I thoroughly reviewed this manuscript. The manuscript titled “Short-term Source Apportionment of Fine Particulate Matter with Time-dependent Profiles Using SoFi: Exploring the Reliability of Rolling Positive Matrix Factorization (PMF) Applied to Bihourly Molecular and Elemental Tracer Data” suggests that short-term PMF analysis with the a-value constraints in SoFi can be utilized to apportion primary sources accurately. This may have implications for short-term pollution episodes management. The topic is interesting and meaningful. However, there are some problems in the article. I would recommend authors concerns on the following comments. Minor revision is recommended.

1. Line 35, “The implementation of stringent control measures since 2013 has led to declining concentrations of PM₅ in many megacities in China”, it is recommended to provide specific data on the decrease in PM_2.5.
2. Line 55, “This limitation explains the common observation that PMF with robust mode tends to underestimate the high concentration data while overestimating the low concentration data.”, it is recommended to modify this sentence to make it coherent with the preceding and following sentences.
3. Line 99, please define SOA when it first appears in the text.
4. Line 121, it is recommended to explain what the meaning of k and j in Eq 4.
5. Line 170, Do you think these “18 days” must be continuous? Does discontinuity have an impact on the results?
6. Line 121, please define Q/Qexp.
7. Line 221, please explain what N_{firework_data}
8. Line 240, pleasedefine
9. Line 273, “Secondary nitrate, SOA_I and SOA_II showed large variations, with an average relative difference of 173%, 162%, and 75%, respectively.”, it is recommended to provide a more detailed explanation of the reasons for the significant differences.
10. Abstract and conclusion of the work can be improved as per the methods applied.

The manuscript titled "Short-term Source Apportionment of Fine Particulate Matter with Time-dependent Profiles Using SoFi: Exploring the Reliability of Rolling Positive Matrix Factorization (PMF) Applied to Bihourly Molecular and Elemental Tracer Data" presents extensive datasets of real-time chemical characterization to employ both traditional and rolling Positive Matrix Factorization (PMF) techniques utilizing the SoFi software. The comparative analysis between traditional and rolling PMF methodologies is particularly intriguing, primarily in the context of effectively modeling primary sources, which exhibit relatively minor variabilities. Conversely, the study reveals substantial variations in the case of secondary factors. This result deserves more explanation and details regarding the relative differences. Furthermore, from what has been presented in the paper, the rolling PMF should always be based on source profiles from traditional PMF (Conclusion lines 300-302).  This prompts the question of how to address scenarios where no source profiles are available, such as for newly emerging sources or in regions lacking local source profiles.

I recommend major revisions before considering the manuscript for acceptance:

Introduction: It is imperative to clarify that the traditional PMF was conducted using the US EPA PMF, as indicated in your prior publication (Wang et al., 2022b). This should be explicitly mentioned unless  this is not the case. 

Section 2.2: This section is lacking essential details regarding the specific species incorporated into the model, the total number of data utilized for traditional and rolling PMF, and the temporal resolution applied (e.g., 2-hour or 1-hour intervals). These specifics are fundamental for a comprehensive understanding of the methodology.

Line 128: The choice of an 18-day duration for data inclusion needs further justification. Is this duration adequate for achieving robust and optimal a-values? 

Line 147: The difference in the correlation is clear for K and levoglucosan but what about K and Pb? They're still well correlated even in the CNY period? 

Line 150: A discrepancy is noted between the number of factors reported in the paper by Wang et al., 2022b (14 factors) and the present manuscript (10 factors). The differences in PMF methodologies should be thoroughly explained and supported with additional details, either integrated into the main text or provided as supplementary information. Additionally, the manuscript should incorporate validation results for traditional PMF, includingo Bootstrap analysis, DISP analysis, reconstruction of species, Q/Qexp... .

Section 3.2: What is the reason behind not adding an a-value to the secondary sources? 

Section 3.4: The variabilities observed in secondary profiles demand a more in-depth exploration. Possible explanations, including the potential impact of poorly resolved profiles in traditional PMF, should be examined and discussed in greater detail.

In the SoFi software, what are the validation methods in order to ensure that you have good results?