Meteorology-based Forecasting of Air Quality Index Using Neural Network

Meteorology-based Forecasting of Air Quality Index Using Neural Network

Mukesh Sharma,* Sachin Aggarwal*, Purnendu Bose* and Ashok Deshpande+

*Department of Civil Engineering, Indian Institute of Technology

Kanpur, 208016, India

                   +Distinguished Professor SIES-Indian Institute of Environment Management,    

                                                      Navi Mumbai, 400 076 India

Abstract: Air Quality Index (AQI), a system for transforming air pollution levels into a single number, aims at providing information about air quality in simple terms to general public. Any advance information about AQI can forewarn the public of unhealthy air and encourage people to voluntarily reduce emissions-producing activities and avoid exposures to polluted environment. Two mathematical models (i) meteorology-based air quality level predictions and (ii) meteorology forecasting, have been developed (based on four year data) using neural network to forecast AQI for following three days. The AQI forecasting model was concluded as being satisfactory and useful for information dissemination to general public.

Introduction and Objective

Air pollutants at ground level can be harmful to human health if their concentrations exceed certain acceptable levels. As pollutants accumulate in or near large metropolitan areas, this typically exposes people to unhealthy pollutant concentrations In light of the health effects of air pollutants, environmental agencies (e.g. USEPA, 1998) have been using air quality index (AQI) for public information (about air quality) and data interpretation. AQI is defined as an overall scheme that transforms the weighted values of individual air pollution related parameters (e.g. SO₂, CO, visibility, etc.) into a single number or set of numbers. An AQI is developed in Indian context (Sharma et al (2001)) which include three pollutants (sulphur dioxide (SO₂), nitrogen dioxide (NO₂) and Suspended Particulate Matter (SPM)). Segmented linear functions are used for relating the actual air pollution concentrations (of each pollutant) to a normalized number. The categories of index system are:

 0-100; Good, 101-200; Moderate, 201-300; Poor, 301-400; Very poor, 401-500; Severe

Ideally, the developed AQI should have accurate forecasting capability. Any advance information about AQI can forewarn the public of unhealthy air and encourage people to voluntarily reduce emissions-producing activities and avoid exposure to polluted environment. The purpose of this paper is to establish a mathematical model between the metrology and air quality in terms of air quality index and develop a forecasting mechanism using artificial neural network for the developed AQI.

A scheme utilizing neural network simulation is developed in the course of this work for prediction of AQI for a location in the city of Kanpur (longitude 88° 22' E and Latitude 26° 26' N), India. The neural network model is trained and tested using the air quality and meteorology data. Two neural network models were developed one for developing a meteorology -air quality linkage and second for forecasting the meteorology three days in advance. Sampling location at Agricultural University (AU), Kanpur was selected for this study as concurrent data of air quality (from National Ambient Air quality Monitoring Programme) and meteorology were available for last several years at this site, which were essential for this study.

Meteorology-Air Quality Model Development

Air pollution system has three components: emission source, transport medium (atmosphere) and receptor. Pollution reaching a receptor depends not only on the emitted quantity but also on the atmospheric dynamics. The impact on receptor can be estimated by developing source-receptor linkages through atmosphere.

A modeling system (source-receptor linkages) is necessary to predict impact on the receptor and translate the impact into an air quality index.

Steps in Model Formulation

The impact on receptor at any point is a function of source strength, and meteorological parameters like wind speed, wind direction, temperature, relative humidity, stability, rainfall etc. Thus, the AQI forecasting modelling completion is a three-step process.

(i) Air Quality Models: Establish the model that can predict the air quality for various pollutants in terms of meteorology

(ii) Meteorology Forecasting Models: Forecast meteorology for coming next days and combined the forecasted meteorology to forecast air quality using the model developed in step (i) above

(iii) AQI Forecasting: Estimate and forecast the air quality index based on forecasted air quality levels in step (ii) above.

Air Quality Models  

Since air quality data were available on three pollutants, SPM, SO₂, and NO₂, these pollutants are selected for model prediction and index formulation.

 Four-year air quality data for the site (Agriculture University, Kanpur) for the years 1997 to 2000 were collected from Central Pollution Control Board, Delhi. Similarly, meteorological data for the same time period and at the same location were also collected. Meteorological data included temperature, wind speed and direction, relative humidity and rainfall.

Initial data analysis showed significant seasonal variability in air pollutant concentrations. Monsoon period was the cleanest and winter months were the most polluted. Therefore, meteorological and air quality data were segregated as per the following seasons:

Winter: December, January, and February, march

Summer: April, may, June

Monsoon: July, August, September and

Post monsoon: October, November

In all, twelve models need to be developed for predicting air quality concentrations for each set of meteorological parameters. The twelve models include three models (for SO2, NO2 and SPM) for four seasons (winter, summer, monsoon and post monsoon). In other words, for a given season and meteorological data, the model should predict concentration of the pollutants. For this purpose, a neural network model (presented below) was developed.

Artificial Neural Network (ANN) Model for Air Quality Modelling

For construction of neural network model, the air pollution system is looked upon as a system that under varying sets of meteorological inputs (e.g. weather conditions) will respond by producing different sets of output. Such a model presupposes no prior knowledge about the structure of relationship that exists between input and output variables (e.g. pollutant concentrations).

