Round 1

Reviewer 1 Report

The topic of this paper is interesting. The paper achieves the identification and change detection of building area and height using multi-source high resolution imagery for long term monitoring of urban regrowth. The paper's ideas are well presented and organized. The following suggestions are available for reference.

1. Abstract section. The abstract needs to be further divided into clear methodology and results of the study. The first sentence of the study is far less valuable than the introduction section. In addition, it is suggested to add a sentence on the research implications or merits.

 

2. The Introduction section, the review of studies is a bit weak and needs to be further supplemented with similar studies. A review of studies that share this same idea is suggested.

 

3. The methodology section suggests adding a general technical flowchart and some corresponding descriptions that can make the reader more clear.

 

4. Results section. In paragraph 4 of the Introduction, it is mentioned that the study area is exposed to high seismic risk. Is it possible to add the corresponding statistical results to be able to highlight the value of this paper.

Author Response

Dear reviewer,

Thank you for your comments on our manuscript. We have addressed your comments below. Please find our comments in bold, and new text and references in bold italics.

Sincerely,

Scott Watson (on behalf of all co-authors)

1. Abstract section. The abstract needs to be further divided into clear methodology and results of the study. The first sentence of the study is far less valuable than the introduction section. In addition, it is suggested to add a sentence on the research implications or merits.

We have added the following text to introduce the methods before presenting the study results in the abstract:

First, the urban sprawl of Bishkek (1979–2021) was quantified using built-up area land cover classifications. Second, a change detection methodology applied to a declassified KeyHole Hexagon (KH-9) and Sentinel-2 satellite image to detect areas of redevelopment within Bishkek. Finally, vertical development was quantified using multi-temporal high-resolution stereo and tri-stereo satellite imagery, which were used in a deep learning workflow to extract buildings footprints and assign building heights.

We have added the following sentence summarising the research merit of deep learning methodologies:

Deep learning methodologies applied to high-resolution imagery are a valuable monitoring tool for building stock, especially where country-level or open-source datasets are lacking or incomplete.

2. The Introduction section, the review of studies is a bit weak and needs to be further supplemented with similar studies. A review of studies that share this same idea is suggested.

 We have added new text and references to the introduction to strengthen the review in several places. Please see below.

We have added references to other studies using lidar data and high-resolution DEMs for building height extraction:

…however, retrieving the building-level 3D structure of a city generally requires expensive aerial imagery or light detection and ranging (LiDAR) surveys [20,21] or high-resolution digital elevation models (DEMs) [22].

Priestnall, G.; Jaafar, J.; Duncan, A. Extracting urban features from LiDAR digital surface models. Computers, Environment and Urban Systems 2000, 24, 65-78, doi:https://doi.org/10.1016/S0198-9715(99)00047-2.

Weidner, U.; Förstner, W. Towards automatic building extraction from high-resolution digital elevation models. ISPRS Journal of Photogrammetry and Remote Sensing 1995, 50, 38-49, doi:https://doi.org/10.1016/0924-2716(95)98236-S.

We introduce an additional method of building height estimation using shadows:

Other approaches use shadow-based height estimation; however, this does not work well in densely urbanized areas with overlapping shadows [22,23].

Xie, Y.; Feng, D.; Xiong, S.; Zhu, J.; Liu, Y. Multi-Scene Building Height Estimation Method Based on Shadow in High Resolution Imagery. Remote Sensing 2021, 13, 2862.

Cheng, F.; Thiel, K.H. Delimiting the building heights in a city from the shadow in a panchromatic SPOT-image—Part 1. Test of forty-two buildings. International Journal of Remote Sensing 1995, 16, 409-415, doi:10.1080/01431169508954409.

We have added an additional reference at Line 55 highlighting the requirements for building footprint datasets with corresponding height information for use in seismic risk assessments:

Mansouri, B.; Ghafory-Ashtiany, M.; Amini-Hosseini, K.; Nourjou, R.; Mousavi, M. Building Seismic Loss Model for Tehran. Earthquake Spectra 2010, 26, 153-168, doi:10.1193/1.3280377.

3. The methodology section suggests adding a general technical flowchart and some correspondigng descriptions that can make the reader more clear.

 We have added a flowchart (Figure 2) providing an overview of our methodologies for (a) the built-up area change analysis, and (b) building inventory analysis.

