Review Reports - New Era for Geo-Parsing to Obtain Actual Locations: A Novel Toponym Correction Method Based on Remote Sensing Images

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Review of the paper “New Era for Geo-Parsing to Obtain Actual Locations: A Novel Toponym Correction Method Based on Remote Sensing Images”

General thoughts

The paper describes a new method for correcting position offsets in geo-parsing applications using additional information gathered from remote sensing data. I see authors made a great job improving text according to reviewer’s remarks, but it needs to be improved in some please before acceptance.

Detailed comments

1. Authors need to very clearly state what their text brings new into remote sensing in general?

2. Part of the abbreviations are not explained

3. Text is not prepared according to the instruction for authors

4. Part of the figures (e.g. Figure 5) is unreadable, moreover every map needs to have a north arrow and definitely linear scale.

5. In my opinion text needs to be improved by adding a couple of references regarding determination of the points in a fields using GNSS, LiDAR and InSAR. I recommend:

5.1. Benoit L, Briole P, Martin O, et al (2015) Monitoring landslide displacements with the Geocube wireless network of low-cost GPS. Eng Geol 195:111–121.

5.2. Carlà T, Tofani V, Lombardi L, et al (2019) Combination of GNSS, satellite InSAR, and GBInSAR remote sensing monitoring to improve the understanding of a large landslide in high alpine environment. Geomorphology 335:62–75.

5.3. Chwedczuk K, Cienkosz D, Apollo M, et al (2022) Challenges related to the determination of altitudes of mountain peaks presented on cartographic sources. Geod Vestn 66:49–59.

6. Table 1 needs to be transformed into a map or text. Looking at coordinates you shoed a 2 points?

7. Table 2 location as above

8. Figure 7 is unreadable. Maybe you should add it enlarged at the end of the paper as attachment?

After authors’ improvements text will be accepted for a publication.

Author Response

Reviewer 1

General thoughts

A: Thank you very much for your approval of our revisions and your detailed suggestions to help us improve the quality of this manuscript further. We addressed all the commons point by point.

Detailed comments

Authors need to very clearly state what their text brings new into remote sensing in general?

A: Thanks for your question. We apologize for not making our core contribution clear to readers. Thus, we did the following supplements.

First, we clearly state the core contribution of this manuscript after our assumption descriptions in the introduction section (Line 84-87). It will let the readers understand what and how we want to fix.

Second, we emphasize the current offset values in the beginning example in section 2.2 (Line 141). This can make the readers understand the real challenges in geo-parsing results. Moreover, we compared the offsets (50 km) with the spatial resolution of satellite remote sensing images (1m-30m). We supplement these descriptions in section 3.1 (Line 153-156).

We hope these supplements will let the readers understand why and what we improved in the TC-RSI method. That is the key to what the text brings new.

Part of the abbreviations are not explained

A: Thanks for reminding us the abbreviations. We ignored the full descriptions of the UAV, CNN, DBN, and RNN, which are popular in the field of remote sensing.

To ensure that all readers can easily understand our manuscript, we have added the omitted abbreviations in Section 1 Line 73 and Section 2 Line 101-102.

Text is not prepared according to the instruction for authors

A: Thanks for reminding us the text formatting. After careful comparing the manuscript with the template of the instruction for authors, we find the text formatting of figures and tables are not suitable. Thus, we have adjusted these texts in Line 147-148, Line 165-168, Line 189, Line 247, Line 271-272, Line 273, Line 283, Line 299, Line 313, Line 336, Line 357, Line 365-366, Line 395-396 and Line 422-424.

Part of the figures (e.g. Figure 5) is unreadable, moreover every map needs to have a north arrow and definitely linear scale.

A: Thanks for your commons. The scale bar and north arrow are important to the map to provide the spatial information of the direction and the length. According to the complexity of our maps in Figures 1, 5, and 6, we use the latitude-longitude grid to present the spatial information of the direction and the length. The latitude-longitude grid can replace the scale bar and north arrow to express the direction and length. Thus, we do not add the north arrow and definitely linear scale. We hope the figures with latitude-longitude grid may be more simple and beautiful to read.

