Quality and performance of a PM10 daily forecasting model

doi:10.1016/j.atmosenv.2007.10.073

Atmospheric Environment

Volume 42, Issue 6, February 2008, Pages 1098-1109

https://doi.org/10.1016/j.atmosenv.2007.10.073 Get rights and content

Abstract

Particularly in the cold season unfavorable dissemination conditions of the ambient air lead to higher-than-average PM10 concentrations in parts of the western Alpe-Adria-Region, covering the provinces South Tyrol, Carinthia and Styria. Therefore, EU pollution standards cannot be met in the cold season and partial traffic regulation measures are taken in Bolzano, Klagenfurt and Graz, the three capitals in this region. Decision making for these regulations may be based on the average PM10 concentration of the next day provided that reliable forecasts of these values can be offered. In the present paper we show how multiple linear regression models combining information of the present day with meteorological forecasts of the next day can help forecasting daily PM10 concentrations for sites located in the three cities. Special emphasis is given to an appropriate selection of the regressor variables readily available as measured values, factors or meteorological forecasts suitable in operational mode. To reflect the quality of the forecast properly, we define a quality function where prediction errors near the threshold PM10 of $50 μ g m^{- 3}$ are assumed to be more severe than errors in regions that are either far below or above the threshold. Since December 2004, the forecasts are used as a monitoring and information tool in Graz. Our daily forecasts have been carried out in cooperation with the meteorologists from the ZAMG Styria (Styrian meteorological office). The investigations in terms of the quality function and according possible decision rules show that our prediction models may support future decisions concerning possible traffic restrictions not only in Graz, but also in Bolzano and Klagenfurt.

Introduction

Particularly during the winter season, the basin areas of the Alps (including the cities of Graz, Klagenfurt and Bolzano) are exposed to weather conditions such as stationary temperature inversions, a low amount of precipitation and low wind velocities. These special weather conditions cause an extensive load of particulate matter (PM) in ambient air. The issue of PM/fine dust has recently caught remarkable attention and is still a very present and explosive topic in science and politics. The PM10 (particles with an aerodynamic diameter $< 10 μ m$ ) concentration is measured in units of $μ g m^{- 3}$ . According to the EU framework directive 1999/30/EC (European Community, 1999) the limit value for the daily PM10 average is $50 μ g m^{- 3}$ and must not be exceeded on more than 35 days of the year (valid for the years 2005–2009); a reduction to 7 days in 2010 is envisaged. Additionally, the annual PM10 average must not exceed the limit of $40 μ g m^{- 3}$ ( $20 μ g m^{- 3}$ starting with 2010). However, at the test point GT1 in Graz (near the pedestrian zone in the center of the city) we observed in the period 2003–2006 between 90 (2004) and 137 (2003) exceedances of the daily PM10 limit and registered high annual averages between 41 and $49 μ g m^{- 3}$ . The physical and chemical composition of the particles is very complex. There are natural sources like pollen or crushing and grinding rocks and soil (primary particles). Contrarily, there are particles which arise from aerially pollutants (secondary particles). Anthropogenic particles are produced by traffic, domestic fuel and industry. They may be directly exhausted by burning processes or arise from mechanical abrasion of tyres, brakes, tarmac, etc. The coarse particles (PM10 –2.5) are composed of smoke, dirt and dust, the fine particles (PM2.5) are rather toxic organic compounds or heavy metals. Estimates about the proportion of the PM polluters differ widely and have to be related to the specific environment (urban, rural, seaside, etc.). For further information on aerosol source analysis in urban and rural areas of Austria we refer to the Aquella project http://www.iac.tuwien.ac.at/environ/aquella.html, which is still in progress.

Negative health effects caused by PM have been analyzed in many epidemiological and toxicological studies. An extensive general review can be found in Pope III and Dockery (2006). A description of the Austrian study AUPHEP is given in Hauck et al. (2004), while Schwarze et al. (2006) is an excellent review including 210 references of studies carried out in different regions throughout the world. In general, fine particles (PM2.5) are more likely to be toxic since they often consist of heavy metals and carcinogenic organic compounds. Furthermore, they are inhaled into the trachea and the respiratory system in general. In Klagenfurt and Bolzano where measurements for PM2.5 are available, it can be observed that the ratio PM2.5/PM10 is relatively constant. Hence investigations on PM10 in this region allow to draw some conclusions on finer PM fractions as well. In Bolzano about 40% of PM10 belong to the fine fraction, while the percentage in Klagenfurt is approximately twice as high. In contrast to the rare PM1 and PM2.5 data sources, extensive data on PM10 were available in all three cities. Furthermore, since the EU limits refer to PM10 we based our research on this specific measure.