4. Results section. In paragraph 4 of the Introduction, it is mentioned that the study area is exposed to high seismic risk. Is it possible to add the corresponding statistical results to be able to highlight the value of this paper.

We have added an additional reference supporting high seismic risk in the city of Bishkek at Line 76.  

Global Earthquake Model. Country profile for Kyrgyzstan Version 1.2. [online]. Accessed 28th October 2022. Available from: https://downloads.openquake.org/countryprofiles/KGZ.pdf. 2022.

We have added the text below to give an indication of the magnitude of damages expected, noting that these values are variable depending on the future scenario.

L76. A future earthquake in Bishkek could lead to thousands of fatalities and require hundreds of millions of dollars rebuilding costs [33,34].

Reviewer 2 Report

Some figures and tables are marked with an extra S in the text.

Figure 1 has unclear legends. It should be improved

Several references to tools, repositories... are needed, for example:

- l84: EROS center o whatever

-l94 Orfeo-toolbox.org?

- l 132 whitebox tools

- l 204-206, 212- 213-----

It woild be needed a better description (with more details) of the used network and its operation. Then, it is possible to analize the reached accuracy

Author Response

Dear reviewer,

Thank you for your comments on our manuscript. We have addressed your comments below. Please find our comments in bold, and new text and references in bold italics.

Sincerely,

Scott Watson (on behalf of all co-authors)

Some figures and tables are marked with an extra S in the text.

These figures and tables refer to the supplementary figures and tables. We believe these correspond to the journal requirements to have not changed the labels.

Figure 1 has unclear legends. It should be improved

We have increased the label text size on panel (a). We have changed the colour of the leader lines on panel (b) to show corresponding polygons and data sources. We have also modified the text to clarify the elements on panel (b). The caption for panel (b) now reads:

(b) Analysis extents capturing Bishkek’s designated districts (grey shading) based on the spatial availability of each data source shown by colored polygons (colored lines for ICESat-2). The hashed blue polygon shows the intersecting WorldView-2 and Pleiades extents

Several references to tools, repositories... are needed, for example:

- l84: EROS center o whatever

We have added the following text and reference:

We downloaded a stereo pair of KH-9 satellite images (Table S1) from USGS EarthExplorer [39].

USGS. Earth Explorer. [online]. Accessed 20th May 2021. Available from: https://earthexplorer.usgs.gov/. 2022.

-l94 Orfeo-toolbox.org?

We have added the Orfeo Toolbox citation at the point of first mention:

Orfeo Toolbox. User Guide. [online]. Accessed 07 January 2022. Available from: https://www.orfeo-toolbox.org/CookBook/C++/UserGuide.html. 2021.

- l 132 whitebox tools

We have added the citation:

Lindsay, J. The Whitebox Geospatial Analysis Tools project and open-access GIS; 2014.

- l 204-206, 212- 213-----

It woild be needed a better description (with more details) of the used network and its operation. Then, it is possible to analize the reached accuracy

We are not clear on the specific area requiring clarification. However, we have clarified the description of the Mask R-CNN.

A Region-based Convolutional Neural Network Mask R-CNN deep learning model was trained to detect buildings in the high-resolution satellite imagery. Mask R-CNN is an in-stance segmentation model that provides a segmentation mask and output polygon for each instance of a building detection. Mask R-CNN was chosen owing to its efficient building extraction capabilities applied to high-resolution satellite imagery [30,62,63].

Reviewer 3 Report

Dear authors, thank you for an interesting contribution.


I have some remarks:

By 2050, ~68%, better dont use the ~

the same row 85 at ~6–9 m
row 236    ~2016

 

row 84  image: 21 June 1979 KH-9 ...describe better these data  (resolution, etc.), this is insufficient

row 88 you write: To classify built-up areas in a 21 June 1979 KH-9 image, we first orthorectified and 88 georeferenced the image using Agisoft Metashape v1.7.2 with predefined camera param-89 eters obtained by Dehecq [33] and 11 ground control points obtained from Google Earth 90 on static features.   however, you need a DEM for orthorectification. Which one?


row 164 , row 168 etc...We used  please, don't use in scientific text "we", better passive

(for example 2.3.1 Building polygons....too much "we"


