An official website of the United States government
Despite improvements in safe water and sanitation services in low-income countries, a substantial proportion of the population in Africa still does not have access to these essential services. Up-to-date fine-scale maps of low-income settlements are urgently needed by authorities to improve service provision. We aim to develop a cost-effective solution to generate fine-scale maps of these vulnerable populations using multi-source public information. The problem is challenging as ground-truth maps are available at only a limited number of cities, and the patterns are heterogeneous across cities. Recent attempts tackling the spatial heterogeneity issue focus on scenarios where true labels partially exist for each input region, which are unavailable for the present problem. We propose a dynamic point-to-region co-learning framework to learn heterogeneity patterns that cannot be reflected by point-level information and generalize deep learners to new areas with no labels. We also propose an attention-based correction layer to remove spurious signatures, and a region-gate to capture both region-invariant and variant patterns. Experiment results on real-world fine-scale data in three cities of Kenya show that the proposed approach can largely improve model performance on various base network architectures.
more »
« less
Deep Transfer Learning on Satellite Imagery Improves Air Quality Estimates in Developing Nations
Urban air pollution is a public health challenge in low- and middle-income countries (LMICs). However, LMICs lack adequate air quality (AQ) monitoring infrastructure. A persistent challenge has been our inability to estimate AQ accurately in LMIC cities, which hinders emergency preparedness and risk mitigation. Deep learning-based models that map satellite imagery to AQ can be built for high-income countries (HICs) with adequate ground data. Here we demonstrate that a scalable approach that adapts deep transfer learning on satellite imagery for AQ can extract meaningful estimates and insights in LMIC cities based on spatiotemporal patterns learned in HIC cities. The approach is demonstrated for Accra in Ghana, Africa, with AQ patterns learned from two US cities, specifically Los Angeles and New York.
more »
« less
- Award ID(s):
- 1735505
- PAR ID:
- 10336193
- Date Published:
- Journal Name:
- ArXivorg
- ISSN:
- 2331-8422
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Floods cause extensive damage in high-income countries, including the United States, but problems are more severe in low- and middle-income countries (LMICs) that lack preventative and mitigating infrastructure. Marginalized children’s education in LMICs might be particularly vulnerable. Using the Indian Human Development Survey, we investigate flood exposure implications for the education of school-age rural children, paying particular attention to children from marginalized groups. Results show that lower-caste Hindu, Muslim, and poorer children with less-educated parents in agricultural households are more likely to experience flooding. Interactions between flooding and marginalization characteristics indicate that flood exposure is associated with disproportionately negative learning outcomes for girls and that economic resources may mitigate flood exposure effects on delayed school progress. While greater exposures for marginalized groups are concerning, the limited number and modest magnitudes of documented negative effect heterogeneities for marginalized children are somewhat better news.more » « less
-
One-third of children globally have blood lead levels (BLLs) exceeding the (former) US CDC reference value of 5 μg/dL; this value may be as high as one-half for children in low- and middle-income countries (LMICs). Lead exposure occurs through a variety of routes (e.g., water, dust, air), and in LMICs specifically, informal economies (e.g., battery recycling) can drive lead exposures due, in part, to absent regulation. Previous work by our team identified a ubiquitous source of lead (Pb), in the form of Pb-containing components used in manually operated pumps, in Toamasina, Madagascar. Characterization of BLLs of children exposed to this drinking water, and identification of additional exposure routes were needed. BLLs were measured for 362 children (aged 6 months to 6 years) in parallel with surveying to assess 14 risk factors related to demographics/socioeconomics, diet, use of pitcher pumps, and parental occupations. BLL data were also compared against a recent meta-review of BLLs for LMICs. Median childhood BLL (7.1 μg/dL) was consistent with those of other Sub-Saharan African LMICs (6.8 μg/dL) and generally higher than LMICs in other continents. Risk factors significantly associated (p < 0.05, univariate logistic regression) with elevated BLL (at ≥ 5 μg/dL) included male gender, living near a railway or major roadway (owing potentially to legacy lead pollution), having lower-cost flooring, daily consumption of foods (beans, vegetables, rice) commonly cooked in recycled aluminum pots (a previously identified lead source for this community), and a maternal occupation (laundry-person) associated with lower socioeconomic status (SES). Findings were similar at the ≥ 10 μg/dL BLL status. Our methods and findings may be appropriate in identifying and reducing lead exposures for children in other urbanizing cities, particularly in Sub-Saharan Africa, where lead exposure routes are complex and varied owing to informal economics and substantial legacy pollution.more » « less
-
Automatic satellite-based reconstruction enables large and widespread creation of urban areas. However, satellite imagery is often noisy and incomplete, and is not suitable for reconstructing detailed building facades. We present a machine learning-based inverse procedural modeling method to automatically create synthetic facades from satellite imagery. Our key observation is that building facades exhibit regular, grid-like structures. Hence, we can overcome the low-resolution, noisy, and partial building data obtained from satellite imagery by synthesizing the underlying facade layout. Our method infers regular facade details from satellite-based image-fragments of a building, and applies them to occluded or under-sampled parts of the building, resulting in plausible, crisp facades. Using urban areas from six cities, we compare our approach to several state-of-the-art image completion/in-filling methods and our approach consistently creates better facade images.more » « less
-
Spatial data are ubiquitous and have transformed decision-making in many critical domains, including public health, agriculture, transportation, etc. While recent advances in machine learning offer promising ways to harness massive spatial datasets (e.g., satellite imagery), spatial heterogeneity -- a fundamental property of spatial data -- poses a major challenge as data distributions or generative processes often vary over space. Recent studies targeting this difficult problem either require a known space-partitioning as the input, or can only support limited special cases (e.g., binary classification). Moreover, heterogeneity-pattern learned by these methods are locked to the locations of the training samples, and cannot be applied to new locations. We propose a statistically-guided framework to adaptively partition data in space during training using distribution-driven optimization and transform a deep learning model (of user's choice) into a heterogeneity-aware architecture. We also propose a spatial moderator to generalize learned patterns to new test regions. Experiment results on real-world datasets show that the framework can effectively capture footprints of heterogeneity and substantially improve prediction performances.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.48550/arXiv.2202.08890
