Predicting Neighborhood-Level Residential Carbon Emissions from Street View Images Using Computer Vision and Machine Learning

Round 1

Reviewer 1 Report

The authors investigated Neighborhood-Level Residential Carbon Emission prediction from Street-view Images Using Computer Vision and Machine Learning. This work is very important and also relevant for China. Though the work is largely well-written, the article still needs significant improvement in the introduction, methodology, results, and discussion sections. Considering my observations as follows, I suggest minor revisions before considering it for publication.

Comments:

·       lease omit or delete the sub heading (1.1, 1.2 1nd 1.3) of ‘section 1’ in the introduction.  I would suggest to write the introduction into three paragraphs, highlighting the basic content of the research field in the first paragraph. Then, review the research progress of the literature in the second paragraph, and in the third paragraph, analyse the limitations of past research and clarify the innovation of your own research.

·       The flow chart in the page 23 is part of Figure 2’? If not please provide the caption.

·       In case of RF how was the model built? How was it calibrated? The most important parameters and the choice of values for the model were not explained.

·       Section 4, please divide this section “Results and Discussion” into two separate sections: “Results” and “Discussion”, which will provide readers a clearer understanding.

·       Please check the English language, Grammar. 

Minor editing of English language required

Author Response

Thanks for your careful and constructive comments, we've learned a lot through this review and revision process. Please see the attached document for detailed responses.

Author Response File: Author Response.docx

Reviewer 2 Report

Dear Authors,

Please see my review of the manuscript titled ‘Predicting Neighborhood-Level Residential Carbon Emission from Street-view Images Using Computer Vision and Machine Learning’.

Title and the manuscript, abstract, keywords:

The title expresses well what the manuscript is about.

Abstract:

English must be improved, but it is the general recommendation. The abstract has a good structure, however, there is a serious lack of numbers by which the “effectiveness” of the framework they made to predict residential CE in urban areas (LINE 16), or the correlation between the mentioned parameters (LINE 23). How would you justify the statement of LINE 28-29? This is, in its present form a too general statement not fitting into academic articles. Please, re-write this part based on the comments, but keep the structure.

Instead of 1KM grid expression use e.g. 1 km-grid or 1km grid. 1KM is not correct. Correct it everywhere, please.

Keywords:

These are expressing adequately the most important words occurring in the manuscript. Please, check again the grammar and apply the correct version in the entire manuscript (e.g. Street view images or street-view images).

Introduction:

LINE 35: paraffin and gas are natural gases both, so please, correct this list adequately.

LINE 40: What is it supposed to mean: (Joint Research Centre (European Commission) et al., 2003) ? Please, correct it!

I will not mention them one by one, but in-text citations are not in the required form, please, correct them throughout the entire manuscript!

Please, avoid general statements in sentences. Try to provide numbers, and scales when it is about e.g. rapid growth of urbanization (where, which period, what rate?).

Please, try to re-phrase sentences where you start the sentence with ‘First, ….’ or ‘Second,…’. It is not a good style, and is over-used in this manuscript. Section 1.2 Knowledge Gap should be restructured and re-phrased accordingly. What is the justification for the statement in LINES 96-99?? Too general statement, not scientifically supported here. Please, correct it.

My question is: Is high-resolution data always better? Think about that, and if for the problem about what you provide a solution, this is favourable, then highlight that.

Literature Review:

LINE 143: CO2 must be CO₂ . Please, correct it everywhere in the manuscript.

LINE 230: What is CBD? It appears 2 times, here and in LINE 502.

Data and Method:

Section 3.1.2: The PlanetData link does not work, hence the RS CE data is not available or checkable. I cannot accept this in its present form.

Fig. 2 is divided at the moment, but even in one part, they are not expressing well the steps you made. The texts are barely readable. Please, modify it. Improve both its editing and structure. In addition, please, provide a better description in the section.

Section 3.2: Line 287 and Line 290, moreover, line 315 are the same information, please reduce repeats. http://api.map.baidu.com/panorama/v2 link does not work, hence it is not acceptable. Remove it, or correct it.

Table 1 is supposed to be result? However, please clarify what it is! In this manuscript, there is no written text about Table 1. However, Category ‘earth’ must be not correct, please, rephrase it, e.g., to soil or bare ground.

There is no solid information on the exact days/periods involved in the analysis. This is important to see which seasons were involved. It is important to state and take into account whether the vegetation is green or not resulting in differences in SVI analysis and in CE rate.

If the CE values from RS data and the CE values you estimated do not cover the same time period of a year, then your method is wrong. You cannot compare CE data from winter or annual averages and CE data calculated based on summer data. This must be corrected because you can get results showing less valid data. Either way, this also must be described here.

Fig. 4 needs modification: barely readable texts cannot remain. What are the numbers 2 and 1 in the centre of Beijing, on the Fig. 4. (4) submap.

