AIR POLLUTION PREDICTION MODEL USING GIS
Anil Chitade Dr. S. K. Katiyar
Associate Professor Professor
Deptt. Of Civil Engg. Deptt. Of Civil Engg.
Rajeev Gandhi College of Engg & Research, Maulana Azad National Institute of Technology
Chandrapur (M.S.), India Bhopal (M.P.), India
Email: Email:
ABSTRACT:
The increasing demand of coal for energy generation and sequentially increase in coal mines puts the surrounding environment under considerable pressure. Ecosystem contamination caused by coal mining can be particularly dangerous when the geographical context renders the mine environment very sensitive to increase in pollution. In this connection, the development of operational systems and services for monitoring and forecasting of pollution in the mining environment, which provide near real-time information in a user-friendly way, is almost a compulsory need. Existing conventional techniques used to monitor air quality involves manually measured pollution concentrations within the area of measuring stations. Geographical information system (GIS) is the new technique by which the air pollution can be predicted throughout the mining region where continuous air quality monitoring (CAQM) stations are not available.
This research presents the process of preparing the air pollution prediction model for respiratory suspended particulate matter (RSPM) concentrations in air. RSPM sample data in the study area was collected from Maharashtra pollution control board (MPCB). Arc-GIS Geostatistical analyst tool with ordinary Kriging interpolation method was used to generate the prediction model of the area. Accuracy of the results was assessed using RMSE technique. It was observed that the presented prediction model using GIS allows a more detailed assessment and more realistic distribution of RSPM concentration of air quality within mining areas with limited CAQM stations. The concentrations of RSPM are highest in the mining zones and surrounding region of the study area whereas forest area represents the less concentration. This method can be used by environmental managers and local authorities to continually monitor air quality in mining areas. It is also useful to study the changes in level of air pollution in the region.
Keywords: Geographical information system (GIS), Continuous air quality monitoring (CAQM) station, RSPM, Ordinary Kriging, Geostatistical analyst.
I. INTRODUCTION:
1 | Page
The conflict between mining activities and environmental protection has intensified over recent years, emphasizing the need for improved information on the dynamics of impacts at regional and local scales (Latifovica et al., 2004). A limited number of air quality monitoring stations limits the initial strategy of pollution prevention program in the areas at micro-scale level, where air pollution is a serious threat to the community. Air pollution in coal mines is mainly due to the fugitive emissions of particulate matter and gases including methane, sulphur dioxide, oxides of nitrogen and carbon mono-oxide (AMA, 2007). Opencast mining creates much more air quality deterioration in respect of dust and gaseous pollutants (Holgate, 1999). In mining areas generally the air content particulate matter such as RSPM and SPM, if excessive then it has the adverse effect on human health. Harmful effects have been associated with RSPM exposure levels that are below the ambient air quality guideline levels set by the world health organization (Mukala et al., 2000) and central pollution control board (CPCB), of India.
Most of the epidemiological studies related to air pollution are based on RSPM concentrations at fixed ambient air quality monitoring sites (CPCB, 2006). However, the measurement data from these stations do not necessarily represent areas beyond their immediate vicinity, as the concentrations of pollutants in urban areas may vary by orders of magnitude on spatial scales varying from tens to hundreds of meters. Mathematical modeling technique can provide quantitative estimate of air pollutants in the desired areas, where no stations are available for monitoring the air quality. Modeling is the process of producing a model, which is a representation of the working of some system of interest, and in air quality prediction it involves application of emission inventories, combined with atmospheric dispersion and population activity modeling.
Models for evaluating exposure to air pollutants have been classified as statistical, mathematical and mathematical-stochastic (Bin Zou et al., 2008). The statistical approach involves the statistical determination of the measured exposures in terms of the factors that are assumed to influence these exposures. The stochastic approach attempts to include a treatment of the inherent uncertainties in the model, e.g., those caused by the turbulent nature of atmospheric flow. Mathematical modeling utilizing emission and dispersion models is also called deterministic modeling. Source apportionment methods can be used in order to analyze the contribution of various emission categories to the total human exposure (Kousa et al., 2002).
