This manuscript presents a Bayesian modeling approach to understanding factors affecting temporal variability in stream water quality. Overall, I think the manuscript is well written and will become a worthwhile contribution to the hydrological community after moderate revisions. Below I provide some comments to the author, which I hope can help improve the manuscript. 

1. The authors have made it explicit that the current work follows previous study investigating water quality variability in the same region (Liu et al., 2018). There are also other publications from these authors, e.g., Guo et al., 2019, 2020. The discussion section seems not provide much comparison or synthesis of the results from these different but related studies, which appears to be a missed opportunity. I am aware some of these studies focused on temporal patterns and some on temporal patterns. It can potentially become a nice addition to the manuscript and a contribution to the community if the authors can provide some reflection on what different modeling techniques they have used and what new insights on water-quality patterns they have learned from those techniques. 

2. The authors have analyzed nine water quality constituents. While I do appreciate the amount of efforts the authors invested in data analysis and modeling, I wonder if it helps everyone stay focused if the authors were more selective on the constituents. Since a key message from this work is on the different drives of particulate and dissolved constituents, it may be sufficient to select two constituents from each category, as opposed to showing data and results for all nine constituents. 

3. Of the two clusters of sites (Figure 2), Cluster 1 sites are quite concentrated, whereas Cluster 2 sites are much more scattered. Also, there seems to be more sites in cluster 2 than cluster 1. I noted that the Bayesian modeling framework was applied to the two clusters independently, I wonder if any of these two aspects (geographical proximity and number of sites) could potentially affect your models and comparison of results between the two clusters. In addition, have you considered developing a single Bayesian model on all sites with the cluster assignment has an explanatory variable?

4. Line 35: In addition to sources, mobilization, and delivery, “transformation” should be included. 

5. Section 2.2.3: The authors have quantified the correlation between explanatory variables (Figure B1). Have you considered excluding some variables based on the correlations? If any two variables are highly correlated, it may be wise to keep just one of them in the models.

6. BMA model coefficients plots (Figure 5 and other related figures in the SM): I found it difficult to compare the patterns across clusters or among constituents because the variables are not displayed in the same order in these panels. 

7. Predictive model performance (Section 3.2 and Table 4): The NSE values are not high, some are very low. This seems to limit the utility of the proposed Bayesian approach, which the authors should discuss and defend against. 

8. Line 415: Again, the effect is not only on transportation but also on transformation. Specifically, temperature is expected to affect the intensity of biological processes, e.g., denitrification.  

Liu et al. conducted an improved Bayesian approach to evaluate the temporal variability in stream water quality and the related key factors. This study aimed to: i) identify the key influencing factors, and ii) predict the temporal variation, taking advantages of multiple locations and multiple water quality monitoring data. In addition, authors divided the study sites into two clusters and analyzed separately, which might avoid potential uncertainty issues caused by a single model, and improve the scientific and reliability of the modelling results.

This study is an interesting topic and generally well written. It contributes to our knowledge of both the further application of the developed Bayesian model framework and the understanding the temporal water quality variability in the Great Barrier Reef catchments. In general, this piece of work could be considered for publication after some unclear concerns were addressed.

Major Comments:

1. Section 2.2.2   The authors gave a detailed process of data extraction and processing. Among them, it was noticed that “The start and end points of a specific event were determined by using a local minimum method that calculates the first derivative of the streamflow record (separated from baseflow)”. Basing on your data processing method, when can be identified as the start or end points? I think more details of the key standard or parameter maybe better for the readers to further understand your approach.
2. The authors divided the site locations of the GBR catchments into two clusters (wet and dry), and modelled separately. The advantages of the subsequent result are obvious, i.e., pertinence, reliability and so on. However, whether the strong pertinence will reduce the universality of this approach and limit its universal application? And if it is necessary to add the model and discussion of all sites?
3. The authors targeted nine common water quality indicators, including sediments, nutrients and salinity. But in the nutrients part, they only focused on N and P, without any constituents about Carbon studied. Why? Please explain it.
4. 2.2.2 again “The event-mean concentration (EMC) was then calculated for each event that had at least two samples on each of the rising and falling limbs of the hydrograph.” Table C2 showed the Number of EMCs for each constituent. So what is the approximate amount of data per event? Why you set two samples as the minimum limitation? whether two samples are too few?
5. I also noticed that you normalized the data of each event first and then calculate the Event mean concentration. If this process is necessary?

Minor comments:

1. Fig 1d  six_NRM regions
2. Table 1  Delete the comma at the end of the sentence in the item “Land use/land cover” of Cluster 2
3. L274 Delete the full name of “MCMC”, which has appeared in the line 261.