Exploring the Relationship between Temporal Fluctuations in Satellite Nightlight Imagery and Human Mobility across Africa

Round 1

Reviewer 1 Report

VIIRS observes earth at 1.30am, in contrast to DMSP which observed in the early even (and getting earlier as the satellite aged). The reference you give for the claim that small town lights will get brighter as migrant populations arrive is from 2011 (so pre-VIIRS). I would like you to explain what sort of lights you believe that VIIRS is detecting, and why you believe that these lights at 1.30am in the morning will fluctuate with new arrivals (as opposed to street lights or other public infrastructure lights that are unlikely to be short-term responsive to population arrivals).

I am also puzzled why you spent so much of the paper talking about your processing of the monthly lights while indicating that the VNL annual composites were not available when you wrote the paper. So just how old is this paper, because the VNL annual composites have been available for over two years. Moreover, you could always use Black Marble which are available annually and monthly.

I also found the comparison of DMSP and VIIRS spatial resolution on page 2 (with a citation to Elvidge et al 2013) to be misleading. This is the output grid, but not the resolution of the underlying sensor. Rather than a factor of two, as your sentence implies, Elvidge shows that VIIRS is about 45 times more spatially precise.  

Author Response

Reviewer 1’s Comments and Suggestions for Authors

VIIRS observes earth at 1.30am, in contrast to DMSP which observed in the early even (and getting earlier as the satellite aged). The reference you give for the claim that small town lights will get brighter as migrant populations arrive is from 2011 (so pre-VIIRS). I would like you to explain what sort of lights you believe that VIIRS is detecting, and why you believe that these lights at 1.30am in the morning will fluctuate with new arrivals (as opposed to street lights or other public infrastructure lights that are unlikely to be short-term responsive to population arrivals).

Response: In addition to the pre-VIIRS reference pointed out by the Reviewer in the above comment (ie, Bharti et al., 2011), we have also provided more recent, post-VIIRS, references investigating the relationship between NTL and population mobility using VIIRS data: “Previous research has highlighted the potential of multi-temporal NTL imagery for measuring changes in population presence and density over time as a result of mobility “this has included seasonal labour migration into towns and cities in the Sahel region of Africa ​(Lai, Farnham, et al., 2019)​, and its impact on infectious disease dynamics ​(Bharti et al., 2011)​, seasonal flows of tourists ​(Stathakis & Baltas, 2018; Tselios & Stathakis, 2020)​ and induced displacement ​(Lu et al., 2016).​”, all showing promising results worth to be further investigated (Dickinson et al., 2020). We have now also added in the introduction the two following references investigating the relationship between NTL and human mobility: Xu etal., 2021 and Chen, 2019 (respectively investigating the relationship between VIIRS NTL changes and the net migration at the NUTS III level in Europe, and the COVID-19 lockdown induced night-time light dynamics in global megacities).

According to Eldvige et al. (2013), we are aware that, in contrast  to the DMSP overpass time which is near 7.30pm, the SNPP overpass time is near 1.30am and that peak lighting is prior to 10pm (after which there is some decline in the quantity of outdoor lighting, but we also agree with Eldvige et al. (2013) that VIIRS data strongly indicate that there is still plenty of lighting being detected after midnight which may or may not only linked to public infrastructure lights. We did not believe a priori that these after 1.30am lighting is  fluctuating with new arrivals, but rather this is exactly what we wanted to investigate/have investigated by comparing the NTL and GAMRD data.

After using the annual composites for removing ephemeral lights (unrelated to electric lighting) and background (non-lights) from monthly composites which were already processed for removing persistent gas flares, as well as the impact of sunlit, moonlit, stray lights, lightening, high energy particle, overglow and cloud-cover, the monthly composites should only include electric lights which may or may not be related to population presence, and thus be affected by human mobility in different way in different context (ie, urban, peri-urban, vs rural) - which once again was the scope of this study..

I am also puzzled why you spent so much of the paper talking about your processing of the monthly lights while indicating that the VNL annual composites were not available when you wrote the paper. So just how old is this paper, because the VNL annual composites have been available for over two years. Moreover, you could always use Black Marble which are available annually and monthly.

