Reducing GPS Error for Smart Collars Based on Animal’s Behavior
, Otabek Sattarov 1, Cheol Won Lee 1, Hyun Kyu Kang 2, Myeong-Cheol Ko 2, Ryumduck Oh 3, Junho Ahn 3, Hyung Jun Oh 4 and Heung Seok Jeon 1,*Round 1
Reviewer 1 Report
The authors have tested using machine-learning algorithms of animal movement combined with GPS collars to provide a more accurate GPS locations. They tested the method on goats in a 2 ha meadow (although looks like a playing field in the figures). The authors claim the method increases accuracy, although it is difficult to ascertain in the way the results are presented.
The results need some form of statistical test, to demonstrate how the accuracy is improved. It is not very evident from the tables or figures. The data is also not summarised in a way so the reader can easily see the improvements in accuracy.
Other comments include vague statements in the Abstract about the significantly further differences of the unfiltered locations compared to the actual results. These need to be more specific, and significance needs statistics.
Author Response
Cover letter for Reviewer 1
Reviewer 1: The results need some form of statistical test, to demonstrate how the accuracy is improved. It is not very evident from the tables or figures. The data is also not summarized in a way so the reader can easily see the improvements in accuracy.
Answer: We changed the tables and graphs to show average as you mentioned. In addition, we have added some more explanation about results. Table 5 shows the maximum and mean of the error passed distances of the mean among the tests. This result shows that when using the proposed method, the maximum error of running, walking, and resting behavior can be reduced from 153 to 20 m, 190 to 8 m, 162 to 0 m respectively; the average error is from 148.8 to 16.2 m, 182.7 to 5 m, 136.2 to 0 m as well.
Reviewer 1: Other comments include vague statements in the Abstract about the significantly further differences of the unfiltered locations compared to the actual results. These need to be more specific, and significance needs statistics.
Answer: We revised the Abstract to be more specific as you mentioned.
Author Response File: 
Author Response.docx
Reviewer 2 Report
This is an effort of providing solution to positioning related problems for animal tracing. I have the following concerns:
the authors should give some more information regarding the use of Eq. (1)
They should elaborate Eq. (2) in order to be clear to the readers. They should explain the SVM as srategy. What is the meaning of the terms in (2)? What is the F-score? Clearly this part should be explained in detail.
Do they use other type of observations (acceleration?). This is not clear.
Regarding the distance. Why the authors solve a spherical traingle? In cases of <10 km the distance could be easily estimated through the following concept :
x=Rcosφδλ
y=Rδφ where δλ δφ are the differences wrt a reference meridian and parallel.
Page 7, equations (6) and (7), The explanation of the terms unfiltered and filterd biases are poor. The authors should pay greater attention to their clarification.
Finally, what is the acceptable accuracy for this kind of applications? E.g. 10-20 meters?
Author Response
Cover letter for Reviewer 2
Reviewer 2: The authors should give some more information regarding the use of Eq. (1)
Answer: We added more information about the Eq. (1) as you mentioned.
Reviewer 2: They should elaborate Eq. (2) in order to be clear to the readers. They should explain the SVM as srategy. What is the meaning of the terms in (2)? What is the F-score? Clearly this part should be explained in detail.
Answer: We added more explanations about Eq. (2), SVM, and terms as you mentioned.
Reviewer 2: Do they use other type of observations (acceleration?). This is not clear.
Answer: We have considered the issues above and we specified that we mean accelerometer sensor raw data. And we made some changes in the current article.
Reviewer 2: Regarding the distance. Why the authors solve a spherical traingle? In cases of <10 km the distance could be easily estimated through the following concept:
x=Rcosφδλ
y=Rδφ where δλ δφ are the differences wrt a reference meridian and parallel.
Answer: You are absolutely right! The calculated distance is almost same when we use the Haversine or meridian and parallel algorithm in case of the distance is less ten km. When we chose an algorithm to measure the distance between two geolocation, we tried to choose the optimal algorithm which calculates with higher precision for all situation. The Haversine algorithm calculates the distance between two geolocation with higher precision than other algorithms. Another reason for choosing the Haversine algorithm: As we motioned in the current article, this article is continuing our previous study on virtual fencing for goats. In the virtual fencing application, it is very important to use Haversine algorithm for making virtual fences.
Reviewer 2: Page 7, equations (6) and (7), The explanation of the terms unfiltered and filtered biases are poor. The authors should pay greater attention to their clarification.
Answer: We added more explanation about equations (6) and (7) as you mentioned:
Reviewer 2: Finally, what is the acceptable accuracy for this kind of applications? E.g. 10-20 meters?
Answer: We think there is no acceptable threshold for this application. The smaller the error, the wider the area could be used for grazing. We added more information to explain the performance of our idea.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The authors have made suggestions improving the clarity of the manuscript. I have no further suggestions.