In my opinion text needs to be improved by adding a couple of references regarding determination of the points in a fields using GNSS, LiDAR and InSAR. I recommend:

5.1. Benoit L, Briole P, Martin O, et al (2015) Monitoring landslide displacements with the Geocube wireless network of low-cost GPS. Eng Geol 195:111–121.

5.3. Chwedczuk K, Cienkosz D, Apollo M, et al (2022) Challenges related to the determination of altitudes of mountain peaks presented on cartographic sources. Geod Vestn 66:49–59.

A: Thanks for your suggestions. These papers bring more application details to our manuscript. We have added these references in our reference list (reference 47-50). Moreover, these literatures inspire us to understand that different data sources (such as GNSS, satellite InSAR, GBInSAR, and etc.) have different spatial and temporal coverages and resolutions. Thus, we added more descriptions in discussion section 5.3 potential future works (Line 471-474).

Table 1 needs to be transformed into a map or text. Looking at coordinates you shoed a 2 points?

A: Thanks for your question. Yes, the coordinates of the examples in Hefei locate in two points. The question you mentioned we had discussed when we starting this manuscript. After representing this information in different forms, such as texts, figures, and tables, we finally found the best form to represent these examples is table. There are two key reasons.

First, table can easily list all the attributes information of the examples, such as names of forest ecological pattern and their corresponding toponym and locations. Text, however, cannot show them clearly.

Second, the points will be overlaid when they have the same locations. Thus, the figure with only two points must contain a complex legend or even a legend table. A brief preview sample can refer the Figure 5 (b) and Figure 6.

According to these reasons and failed attempts, we finally choose table to represent the examples of forest ecological patterns in Hefei. To make the readers understand the two points clearly, we added a note after the Table 1 (Line 274).

Table 2 location as above

A: Thanks for your question. After we double checked the Table 2, we do not find any location in Table 2. We though you may mean the Table 3, which contains lots of original locations and the corresponding corrected locations. Thus, we also added a note (Line 314) to tell the readers “the correction process for each point can be shown in Figure 6.”. it will help the readers understand the Figure 6 and Table 3 clearly.

Figure 7 is unreadable. Maybe you should add it enlarged at the end of the paper as attachment?

A: Thanks for your suggestions. In fact, Figure 7 presents lots of the correcting information. Due to the size of the image, it is not possible to give readers more detailed textures of the actual image before and after the TC-RSI correction. Consequently, we added an appendix A to show the enlarged corrected locations in details at the end of the manuscript (Line 518-520, Appendix A).

After authors’ improvements text will be accepted for a publication.

A: Thanks for your valuable suggestions. We hope our improvements are enough for the readers understanding our idea in this manuscript, which may change the geo-coding field in the future.

Author Response File: Author Response.docx

Reviewer 2 Report (Previous Reviewer 4)

commends well received

Author Response

Reviewer 2

commends well received

A: Thank you for your approval of our revisions and manuscript. We believe that the proposed TC-RSI method is expected to usher in a new era of geo-parsing with toponym corrections. We hope the relevant researchers can understand our idea as soon as possible, which could improve the geo-parsing and remote sensing application further.

Author Response File: Author Response.docx

Reviewer 3 Report (New Reviewer)

The manuscript discusses an approach to obtain actual location of geographical objects based on the use of satellite images. Despite its efforts spent, the study does not contain any scientific novelty, and the scientific breakthrough has to be explicitly spelled out in many different parts of the manuscript.

Major Correction:

Lines 147-148: the spatial resolution of remotely sensed images is usually very coarse, how do you use these images to identify the actual spatial position of concerned places? Highly unclear and may not be trustworthy.

