Earth Observation for the Implementation of Sustainable Development Goal 11 Indicators at Local Scale: Monitoring of the Migrant Population Distribution

Round 1

Reviewer 1 Report

The paper presents a method for mapping/monitoring migrant population at local scales from EO data. The study adopts the JRC methodology for mapping migrant population density, and offers a vector-based implementation in an open-source platform, QGIS. The results of the study provide updated mapping for 2018 at regional scale, covering Bari and Modugno in Apulia region, Italy, and relevant SDG11 indicators maps. 

The manuscript conveys clearly the analysis/processing steps of the methodology and the results via a good number of maps and well chosen visualisations. 

Some main remarks about the presentation 

The focus of the work is not clear/sharp enough. It seems there are multiple objectives: updating the JRC map(s) of 2011 for the Apulia region; implementing SDG 11 indicators; offering a vector-based implementation for dasymetric mapping. The multiple objectives are somehow weakening the overall contribution and robustness of the work, e.g. a new implementation, vector-based, for dasymetric mapping calls for measures/evidence for the benefits, i.e. performance improvements e.g. in computation time.  The novelty and contributions of this work are not clear or strong enough, e.g. the importance of the output maps comes in sentences scattered in different sections of the manuscript, materials and methods, .. discussions, conclusions. Conclusions are weak. The whole section sounds more like after thoughts, e.g. in data availability, applicability of the methodology. 

Some minor comments

The paper seems to be promising a more continuous temporal dimension, change/monitoring through years. You then finally see it is an update of 2011 maps for the year 2018. Not all equations are clear as to what they measure or how.  You mention that the vector-based method can be used in other studies, and that would be a strong point. Give some more attention as to how, or which studies. 

Author Response

Response to Reviewer 1 Comments

 

Point 1: The focus of the work is not clear/sharp enough. It seems there are multiple objectives: updating the JRC map(s) of 2011 for the Apulia region; implementing SDG 11 indicators; offering a vector-based implementation for dasymetric mapping. The multiple objectives are somehow weakening the overall contribution and robustness of the work, e.g. a new implementation, vector-based, for dasymetric mapping calls for measures/evidence for the benefits, i.e. performance improvements e.g. in computation time. 

 

Response 1: We have extensively revised the text of the paper to clarify and strengthen  the focus of the work. Specifically:

1. We have changed the title of the paper. The new one is: EO for the implementation of SDG 11 indicators at local scale: monitoring of migrant population distribution
2. We have revised the abstract by evidencing the main objective of the work and the working objectives useful to address the main one.
3. We have reorganized the introduction, as well as the discussion and conclusion sections accordingly.

 

Point 2: The novelty and contributions of this work are not clear or strong enough, e.g. the importance of the output maps comes in sentences scattered in different sections of the manuscript, materials and methods, .. discussions, conclusions.

 

Response 2: We have concentrated the importance of our work in the discussion section.

 

Point 3: Conclusions are weak. The whole section sounds more like after thoughts, e.g. in data availability, applicability of the methodology.

 

Response 3: We tried to improve the conclusions. See the new section

 

Point 4: The paper seems to be promising a more continuous temporal dimension, change/monitoring through years. You then finally see it is an update of 2011 maps for the year 2018.

 

Response 4: We have revised the text and evidenced that we detected the change in the regular migrant population density comparing the epochs 2011 and 2018, according to user request.

 

 

Point 5: . Not all equations are clear as to what they measure or how.

 

Response 5: In section 2, we have introduced a sub-section, titled 2.5. Ancillary data for the implementation of SDG 11 sub-indicators. Such sub-section describes the input used for equations 4 to 6. Additional minor corrections have been done.

 

Point 6: You mention that the vector-based method can be used in other studies, and that would be a strong point. Give some more attention as to how, or which studies. 

 

Response 6: We have described additional applications of the dasymetric vector-based approch in the conclusion section

 

Reviewer 2 Report

A brief summary (one short paragraph) outlining the aim of the paper and its main contributions.

improved methods ... replication ... open source for QGIS

Electronic files and software regarding the full details of the calculation and experimental procedure, if unable to be published in a normal way, can be deposited as supplementary material
We also accept manuscripts communicating to a broader audience with regard to research projects financed with public funds

 

Scope

Change detection

Image processing and pattern recognition

Remote sensing applications

 

Section

Urban Remote Sensing

Urban data-processing methods and algorithms

Urban disaster-monitoring and change analysis

Urban climate-change and variation

Space-time analysis of urban environmental parameters

Interdisciplinary urban case studies

 

