe specifies the obligations of Member States regarding fixed measurements of atmospheric pollutants in areas where maximum concentration levels exceed the lower assessment threshold. However, indications as to the siting of atmospheric monitoring stations are less precise and sometimes confusing. Campaigns to measure benzene and other volatile organic compounds in Murcia in 2007 and 2008 revealed that the areas where measurements coincided with the mean concentration for the whole city were always practically the same. Consequently, we propose a method for siting atmospheric monitoring stations in cities for which emission sources remain steady throughout the year, as is the case for cities in most southern European countries, where the most important emission source is traffic. The method is particularly useful for towns in which only one fixed measuring point is compulsory and should help local authorities in making correct decisions in this context.     

A holistic approach for optimal design of air quality monitoring network expansion in an urban area

This paper presents an objective methodology for determining the optimum number of ambient air quality stations in a monitoring network. The methodology integrates the multiple-criteria method with the spatial correlation technique. The pollutant concentration and population exposure data are used in this methodology in different ways. In the first stage, the Fuzzy Analytic Hierarchy Process (FAHP) with triangular fuzzy numbers (TFNs) is used to identify the most desirable monitoring locations. The network configuration is then determined on the basis of the concept of sphere of influences (SOIs). The SOIs are dictated by a predetermined cutoff value (r_c) in the spatial correlation coefficients (r) between the pollutant concentrations at the monitoring stations identified from first step and the corresponding concentrations at neighboring locations in the region. Finally, the optimal station locations are ranked by using combined utility scores gained from the first and second steps. The expansion of air quality monitoring network of Riyadh city in Saudi Arabia is used as a case study to demonstrate the proposed methodology.     

Prediction of hourly O3 concentrations using support vector regression algorithms

In this paper we present an application of the Support Vector Regression algorithm (SVMr) to the prediction of hourly ozone values in Madrid urban area. In order to improve the training capacity of SVMrs, we have used a recently proposed approach, based on reductions of the SVMr hyper-parameters search space. Using the modified SVMr, we study different influences which may modify the ozone prediction, such as previous ozone measurements in a given station, measurements in neighbors stations, and the influence of meteorologic variables. We use statistical tests to verify the significance of incorporating different variables into the SVMr. A comparison with the results obtained using a neural network (multi-layer perceptron) is also carried out. This study has been carried out in 5 different stations of the air pollution monitoring network of Madrid, so the conclusions raised are backed by real data. The final result of the work is a robust and powerful software for tropospheric ozone prediction in Madrid. Also, the prediction tool based on SVMr is flexible enough to incorporate any other prediction variable, such as city models, or traffic patters, which may improve the prediction obtained with the SVMr.     

Regulation of Indirect Sources

The Division of Air Pollution Control, Illinois Environmental Protection Agency, has conducted an ambient air quality monitoring project focusing on carbon monoxide levels in and around several indirect sources. An analysis of the data indicates that highway-type pollutant emissions have the greatest impact on receptors in the vicinity of indirect sources. This implies that the principal, localized constraint on the siting of indirect sources will be the carbon monoxide generated on public roadways servicing those indirect sources. Clearly, adequate procedures must be developed to link such highway-type emissions to pollutant concentrations. An area-source model and a line-source model were tested using the data generated during the monitoring project. Favorable results were achieved using the line-source model. The proper siting of indirect sources involves the allocation of roadway capacity by the governmental units responsible for transportation network design, working in conjunction with regional planning bodies. A regulatory structure is suggested which emphasizes a regional approach, and an example of an air quality allocation scheme is given. The methodology is applicable to all automotive air pollutants although, in general, localized sensitivity is lost for N0₂ and photochemical oxi-dants.     

Pennsylvania’s Compuiterized Air Monitoring System

The Pennsylvania Air Pollution Commission has developed a regulatory program based upon the control of local air pollution problems and reduction of pollutant levels in air basins. The geographical boundaries of 10 air basins have been established. The Commission’s air basin regulations will provide for the reduction of over-all pollutant levels and for emergency procedures in the event of adverse meteorological conditions. The paper discusses the format and objectives of the program.