Figure 2: need to provide some examples on these feature association etc., and some explanations have to be provided with respect to "recognition" and "resolution" - on the way of how toponym and features are associated etc.

Figure 4: The concept of toponym correction algorithm is stated somehow, but without any technical details and motivations behind. Some inputs and ideas have to be established before giving detailed information.

Section 4.3.2: For the section of statistical assessment - missing description of quantities / metrics for assessing accuracy of studies. Improper use of statistical quantity as well.

Lines 412-415: How? No mathematical quantities were used for describing the effects of terrains towards retrieval results.

Line 443: "improve accuracy of geo-parsed locations": The authors didn't mention the numerical perspectives of "accuracy"

Lines 445-446: The authors have not shown the effectiveness of TC-RSI method in practical applications

Lines 453-454: What does it mean by "correlate with the attributes and relevant toponyms"?

Lines 456-457: How should the toponym correction algorithm be adjusted? In particular, which parameters should be adjusted?

Selected Typos

Line 42: mentioned

Figure 9: Mountainous

What does it mean by "effectively decrease" in Line 462? By how much?

Finally, for the algorithm / methodology, the authors should explain in much finer details. They should provide much more technical details within different parts of the manuscript, for example, the calculations, the statistical metrics used for inter-comparison / validation etc.

Author Response

Response to reviewers:

A: We thank the editors and reviewers for reviewing our manuscript and bringing forth valuable comments. Below we list the reviewer’s comments in bold and discuss how we incorporated them into our paper point by point.

Reviewer 3

A: Thank you very much for your approval of our revisions and your valuable suggestions to help us improve the quality of this manuscript further. We will do our best to address all the commons point by point to let the readers understand our idea in TC-RSI method, which can really improve the whole geo-parsing field with Remote Sensing techniques.

Major Correction:

Lines 147-148: the spatial resolution of remotely sensed images is usually very coarse, how do you use these images to identify the actual spatial position of concerned places? Highly unclear and may not be trustworthy.

A: Thanks for your question. Although the spatial resolution of Remote Sensing images is usually very coarse nearly 1m-30m, it is much smaller than the offset distance nearly 50km. These images, however, can provide more spatial information to the current geo-parsing results. To address this spatial resolution issue, we had made the following changes.

First, we emphasize the current offset values in the beginning example in section 2.2 (Line 141). This can make the readers understand the real challenges in geo-parsing results.

Second, we compared the offsets (50 km) with the spatial resolution of satellite remote sensing images (1m-30m). The overall improvement in accuracy can efficiently provide more spatial information. Thus, we supplement these descriptions in section 3.1 (Line 153-156). This will let the readers understand why and what we can improve in this TC-RSI method.

Finally, we also added an Appendix A section (Line 518-520) to show the enlarged corrected locations in Figure 7. With more detailed images, the readers can clearly know the results and the validation process.

Figure 2: need to provide some examples on these feature association etc., and some explanations have to be provided with respect to "recognition" and "resolution" - on the way of how toponym and features are associated etc.

A: Thanks for your question. After reviewing our Figure 2, we also find some examples are required to let the readers understand our idea more clearly. Consequently, we re-draw the figure with an example and added some explanations after the figure as you mentioned.

First, a full example of “forest-herb” mode in Hefei is given in the figure 2 with each process results. The details can be shown in the new version Figure 2 (Line 164).

Second, more explanations are added after the figure 2 (Line 165-168). An example, “forest-herb” mode in Hefei, has been attached to show the results of each process. The detailed data and validation of this example can be found in section 4 with the id 5 forest ecological pattern. The readers can also validate this example with the experiment in section 4.

We hope these supplements can be enough to improve the Figure 2 and give a clear figure to show the basic idea of TC-RSI method.

Figure 4: The concept of toponym correction algorithm is stated somehow, but without any technical details and motivations behind. Some inputs and ideas have to be established before giving detailed information.