Due to the negative health effects caused by PM10, policy had to react (and take drastic measures) against the PM problem. In the winter season 2006/2007, the cities of Bolzano, Graz and Klagenfurt as well as the respective Provinces were forced to take further action against high PM10 concentrations of the ambient air. In general, traffic regulations become effective if the limit values are exceeded for several days and if a reduction on the day in question is “unlikely”. For the appropriate authorities it is necessary to base singular decisions on reliable forecasting models for daily PM10 concentrations. The objective of the present paper is to deliver PM10 forecasting models for the specific sites and to illustrate their practical applicability as a decision making tool for possible traffic restrictions. As the traffic regulations mentioned above regard the period between October and March, we based our models on that specific time of year. Better dissemination conditions for the ambient air during spring and summer lead to considerably lower PM10 concentrations and thus, there is no urgent need for action in that warm season of the year. The investigations were made within the framework of the EU-Life-project KAPA GS (Klagenfurts Anti PM10 Action Programme in cooperation with Graz and South-Tyrol: project duration from 1 July 2004 to 30 September 2007).

Our models are based on multiple linear regression which we found to be the most convenient method. Other typical approaches for PM prediction are neuronal networks (cf. Pérez and Reyes, 2002, Hooyberghs et al., 2005), discriminant analysis (cf. Silva et al., 2001) or Kalman filtering (cf. van der Wal and Jansen, 1999). The demand for our model was simplicity, practical feasibility and sufficient accuracy. Simplicity is guaranteed by the linear structure of the model. To obtain practical feasibility, it is vital to perform a careful choice of parameters. Meteorological parameters, for example, have to be forecasted individually (type-B parameters) in operational mode and thus it has to be assured that this additional uncertainty will not prevail. Hence, the precision of the prediction for a specific day will to some extent depend on the quality of the singular weather forecasts. Our empirical studies showed that temperature inversion, precipitation and wind velocity play an important role at all three sites. After consulting the ZAMG (Styrian meteorologic office) we decided to include variables which are representative for these impact factors and can also be forecasted with sufficient precision. In order to guarantee sufficient accuracy, we included all relevant parameters available or measured at the time when the forecasts were generated (type-A parameters), e.g. the PM10 24 h moving average.

In contrast to many theoretical studies, we are able to present the performance of the model in operational mode. During a three-year trial period in Graz we made PM10 forecasts available at the web site of our project (http://www.feinstaubfrei.at). For the generation of the predictions we used meteorological forecasts from ZAMG Styria where the meteorologists provided us with analyzed simulation data by the systems ECMWF and Aladin. We observed that PM10 forecasts for several days do not loose much quality if the type-B parameters were known. However, we found that forecasts of two or more days in advance will become unreliable in practise.

By virtue of the high dispersion of PM10 data and our specific requirements we found that commonly used measures do not reveal the quality of the forecasts with respect to our needs. A forecast of $15 μ g m^{- 3}$ for an observation of $30 μ g m^{- 3}$ , for example, yields a 100% relative error, though both the observation and the prediction are clearly below the limit value. On the other hand we observe peaks with values $> 150 μ g m^{- 3}$ . Typically, the model under-estimates jerky leaps and in this case a prediction of $110 μ g m^{- 3}$ (say) is not bad, even though the absolute error is high; the forecast value still indicates alert status. Concerning decision making for traffic restrictions, we have to concentrate on avoiding errors leading to unjustified measures. In order to incorporate these specific requirements we used—besides standard measures like correlation or mean squared error—a quality function assigning a meaningful rating to each pair of observation and forecast value.

In the next section we describe the sites, databases and input parameters. Section 3 contains the methodology of our study and quality issues are discussed in Section 4. Results of test runs are analyzed in Section 5 and in the final section we summarize our findings and express our conclusions.

