Time-series of half-hourly PM10 concentrations and meteorological data measured near industrial sites where dusty materials are processed can be used to estimate PM10 source strengths for these activities using reverse modelling. Reverse modelling starts with building an over-determined system of linear equations:
+ Σ i=1, NComputed_C
where the N values of Q
are determined by least squares regression. The paper focuses on regression solutions tainted by noise fitting that are physically meaningless, on rules to reduce the risk for noise fitting, and on how to use cumulative frequency distributions, time series for different averaging times and pollutant roses to obtain a physically sound solution.
Keywords: reverse modelling, fugitive sources, least squares regression, colinearity, noise fitting, cumulative frequency distribution, time series, averaging time, pollutant rose, IFDM, PM10, air quality, air pollution