In order to effectively enforce the air basin regulations and maintain the necessary surveillance of the state’s air quality, a "computerized real time on-line integrated air monitoring-data handling system" has been designed. The system will incorporate a network to constantly monitor the air in each air basin.The primary objectives of the system are: 1. Constant surveillance of air pollution in the air basins.

2. Provide information on air pollution potential alerts.

3. Aid in further development of air quality criteria and regulations.

The air monitoring network is estimated to include approximately 25 remote stations. Each remote will contain air pollution and meteorological sampling equipment and hardware to telemeter to a central station. The data will be transmitted over leased telephone lines. The central station in Harrisburg will contain the necessary hardware to receive and process data, calculate and display results and permit supervisory control of the network. Output options will include immediate display of edited data, command and alarm information, and presentation of statistical results.

Although the air monitoring system is one of the principle ingredients of the program, the air basin concept encompasses other component systems designed to knit together the entire air pollution control program in Pennsylvania.     

环境空气质量自动监测站的管理与维护

随着工业化、城镇化的深入推进,二氧化硫、氮氧化物、烟粉尘和挥发性有机物等各类污染物排放到环境中,致使中国大气受到严重污染,给人体的健康、动植物的生长、发育和繁殖等带来负面的影响。为实时监测环境空气质量,建立环境空气质量自动监测站逐渐成为大气污染防治的主要手段。文中以环境空气质量自动监测站为研究对象,提出环境空气质量自动监测站管理与维护面临的问题,探讨相应的解决措施,以期为环境空气质量自动监测站的管理与维护提供参考依据。     

Air monitoring network design using Fisher's information measures—A case study

Statistical technique based on Fisher's Information Measures is applied to design an optimal air monitoring network for the Eastern Province of Saudi Arabia. The methodology presented in this study deals with both randomly distributed as well as autocorrelated data series. The study area, which has boundaries 23–30° latitude and 44–53.39° longitude, is divided into 15 × 10 equal grids. The major pollution-causing sources are identified and their emission inventories are compiled. Taking six-hourly meteorological records, collected at the meteorological stations as input to a long-range transport model, time series data on SO₂ concentration are simulated. The information content, which is the reciprocal of the variance of the parameters being estimated, is calculated for station pairs. Using an optimization algorithm, an optimum set of stations transmitting the maximum amount of information are selected for monitoring purposes.     

Evaluation of precipitation chemistry siting criteria using paired stations from northern Maine and southeastern Texas

Two-year precipitation chemistry data records from each of the two paired stations are compared to determine whether the precipitation chemistry is different. In each pair, one of the stations (Caribou, Maine and Victoria, Texas) violates National Atmospheric Deposition Program (NADP) siting requirements, while the other station (Presque Isle, Maine and Beeville, Texas) is in compliance. Methods employed to carry out this study include the calculation of precipitation weighted statistics, the study of logarithmic distributions, the study of regression line residuals, the study of relative sample differences and the use of the non-parametric Wilcoxon test for two matched samples. Results indicate no clear differences occurred between paired stations over a seasonal or longer period of time; however, considerable variation was observed on a weekly basis. Maine stations were in excellent agreement for all ions; Texas stations showed small differences between ions typically associated with soil dust and between ions typically associated with anthropogenic activity.     

Site Redundancy in Urban Ozone Monitoring

This article describes two statistical methods that enable air pollution control agencies to assess the effectiveness of the spatial distribution of current stationary ozone monitoring networks by providing measures of site redundancy. These methods analyze site redundancy by determining the degree to which ozone measurements at one site can be successfully predicted from data collected at other monitoring sites. The first method, the similarity (SIM) measure, calculates redundancy based on the percentage of common operational days during which two monitoring stations report similar daily maximum ozone concentrations. The second method, a modeling technique, relates site redundancy in ozone measurement to an R-squared statistic from an autoregressive model. The model uses meteorological components recorded at a central location and ozone concentrations reported by the network’s other monitoring stations. Both techniques can assist in effective allocation of limited monitoring resources and offer a statistical approach to ambient air monitoring network design. The techniques are applied to data collected at six ozone monitoring stations in Harris County, Texas, during an eight-year period in the 1980s. The methods identified two sites in the six-site network that exhibit a high degree of redundancy.     