Section snippets

Sites and database

Our investigations took place of sites within the cities of Bolzano, Klagenfurt and Graz. The three cities are located in basin areas south of the main Alpine crest and show similar climatical characteristics. Rain clouds from the Atlantic are kept away by the Alps implying low precipitation and low wind velocities during the cold period. Furthermore, stationary temperature inversions frequently occur at that time of year (due to the basin location). Some key data, listed in Table 1, Table 2,

Multiple linear regression

Our goal is to model the daily PM10 averages by a linear model which will be the basis for our forecasting models. The regression models are conceived for the dependent variable $\sqrt{PM 10}$ . A square root transformation is necessary to assure a constant error variance and to avoid a violation of the model assumptions. It should be noted that the standard transformation used in this context is rather the log-transformation (see e.g. Lonati et al., 2006). However, in our case the model diagnostics showed

The forecasting models

The multiple linear regression models obtained by means of the procedure described in Section 3.1, are now used as the basis for the forecasting models. As a first step we used historical data to compute the estimates ${\hat{β}}_{i}$ for the linear model. This can be done with any standard statistics package (we employed SPSS 14.0) and yields an estimate for $\sqrt{PM 10}$ . The simplest way to get an estimate for PM10 is to take the square of the predicted $\sqrt{PM 10}$ . The resulting bias is negligible for our purposes (and

Quality function

To measure the quality of our forecasts we need a reasonable rating system different from usual measures as e.g. the absolute error which is large if the forecast is $120 μ g m^{- 3}$ and the observation is $170 μ g m^{- 3}$ . However, this forecast value will indicate a right decision when partial traffic regulation measures are taken whenever $PM 10 > certain limit$ (e.g. $50 μ g m^{- 3}$ ). Contrarily, if the prediction is $55 μ g m^{- 3}$ and the observation is $40 μ g m^{- 3}$ a traffic regulation may not be justified and the relatively

Theoretical performance

The investigations below are based on all available data for the cold periods at the sites BT1, KT1 and GT1. To check the theoretical performance of our models, we computed the coefficient estimates ${\hat{β}}_{i}$ by excluding the last cold period available, which shall be our test period. Then we calculated the daily PM10 predictions of the test period with these estimated coefficients. Instead of the forecasted ${\hat{z}}_{d}^{(i)}$ ( $i = 8, 9, 10$ ) we used the measured values $z_{d}^{(i)}$ , referred to as “exact” meteorological

Meteorological forecasts

In Graz, the daily meteorological forecasts delivered by the ZAMG Styria started on 15 December 2004 and includes the variables wind, $Δ_{temp}$ and prec. The first winter period 2004/2005 was used as pilot study to test the performance of the model with actual weather forecasts. The quantitative forecast of the wind velocities was a particularly difficult task for the meteorologists. At the beginning of the test run the scale was systematically too high. The reason for this was that the test point

Quality in operational mode

During our three test periods in Graz (15 December 2004 to 31 March 2005; 17 October 2005 to 31 March 2006; 2 November 2006 to 31 March 2007) we fed our linear models with the meteorological forecasts of wind, prec and $Δ_{temp}$ . Before each season the parameters of the models were estimated anew from the updated data. As a matter of fact replacing the observed by the forecasted values causes an additional error and the question is how reliable the model remains. Fig. 4 compares the observed and

Summary and conclusion

We show that PM10 forecasting models based on linear regression for Bolzano, Klagenfurt and Graz provide suitable results. Due to the simple and transparent character of the model we find that more complicated black box approaches are not necessary in this case. The input variables are selected in order to represent both meteorological and anthropogenic parameters. In general, we lay special emphasis on the practical performance and the treatability of the model. For the operational mode it is

References (11)

H. Hauck et al.
AUPHEP—Austrian project on health effects of particulates—general overview
Atmospheric Environment
(2004)
J. Hooyberghs et al.
A neuronal network forecast for daily average PM10 concentrations in Belgium
Atmospheric Environment
(2005)
G. Lonati et al.
The role of traffic emissions from weekends’ and weekdays’ fine PM data in Milan
Atmospheric Environment
(2006)
P. Pérez et al.
Prediction of maximum of 24-h average of PM10 concentrations 30 h in advance in Santiago, Chile
Atmospheric Environment
(2002)
Council Directive 1999/39/EC of 22 April 1999 regulating to limit values for sulphur dioxide, nitrogen dioxide and oxides of nitrogen, particulate matter and lead in ambient air
Official Journal of the European Communities
(1999)

There are more references available in the full text version of this article.

