Toward Model-Generated Household Listing in Low- and Middle-Income Countries Using Deep Learning

Abstract

While governments, researchers, and NGOs are exploring ways to leverage big data sources for sustainable development, household surveys are still a critical source of information for dozens of the 232 indicators for the Sustainable Development Goals (SDGs) in low- and middle-income countries (LMICs). Though some countries’ statistical agencies maintain databases of persons or households for sampling, conducting household surveys in LMICs is complicated due to incomplete, outdated, or inaccurate sampling frames. As a means to develop or update household listings in LMICs, this paper explores the use of machine learning models to detect and enumerate building structures directly from satellite imagery in the Kaduna state of Nigeria. Specifically, an object detection model was used to identify and locate buildings in satellite images. In the test set, the model attained a mean average precision (mAP) of 0.48 for detecting structures, with relatively higher values in areas with lower building density (mAP = 0.65). Furthermore, when model predictions were compared against recent household listings from fieldwork in Nigeria, the predictions showed high correlation with household coverage (Pearson = 0.70; Spearman = 0.81). With the need to produce comparable, scalable SDG indicators, this case study explores the feasibility and challenges of using object detection models to help develop timely enumerated household lists in LMICs.

Global Ambitions, Local Contexts: Alternative Ways of Knowing the World

“…Further, for several indicators the production of data is not yet clear. A number of SI contributions [30,[37][38][39][40][41][42] engage with the production phase, in particular with sourcing or creating data that can support the measurement and monitoring of conceptually clear indicators, albeit allowing for additional refinements.…”

“…Chew at al. [37], for instance, focus on indicators requiring household listings for their production, for instance to determine progress on adequate housing (SDG indicator 11.1.1). Such data are usually derived from household questionnaires such as those collected by a census or other type of survey.…”

“…However, especially in low-and middle-income countries, only areas recognized by the government are surveyed, while households in 'invisible' areas are excluded from such data collection campaigns. In order to ensure complete coverage and generate support for targeted and inclusive household sampling, Chew et al [37] develop an object detection model based on EO data (Bing maps) to detect housing structures in Kaduna state in Nigeria. To do so, they employ machine learning, specifically deep learning, a promising method to directly enumerate housing structures in EO data for compiling household lists as input for household sampling [27].…”

“…To do so, they employ machine learning, specifically deep learning, a promising method to directly enumerate housing structures in EO data for compiling household lists as input for household sampling [27]. While Chew et al [37] rely on EO data to account for the unaccounted, Ulbrich et al [30], in this issue, advocate participatory geo-spatial methodologies and citizen-generated data practices to collect data on multi-dimensional urban issues of marginalized urban communities and complement national statistics.…”

“…In Table 2, we summarize the different contributions, their link to policy making as well as the critical insights offered, where applicable. Green space is lost due to densification, while the aim is to increase green space [38] Seeing the methodology as complementary/supportive [37,38,40] Spatial knowledge of informal housing (location, degree of deprivation) to inform pro-poor development strategies [27] Considering the importance of local differences/goals [25,27,36] Emphasizing potential societal effects [27]; ethical implications, possibility of multiple understandings/framings [27] Identification of underrepresented groups to leave no one behind [37] Considering shortcomings of conventional population statistics [37] Monitoring forest cover trends in support of forest conservation [40] Warning of 'uncertainty in values' [27,40] Consistent information on urban built-up area to inform land-use efficient urban development strategies [41,42] Addressing shortcomings of administrative boundaries [41,42] Geographic information analysis and maps (if suitable spatial data are available) can adequately capture urban reality (land use, population distribution, accessibility, etc. )…”