Neural network comprises a number of interconnected entities, similar in many ways to biological neurons. The choice of the architecture of the network depends on the task to be performed. For modelling physical system such as air pollution system, a feed-forward layer is normally employed (Wasserman, 1989). It consists of layers of input neurons, and one and more hidden layers. For this study, a software "Neuro Genetic Optimizer" (NGO, version 2.6) is used. The modelling results are shown in Figure 1. The other details of model are given below.

ANN model  - Steps Followed

1.       The air quality data were categorized into four seasons as explained earlier.

2.       The meteorology data were regressed with air quality data. Only average wind speed and average temperature of the day showed significant correlation with air quality. Hence, average temperature and wind speed of day was only taken as input layer. Output layer consisted of concentration of SO₂, NO2, and SPM concentration.

3.       Four models one for each season were developed for each pollutant. For this purpose, data set of each season for four years was randomly divided into training and testing record (50% for each).

4.       Input data was normalized between -1 and 1 and output was subsequently de-normalized

5.       Weights were randomly initialized between +0.3 to -0.3.

6.       Momentum constant was selected in the range 0.1 to 0.3

7.       Multiple hidden layers were used; their number was selected using genetic algorithms.

8.       Selection of hidden layer and number of hidden neurons was decided using genetic algorithms

9.       Limit on the number of hidden neurons was set as eight.

10.    The minimum number of passes for each network was 20 and maximum was 50.

11.   

Learning rate constants were selected in the range 0.1 to 0.4.

Figure 1 Neural Network Modelling of NO₂, SO₂ and SPM

Figure 1 shows model performance and values of coefficient of correlation (R). The R-values were found significant at 1% level of significance in a statistical sense and model was accepted for further analysis.  It can be concluded that the developed neural network model establishes a reasonable relationship between meteorological inputs and air concentrations.

Meteorology Forecasting Model

As identified that for predicting air concentration only average temperature and wind speed of the day were adequate for predicting air concentrations. It was decided that attempt should be made to forecast AQI for next three days to provide enough warning period to people likely to be exposed from air pollution. To this end, two neural networks were developed one for forecasting mean temperature and another for wind speed. It may be stated that meteorology is a very complicated phenomenon and explaining it on basis of simple mathematical models is very difficult and one has to resort to ANN. The following steps were undertaken to develop two neural networks-based forecasting model one each temperature and for wind speed.

·         Collection of data : four years of meteorological data for various parameters like  temperature,  wind speed and direction, relative humidity, rainfall etc were collected,

·         Preparation of data: for advance forecasting for three days, input data for each set of output was prepared. A value of previous four days was found adequate to forecast temperature and wind speed for next three days. Therefore, the data were arranged in a set of four; total 1100 such sets were made and fed to the ANN model, and

·         Other steps were identical to those presented in the section on air quality modelling.

Based on the above exercise two models were developed. Model for temperature showed much higher accuracy in prediction than wind speed (Figure 2)

The temperature forecasting results are excellent, and it demonstrates the high predictive capacity of ANN. Both for temperature and wind speed, the value of R was found significant at 1% of level of significance, and model was accepted for further analysis.

AQI Forecasting: Results and Discussion

Having established two models one fore predicting pollutant concentration based on wind speed and temperature and other model for forecasting wind speed and temperature, the values of model-computed AQI can then be estimated for all days of air quality monitoring. Figure 3 compares the AQI values based on actual measurements vis-à-vis model computed AQI.

Figure 2 Neural Modelling results for temperature and wind speed forecasting

The model does not show very high correlation between computed and observed AQI (Figure 3). The probable reason for such a behavior can be attributed high background concentrations and low values of pollutant concentration that may be fluctuating from one day to another and model is not capable of picking these subtle changes. However, the aim of this study is to develop a method, which could predict AQI into a band like good, moderate, poor, severe etc. The model strength must be tested on predictability of AQI in the right band rather than in terms of absolute values. 

Figure 4 compares the model-computed AQI band vis-a-vis observed AQI bands. Considering the AQI bands, it can be stated that the performance of the model improves and it can be used for public information as an advance warning system. Results of other seasons are not shown here. It may be mentioned that for summer and winter the model has performed better than monsoon season (Aggarwal, 2001).

                                                                                                Figure 3 Comparison of AQI predicted and Actual        

                                                                                          for Monsoon.

Figure 4. Comparison of Observed and predicted AQI Bands For Monsoon season.

References

Aggarwal Sachin (2001).  Application of Neural Network to Forecast Air Quality Index. Thesis submitted in partial fulfillment of requirements for a degree in Bachelor of Technology, April 2001.

NGO (version 2.6) (2000). Neuro-Genetic optimizer URL www.bio-comp.com

Sharma M., Sengupta, B., Shukla, B.P. and Maheshwari, M (2000). Air Quality Index for Data Interpretation and Public Information. Presented in International Conference, Centre for Science and Environment, New Delhi, June 6-8, 2000.

USEPA (1998). Federal Register Vol.63 No. 236/Wednesday, December 9, 1998

Wassermann P.D. (1989). Neural Computing theory and Practice. New York Van Nostrand Reinhold.