Fig. 5: What is the legend for the colours (b)?

Principal component analysis and cluster analysis are recommended to be performed on your data.

Table 2 is informative, but correct km² instead of km2.

R² grammar must be checked, provide a more adequate description here, and modify 3.3.2. accordingly.

Results and Discussions:

LINE 361 the range 950-1056 is greater than any number on the maps later.

Fig. 6 is done for what reason? Similarly, Table 3 was edited, but the former one is Figure, while the latter one is Table. Moreover, why do you name differently in the titles?

Impact ranking and Feature importance are important results but there is no description of them in the methodology previously.

In the description, there is no proper description of these results. Figure titles are not good.

You mention that the most significant correlations are (LINE403-405):

CE – sidewalk: -0,04

CE – fence or wall: -0,03

CE – buildings: 0,12.

Are they really high correlation? I cannot accept them. Please, re-think and re-do these results.

Try to plan more adequately which SVI element may influence potentially CE! If an element, which can be delineated very well on SVI, does not seem to be a CE factor, then stop taking it into account, simply.

Fig. 9 is very complex, and a good idea, but contains many methodological errors: symbology is presented as continuous but on the map, the grid's colouring seems discrete. Please, use unit names as well, not only numbers. The histograms are good, but the break values should be applied to the symbology of the map in the legend. Provide an explanation for the different characteristics of the predicted and the observed residential CE.

Please, provide a more adequate Conclusion part and write the limitation part using scientific proofs, based on your results, mainly.

References:

The authors do not follow the formal requirements, please, redo the list. One by one check how to cite properly here.

There is a need for a grammar check.

Author Response

Thanks for your careful and constructive comments, we've learned a lot through this review and revision process. Please see the attached document for detailed responses.

Author Response File: Author Response.docx

Reviewer 3 Report

The article is current and researches important aspects for solving everyday problems.

The authors show in the introduction that the measurement and estimation of carbon emissions (CE) is important to allow solving various urgent environmental problems, including global warming. In previous studies, the impact of urban landscapes at the micro level was not fully incorporated, which could lead to biased and incomplete predictions.

To fill the gap, the authors of the paper developed an efficient framework to predict residential CE in urban areas from widely available and publicly available street view imagery (SVI) using machine learning. First of all in the conducted study, the authors used a semantic segmentation algorithm to classify more than 30 streetscape elements derived from SVI images to describe the built environment whose characteristics could affect residential and transport CE. Second, based on the quantified streetscapes, the authors of the article trained a cross- 20 10-fold validation method with various machine learning models to predict CE at the 1KM grid level using CE data from Planet Data.

To begin with, the authors of the article first found the characteristics of the built environment, such as sidewalks, roads, fences, buildings and walls, which are significantly correlated with residential CE. Second, the presence of buildings and subtle streetscape features (eg, walls, fences) indicate higher density residential areas that are related to more residential CE. Third, vegetation (eg, trees and grass) is inversely related to residential CE.

The findings in the present study by the authors of the article shed light on the feasibility of using a single and open data source (ie, SVI) to effectively model CE at the neighborhood level for regions of diverse urban forms. The framework presented and developed by the authors of the study, in the present article, is useful for urban planners to inform the development of new cities and urban regeneration towards low EC objectives

The conclusions that emerge from the work presented are mainly given by the prediction model developed by the authors of the article, which is based on SVI and is a new tool, to predict the residential CE in the urban area according to publicly available Street View images (Google /Baidu).

The model can track temporal and spatial changes well to predict. The model can be used for data visualization and data prediction, can effectively provide CE data for decision makers and urban planners in public environmental protection. The model reflects the influences of specific characteristics on residential CE, so that it could provide urban signers with simulation experiments on specific factors of environmental influence.

For the multitude of researchers, the authors' approach in the present article presents a new perspective for predicting data and increases the application of machine learning in multidisciplines. In addition, the visual expression of the model also provides the opportunity for ordinary citizens to participate in public decision-making activities and choosing where to live.

  This study, presented by the researchers, brings a series of contributions to the topic studied and presented.

I congratulate the authors of the presented article for the work done in carrying out the study.

The bibliography is rich and reflects the work done and the multitude of sources that have been researched.

Minor English editing required

Author Response

Thanks for your careful and constructive comments, we've learned a lot through this review and revision process. Please see the attached document for detailed responses.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Dear Authors,

I accept most of your answers, thank you for the detailed reply and the work. However, please, work more on the figures since the annotations, legends and other text elements are barely visible without 300x zoom. And such high-quality journals cannot let undemanding parts be published.

Regards,
 Reviewer

A minor check is recommended.

Author Response

Thanks and we've revised most of the figures. Please see attached response letter and manuscript for more details.

Author Response File: Author Response.docx