Evolution Stage Identification of Haze Pollution Episodes Over the North China Plain Using Constrained Dynamic Time Warping and Multiway Principal Component Analysis

To divide haze pollution episodes into different types, dynamic time warping (DTW) distance is employed to calculate the dissimilarities of different episodes. Then, a multiway principal component analysis (MPCA) method based on the constrained DTW, which is used to adjust episodes of different duration to the same length, is carried out to identify different evolution stage of each episode automatically according to the change rate of the first principal component in MPCA model. Contribution rates of meteorological factors and other air pollutant to the fluctuation of PM2.5 concentrations in different evolution stages are also calculated using the contributions to the principal component subspace. Results show that each episode consists of an early formational stage, an accumulation stage, a steady stage, and a clearing stage.