A: Thanks for your question. The figure 4 shows the core flow of toponym correction algorithm based on remote sensing images. We need to make sure everyone can understand it clearly. Thus, we adjusted two aspects.

First, we extracted the AI value part from the Figure 4 and added the descriptions of how to calculate it in formula (4). This is because calculating AI value is one of the key processes during the TC-RSI method. This will let the readers understand more clear.

Second, we double checked the previous paragraph that every inputs and outputs are described in section 3.2 (Line 225-245).

We hope the adjusted two aspects are enough for the readers.

Section 4.3.2: For the section of statistical assessment - missing description of quantities / metrics for assessing accuracy of studies. Improper use of statistical quantity as well.

A: Thanks for your question. The statistical assessment section is not perfect. Thus, we adjusted the section 4.3.2 as your commons.

First, we added the metric to show how we calculate the geo-parsed locations and get the average offset value. More details can be shown in Formula (5) and the relevant text in Line 350. Second, we also supplement more descriptions about the provincial statistical assessment in the section 4.3.3. Considering the provincial assessment belongs to the robustness assessment, we added some intro text in Line 354-355.

We hope these supplements can improve the statistical assessment section.

Lines 412-415: How? No mathematical quantities were used for describing the effects of terrains towards retrieval results.

A: Thanks for your question. Though there is no mathematical metric to evaluate the effects, we also have some statistical quantities. For example, 70% of the records (ID 1-65) with small offset distances are plain areas, which have lower reliefs in their terrains (Line 412-414). Moreover, the offset distance curve is also shown in Figure 9 (d), which illustrates all the offset distances. The figure can clearly show the changes and the types (the plain area or the mountainous area).

After the discussion by all the authors, we trust the graphic expression is better than the text with numbers. Furthermore, the Figure 9 (Line 421-423) we changed also have mathematical quantities to enhance the expression of terrain impact.

Line 443: "improve accuracy of geo-parsed locations": The authors didn't mention the numerical perspectives of "accuracy"

A: Thanks for your question. We are sorry for this confusions to the readers. The readers may not understand the underlying meanings, such as “improve accuracy of geo-parsed locations”. This is because Section 5.3 belongs to the discussion part, which closely related to the previous experimental section. No matter the terrain impact, method limitation, or the potential future works, we are high correlated with our experiments.

To the issue of “improve accuracy of geo-parsed locations”, we also have numerical details in pervious experiment sections that have been verified carefully. As the readers require the basic numerical details, we added more verified numeric details about the improved accuracy in Section 5.3 (Line 456-457).

Lines 445-446: The authors have not shown the effectiveness of TC-RSI method in practical applications

A: Thanks for your question. That’s true, Lines 445-446 do not show the application details. This is because section 5.3 focuses on the potential future works. All these discussions are based on the previous series of experiments in section 4. Therefore, the effectiveness of TC-RSI method in practical applications can be shown in the section 4 experiments and results. There are lots of applications in different conditions.

Lines 453-454: What does it mean by "correlate with the attributes and relevant toponyms"?

A: Thank for your question. We do not state very clearly about “correlate with the attributes and relevant toponyms”. Thus, we re-written the whole paragraph and added an example to supplement this sentence (Line 469-472).

Satellite remote sensing images can provide the extra spatial information for the geo-parsing. This inspires us other data sources may also provide extra spatial information for the geo-parsing. For example, the sentence “Nanjing city constructs lots of agriculture parks with the agriculture-park ecological pattern, such as Guli Modern agriculture park, …” contains the attribute information like “Guli Modern agriculture park”, which may record in the domain gazetteers that store the actual locations. Moreover, other sources may also have the information. Thus, domain gazetteers, low-altitude UAV images, Wikimedia, streetscapes, GNSS, InSAR, and hybrid sources may also be the data sources to correct geo-parsed locations. Only they need to do is correlating the attributes with the relevant toponyms. For example, correlating “Nanjing” with “Guli Modern agriculture park”.