You use Agisoft Metashape for creating of DEM from VHR satellite data? Why this software? It is typically use for photographic (central projection) images; satellite images are made using moving satellite optical scanner (HRV etc.), it is not a central projection, only in one row

2.2 If you use terminus DEM, define it; it should be a interpolated grid. From aerial and VHR satellite images you get DSM (digital surface model)

 

you use a mixture of DEM, DSM, DTM; it must be explained or joined; however these terms have different definitions

 

row 259 Building height validation
We used 11 field-measured building heights and ICESat-2 altimetry data as an inde-260 pendent check of our satellite DEM-derived building heights. They were obtained using 261 a laser range finder  - there should be a link to the results, e.g. a table



Materials and methods

This part is in comparison with results very short and not easy to understand, how the data were processed. Please improve the text for better understanding and add a flowchart to show the processing procedure

How do you combine data with very different parameters? KH, Landsat, Sentinel, WV, Pleiades, TanDEM X, PlanetScope - these are very different systems with different resolutions. It's not entirely clear from the text what you used and when. Maybe some processing scheme would improve this.

Add more citations on historical KH images; it is a valuable source of information in countries, for which the data are not available (for example):

A pipeline for automated processing of Corona KH-4 (1962-1972) stereo imagery, 10.1109/TGRS.2022.3200151

Study of Erbil Al-Qala citadel time changes by comparison of historical and contemporary image data, 10.1080/22797254.2018.1531683

Glacier Mass Loss during the 1960s and 1970s in the
Ak-Shirak Range (Kyrgyzstan) from Multiple Stereoscopic Corona and Hexagon Imagery, https://doi.org/10.3390/rs9030275

Positional Accuracy Evaluation of Declassified Hexagon KH-9 Mapping Camera Imagery,10.14358/PERS.76.5.603

 

 

Author Response

Dear reviewer,

Thank you for your comments on our manuscript. We have addressed your comments below. Please find our comments in bold, and new text and references in bold italics.

Sincerely,

Scott Watson (on behalf of all co-authors)

I have some remarks:

By 2050, ~68%, better dont use the ~

We have replaced ‘~’ with ‘an estimated’.

the same row 85 at ~6–9 m
row 236    ~2016

We have removed these occurrences of ‘~’. 

row 84  image: 21 June 1979 KH-9 ...describe better these data  (resolution, etc.), this is insufficient

Please see our response to your comment below.

row 88 you write: To classify built-up areas in a 21 June 1979 KH-9 image, we first orthorectified and 88 georeferenced the image using Agisoft Metashape v1.7.2 with predefined camera param-89 eters obtained by Dehecq [33] and 11 ground control points obtained from Google Earth 90 on static features.   however, you need a DEM for orthorectification. Which one?

We have clarified the use of the KH-9 imagery (Section 2.1.1) in response to this and your previous comment. We now refer to a stereo pair of KH-9 images, which were detailed in Table S1, but were not mentioned in the main manuscript. The stereo pair was used to generate a DEM, which was then used to orthorectify the imagery.

KH-9 satellite imagery was collected by United States reconnaissance programs (1973–1980) at 6–9 m ground resolution and has been used in a range of applications. A pair of KH-9 satellite images (21 June 1979) (Table S1) were downloaded from USGS EarthExplorer [39]. To our knowledge, no studies have applied image classification techniques to semi-automatically extract built-up areas from KH-9 imagery.

To classify built-up areas in a 21 June 1979 KH-9 image (Figure 2a), the imagery was first used to generate a DEM in Agisoft Metashape v1.7.2 with predefined camera parameters obtained by Dehecq [40] and 11 ground control points obtained from Google Earth on static features. The DEM was used to orthorectified and georeferenced the imagery, which was output at a resolution of 4 m.


row 164 , row 168 etc...We used  please, don't use in scientific text "we", better passive

(for example 2.3.1 Building polygons....too much "we"

We have removed some occurrences of ‘we’ but prefer not to change to passive throughout.

