Abstract
New monitoring technologies have now permitted the measurement of a variety of chemical species in airborne particulate matter with time resolution as high as 10min to 1h. There are still species that are measured with longer integration periods such as several hours to a day. These data from different measurement methods produce a data set of mixed time resolution. Traditional eigenvalue-based methods used in solving multivariate receptor models are unable to analyze this kind of data set since these data cannot form a simple matrix. Averaging the high time resolution data or interpolating the low time resolution data to produce data on the same time schedule is not acceptable. The former method loses valuable temporal information and the latter produces unreliable high resolution series because of the invalid assumption of temporal smoothness. In the present work, a solution to the problem of multiple sampling time intervals has been developed and tested. Each data value is used in its original time schedule without averaging or interpolation and the source contributions are averaged to the corresponding sampling interval. For data with the highest time resolution, the contributions are not actually averaged. The contribution series are smoothed by regularization auxiliary equations especially for sources containing very little high resolution species. This new model will be explored using data from the Pittsburgh supersite.
| Original language | English (US) |
|---|---|
| Pages (from-to) | 4909-4920 |
| Number of pages | 12 |
| Journal | Atmospheric Environment |
| Volume | 38 |
| Issue number | 29 |
| DOIs | |
| State | Published - Sep 2004 |
| Externally published | Yes |
Keywords
- Aerosol
- Factor analysis
- Multilinear Engine (ME)
- Multiple time resolution
- Non-parametric regression (NPR)
- Source apportionment
ASJC Scopus subject areas
- General Environmental Science
- Atmospheric Science