Assessment of the impact of human activities on the environment requires qualitative and quantitative information on the spatial distribution and dynamics of natural resources. In recent years, GIS has emerged as a powerful tool to handle voluminous data (both spatial and non-spatial). In the literature, various researchers have used different air pollution models and the Gaussian based air pollution model using GIS tools was found suitable for the prediction of air pollutants (Gupta & Bariar, 2006; Khan, S., & Toleti, B.R., 1993). GIS helps to integrate various types of information spatially to aid in characterization and decision making (Rohayu & Wan, 2010). GIS can also be an effective tool for the determination of the areas that are contaminated as a result of contaminant transport via surface runoff (Yenilmez, 2011). GIS based zonal statistics can be used to summarize the environmental risk surfaces (ERS) model statistics by conservation planning unit or habitat polygon. By characterizing human disturbance using GIS-based risk indices in combination with the distribution of biodiversity feature occurrences, the most promising areas to sustain biodiversity to meet national and CBD biodiversity conservation goals can be identified (McPherson et al., 2008).
A limited number of air quality monitoring stations limits the initial strategy of pollution prevention program at micro-scale level. Existing techniques used for monitoring air quality involves, manually measured pollution concentrations within the area of measuring station. GIS is the technique by which the air pollution can be predicted throughout the region where continuous air quality monitoring stations (CAQM) are not available. Arc-GIS Geostatistical analyst tool of Arc-GIS provides a suite of statistical models and tools for spatial data exploration and surface generation. Using Arc-GIS Geostatistical analyst tool, we can create a statistically valid prediction surface, along with prediction uncertainties, from a limited number of data measurements. With Arc-GIS Geostatistical analyst, one can
(i) Explore data variability and spatial relationships and look for unusual data values and examine global and local trends.
(ii) Utilize multivariate analysis to create optimal statistical models to produce reliable maps of predictions, prediction errors, quantiles and probabilities for improved decision making.
(iii) Modify model parameters interactively or automatically optimize them using cross validation.
(iv) Determine optimal locations to create or update a monitoring network.
(v) Prepare for worst-case scenarios by simulating many possible realizations of an environmental process.
In this research paper, the prediction model was designed for the RSPM contents in the study area using GIS techniques.
II. STUDY AREA
Chandrapur district of Maharashtra state in India is famous for its sprawling coal mines and Tadoba wildlife sanctuary, which is an important tiger destination in the country (www.chanda.nic.in). The mineral based industrial development and rapid urbanization in this district has albeit resulted in pollution and environmental degradation and its effects are being felt on a large scale. The industrial and other activities in and around Chandrapur have extensively contributed to pollution and there is a considerable rise in the associated health problems in the local population (MPCB report, 2007). It is perceived that mainly air pollution is a serious threat to environment and public health. As per Central pollution control board (CPCB) of India statement reported in the year 2010, the Chandrapur has been ranked as the fourth most polluted place in the country. Hence, part of this area has been taken for this research paper investigations and this area is schematically shown in figure 1.
The study area is a part of Chandrapur district, the eastern edge of Maharashtra state of India. It is located between 19045’ N to 20000’ N latitude and 79015’ E to 79037’E longitudes. The area is situated within the Wainganga and Wardha river basins respectively, which are the tributaries of Godavari River. This area is abundantly endowed with rich flora and fauna, water resources and mineral wealth. It is spread over about 390.52 sq.km areas.
1 | Page
(Source- www.chanda.nic.in)
Figure 1: Map of study area
III. DATA RESOURCES AND SOFTWARE USED
For the preparation of prediction model, the air content of RSPM data was collected from pollution control board of Maharashtra state of India. Measurements of RSPM were taken twice a week of 24 hours duration and the values of RSPM were calculated as the average of the sampling data collected monthly for entire year. The model was prepared using Arc-GIS 9.2 software.