Response: At the time of the study design and data analysis (mid-2019), VIIRS annual composite data  from either the Earth Observations Group (EOG) at The National Oceanic and Atmospheric Administration (NOAA)/National Centers for Environmental Information (NCEI) or the EOG at the Payne Institute for Public Policy of the Colorado School of Mines were not available for all three years of 2018-2020 (https://www.mdpi.com/2072-4292/13/5/922). This is now better clarified in the revised version of our manuscript. Similarly, we are not sure whether VIIRS annual composite data (ie, VNP46A4 products for 2018, 2019 and 2020) for the three years were available through the NASA Black Marble Archive at the time this study started (https://www.sciencedirect.com/science/article/pii/S003442571830110X). This work conducted during 2019-2021 was significantly delayed by team member changes and the COVID-19 pandemic-related work, as staff time shifted to support mobility and intervention effectiveness analysis for COVID-19 responses. It also took several months to get permission from Google Inc. to publish the research, which also delayed the submission to the journal.

I also found the comparison of DMSP and VIIRS spatial resolution on page 2 (with a citation to Elvidge et al 2013) to be misleading. This is the output grid, but not the resolution of the underlying sensor. Rather than a factor of two, as your sentence implies, Elvidge shows that VIIRS is about 45 times more spatially precise.  

Response: We definitely agree with the Reviewer that our statement, in terms of improved spatial resolution, could have been misleading - we have updated our statement accordingly by explicitly referring to the  spatial resolution of both the Ground Instantaneous Field of Views and the corresponding generated global grids. 

Reviewer 2 Report

Authors integrated VIIRS NTL nighttime datasets to explore short-term and intra-annual human mobility. After proprocessing, NTL-GAMRA correlation was done to found the potentially weaker areas, in addition, the highlights displayed that these available datasets were useful for human mobility mapping and analysis. Generally, this manuscript had good novelty, however, i have few comments that may be useful for authors.

NTL data were seldom used for human mobility. However, for population spatial distribution mapping has been for a long time. Is it available to detect human mobility? I suggest author to make this more clear.

Please improve the dpi of your figures. All tables should follow the format of journal's standard.

Please add more discussion content. At least, authors should compare your results with similar study.

Author Response

Reviewer 2’s Comments and Suggestions for Authors

Authors integrated VIIRS NTL nighttime datasets to explore short-term and intra-annual human mobility. After proprocessing, NTL-GAMRA correlation was done to found the potentially weaker areas, in addition, the highlights displayed that these available datasets were useful for human mobility mapping and analysis. Generally, this manuscript had good novelty, however, i have few comments that may be useful for authors.

Response: Thank you for the very positive comments. We have addressed the comments one-by-one below.

NTL data were seldom used for human mobility. However, for population spatial distribution mapping has been for a long time. Is it available to detect human mobility? I suggest author to make this more clear.

Response: Thanks. As pointed out by the reviewer, NTL data have been used for exploring and mapping population distribution for over a decade, but the association between changes in nighttime brightness and mobility has been rarely explored. The longitudinal Google Aggregated Mobility Research Dataset (GAMRD) and VIIR NTL data for Africa in 2018-2020, according to the degree of urbanisation, provide a good opportunity to improve our understanding of the value of NTL data for assessing human mobility and the associated changes in population presence in low- and middle-income countries. Our results demonstrated the high variability in correlations between NTL radiance values and GAMRD flow metrics across a broad geographic range and within different rural/urban classifications. Administrative units classified as rural and semi-rural were shown to have on average the highest NTL-GAMRD correlation whilst administrative units classified as “urban centres” had the lowest (not including the peri-urban class which had several low p-values). 

Most noticeable in the study was the significant difference in correlation strength between the two periods 2018-19 and 2020. Correlations across most rural/urban classifications and particularly “urban centres” were considerably less in 2020 than in 2018-19. Whilst the variation of the NTL SoL radiance values across all urban classes remained relatively stable throughout the year, the GHL flow metrics referring to 2020 showed that the corresponding flow values were far more erratic than in 2018-19, with the months of April and September most prominent in their deviation and created a consequential effect for the NTL-GAMRD correlations during these months, which might be attributed to the implementation of  COVID-19 lockdown measures, especially affecting the urban areas. 

