Rolling PMF means spanning a PMF window and moving/rolling it over the entire dataset in order to account for changing factor profiles (fingerprints).
The settings for the rolling PMF can easily be defined in the PMF options (orange) panel. The rolling PMF window approach is controlled by the following parameters:
- “window width”: size of the PMF window expressed in days
- “window shift”: shift of the PMF window expressed in days
- “nb. of repeats”: number of repeats per PMF window
The PMF window must be smaller compared to the entire dataset, e.g. several days if the total PMF input comprises 1-2 years. More importantly, the window length should be chosen as such that the assumption of static factor profiles is legitimate. It has been shown that for urban ACSM datasets a window size of 14 days, a shift of 1 day and 50 repeats per PMF window works well.
Canonaco et al., 2021: https://doi.org/10.5194/amt-14-923-2021
Chen et al., 2022: https://doi.org/10.1016/j.envint.2022.107325