WIT Press


Air Quality Forecasting In A Large City

Price

Free (open access)

Paper DOI

10.2495/AIR080031

Volume

116

Pages

8

Page Range

21 - 28

Published

2008

Size

385 kb

Author(s)

P. Perez

Abstract

We describe the different air pollution statistical forecasting models that have been used in Santiago, Chile during the fall/winter period for the last ten years. Effort has been concentrated on particulate matter PM10 for which a standard of 150 µg/m3 for the 24 h average is currently established. Inputs to the models are concentrations measured at several monitoring stations distributed throughout the city and meteorological information in the region. Outputs are the expected maxima concentrations for the following day at the site of the same monitoring stations. Forecast values using neural network models are compared with the results obtained with linear models and persistence. Recently, a clustering algorithm has appeared as a potentially useful tool to detect high concentration episodes in advance. Keywords: particulate matter forecasting, neural networks, linear models. 1 Introduction Air pollution has been a major concern in the metropolitan area of Santiago, the capital of Chile during the last 15 years. Together with Sao Paulo, Mexico City, and some Chinese cities it is considered as one of the most polluted in world. Several factors concur to create unfavorable conditions for air pollutant dispersion. The city is located in a valley that has an extension between 70 and 80 km in the north-south direction and approximately 40 km in the east-west direction. To the west we find the Andes Mountains and to the east a coastal range. Some elevations to the north and south trap the air and air pollutants in a region of poor air circulation, which is enhanced during fall and winter when strong thermal inversions prevent vertical dispersion. During this period of the year, the 24 hour moving average (24MA) of PM10 is used as an indicator of air quality.

Keywords

particulate matter forecasting, neural networks, linear models.