With a coherent example, we hope the readers can understand the TC-RSI method and the underlying broad application prospects.

Lines 456-457: How should the toponym correction algorithm be adjusted? In particular, which parameters should be adjusted?

A: Thanks for your question. This question is one of the key question in the future. We are also interested in how should the toponym correction algorithm be adjusted according to complex and diverse data sources. This is why we put it into the potential future works that means we have started the following researches and do not solved it.

According to the readers want to know more, we added the potential aspects in the sentences (Line 474-475). We are glad to hear that global researches also focus on this question. And we also expected to see some researches can fix it well with brilliant idea and method.

Selected Typos

Line 42: mentioned

A: Thanks for reminding us. We have corrected the word “mentions” in Line 42.

Figure 9: Mountainous

A: Thanks for reminding us. We have corrected the word (Line 412) and redraw the Figure 9 (Line 419-421).

What does it mean by "effectively decrease" in Line 462? By how much?

A: Thanks for your reminding. Generally, the first paragraph of the conclusion section summarizes the core contributions and improvements with refined vocabularies. Thus, “effectively decrease” shows in this paragraph to give a general description. The detailed supplements illustrate in the next paragraph “Using the TC-RSI method, the offset distances of the current geo-parsed locations can decrease from nearly 50 km to nearly 1 km level. The dramatic improvement shows that the remote sensing images have powerful accurate spatial information, which can be used to correct geo-parsed locations. This remarkable improvement will allow text mining to find more accurate geographical discoveries with lower offset distances.” (Line 482-487). We don't let that uncertainty descriptions occur in our manuscript text.

A: Thanks for your suggestions. We had made lots of efforts to improve the details for reading it easily. The adjustments are shown below.

First, we added more descriptions in section 3.2, such as General geo-parsing procedure, Attribute associating procedure, RS feature associating procedure, RS area associating procedure, and Location correction procedure (Line 190-245).

Second, an example is attached in the methodology section. Each procedure will be described with an example, for example, the formula (1), (2), and (3).

Moreover, other statistical metric is supplemented in our manuscript, for example, the Formula (5) in section 4.3.2.

Finally, we re-draw the figures (Figure 2 & 9) and added some necessary descriptions in our manuscript (Line 458-477). we hope all these changes can help the readers understand our idea clearly. And we do trust this idea can soon lead lots of breakthroughs in geo-parsing and remote sensing fields.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report (Previous Reviewer 1)

Authors improved text according to reviewer's remarks and is accepted for a publication in RS.

Author Response

Authors improved text according to reviewer's remarks and is accepted for a publication in RS.

Reviewer 3 Report (New Reviewer)

Thanks the authors for the detailed revision. The manuscript looks much better in overall now. There are still some minor points to go:

Figure 4: The overall framework and its main ideas can be described in words in a new paragraph, and this table can be moved to Appendix.

Table 2: For the Forest main species types, it will be great if the authors can include some figures of these, then it looks clearer to readers.

Section 4.2.2: Statistics should have some more detailed statistical description, otherwise the main contents do not match with the sub-heading.

The authors have attempted to use different statistical approaches for quantifying the accuracy of the study, but which one is the most reliable in general? Why not other statistical quantities in remote sensing? A brief discussion should be included before any validation is conducted.

Further, the connections of the use of satellite remote sensing with environmental remote sensing should also be mentioned in the manuscript.

Author Response

Reviewer 3

Thanks the authors for the detailed revision. The manuscript looks much better in overall now. There are still some minor points to go:

A: Thank you very much for your approval of our revisions and your detailed suggestions to help us improve the quality of this manuscript further. We addressed all the commons point by point.

Figure 4: The overall framework and its main ideas can be described in words in a new paragraph, and this table can be moved to Appendix.

A: Thanks for your question. Yes. The overall framework of the toponym correction algorithm can be described in a new paragraph, which can be understood more easily. As the Toponym Correction Algorithm is the key to the TC-RSI method, we prefer reserve this Figure in the methodology section rather than in the appendix.