Cited by (111)

Spatiotemporal informer: A new approach based on spatiotemporal embedding and attention for air quality forecasting
2023, Environmental Pollution
Accurate prediction of air pollution is essential for public health protection. Air quality, however, is difficult to predict due to the complex dynamics, and its accurate forecast still remains a challenge. This study suggests a spatiotemporal Informer model, which uses a new spatiotemporal embedding and spatiotemporal attention, to improve AQI forecast accuracy. In the first phase of the proposed forecast mechanism, the input data is transformed by the spatiotemporal embedding. Next, the spatiotemporal attention is applied to extract spatiotemporal features from the embedded data. The final forecast is obtained based on the attention tensors. In the proposed forecast model, the input is a 3-dimensional data that consists of air quality data (AQI, PM_2.5, O₃, SO₂, NO₂, CO) and geographic information, and the output is a multi-positional, multi-temporal data that shows the AQI forecast result of all the monitoring stations in the study area. The proposed forecast model was evaluated by air quality data of 34 monitoring stations in Beijing, China. Experiments showed that the proposed forecast model could provide highly accurate AQI forecast: the average of MAPE values for from 1 h to 20 h ahead forecast was 11.61%, and it was much smaller than other models. Moreover, the proposed model provided a highly accurate and stable forecast even at the extreme points. These results demonstrated that the proposed spatiotemporal embedding and attention techniques could sufficiently capture the spatiotemporal correlation characteristics of air quality data, and that the proposed spatiotemporal Informer could be successfully applied for air quality forecasting.
A statistical learning framework for spatial-temporal feature selection and application to air quality index forecasting
2022, Ecological Indicators
Accurate air quality index (AQI) forecasting makes a difference to public health, local economic development, and ecological environment. As a typical geographical datum, the spatial autocorrelation (SAC) of the AQI is often ignored, which may violate the assumptions of some models, such as machine learning which requires variables to be independent and identically distributed. Considering the strong SAC of the AQI, this study proposes a novel statistical learning framework integrating SAC variables, feature selection, and support vector regression (SVR) for AQI prediction in which correlation analysis and time series analysis are used to extract the spatial-temporal features. In addition, the historical AQI series of the target site is adjusted by using trigonometric regression to eliminate the non-stationarity. To further improve prediction accuracy, a feature selection method combining reinforcement learning with a heuristic algorithm is adopted. To demonstrate the effectiveness of our proposed framework, we select the AQI data of 34 cities from the Yangtze River Delta, which is one of the most polluted areas in eastern China, and focus on the three largest cities, Nanjing, Hangzhou, and Shanghai. We compared the proposed framework with several baselines, and the experiment illustrates that the forecasting accuracy of the proposed framework is significantly better than the baselines at all selected key sites that can provide accurate predictions for air quality.
Air quality prediction using spatio-temporal deep learning
2022, Atmospheric Pollution Research
With the rapid urbanization, many developing countries are suffering from heavy air pollution. Predicting air quality accurately is critical for people’s outside planning and governments’ policy-making in these countries. This paper studies the air quality dataset collected in 22 stations of Beijing city, China, and reveals the spatial correlations, the temporal dependencies and feature correlations hidden in the dataset. Different from existing shallow or deep learning models that do not fully consider these correlations, we propose a spatio-temporal deep learning model called Conv1D-LSTM, which unifies 1-dimensional convolutional neural network (1D CNN) and long short-term memory (LSTM) for spatial and temporal correlation feature extraction, and uses a fully connected network (FCN) to exploit these features for air quality prediction. Moreover, this study investigates the problem of data missing in the air quality dataset and impute the missing values from spatial and temporal views. Extensive experiments are conducted by applying the proposed model to hourly PM_2.5 and PM₁₀ with single and multi-step predictions over the study area during 2014–2015. The results show the effectiveness of the proposed model and its superiority over other baseline models.
Forecasting: theory and practice
2022, International Journal of Forecasting
