Source emissions with high covariance degrade the performance of multivariate models, and often highly-time resolved data is needed to accurately extract the contribution of different emissions. Here, we use highly time-resolved size segregated elemental composition data to apportion the sources of the elemental fraction of PM in Zürich (May 2019-May 2020). For data collection, we have used an ambient metals monitor, Xact 625i, equipped with a sampling inlet alternating between PM_2.5 and PM₁₀. By implementing interpolation and a newly proposed uncertainty estimation methodology, it was possible to obtain and use in PMF a combined dataset of PM_2.5 and PM_coarse (PM_10-2.5) having data from only one instrument. The combination of the inlet switching system, the instrument's high time resolution, and the use of advanced source apportionment approaches yielded improved source apportionment results in terms of the number of identified sources, as the model, additionally to the diurnal and seasonal variation of the dataset, also utilizes the variation from the size segregated data. Thirteen sources of elements were identified, i.e., sea salt (5.4%), biomass burning (7.2%), construction (4.3%), industrial (3.3%), light-duty vehicles (5.4%), Pb (0.7%), Zn (0.7%), dust (22.1%), transported dust (9.5%), sulfates (15.4%), heavy-duty vehicles (17%), railway (6.6%) and fireworks (2.4%). The Covid-19 lockdown effect in PM sources in the area was also quantified. High-intensity events disproportionally affect the PMF solution, and in many cases, they are getting discarded before analysis, removing thus valuable information from the dataset. In this study, a three-step source apportionment approach was used to get a well-resolved unmixed solution when firework data points were included in the analysis. This approach can also be used for other sources and/or events with very high contributions that distort source apportionment analysis. Optimized source apportionment techniques are necessary for effective air pollution monitoring.