Laboratory Development of an AI System for the Real-Time Monitoring of Water Quality and Detection of Anomalies Arising from Chemical Contamination

Round 1

Reviewer 1 Report

This manuscript reports a machine learning approach to create a water quality warning system by detecting anomalies. This is a very important topic in water research. This manuscript mainly focuses on the machine learning aspect with less emphasises on the water quality, laboratory experimental and analytical chemistry aspects. I recommend improvement in those aspects. In addition, the experimental design wasn’t focussing on getting high quality analytical data. Using a running tap, the supplying water quality can change on top of the chemical spikes, hard to conclude the changes (anomalies) are caused by the spike chemicals or water quality change from the tap water, on top of that turbidity and particles can affect absorbance which added another variable of uncertainty. The experimental design should be discussed and what assumptions have been made are required. I understand the focus of this paper is about machine learning / data analytics aspect, still it is important to ensure to have a sound analytical chemistry component.

Some items require attention.

1. Line 89, UV-vis spec., it should be UV spec., 200 - 360 nm is not covering the visible region. In this paper, it should be UV data not UV-vis data. Need correction.

2. Line 95, not clear how the continuous flow-through system was setup, the sensors were in the water bath, 3 pumps to circulate the water in the water bath, how was the water flow through the sensors? If the water only circulating in the water bath, it is not a flow through system? I would expect a flow cell configuration to call flow-through, otherwise just circulating.

3. Line 102, as per above, if the water is flow-through over a few days, one pass or recycled back. I believe the authors used a large container to hold the test water sample spiked with chemicals and flow it to the water bath, it wasn’t clear from the procedure that the test sample is circulated back to the container or discarded. Need to clarify in the section.

4. After I read your fig 1, I believe the water source is the tap water (a running tap), I initially read it as a batch of water collected for a source, as you have info about the lake etc. Perhaps if it can be clarified in the procedure it would be better. In addition, how the chemicals were spiked? Not clear in the procedure. I assume just spike in the running water, one dose or multiple doses? Concentration? Over a period?

5. Line 121, I believe the authors need to provide evidence that the rapid increase is caused by biofouling, should also include details of the correction procedure using polynomial equation. How was the polynomial equation found? This is important for the accuracy of the analysis. In addition, if cleaning can avoid the issue, why not doing it for this experiment. My comment is about how the authors can valid the data obtained is the true water quality change.  

6. Fig 1, I suggest to use day as unit rather than minute. 

7. Fig 1, are the data corrected using the polynomial described in the procedure? How were turbidity and particles compensated, the probe has built in algorithm, or no correction needed because the tap water has low turbidity that not affecting the absorbance readings? This needs to be addressed.

8. Fig 1, base line representing the water quality at that time, is the changes caused by water quality, biofouling on the lens or change in turbidity / particles affecting the absorbance? This needs to be discussed.

9. Fig 2, two concentrations were used, the procedure of how the solutions were prepared should be stated in the method section. How the measurement was conducted, chemical spiked in the flow system while flowing or static measurement, if it is flowing how the concentration was determined? All need in the procedure.

10. Line 199, it stated the fluorescence data are normally distributed, why it was normally distributed, need explanation. And also what is normally distributed? You were monitoring a single fluorescence point over a period of time, how they were normally distributed.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Journal Name: Water

 

Title:  Laboratory development of an AI system for real-time monitoring of water quality and detection of anomalies arising from chemical contamination

 

Monitoring water quality is very important in this era, the authors used an AI system for this purpose. It’s an interesting and suitable research article for the Water MDPI journal. The article fits the scope of the journal. I recommend its publication with minor revision. Here I am mentioning my detailed comments. 

 1.      Introduction section looks very weak authors need to improve and discuss how to relate AI with UV-vis and fluorescence.

2.      The quality of Fig 4 has to be improved

3.      I suggest giving some reference books for equations 1 and 2

4.      What is the error percentage of work that has to be mentioned?

5.      What are all the drawbacks to using AI with UV-vis and fluorescence

6.      Can the result be reproducible? need to be explained

7.      Conclusion is too short and has to be improved with deep insights 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

I satisfy with the modified version