Comments  and Suggestions for Authors

The article presents unique work that tries to bridge EO image processing with demographic geostatistics for the purpose of SDG indicators application. The effort behind the presented materials, methods and results seems to have been serious and with experimental character. It builds upon the quoted previous attempt of the European Joint Research Center (JRC) and its Data for Integration initiative (D4I) with the usage of similar datasets and the dasymetrc method but improving it. The D4I itself had been opened to various contributions from many research centers at the time of the elaboration of the basic dataset. This gave impetus for further valuable analyses from diverse research institutes. This later contribution described in the article tries to replicate the basic results by upgrading the standard EO image processing approach through optimization with the help of vector based technique which is claimed to preserve more rich attribute data, to save computing resources compared to the raster one and to be in agreement with the SDG metadata files for Indicator 11.1.1: Proportion of urban population living in slums, informal settlements or inadequate housing and Indicator 11.3.1: Ratio of land consumption rate to population growth rate. Along with that there is a promise that the processing chain “will be 100 made available on free access repository”. The article was funded by public sources and there are also messages directed to the proper management of demographic data which is a significant issue raised by the authors. All these aims and contributions of the article are in line with the journal’s aims and two of its three “unique features”. As a scope it seems to falls in the “Change detection”, “Image processing and pattern recognition” and “Remote sensing applications” categories and can be attributed to several topics in the “Urban Remote Sensing” section of the journal.

The bridging of the Sentinel EO data through the actualized and improved application of the dasymetric method with the challenging for standardised processing but more actual demographic data has been well articulated in the context of the SDG indicators 11.01.01 and 11.03.01. The demonstrated approach may be helpful in broader terms, especially in areas with lack of well structured and standardised data and in situations when time is factor for both gathering of more reliable and verified census data as well as its processing. This is especially true for many localities experiencing intensive migration flows and makes the article interesting for readers working for such places as well as raises the significance of the content. The quality of the presentation can be improved in terms of more readable legends and symbol levels at some of the figures. A more scientifically sound approach would require examples for control of all of the procedures through more detailed checks, estimation of errors and samples for verification.

Several more specific comments and questions can be outlined about the following:

The sentence in the abstract at lines 22-24 does not provide clear idea which of the two methods used is preferred no matter of the discussion that the knowledge-driven automatic classification technique may be used when there is lack of more reliable data driving the automatic classification. At Figure 3 many census zones are missing inside of the town of Modugno. Is this explained by the lack of migrant population or there are gaps in the census tracts graphic used? At Figure 4 the category “residential” is not correct interpretation of the original CLC dataset used which is associated with continuous and discontinuous urban fabric which includes predominantly residential areas but also many public amenities and open spaces so it is not homogenous and can be interpreted as predominantly residential urban fabric. The weighting procedure with the correction factor at Table 1 is replication of the JRC approach but for local scale purposes it is not that sound and other approaches can be discussed (such as the known number of households, residential area, etc. from the standard censuses bound to census tracts, buildings or even addresses (entrances)) in comparison between the cadastral-based dasymetric approach and the EO-based when dealing with aggregation/disaggregation of census data. It is true that sometimes cadastral and census addresses differ which may hamper or make hard the attribution of the latter to the first. At line 594 in the discussion the INSPIRE Directive at least for the EU can be mentioned in terms of existing requirement for buildings information, along with that the Copernicus building heights dataset (which currently available only for the capital cities) can be commented in terms of its replication which can improve the basic EO data according to which the dasymetric method is being applied At Figure 6 the graphic starts with output at the top which seems illogical At Figure 12 the lines of the grid cells can be made more transparent in order see the changes more clear A Figure 14 the legend is much blurred At both Figures 3, 10, 11, 12, 15a and 16 the measurement unit should be specified in the legend or in the title of the figures (e.g. number, percent, coefficient m2/capita, etc.)

Author Response

Response to Reviewer 2 Comments

Point 1: The quality of the presentation can be improved in terms of more readable legends and symbol levels at some of the figures.

Response 1: We have revised the figures and legends according to the reviewer ‘suggestion.

 

Point 2: A more scientifically sound approach would require examples for control of all of the procedures through more detailed checks, estimation of errors and samples for verification.

Response 2: Due to the confidentiality nature of some information (e.g., the exact address number) it was not possible to use reference samples for validation. However,

in order to handle the dasymetric modelling uncertainty related to the output map, two indicators have been introduced, i.e., the Mean Absolute Error (MAE) and the Total Absolute Error (TAE) described in references 10 and 34. These two indicators are described in section 2.4 and the values obtained for the output of the dasymetric approach have been provided in the result section.

 

Point 3: The sentence in the abstract at lines 22-24 does not provide clear idea which of the two methods used is preferred no matter of the discussion that the knowledge-driven automatic classification technique may be used when there is lack of more reliable data driving the automatic classification.

Response 3: We have revised the abstract. The SVM supervised classifier was preferred, but we demonstrated the feasibility of the knowledge driven approach by comparing the two approaches for the study areas.

 

Point 4: . At Figure 3 many census zones are missing inside of the town of Modugno. Is this explained by the lack of migrant population or there are gaps in the census tracts graphic used?

Response 4: It is explained by the lack of migrant population. In addition, Modugno is characterized by large industrial areas with low population density. The comment has been introduced in Figure 2, which corresponds to old Figure 3.

 

 

Point 5: . At Figure 4 the category “residential” is not correct interpretation of the original CLC dataset used which is associated with continuous and discontinuous urban fabric which includes predominantly residential areas but also many public amenities and open spaces so it is not homogenous and can be interpreted as predominantly residential urban fabric..