IV. METHODOLOGY
The Air quality data for the contents of RSPM values at 44 different locations in study area was collected from Maharashtra pollution control board (MPCB). Geostistical analyst tool available in ArcGIS 9.2 is used to assess the statistical properties of RSPM data such as spatial data variability, spatial data dependence and trends. Prediction model of RSPM concentration in air in the study area was prepared using Geostatistical analyst tool of Arc-GIS 9.2. Methodology followed for preparation of model is described as below:
Data layers of RSPM concentration collected at different locations were added to ARC-Map in the form of shape file using UTM projection with WGS-84 coordinate system. Using Geostatistical analyst tool Kriging interpolation was carried out to know the RSPM concentration and surface map of RSPM concentration was generated. The Histogram tool plots frequency histograms for the attributes in the dataset, enabling us to examine the univariate (one-variable) distribution for each attribute in the dataset hence frequency histogram for earlier created surface map was plotted for the attributes in the dataset.
Normal QQplot ("Q" stands for quantile) is a probability plot, which is a graphical method for comparing two probability distributions by plotting their quantiles against each other was generated to check normality of RSPM data. The QQ plot is where we compare the distribution of the data to a standard normal distribution, providing another measure of the normality of the data. The closer the points are to the straight line in the graph, the closer the sample data follows a normal distribution.
A global trend was then identified to check the presence/absence of trends in the input dataset. If a trend exists in your data, it is the nonrandom (deterministic) component of a surface that can be represented by a mathematical formula. RSPM concentration model was then generated to know the trend of air pollution among the study area. The generated model was cross validated to compare the measured and predicted values. The probability of RSPM concentration exceeding a critical threshold values were mapped. Final prediction map of RSPM concentration was then generated in Arc-Map.
V. RESULTS AND DISCUSSIONS
Surface map of RSPM concentration using the above stated tool was prepared using the RSPM concentration values collected from the field with Kriging method of interpolation as per the methodology discussed above. The output surface map of RSPM concentration obtained is shown in figure 2. During the process semivariogram and covariance values were observed and close spatial relationship between measured points are examined. The Semivariogram/Covariance Modeling dialog box allows us to examine spatial relationships between measured points. We assume things that are close are more alike. The semivariogram allows us to explore this assumption. The process of fitting a semivariogram model while capturing the spatial relationships is known as variography.
Figure 2: Surface map of RSPM concentration
For examination of the distribution of RSPM data trends and to understand the spatial auto-correlation and directional influences, the histogram is helpful. This plots frequency histogram for the attributes in the dataset enabling to examine univariable (one-variable) distribution of the dataset for each attribute. The distribution of RSPM is depicted by a histogram with range of values separated into ten (10) classes. The relative proportion (density) of data within each class is represented by the height of each bar as shown in figure 3. Blue dots in figure 3 represent the location of stations where values of RSPM ranges from 254 to 292.
Figure 3: Histogram of RSPM concentration
Trend analysis was carried out after plotting the normal QQplot as shown in figure 4.
Figure 4: Normal QQPlot for RSPM concentration
As there were no outlier (or erroneous) sample points in the dataset and distribution was close to normal, RSPM concentration mapping was then carried out. When the model was prepared, its cross validation was necessary. This gives an idea of ‘how well’ the model predicts the unknown values. For all points, cross validation sequentially omits the points, predicts its values using the rest of the data and then compares the measured and predicted values. Prediction error is the difference between the prediction and actual measured values. RMS standardized prediction error was found to be 0.029. RMS prediction error observed was smaller hence this model provides the approximately accurate predictions, as shown in figure 5.
Figure 5: Cross validation comparison result
A critical threshold value of 150 was used to decide if any locations exceed this value which is based on probabilistic approach. The probability map generated is shown in figure 6. Finally the prediction model was prepared as shown in figure 7.