Development of an individual exposure model for application to the Southern California children's health study

The Southern California Children's Health Study (CHS) investigated the relationship between air pollution and children's chronic respiratory health outcomes. Ambient air pollutant measurements from a single CHS monitoring station in each community were used as surrogates for personal exposures of all children in that community. To improve exposure estimates for the CHS children, we developed an Individual Exposure Model (IEM) to retrospectively estimate the long-term average exposure of the individual CHS children to CO, NO₂, PM₁₀, PM_2.5, and elemental carbon (EC) of ambient origin. In the IEM, pollutant concentrations due to both local mobile source emissions (LMSE) and meteorologically transported pollutants were taken into account by combining a line source model (CALINE4) with a regional air quality model (SMOG). To avoid double counting, local mobile sources were removed from SMOG and added back by CALINE4. Limited information from the CHS survey was used to group each child into a specific time-activity category, for which corresponding Consolidated Human Activity Database (CHAD) time-activity profiles were sampled. We found local traffic significantly increased within-community variability of exposure to vehicle-related pollutants. PM-associated exposures were influenced more by meteorologically transported pollutants and local non-mobile source emissions than by LMSE. The overall within-community variability of personal exposures was highest for NO₂ (±20–40%), followed by EC (±17–27%), PM₁₀ (±15–25%), PM_2.5 (±15–20%), and CO (±9–14%). Between-community exposure differences were affected by community location, traffic density, and locations of residences and schools in each community. Proper siting of air monitoring stations relative to emission sources is important to capture community mean exposures.     

The role of a steel plant in north-west Italy to the local air concentrations of PCDD/Fs

Polychlorinated dibenzo-p-dioxins (PCDDs) and polychlorinated dibenzofurans (PCDFs) are ubiquitous contaminants, mainly released into the environment during combustion processes (point sources), but also from other sources (traffic, uncontrolled combustion).This study aims at investigating the contribution of a steel plant in NW Italy (700 000 tons of steel year⁻¹) to the air concentrations of PCDDs/PCDFs at local level, through the analysis of measured, modelled and literature data. The study was carried out in an area of 600 km², using air quality data measured by the institutional monitoring network, data obtained from AERMOD simulations and literature data.The measured air concentrations were consistent with literature values for similar areas, and both the homologue profiles and PCA analyses showed a clear distinction between the monitoring stations and the source profiles.All the previous results were confirmed by the air dispersion model (AERMOD), that predicted PCDD/F air concentrations due to the steel plant from four to two orders of magnitude lower than those measured in the monitoring stations, highlighting the presence of other sources.This study outlines the limited influence of the source in the local PCDD/F air concentrations and at the same time the usefulness of a joint analysis of measured, literature and calculated data to correctly evaluate the role of a source to the local pollution. The study also highlights the usefulness of AERMOD as a complementary tool to define the correct placement of monitoring stations and to locate those areas expected to have the highest air concentrations deriving from a source.     

Revised Air Quality Surveillance Requirements for State Implementation Plans

The National Ambient Air Quality Monitoring Program is carried out by state and local air pollution control agencies in support of their State Implementation Plans (SIP’s). The current EPA regulations which specify the characteristics of these state monitoring programs are undergoing change as a result of a comprehensive review by an independent work group. These revised regulations, which are described in the paper, are intended to improve the quality, timeliness, and usability of the data generated by the states for all data users. In addition, the revised regulations seek to bring about; (a) national consistency in monitoring site locations through standardized siting procedures; (b) improved network operations by means of a minimum quality assurance program; (c) reduced network inflexibility through an annual network review process; and (d) reduced data reporting through changed data reporting procedures.     