You use Agisoft Metashape for creating of DEM from VHR satellite data? Why this software? It is typically use for photographic (central projection) images; satellite images are made using moving satellite optical scanner (HRV etc.), it is not a central projection, only in one row

The latest versions of Metashape are now designed to produce DEMs using satellite imagery provided with rational polynomial coefficients (i.e. not using the WorldView-2 or Pleiades sensor models). This is designed functionality of the software and robust results were observed in our paper (e.g. accuracy statistics in Figure 7). We have also referenced another example of Metashape’s use for satellite data DEM generation:

Lastilla, L.; Belloni, V.; Ravanelli, R.; Crespi, M. DSM Generation from Single and Cross-Sensor Multi-View Satellite Images Using the New Agisoft Metashape: The Case Studies of Trento and Matera (Italy). Remote Sensing 2021, 13.

2.2 If you use terminus DEM, define it; it should be a interpolated grid. From aerial and VHR satellite images you get DSM (digital surface model)

you use a mixture of DEM, DSM, DTM; it must be explained or joined; however these terms have different definitions

We have revised this terminology to be specific and now refer to digital surface models (DSMs) or digital terrain models (DTMs) throughout.

row 259 Building height validation
We used 11 field-measured building heights and ICESat-2 altimetry data as an inde-260 pendent check of our satellite DEM-derived building heights. They were obtained using 261 a laser range finder  - there should be a link to the results, e.g. a table

We now provide this information in Supplementary Table 2.

Table S2: Comparison between field-measured building heights and those derived from the DSM-DTM difference for Pleiades and WorldView-2 data.


Materials and methods

This part is in comparison with results very short and not easy to understand, how the data were processed. Please improve the text for better understanding and add a flowchart to show the processing procedure

We have added a flowchart (Figure 2) providing an overview of our methodology. We have also added more information on data processing, for example in section ‘2.1.1 KH-9 classification’ to detail the imagery processing, and ‘section 2.3.1 Building polygons’, where the deep learning approach is introduced and justified.

How do you combine data with very different parameters? KH, Landsat, Sentinel, WV, Pleiades, TanDEM X, PlanetScope - these are very different systems with different resolutions. It's not entirely clear from the text what you used and when. Maybe some processing scheme would improve this.

We expect that the new Figure 2, which summarises the processing methodology should address this comment by detailing the outputs from each analysis ((a) built-up area change, and (b) Building inventory) and the source data used. The different resolutions determine the applicability of the datasets in our study. Here the data are compared, rather than combined. For example, we show through error statistics (Figure 7) and visual representation (Figure 12) that the resolution and accuracy of the PlanetScope DSM is insufficient to resolve individual buildings.

Add more citations on historical KH images; it is a valuable source of information in countries, for which the data are not available (for example):

A pipeline for automated processing of Corona KH-4 (1962-1972) stereo imagery, 10.1109/TGRS.2022.3200151

Study of Erbil Al-Qala citadel time changes by comparison of historical and contemporary image data, 10.1080/22797254.2018.1531683

Glacier Mass Loss during the 1960s and 1970s in the
Ak-Shirak Range (Kyrgyzstan) from Multiple Stereoscopic Corona and Hexagon Imagery, https://doi.org/10.3390/rs9030275

Positional Accuracy Evaluation of Declassified Hexagon KH-9 Mapping Camera Imagery,10.14358/PERS.76.5.603

We have added the follow references at the start of section ‘2.1.1 KH-9 classification’, which specifically use KH-9 imagery:

Goerlich, F.; Bolch, T.; Mukherjee, K.; Pieczonka, T. Glacier Mass Loss during the 1960s and 1970s in the Ak-Shirak Range (Kyrgyzstan) from Multiple Stereoscopic Corona and Hexagon Imagery. Remote Sensing 2017, 9, 275.

Surazakov, A.; Aizen, V. Positional accuracy evaluation of declassified Hexagon KH-9 mapping camera imagery. Photogrammetric Engineering & Remote Sensing 2010, 76, 603-608.

Dehecq, A.; Gardner, A.S.; Alexandrov, O.; McMichael, S.; Hugonnet, R.; Shean, D.; Marty, M. Automated Processing of Declassified KH-9 Hexagon Satellite Images for Global Elevation Change Analysis Since the 1970s. Frontiers in Earth Science 2020, 8, doi:10.3389/feart.2020.566802.

Round 2

Reviewer 2 Report

In this new version the references are more correct and extensive. The design of the network used in the method is still little explained, in my opinion, with only references included about-

Reviewer 3 Report

Dear authors, ok, thank you. I have no more questions.