Event № 309
Advisor: Assistant Professor Barak Fishbain
Abstract: Air pollution is a significant risk factor for multiple health situations. In addition, it causes many negative effects on the environment. Thus, arises the need for assessing air-quality. Air quality modeling is an essential tool this task and is in use in many studies such as air quality management and control, epidemiological studies and public health. Today, most of air-pollution modeling is based on data acquired from Air Quality Monitoring (AQM) stations. AQM provides continuous measurements and considered to be accurate; however, they are expansive to build and operate, therefore scattered sparingly. As the number of measuring sites is limited, the information obtained from those measurements is generalized with mathematical methods.Here we introduce two methods to improve the spatio-temporal coverage. The first method, a new interpolation scheme, will expand the scope of the spatial coverage in order to infer the pollution levels in the entire study area. The second is a long-term forecasting method, to implement a better and wide perspective of the temporal coverage. Many researches in air quality modeling uses interpolation schemes such as IDW or Ordinary Kriging. Yet, the mathematical basis of those schemes defines that the extremum value obtained at the measuring places (without considering edge effects). In addition, they are not considering the location of pollution source or any physicochemical characteristics of pollution, hence does not reveal the real spatial coverage. Our interpolation scheme takes into account patterns of dispersion and source location. Source detection is achieved through a novel Hough Transform-like technique.Extending the temporal coverage of the measuring array is achieved through long-term forecasting. Nowadays there are only short-term forecasting methods (24-72 hours ahead), no method exists for long-term (e.g. a year) forecasting. Discrete Time Markov Model is a well-known probabilistic model used to describe and analyze stochastic processes. Here we first define and introduce a method for long-term forecasting based on Discrete-time Markov model for a better temporal coverage.These building blocks which, will be presented in the talk, facilitate the future study of spatio-temporal interpolation methods, which improve the current state-of-the-art by devising new source-location based interpolation methods.