Prediction of carbon monoxide acute air pollution episodes. Model formulation and first application in Lombardy

This paper describes a method to forecast the exceedance probability of a fixed threshold for a certain air pollutant concentration. This general approach has been applied in the case of carbon monoxide on 35 traffic monitoring stations in the Lombardy Region. The implemented model has been called FOREPOLL (Forecast Pollution).The model structure, consisting in three basic modules (the deterministic, the stochastic and the Bayesian one), has been thought to be adjustable to different stations of an air quality network and to various pollutants. Forepoll uses the last biennium data set of the station in exam (pollutants and micrometeorological parameters) as a moving learning time and also needs the forecasted synoptic weather type. In the first module the daily maxima of the hourly measured pollutant concentrations are checked in order to eliminate some known physical dependencies, such as the temperature dependence of the emissions. Then the data are divided into subgroups depending on the weather type and in each group fitted by a different Weibull distribution. To provide a first a priori exceedance probability, this distribution is daily rebuilt by taking into account the forecasted parameters. In the last module this probability is enhanced or reduced using simple, experience based, bayesian rules providing the a posteriori exceedance probability.Model validation trials have been carried out on a year of CO forecasted concentrations in different air quality stations; the first results are quite good particularly for the metropolitan areas, because the model seems to work better in case of stronger and more diffuse pollution.     

Interpretation of air quality in relation to monitoring station's surroundings

This study explores the appropriateness of the locality of air monitoring stations which are meant to indicate air quality in the area. Daily variations in NO₂ and PM₁₀ concentrations at 14 monitoring stations in Hong Kong are examined. The daily variations in NO₂ at a number of background monitoring stations exhibit patterns similar to variations in traffic volume while variations in PM₁₀ concentration exhibit less discernible pattern. Principal component analysis (PCA) and cluster analysis (CA) are applied to analyse NO₂ and PM₁₀ measurements between January 2001 and December 2005. The results show that NO₂ concentrations at background stations within the urban area are highly influenced by vehicle emissions. The effect vehicle emission has on NO₂ at stations within new towns is smaller. CA results also show that variations in PM₁₀ concentrations are distinguished by the area the station is located in. PCA results show that there are two principal components (PC's) associated with variations in roadside concentration of PM₁₀. The strong influence of roadside emissions towards concentrations of NO₂ and PM₁₀ at a number of urban background stations may be due to their close proximity to busy roadways and the high density of surrounding tall buildings, which creates an enclosure that hinders dispersion of roadside emissions and results in air pollution behaviour that reflects variation in traffic.     

Spatial variations of ground level ozone concentrations in areas of different scales

The spatial variation of ground level ozone concentrations was investigated for areas of three different scales: (1) an air quality management district (a region about 100×70 km²) in northern Taiwan, (2) the neighborhood (about 2 km in radius) of an air quality monitoring station, and (3) an open field (about 400×600 m²) surrounded by 3- and 4-story buildings in an elementary school. Analysis of data on hourly ozone concentration, obtained at 13 m above the ground at 21 monitoring stations in the air quality management district, showed that the stations downwind of the urban center in the district had significantly higher ozone concentrations. Measurements for 8-h average ozone concentrations at 1.5 m above the ground by passive samplers showed that, in a flat area about 2 km in radius, the ratios of the ozone concentration at open areas to that at the monitoring station (0.86–0.93) were significantly higher than those obtained at areas with higher traffic flow and density of buildings (0.60–0.68). For the open field in an elementary school, the 8-h average ozone concentrations at 1.5 m above the ground at sites less than 10 m from the nearest building were considerably lower than those at sites farther away from buildings. The results indicated that, in areas of small scales, the spatial distributions of ozone concentration were highly non-uniform and there were appreciable day-to-day variability in spatial distribution. Such variability should be taken into account in determining the extent to which an individual is exposed to ozone.