Please improve the dpi of your figures. All tables should follow the format of journal's standard.

Response: 

Updated in the latest manuscript.

Please add more discussion content. At least, authors should compare your results with similar study.

Response: Thanks for this comment. We only found a similar study conducted by Dickinson et al. (Proc. AAAI Conf. Artif. Intell., 2020,34:394–402). Based on linear regression and random forest models, they used Google’s human mobility data to predict VIIRS satellite imagery and then assessed how accurately this simulated global NTL imagery can be used to predict GDP across regions in 2015-2016. They demonstrated that the relationship between human mobility and VIIRS NTL was both nonlinear and varied considerably around the globe. The differences across regions were made clear by the improvement in the model performance when modelling each region independently rather than constructing a single global model. Our study further measured the degree to which this relationship varied across locations with different levels of urbanisation and development in 2018-2020. However, we found that, compared with urban settings, there was a higher association between NTL data and mobility changes in rural and peri-urban areas. In addition, a reduced NTL-GAMRD correlation strength in 2020 was observed, especially in urban settings, most probably because of the monthly NTL SoL radiance values remaining relatively similar in 2018-19 and 2020, but the human mobility, significantly decreasing in 2020 with respect to the previous considered period. We have further discussed in Discussion section.

Reviewer 3 Report

The authors attempt to research an important and interesting topic.

i) However, I assume that they need to write more about the various research limitations involved. I suppose that it should be appreciated that these limitations are basically presented in part of the Discussion. However, I think that much more details should be written about these limitations in the manuscript. Among these points, I consider it particularly important that Arica is currently characterized by relatively low smart device ownership and internet penetration globally. In addition, even within smart devices, not all users in Africa or the countries chosen for examination may use the software mentioned by the authors.
For these reasons, the main variables in the cited section mentioned in the manuscript below (e.g., proportion of people using smartphones, including Google's Location History feature, or the proportion of the urban or more educated population in the chosen countries) should be covered by African country-by-country official statistical datasets examined:

"Indeed, such data are limited to smartphone users who have opted into Google's Location History feature, which is off by default, and thus they may not be representative of the population as a whole. Similarly, their representativeness may vary by location and be particularly low in rural areas characterised by low population densities. Additionally, GAMRD data are still likely to be biased towards educated males living in urban areas (Lai, zu Erbach-Schoenberg, et al., 2019)." (p. 13.)

Obviously, exact national statistics on these variables may not be available everywhere, but even if they were, estimates (partially even based on some kind of market surveys nationally) would have to be published. Otherwise, the authors’ research will not be scientifically representative in these terms.

ii) The GIS databases chosen by the authors are relatively justified (if we consider e.g., cost requirements or accessibility options), I suppose. However, it would be worth explaining somewhat more about the limitations of these databases related to the authors' research questions and methods (e.g., which geoinformatic-geographical phenomena in general and in Africa they are suitable for investigating and which are not).

iii) It would be worthwhile for the authors to briefly explain the scientific arguments used to select the countries they gave for the four categories of countries they analyzed.

iv) The authors mention on page 4 that "Monthly composites were filtered to exclude data impacted by stray light, lightning, lunar illumination, and cloud-cover (Mills et al., 2013)". I think that it would be worth writing a little more about this filtering and the exact method done by the authors of the manuscript in the study.

Please, correct some misspellings in the manuscript. 

Author Response

Reviewer 3’s Comments and Suggestions for Authors

The authors attempt to research an important and interesting topic.

1. i) However, I assume that they need to write more about the various research limitations involved. I suppose that it should be appreciated that these limitations are basically presented in part of the Discussion. However, I think that much more details should be written about these limitations in the manuscript. Among these points, I consider it particularly important that Arica is currently characterized by relatively low smart device ownership and internet penetration globally. In addition, even within smart devices, not all users in Africa or the countries chosen for examination may use the software mentioned by the authors.