 

Response 5: We have indicated the original CLC codes in the table of Table 1. Such Table includes the CLC considered in old Figure 4, new Figure 3.  

 

Point 6: The weighting procedure with the correction factor at Table 1 is replication of the JRC approach but for local scale purposes it is not that sound and other approaches can be discussed (such as the known number of households, residential area, etc. from the standard censuses bound to census tracts, buildings or even addresses (entrances)) in comparison between the cadastral-based dasymetric approach and the EO-based when dealing with aggregation/disaggregation of census data. It is true that sometimes cadastral and census addresses differ which may hamper or make hard the attribution of the latter to the first.

 

Response 6: We added a sentence in the Discussion section: Even though other weights selection approaches should be analyzed for local scale analysis, at the present such selection is beyond the scope of our investigation.  However, this selection will be considered in future developments.”

 

 

Point 7: At line 594 in the discussion the INSPIRE Directive at least for the EU can be mentioned in terms of existing requirement for buildings information, along with that the Copernicus building heights dataset (which currently available only for the capital cities) can be commented in terms of its replication which can improve the basic EO data according to which the dasymetric method is being applied

 

Response 7: We have added the following sentence in the discussion section: The output settlement map from Sentinel-2 data includes only buildings defined in agreement with the INSPIRE directive [35] and the definition adopted for the European Settlement Map in [36].

 

 

Point 8: At Figure 6 the graphic starts with output at the top which seems illogical

 

Response 8:  It was a problem with shades of the same colour used for input and outputs. We have changed them.

 

Point 9: At Figure 12 the lines of the grid cells can be made more transparent in order see the changes more clear

 

Response 9: We have improved the Figure.

 

Point 10: A Figure 14 the legend is much blurred

 

Response 10: We have improved the Figure.

 

Point 11: At both Figures 3, 10, 11, 12, 15a and 16 the measurement unit should be specified in the legend or in the title of the figures (e.g. number, percent, coefficient m2/capita, etc.)

 

 

Response 11: Done. We have improved the Figures.

 

 

 

 

 

Reviewer 3 Report

Very interesting and well written paper. Recently, the problem of immigrants’ settlement in Europe, and especially  Italy, is extremely important. The paper presents a modified dasymetric modelling of immigrant distribution and enables the calculation of indicators showing the implementation of the SDG11 goal on the sub-urban level. Both methodological approach and achieved results are important. However, the issue of dasymetric modelling uncertainty is not stress co

Very interesting and well written paper. Recently, the problem of immigrants’ settlement in Europe, and especially to Italy, is extremely important. The paper presents a modified dasymetric modelling of immigrant distribution and enables the calculation of indicators showing the implementation of the SDG11 goal on the sub-urban level. Both methodological approach and achieved results are important. However, the issue of dasymetric modelling uncertainty is not stress comprehensible. Uncertainty is especially important for decision making, and previously was discussed in many paper, e.g.: B. Calka; E. Bielecka. Reliability Analysis of LandScan Gridded Population Data. The Case Study of Poland, Journal: ISPRS Int. J. Geo-Inf., 2019, 8(5), 222; https://doi.org/10.3390/ijgi8050222 ;

 

 

Author Response

Response to Reviewer 3 Comments

 

Point 1: Very interesting and well written paper. Recently, the problem of immigrants’ settlement in Europe, and especially to Italy, is extremely important. The paper presents a modified dasymetric modelling of immigrant distribution and enables the calculation of indicators showing the implementation of the SDG11 goal on the sub-urban level. Both methodological approach and achieved results are important. However, the issue of dasymetric modelling uncertainty is not stress comprehensible. Uncertainty is especially important for decision making, and previously was discussed in many paper, e.g.: B. Calka; E. Bielecka. Reliability Analysis of LandScan Gridded Population Data. The Case Study of Poland, Journal: ISPRS Int. J. Geo-Inf., 2019, 8(5), 222; https://doi.org/10.3390/ijgi8050222 ;

 

Response 1: We have introduced uncertainty measurements as suggested by this reviewer, as well as new references [10, 32, 34].

Due to the confidentiality nature of some information (e.g., the exact address number) it was not possible to use reference samples for validation. Thus,  two indicators, namely the Mean Absolute Error (MAE) and the Total Absolute Error (TAE), were introduced and  described in section 2.4,  in order to handle the dasymetric modelling uncertainty related to the output map. The value obtained are reported in the result section. In addition, the discussion section reports the following sentence: Uncertainty of the output maps is especially important for decision making, and previously was discussed in many papers [10, 32, 34]. MAE and TAE values were used in our paper to verify that the aggregation of output cell data reproduced the totals for the original census areas [7]. The results obtained, i.e., 0,03 and 40.81, respectively, were in agreement with the input cell values. As discussed in [10], MAE was favored as the main indicator of model performance opposed to the root mean square error (RMSE) because it appears more robust for skewed distribution as in the case of population density.