Profiling and forecasting air pollutant index for Malaysia

Abstract

Detection of poor air quality is important to provide an early warning system for air quality control and management. Thus, air pollutant index (API) is designed as a referential parameter in describing air pollution levels to provide information to enhance public awareness. This study aims to study API trend, time series forecasting methods, their performance evaluations and missing values effect for accurate early warning system using several approaches. First, a calendar grid visualization is introduced to effectively display API daily profiling for the whole of Malaysia in identifying the exact point of poor air quality. Second, comparisons between classical and modern forecasting methods, artificial neural network (ANN), fuzzy time series (FTS) and hybrid are carried out to identify the best model in Johor sampling stations; industrial, urban and suburban. Third, due to the issue of different perfect score in existing index measurement to evaluate forecast performance, a combination index measures is proposed alongside error magnitude measurement. Fourth, decomposition and spatial techniques are compared to find the effect of high accuracy imputations in API missing values. The finding presented that the air quality trend across the day, week, month and year are more significant due to the daily arrangement in the calendar grid visualization. The ANN model gives the best forecasting model of API for industrial and urban area while the hybrid model provide the best forecasting for suburban area. The forecasting performance for industrial and urban areas improve between 14% to 20% and 20% to 55% in error magnitude and index measurements, respectively when high accuracy missing values imputation is conducted. In conclusion, the profiling using calendar grid visualization is useful to guide the control actions of early warning system. Forecasting using modern methods give promising result in API and the improvements in measurements will assist in choosing the best forecasting method. Missing values imputation in data series can enhance the forecasting performance.