For these reasons, the main variables in the cited section mentioned in the manuscript below (e.g., proportion of people using smartphones, including Google's Location History feature, or the proportion of the urban or more educated population in the chosen countries) should be covered by African country-by-country official statistical datasets examined:

"Indeed, such data are limited to smartphone users who have opted into Google's Location History feature, which is off by default, and thus they may not be representative of the population as a whole. Similarly, their representativeness may vary by location and be particularly low in rural areas characterised by low population densities. Additionally, GAMRD data are still likely to be biased towards educated males living in urban areas (Lai, zu Erbach-Schoenberg, et al., 2019)." (p. 13.)

Obviously, exact national statistics on these variables may not be available everywhere, but even if they were, estimates (partially even based on some kind of market surveys nationally) would have to be published. Otherwise, the authors’ research will not be scientifically representative in these terms.

Response: Thanks for the insightful comments. It is important to highlight the limitations and potential representativeness biases of mobility data across locations. Indeed, such data are limited to smartphone and mobile internet usage as well as the coverage of Google users among populations. In addition, these aggregated data were subject to differential privacy algorithms, designed to protect user anonymity and obscure fine detail. However, as mentioned by the reviewer, subnational and up-to-date statistics on smart device ownership and internet penetration and Google Apps usages are not available for the study countries. Only some survey data on mobile usage at household level were collected by the Demographic and Health Surveys (https://www.dhsprogram.com) in different years. 

Based on data released by the Global System for Mobile Communications (GSMA), by the end of 2020, 495 million people subscribed to mobile services in Sub-Saharan Africa, representing 46% of the region’s population – an increase of almost 20 million from 2019 (https://furtherafrica.com/2022/07/19/african-countries-with-the-highest-number-of-mobile-phones/). However, there was diverse coverage across regions. For example, West Africa had a 96% mobile penetration rate (based on SIM cards) while Southern Africa's rate was 163% and East Africa was just 62%. In Nigerian aged 16 to 64, by January 2021, 99.5% owned a mobile phone with 99.2% owning a smartphone (https://www.connectingafrica.com/author.asp?section_id=761&doc_id=767400). By 2025, mobile phone subscribers and smartphone adoption are expected to grow to 50% of the population and 75% of mobile users, respectively (https://furtherafrica.com/2022/07/19/african-countries-with-the-highest-number-of-mobile-phones/), which may narrow down the biases of mobility data’s representativeness among populations.

However, considering potential biases of representativeness among populations, rather than grouping countries together, it is important to conduct mobile phone/smartphone penetration survey and include socioeconomic covariates for each country and subnational region (e.g. administrative unit level 1 or 2) separately in the future. 

1. ii) The GIS databases chosen by the authors are relatively justified (if we consider e.g., cost requirements or accessibility options), I suppose. However, it would be worth explaining somewhat more about the limitations of these databases related to the authors' research questions and methods (e.g., which geoinformatic-geographical phenomena in general and in Africa they are suitable for investigating and which are not).

Response: As a work package of the project for Mapping seasonal denominator dynamics in low and middle-income settings, funded by the Bill & Melinda Gates Foundation, 2019-2021, this study aimed to investigate the degree to which this relationship between mobility and NTL varies across locations and the degree of urbanisation, particularly in low and middle-income countries and at the monthly timescale. In our previous research, using mobile phone call detail records or smartphone/social media geo-tagged data, we explored mobility and potential driving factors for African countries, such as Namibia, Tanzania, and Kenya. However, these studies only obtained and analysed mobility data for a short time period (e.g. 6-12 months) in a single country. Supported by Google, for the first time we obtained multiple-year (2018-2020) and large-scale mobility data at fine spatial resolution across 12 African countries, to investigate the novel research questions about the association between mobility and NTL by the degree of urbanisation. The diversity of the study countries in different regions does ensure that this study contains wide variance in socioeconomic, geographic, and demographic contexts. With several proxies available, it is useful to understand the limitations and accuracy of each dataset that can be used for mobility research, which motivates the current study. 

