Identifying Contaminant Intrusion in Water Distribution Networks under Water Flow and Sensor Report Time Uncertainties
Round 1
Reviewer 1 Report
Their proposed method of detecting a single contamination point in space and time looks promising. However, they failed to discuss what if there are more than one source of contamination. How would they deal with this uncertainty? Is there a way to confirm a single injection point? I know that they made an assumption about that to simplify their model, but in the real world, how feasible would their assumption be that you only have one injection point? Isn't there some possibility that, in some case, the model would locate one incorrect injection point when there were multiple? As far as I can see, this model requires a strong assumption, but it would be difficult to validate it. I believe that this discussion is very critical, but is missing.
I reviewed their methods below.
Figure 1: Define nr.
Figure 2: Did you apply one CV across the entire simulation period or different CVs at different time steps? Don't those time steps with a higher average water consumption have a higher uncertainty? Please clarify this. From Figure 2b, it looks like the variability does not change with the magnitude of water consumption or time.
L147: Did you do a similar MCS for the reporting time?
Figure 5b: What do yellow and black colors mean?
L185: Why not 3, 7, and 10 when they are all upstream of at least one of the three sensors? Is it only because you assumed that there is only one source of injection?
L187: Are these times based on flow velocity information? How did you obtain these times based on the sensor report?
Figure 5b: Again, it is not clear how you calculated those times that are not reported by the sensors.
L207: Define nt.
L209: How do you initially determine the minimum nt?
Eq. (4): Please elaborate more on BV. Why did you take squares? Basically, using BV, you want to exaggerate the dispersion of time when there is less information (smaller nt_j), but how would you justify the use of squares and why BV?
L216: Please clarify that you are also switching the meaning of small (less dispersion) and big (more dispersion) of the dispersion measure to min-max normalized 0 (more dispersion) and 1 (less dispersion) of the score.
Eqs. (5) & (6): Please explain the meanings of LOR_j and MOR.
Table 3: I thought that you already have reported times from different sensors. What are these times in Table 3? Are you even simulating reported times? Are these within the 2 hour max delay you mentioned before from the reported times? If so, did you use a uniform distribution to generate these random times?
Figure 7: Again, how did you calculate arrival times that were not recorded?
Figure 8: Is the average absolute deviation of 00:16 the average of the nine numbers in the second table? Please explain how. Do the red rectangles mean that you discarded those numbers? Please clarify this. Why didn't you discard 00:40 and 00:20 when their absolute values are greater than 00:16? In the 3rd table, you discarded the last column because the standard deviation of 20, 50, and 0 is 25 > 16, but isn't the standard deviation of the 1st column (0, 10, 40) 20.8 (sample standard deviation) or 17 (population standard deviation), which is greater than 16?
Also, the 2nd column seems to use the sample standard deviation of |-10|, 0, and 20, but the 1st column is closer to its population standard deviation 17. Please clarify this.
L304: Please unindent "where [...]".
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments
The manuscript describes the development of a model to identify contamination sources in water distribution systems. Its novelty compared to other studies in the field is that the developed model considers the uncertainties in water flow demand and in the time reported by the sensor.
I find the topic of the manuscript of high importance in the field of water distribution systems.
The manuscript is written very well making it easy to follow and understand.
I have just the following comments:
- In the model developed and the example presented the concentration of the contaminate was not considered. Further, only one single injection point was considered. Please add a discussion about the possibility to add the contaminate concentration and more than a single point of injection into the model.
- The model was applied to one network. Will using a network with other characteristics affect the results presented here? Which characteristics and how? Please refer to that in the text.
- Figure 1 – please refer to the middle and right part of the figure in section 2.2 and 2.3, respectively.
- Line 141: 100 random sets were created. Why 100? Shouldn’t this value depend on the size of the network? Please explain in the text.
- In section 3 (application). How were the 3 scenarios chosen? Why were the specific injection points chosen? Will choosing different points affect the results? Please explain in the text. If differences are expected, then please add more scenarios to present these effects.
I suggest accepting the manuscript for publication in Water journal, provided that the appropriate modifications, as concerning the comments described above will be made.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This article by Marlim and Kang develops a model to assess the uncertainties in monitoring contaminants in water distribution network; Contamination-probable setting identifier (CPSITE) . The authors validated the model by considering water flow and sensor report time in a water network system and applying Gausian and Monte Carlo simulations. Although the authors did not compare the model with other methods of assessing uncertainties in intrusion of pollutants in water flow system and report time, the article has scientific merit. It is well written and presented.
Author Response
Thank you for the positive comments. We ask your understanding that we could not conduct the comparative analyses with other models because the models are not available or replicable, and they dealt with different uncertainties.
Round 2
Reviewer 1 Report
Most of my comments have satisfactorily been addressed except for Eqs. (6) and (7). The authors claimed that they followed the journal guideline, but indentation for the "where" clause after display equations is inconsistent across the manuscript. This is a minor stylistic issue.