According to the above adjustment idea, we supplemented some descriptions in the following paragraph in Line 245-252 to let the readers understand it more clearly.

Table 2: For the Forest main species types, it will be great if the authors can include some figures of these, then it looks clearer to readers.

A: Thanks for your commons. You are right. The forest main species with Latin names are difficult to understood by readers. Based on your advice, all the authors agree to add an appendix to show the images of forest main species. Consequently, the additional Appendix A (Line 532-534) has been supplemented.

Section 4.2.2: Statistics should have some more detailed statistical description, otherwise the main contents do not match with the sub-heading.

A: Thanks for your question. Yes, “Statistics” should contain more statistical descriptions, such as mean, variance, distribution, etc. In section 4.2.4, the manuscript wants to show correction ranges of the 11 samples to let readers understand the performance of the TC-RSI method. We found the sub-heading of this section is not appropriate. Thus, we changed the sub-heading of section 4.2.2 to “Correction ranges” (Line 308). We hope this sub-heading is more clear to the readers.

A: Thanks for your suggestion. You are right. Three types of correction evaluation approaches were used in section 4.3 for quantifying the accuracy of the study. We need to tell the readers why we choose these approaches than others. Thus, we added a description paragraph in section 4.3 (Line 325-329) to tell the readers the reason of choosing visual validation, statistical assessment, and robustness assessment.

Further, the connections of the use of satellite remote sensing with environmental remote sensing should also be mentioned in the manuscript.

A: Thanks for your suggestion. We should add more connections of the use of satellite remote sensing with environmental remote sensing in the manuscript. In fact, remote sensing images can directly recognize correct geographic locations with apparent spatial features. Especially, satellite remote sensing images can acquire spatial information about large-scale urban areas and their surrounding environments. Thus, we added more descriptions in section 1 (Line 78-79) to describe the connections.

In summary, we hope all the adjustments are enough for the readers understanding our idea clearly.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

In general I do like this paper. It does deliver a bit less than it promises but I do believe this is a nice experimental method. How doesit correspond to more classical methods like IR images interpretation is unknown. But I do recognize that not everything can be explained in one paper.

I do have a few questions (I would like them to be referred to also in the main text)

What is the time for correction effecnty and how ‘manula’ is this taks
How were the airborne LIDAR and UAV images used
How do you mitigate for significant scale differences between satellite and plane/UAV images.
Would it make sense to ad as first initial layer a topographic map of the region (it could be an old or current map)?

L14 „research hotspot” sounds a bit unprofessional – perhaps ‘not fully analyzed field’

L18 “occurrence of locations”

L24 specifie what Remote Sensing Images

L24 minor scales?

L110 “remote sensing images, Lidar, and UAV data” UAV produces either LIDAR or images unless you mean something deferent? Also it is an acronym needs explanation

Reviewer 2 Report

The paper describes a new method for correcting position offsets in geo-parsing applications using additional information gathered from remote sensing data.

In my opinion the manuscript has several flaws and issues and should not be considered for publication:

Presentation: the structure of the manuscript should be thoroughly revised, being many sections misplaced or overlapping. Both the methodology and the results are not adequately explained and supported. Extensive editing of English language and style is required: in its current form the manuscript has several sentences that might result hard to understand due to this issue.

Introduction: Introduction should be carefully reworked to critically present the current state-of-the-art, which is just addressed from line 52 to line 77. The cited methodologies are just briefly listed, sometimes without a direct cross-reference, and should be better described assuming not all the readers are knowledgeable in this field. The amount of supporting references is quite good, but without a critical review and discussion by the authors, the reader, unless expert in this field, should spend a lot of time trying to build a sufficient knowledge base to fully understand the differences of the several approaches reported.