However, considering potential biases of representativeness among populations, rather than grouping countries together, it may be of interest to analyse each country separately to avoid generalisations in the future. In addition, as the NTL-GAMRD correlation was found to be potentially weaker in “urban centres”/areas, this highlights the importance of integrating additional geospatial datasets able to capture different scales of variation into a larger multivariate mode, instead of our current simple modelling framework. As refinements in NTL technology become available and new datasets released with higher spatial resolution and enhanced post processing, it is hoped these limitations may be overcome. Despite the demonstrated efficacy of GAMRD in “urban centres”/areas and during lockdown periods, with NTL data continuing to be publicly available with wide geographic coverage, its use can remain important as a proxy of human mobility and the associated population presence and density changes, until alternative datasets, such as mobile phone locations, can be more easily accessed by the scientific and operational communities. We have discussed these in the limitation and Conclusions sections.

iii) It would be worthwhile for the authors to briefly explain the scientific arguments used to select the countries they gave for the four categories of countries they analyzed.

Response: Thanks for the comment. A previous global study (Dickinson et al. Proc. AAAI Conf. Artif. Intell., 2020) found that, in different parts of the world, the relationship between mobility and light production differed considerably, and analyses must account for such regional variations. Since the twelve study countries spanned a broad geographic range across the African continent, we simply categorised countries into four groups, according to the United Nations Geoscheme for Africa [27] which separates African countries according to cardinal direction. We have further clarified this in the Materials and Methods section.

1. iv) The authors mention on page 4 that "Monthly composites were filtered to exclude data impacted by stray light, lightning, lunar illumination, and cloud-cover (Mills et al., 2013)". I think that it would be worth writing a little more about this filtering and the exact method done by the authors of the manuscript in the study.

Response: 

Text now updated: 

Monthly composites were filtered to exclude data impacted by stray light, lightning, lunar illumination, and cloud-cover where the monthly series is run globally using two different configurations. The first excludes any data impacted by stray light. The second includes these data if the radiance values have undergone the stray-light correction procedure. These two configurations, one of which includes the stray-light corrected data, will have more data coverage toward the poles, but will be of reduced quality with the decision on which configuration to use dependent on the context. For each of the months from 2012 - 2020, for the monthly non-tiled versions, the annualmasks for each year were applied to all the months for that year. For example, the 2020 lit mask was applied on all the months of 2020. 

Round 2

Reviewer 1 Report

Thanks for dealing with my previous comments. The revised text does a better job of distinguishing between DMSP and VIIRS resolution.

However, the issue of the timing of the observations, and the activity on earth they would correspond to at 1.30am, is still not fully dealt with, and I believe that you could bring some of what you wrote in your response letter into the main text.

You would provide an enormous service to your readers to clearly state the time of observation, the type of lights observed, and what changes on the ground should be detected. For example, we know from the experiments of Tuttle and colleagues that for DMSP to detect lights in a previously large place, it took a large bank (I think at least nine, mounted on a pickup truck and powered by large generators) of 1000w gas-halogen lamps, modified by adding aluminium reflector shields to direct the light more efficiently upwards. These types of lamps weigh about 25kg each. And even with the bank of them, in an otherwise totally dark area, DMSP only detected them on about half the nights they operated.

Obviously you are using VIIRS not DMSP, but the same sort of real world practical example of what sort of light is detected helps readers to better understand the potential, but also the limits, of satellite-detected luminosity data. This would be far more effective than simply referencing more and more studies.

Author Response

Thanks for dealing with my previous comments. The revised text does a better job of distinguishing between DMSP and VIIRS resolution. 

Thank you for your feedback. 

However, the issue of the timing of the observations, and the activity on earth they would correspond to at 1.30am, is still not fully dealt with, and I believe that you could bring some of what you wrote in your response letter into the main text. 

1. We could not find any published article describing similar experiments to the one mentioned by the Reviewer referring to VIIRS 
2. If the Reviewer could point us to any reference providing such practical example referring to VIIRS, we would be happy to add the corresponding insights in a further revised version of our manuscript - furthermore, as mentioned in the previous rebuttal,  
3. We would also reply that investigating the exact nature of the of lights that can be practically/actually detected using VIIRS is somehow far beyond the scope of our manuscript (which main aim is to compare the NTL and GAMRD data to investigate whether 1.30am lighting is fluctuating according to the observed change in monthly mobility). 

Reviewer 3 Report

The manuscript has improved, I accept the answers.

The manuscript reads relatively well. 

Author Response

Thank you for comments.