Department of Systems and Control - Jožef Stefan Institute

Development and implementation of a method for on-line modelling and forecasting of air pollution

Principal investigator:

Duration:

August 1, 2013 – July 31, 2016

Funding:

ARRS - Slovenian Research Agency, L2-5475 (C), co-financed by Meis storitve za okolje d.o.o.

Abstract:

Air pollution is a serious environmental problem in the world, as well as in Slovenia. To monitor the current condition of pollution, data from measuring stations and model spatial calculations are used; to predict pollution in the future, only the models which can be divided into still unreliable three-dimensional pollution representation models, and raster location pollution prediction models, can be used.

The purpose of the research project is to develop a Gaussian-process modelling method and model for accurate ozone predictions at the most heavily burdened locations in Slovenia. To this end, we will combine the scientific experience of two groups that have extensive references in the development of such models. The main approach consists of real-time learning of a temporally variable model. For this purpose, Gaussian-process models (GPMs) will be used. This method is appropriate for the identification of very complex nonlinear processes according to the black-box method, and has proven extremely efficient in the field of modelling of complex, nonlinear dynamic systems. The second approach deals with models based on a multilayer perceptron artificial neural network (MLP) which is proven to be a universal approximator for a nonlinear system of functions of several independent variables. The MLP application methodology for the field of air pollution prediction has already been developed. For the GPM, we will redefine and upgrade the methodologies developed by the MEIS team for the purposes of MPL. The implementation of a new GPM development methodology in the field of atmosphere processes will represent the main scientific result. This method enables dynamic adjustment to the process, and has so far never been used in this field, except in a preliminary study performed by the Jožef Stefan Institute (IJS) which gave very good results.

The applied project result will be the test environment - test bed - used for the elaboration and testing of prediction models. For real-time application, an efficient ozone-concentration prediction system for selected locations across Slovenia will be elaborated. The developed method and the resulting algorithm will be evaluated mainly based on the measurement data from the coastal ozone measurement stations, where pollution is the most problematic. New efficient models developed within the scope of the project will be used for on-time and efficient alerting, which will result in better healthcare prevention measures and compliance to EU directives.

The IJS and MEIS project consortium combines knowledge in the field of Gaussian-process modelling, experience in the field of air-pollution modelling, neural networks, and extensive experience in the field of environmental measurements. It possesses the required computer equipment, and the measurement data is publicly accessible through the Slovenian Environment Agency; and MEIS regularly produces its own detailed weather forecast.

Links:

Ozone forecast for Slovenia: http://www.meis.si/ozon/
Air-quality and weather forecasts for Slovenia: http://www.kvalitetazraka.si/zasavje/index.php

Project partners:

Publications:

KOCIJAN, Juš, GRADIŠAR, Dejan, BOŽNAR, Marija, GRAŠIČ, Boštjan, MLAKAR, Primož. On-line algorithm for ground-level ozone prediction with a mobile station. Atmospheric environment, 2016, vol. 131, 326 - 333.

KOCIJAN, Juš, HANČIČ, Marko, PETELIN, Dejan, BOŽNAR, Marija, MLAKAR, Primož. Regressor selection for ozone prediction. Simulation modelling practice and theory, may 2015, vol. 54, p. 101-115.

BOŽNAR, Marija, MLAKAR, Primož, GRAŠIČ, Boštjan, CALORI, Giuseppe, D'ALLURA, Alessio, FINARDI, Sandro. Operational background air pollution prediction over Slovenia by QualeAria modelling system - validation. International journal of environment and pollution, 2014, vol. 54, no. 2/4, p. 175-183.

PETELIN, Dejan, MLAKAR, Primož, BOŽNAR, Marija, GRAŠIČ, Boštjan, KOCIJAN, Juš. Ozone forecasting using an online updating Gaussian-process model. International journal of environment and pollution, ISSN 0957-4352, 2015, vol. 57, no. 3/4, str. 111-122, doi: 10.1504/IJEP.2015.074494.