Forecasting has always been at the forefront of decision making and planning. The uncertainty that surrounds the future is both exciting and challenging, with individuals and organisations seeking to minimise risks and maximise utilities. The large number of forecasting applications calls for a diverse set of forecasting methods to tackle real-life challenges. This article provides a non-systematic review of the theory and the practice of forecasting. We provide an overview of a wide range of theoretical, state-of-the-art models, methods, principles, and approaches to prepare, produce, organise, and evaluate forecasts. We then demonstrate how such theoretical concepts are applied in a variety of real-life contexts.
We do not claim that this review is an exhaustive list of methods and applications. However, we wish that our encyclopedic presentation will offer a point of reference for the rich work that has been undertaken over the last decades, with some key insights for the future of forecasting theory and practice. Given its encyclopedic nature, the intended mode of reading is non-linear. We offer cross-references to allow the readers to navigate through the various topics. We complement the theoretical concepts and applications covered by large lists of free or open-source software implementations and publicly-available databases.
A short-distance healthy route planning approach
2022, Journal of Transport and Health
With the development of the economy and the accumulation of social wealth, urban residents have begun to give more attention to quality of life than to material needs. Consequently, environmental factors that affect human health, such as air quality, have become a new focus when traveling. A travel scheme with relatively low pollutant exposure to travelers can not only improve their health and satisfy their goals but also benefit social stability and sustained progress. However, low spatiotemporal resolution and coarse spatial details of the distribution of PM2.5 (particles with an aerodynamic diameter of 2.5 μm or less) educe the success rate of short-distance healthy travel route planning. This paper proposes a short-distance healthy route planning approach that is based on PM2.5 retrieval with high spatiotemporal resolution and a dynamic Dijkstra algorithm. First, fine spatial resolution images, meteorological data, and socioeconomic data are used to retrieve the spatial distribution of PM2.5 concentration in hourly intervals via a back-propagation neural network (BPNN). Second, a PM2.5 concentration value is obtained for each road section, and the harm degree to the human body is calculated as the weight of each road section. Then, the healthiest route is obtained based on the Dijkstra algorithm. Finally, the route planning effectiveness is verified by comparing the PM2.5 potential dose descending rate between the healthy route and the shortest route. The results show that the coefficient of determination (R2) of the PM2.5 retrieval approach that is based on multisource data and BPNN is 0.85, which can ensure the accuracy of the PM2.5 data at the street level. On this basis, the potential dose reduction rate of the healthy route can reach up to 20%, which proves that our approach can perform well. It can effectively improve the safety of travel and alleviate the anxiety that is caused by air pollution. In addition, it provides an easy implementation strategy for software for health management.
Air pollution forecasting with multivariate interval decomposition ensemble approach
2021, Atmospheric Pollution Research
As the air pollution particulate matter (PM₁₀) is affected by a variety of factors, especially meteorological factors, resulting in a high degree of complexity and volatility of PM₁₀. However, the existing research does not well deal with the multi-factor PM₁₀ forecasting and is limited to the deterministic point forecast of PM₁₀ concentration and does not consider the interval forecast related to uncertainty. To solve these issues, a novel multivariate interval decomposition ensemble approach is proposed for PM₁₀ concentration forecasting, integrating multi-factor selection, data decomposition, intelligent forecasting network and evaluation system. The stability and robustness of this approach have been tested in three cities with different economic characteristics in China, and the results show that our proposed approach is superior other benchmark models in point forecasting and 80% interval forecasting. Our proposed approach is a promising PM₁₀ concentration forecasting approach, which can enable the government to accurately forecast PM₁₀ concentration and make more effective measures to control and manage the adverse effects of air pollution on the current economy and health.

View all citing articles on Scopus

^☆: Research supported by the EU-Life-project KAPA GS LIFE04 ENV/AT/000006.

View full text

Quality and performance of a PM10 daily forecasting model☆

Abstract

Introduction

Section snippets

Sites and database

Multiple linear regression

The forecasting models

Quality function

Theoretical performance

Meteorological forecasts

Quality in operational mode

Summary and conclusion

Atmospheric Environment

Atmospheric Environment

Atmospheric Environment

Atmospheric Environment

Council Directive 1999/39/EC of 22 April 1999 regulating to limit values for sulphur dioxide, nitrogen dioxide and oxides of nitrogen, particulate matter and lead in ambient air

Official Journal of the European Communities