Updating of Land Cover Maps and Change Analysis Using GlobeLand30 Product: A Case Study in Shanghai Metropolitan Area, China

Round 1

Reviewer 1 Report

In general, topic of updating of landcover maps is very interesting for the readers. However, updating of global LC product on 1 local study region is not enough to reach the readers from the global LC community. Therefore, I recommend reformulating the objectives and clear focus on "using of global LC products for local studies", in this case: urbanization in Shanghai metropolitan area. Thus, more readers could be interested for potential reuse and upgrade of the described approach other parts of the world. In the following part, I describe my crucial recommendations and comments:

• I recommend reformulate the title to emphasise application urbanization context and skip “remote-sensing-based change detection and transfer learning”.
• Please describe in introduction more clearly to readers general context of updating of land cover products and if possible, try to clarify main terminology in this context (domain adaptation, transfer learning, signature extension,…) and if possible classify main approaches, eg. those using change detection and those not. Might be, it could be enough that instead of only listed the references, authors will summarize the main conclusions from those studies ending with how the current will should contribute to the topic.
• Please consider whether the final objective: “some factors that affect updating results are explored” should be define as particular objective. If so, the section that reflects this objective should be replace to the result section. In the current manuscript this part is in the discussion section which is not correct.
• In study area description (line 116) please, provide LC distribution statistics based on the GlobaLand30 and if possible of some other ancillary data sets, in order to describe proportion of metropolitan land use. What kind of cultivated land use is there?
• I missed final product (Updated 2011 product) in the framework figure 2. I would remove the “Factors affecting the updating accuracy” from the graph
• Line 262: SVM instead of SAM
• Accuracy assessment: please provide sample distribution map and clarify whether it was pixel based or object based visual check. If strictly pixel based explain why, since you used object-based approaches.
• In result section (lines 421-441), please provide some tables and figures that could be more effective than textual description of LC transition and accuracies. Please provide comparison as in Table 3, for LC transition quantities, e.g. using GlobeLand30 2000/2010 and GlobeLand30 2010/updated 2011.  
• Discussion with other studies is completely missing. Please provide comparisons with other studies in order to get idea about the level of satisfaction of the approach and results.
• To me, as stated at the beginning, the usability of global LC in local context is one of the value of this study. In this regard, the approach seems to be successful if reveals main types and place of the changes in the metropolitan region. This should be clearly stated and discussed with another studies. I am sure there are many studies on land cover and use change in the region, for example as it is cited: “water bodies and wetland areas are also highly dynamic (ref. 37)” please provide what does this mean in particular land cover transitions.   

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

„GlobeLand30 product updating and change analysis using remote-sensing-based change detection and transfer learning A case study in Shanghai, China” is an interesting article with a suitable structure and proportions. The Authors set research goal to update  GlobeLand30 land cover product from 2000 to 2011 on the example of Shanghai, China. The applied methodology (object-based extended change vector analysis – ECVA_OB) is adequate and presented understandably. The results are elaborated in enough detail. I find the paper acceptable for publication in its present form with a minor revision.

Some remarks & questions:

Why did the Authors decide to use Landsat data from 2011? Wouldn't it be better to perform the classification for 2010 to compare the accuracy of the methods? With the dynamic development of Shanghai, could changes in land cover that occurred during the year affect the results of the comparison between ECVA_OB method and GlobeLand30 2010?

There are strange references in the "Introduction" part, for example row 59 "Literature [25]", row 70 "literature [1]" or row 70 "Literature [27]", and so on. Wouldn't it be better to enter the names of the authors quoted?

In the description of the research area the attached map (Figure 1) actually shows change detection check points and classification check points, not the area under study. I suggest placing this map in section "3.4 Accuracy assessment" and presenting the study area in relation to major objects such as water or major roads.

Figure 10 page 19, the map lacks the scale, boundaries of the study area and the marking of areas of unchanged land cover in the legend.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

I agree to publish.