From line 78 to the end of Section 1, the authors discuss the methodological issues underlying the paper only with reference to their specific case study. A general overview of state-of-the-art methodologies and open questions should be given in the introduction. Otherwise, the reader is not able to understand whether the relevance of the problem is general, relative to certain applications or specific to the case study.

Methodology: the methodology section provides very few details about the implemented method, usually without an in-depth presentation of the different processing stages and without proper discussion on the choices made by the authors.

What inputs are required? Which remote sensing datasets are used? Are they georeferenced? How are clusters obtained from remote sensing data? How are the classes gathered from remote sensing data associated with the parsed attributes? … and so on.

The flowcharts (figures 2 and 3) are, honestly, a little obscure and should be properly commented on, describing how the algorithm works step by step. The code provided in figure 4 is not self-explanatory as well and should be described, including comments on the meaning of the variables and parameters values. For instance, the significance of the Attractive Index (AI) should be clarified in this section, especially since a location error determined by the AI is then highlighted in the results.

Furthermore, it is hard to understand whether example sentences are provided to support the methodology explanation (e.g. lines 134-136) or not, since no inverted commas are used. For a better comprehension of the methodology, it might be helpful to provide an example sentence (in inverted commas) on which to show how the algorithm works step-by-step.

Experiment: the experiment section is confusing and overlaps with the previous sections of Introduction and Methodology. The description of the case study area is provided here, but the case study was already mentioned from the Introduction. The input dataset and the remote sensing data used for location correction should be better described, assuming not all the readers are familiar with Chinese datasets. Additional information on the application of the methodology to the case study is needed.

Results and discussion: the results are not supported by the experiments and very little information is given on how they are evaluated. Is it just a visual and manual evaluation done on only 11 items? Are they statistically relevant?

Section 3.3.3 (Robustness assessment) refers to an additional dataset that has never been introduced or described before; the Attractive Index (line 288) is mentioned as the cause of location error, but it has not been explained before; the impact of terrain is introduced: on which dataset was it assessed?

In the end, the paper in its present form is not capable of ensuring that the experiments conducted are scientifically sound and correctly performed from a methodological point of view. Due to these serious flaws I do not think that providing some more detailed (i.e. line by line) remarks would be of any help.

Reviewer 3 Report

For Figure 1,5 and 6, you need to add a scale bar and north arrow to the map.
The method of using satellites can be helpful in correcting the position on small-scale maps such as Fefei and Feixi, but It is questionable how accurate the method presented in this study can be even on large-scale maps such as small town level .That is, according to the scale it is thought that the method presented in this paper may be helpful or may make an error.
In geo parsing, there are not many cases where the user expresses the location by using the land use state of the satellite map as an attribute value. In this regard, it is necessary to consider how much the method presented in this study will be helpful for geoparsing.
These shortcomings and limitations of the study should be specified in the conclusion section.

Reviewer 4 Report

The initial idea is great and the expected impact of the research is high. However, the language is weird in many places which makes the text incomprehensible and thus unreachable to readers. Moreover, the methodology is not explained in a way that can be fully understood, validated and reproduced. Using only 1 case study and generalize the resuts does not improve the situation.

I think it is unfair for such a good idea not to be able to reach a good number of readers, due to not proper presentation.

I suggest to greatly improve the text and resubmit as a brand new draft

Reviewer 5 Report

This paper proposes a method to improve geocoding from toponyms in text. Text is parsed and improvement is based on addressing metainformation in the sentence to get qualitative information that can be joined to remote sensing images via labelled pixels and successive clustering and spatial analysis to improve the initial geolocation.

There is little novelty related to remote sensing, I therefore suggest to submit to a different journal, and also to drastically improve the clarity of writing.

Below some suggestions.

Only 11 points are tested, not a very significant number.

Figure 5 and figure 8 need that the legend be improved, the density value needs a correct number of decimal digits.

Table 1 - location should have degrees symbol

Figure 9 - 70% plain (flat?) area and 70% mountain? maybe you mean 30%