Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Fitting a lognormal distribution to air quality data observed with measurement error

Thesis/Dissertation ·
OSTI ID:6970531
A model based on a lognormal distribution of air pollutant concentrations coupled with a normal distribution of measurements of these concentrations is examined. The distribution of the measurements is constructed assuming unbiased measurements with a constant coefficient of variation. Various methods of fitting the model are examined and tested. Conditions under which the method of moments can be used are indicated. Minimum distance methods were found to require large amounts of computer time and were discovered to be insensitive to the value of the coefficient of variation. Maximum likelihood methods provided reasonable estimates but also required large iterative, numerical efforts to calculate these estimates. The EM algorithm offered no improvement to this situation. A method combining maximum likelihood and the method of moments was discussed and found to be the most promising. The model was fit to simulated data and to particulate data from air monitors in Philadelphia. The simulated data gave an indication of the variance and bias of the estimators. The bias and variance exhibited by the simulated data indicated that the parameters estimated from the particulate data were reasonable. Numerical integration methods were also discussed along with Peano kernal techniques used to estimate quadrature errors generated in parameter estimation.
Research Organization:
North Carolina State Univ., Raleigh, NC (United States)
OSTI ID:
6970531
Country of Publication:
United States
Language:
English