Temporal characterization of minute-level PM2.5 variation within a local monitoring network using DWT-DTW

Abstract

Meticulously understanding the local processes of particulate pollution can help evaluate and mitigate its impact on residents. However, this issue is not sufficiently addressed on intra-community scales, as the characterization and comparison of short-term variations in urban environment is challenged by dynamic misalignment. Our study integrates discrete wavelet transforms (DWT) and dynamic time warping (DTW) to tackle this problem, and formulate practical pollution process indicators from a time-domain perspective. Specifically, original PM2.5 series are decomposed into multi-scale wavelet approximates, which are temporally realigned using DTW. Then, background variation series is estimated by seeking commonality among the majority of stations on time and frequency domains. Indicators are calculated by comparing local PM2.5 variation and estimated background variation, thus describing the local processes of background pollution episodes and discovering possible local pollution incidents. On this basis, empirical analysis in a campus network identified distinct information conveyed on different temporal scales. Case study discovered four phases within a 3-day period based on changes in dominating temporal scale of the background series, which is consistent with known pollution processes. Respective investigations of each phase show good capability in reflecting the influence of terrain, meteorology and dominant emission sources. Inter-station comparisons suggest significant influence from micro-scale spatial environment even when subject to exterior emissions. In general, the method can provide pertinent indicators for evaluation, prediction and optimization efforts in local pollution mitigation. The empirical results imply potentials of urban design measures for such mitigation, on which our future studies will focus.