Leaking Gas Source Tracking for Multiple Chemical Parks within An Urban City
Round 1
Reviewer 1 Report
This study simulated diffusion of 79 potential leaking sources with consideration of different weather conditions and complex urban terrain by AERMOD and used 61 sensors to monitor the gas concentration within such a large scale. I am wondering why should 61 sensors be used to identify a pollution source, how effective would it be if only 10 or fewer sensors were used? Some other comments as follows:
1. Page 1, line 27. The authors said "Many researchers use CFD to take complex terrains into consideration while ordinary Gaussian diffusion model cannot do such things.". Why don't you use CFD method?
2. Page 2, line 57-69. The authors summarized the 5 contributions of this paper. However, I think at least # 1 and # 2 are not contributions.
3. Page 4, line 127. What is the data table? How did "10296" come about?
4. Page 11, line 284. By the hybrid training strategy, the accuracy improves from 98.86% to 99.14%. What is the significance of this increase in accuracy?
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The paper entitled “Leaking Gas Source Tracking for Multiple Chemical Parks within An Urban City” shows a sensor-based fully connected model trained with a hybrid strategy to track the leaking source in chemical parks within an urban region.
The scientific quality is excellent and the explanations are clear. The paper deserves to be published in MDPI Algorithms. You will find here below some points of enhancement:
- It exists a large variety of sensors. Depending on the detected gas and the type of sensor, the quality and the limit of detection can be very different. The authors should consider this aspect and precise what type of sensor are expected and detail the limitation due to the monitoring devices.
- Figure 11, the addition of the standard deviation on the diagram is an interesting complementary information.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors utilized a machine learning model to estimate the location for gas leaking based on simulation data. It is not the difficulty to apply the machine learning model to solve the source estimation problem. The most difficulty for this issue is that how to verify the model with experiment data and how to make sure the model has good generalization performance due to the variation of the atmosphere. The dispersion data from simulation model especially with Gaussian form， do not present the actual scenarios. All these have not been answered in this research. further , there are still following comments:
1. The literature review is not enough. Some origin and important researches were not discussed. The discussion aboout diffent methods are not enough.
2. The contributions of the authors said in section 1 is not innovative. Hybrid taining is a common strategy in machine learning. The sensor data failure have been considered in other researches.
3. What is the assumption for the simulation in this research?
4.In this simulation scenarios, how to determine the potential sources?
5.The detail about the simulation data is absent? The information about the wind speed, wind direction, atmosphere stability, leaking rate and other more parameters are not been found in the manuscript.
6.The detail about training and test data set and process were absent. What are the inputs for the CMM model?
7. Why 10% of sensor fail is selected in the research? There is not clear explanation for the value of 10% . Besides the sensor fail, a more problem for gas sensor is that many more sensor can not capture the valid concentrations in actual urban scenarios due to the atmosphere and environment conditions, which should be considered in the research.
8. In section 4.4, the test accuracy of SVM is only 28%, which seems not reasonable. Why? The difference between the sensor and potential source may not the only reason. Data source and model structure should be discussed and analyzed.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
I don't have more comments.
Reviewer 2 Report
na
