Analysis of the temporal variability of indoor particulate matter concentrations using Blind Source Separation methods: a comparative study
Résumé
This study explores the possibility to identify the main indoor sources of particulate matter in two offices using statistical methods commonly called Blind Source Separation (BSS) techniques. Two monitoring campaigns of particulate matter were performed. The first one was carried out in a cellular office mainly occupied by a single person; it lasted 45 days with a sampling time-step of 1 minute. The second one was performed in an open-plan office occupied by 6-8 persons during 155 days with an hourly time step sampling. The indoor air particulate matter was sampled using an optical particle counter (Dust Monitor 1.108, Grimm) which provided the number of particles per liter, for 15 size bins within a size range of 0.3-20µm. In addition, several other parameters were measured during the two campaigns. The concentration of CO2 was monitored as a tracer of the human activity in both cases, and indoor and outdoor climatic parameters for the open-plan office. The sources of variability of the concentration of particles are unknown in both cases; we attempt to estimate them by inverse modeling using only receptor information (without a priori knowledge of the pollutant transformation or transportation). The studies on outdoor environment use mainly chemical mass balance (CMB) or Positive Matrix Factorization (PMF) to assess the contribution of the various sources. Some other methods were developed in the field of Signal Processing for the same purposes, e.g. Independent Component Analysis (ICA) or Non-Negative Matrix Factorization (NNMF). The latter methods have not been used in the field of indoor air quality for source identification and apportionment. We attempt to evaluate their potential in the indoor environment. In this study, we compare the results obtained using four different methods: principal component analysis (PCA), ICA, PMF and NNMF. These results obtained are both the time variation of identified sources and their contribution for each particle size. The interpretation of the source temporal profiles is realized with the help of the other parameters such as CO2. For example, the first principal component of PCA is highly correlated to CO2 level, and indicates a human activity source profile. In conclusion, this study is a preliminary analysis on indoor particle source apportionment to identify the most adapted method for blind source separation of particles indoors.