Slums, or deprived urban settlements, are characterized by inadequate housing, limited access to essential services, and overcrowding. Therefore, monitoring the populations living in these areas is crucial for promoting sustainable urbanization and implementing effective, targeted policy interventions. Deep learning-based remote sensing technologies applied at sub-meter resolution can detect informal settlements with high accuracy. However, scaling to multi-temporal and cross-national contexts remains challenging. Here, we developed a scalable computer vision model that detects slums across diverse geographies and timescales, using 60-cm-resolution images and minimal labeled data. We applied the model to 12 cities in low- and middle-income countries from 2014 to 2024. Our longitudinal analysis indicates that informal settlements expanded during the COVID-19 pandemic, even as global estimates from UN-Habitat suggest an overall long-term decline in slum prevalence. Our results demonstrate that redevelopment programs intended to improve slum conditions are associated with unintended spatial spillovers, including new settlement growth in surrounding areas. During the COVID-19 pandemic, informal settlements increased and redevelopment programs intended to improve slum conditions were associated with unintended spatial spillovers in 12 cities in low- and middle-income countries, according to a deep learning-based analysis

Background Lack of access to Water, Sanitation, and Hygiene (WASH) services is a major public health concern in refugee camps, where extreme crowding accelerates the spread of communicable diseases. The Rohingya settlements in Cox’s Bazar, Bangladesh, exemplify these conditions, with large populations living under severe spatial constraints. Methods We develop a semi-supervised segmentation framework using the Segment Anything Model (SAM) to map shelters from multi-temporal sub-meter remote sensing imagery (2017–2025), improving detection in complex camp environments by 4·9% in F1-score over strong baselines. Findings The detected shelter maps show that shelter expansion stabilized after 2020, whereas continued population growth reduced per capita living space by ∼14% between 2020 and 2025. WASH accessibility, measured with an enhanced network-based two-step floating catchment area (2SFCA) method, declined from 2022 to 2025, increasing facility loads and exceeding global benchmarks. Gender-disaggregated scenarios that incorporate safety penalty further reveal pronounced inequities, with female accessibility ∼27% lower than male. Interpretations Our results demonstrate that remote sensing–driven AI diagnostics can generate equity-focused evidence to prioritize WASH investments and mitigate health risks in protracted displacement settings.

Satellite-based slum segmentation holds significant promise in generating global estimates of urban poverty. However, the morphological heterogeneity of informal settlements presents a major challenge, hindering the ability of models trained on specific regions to generalize effectively to unseen locations. To address this, we introduce a large-scale high-resolution dataset and propose GRAM (Generalized Region-Aware Mixture-of-Experts), a two-phase test-time adaptation framework that enables robust slum segmentation without requiring labeled data from target regions. We compile a million-scale satellite imagery dataset from 12 cities across four continents for source training. Using this dataset, the model employs a Mixture-of-Experts architecture to capture region-specific slum characteristics while learning universal features through a shared backbone. During adaptation, prediction consistency across experts filters out unreliable pseudo-labels, allowing the model to generalize effectively to previously unseen regions. GRAM outperforms state-of-the-art baselines in low-resource settings such as African cities, offering a scalable and label-efficient solution for global slum mapping and data-driven urban planning.

Recent advancements in computer vision have significantly improved slum detection using very-high-resolution satellite imagery. However, current algorithms are limited to identifying slums based on single-temporal labels and lack the capability to perform multi-temporal analysis. Here we present a supervised learning model trained on multi-temporal labels, specifically designed to maintain consistent performance across temporal variations. We evaluate our model against baseline approaches trained on singe-temporal labels. Case studies in two cities, Caracas and Karachi, demonstrate that incorporating additional temporal satellite imagery during training produces more consistent and reliable results for multi-temporal analysis. Our study suggests new ways to leverage temporal data in slum detection to effectively monitor urban poverty and track dynamics of informal settlements over time.

Deep learning applied to very high resolution (VHR) satellite imagery has become a powerful method for analyzing urban poverty at scale. Traditionally, these models have required extensive labeled datasets for training. In this study, we propose a semi-supervised learning method that leverages vast amounts of unlabeled satellite images to detect slum settlements across diverse geographical regions and multiple time periods. Our approach can significantly reduce the dependency on manual data annotations. We showcase results based on four cities in sub-Saharan Africa and map slum settlements near a decade period between 2015 and 2024. The findings reveal localized settlement changes in slum areas at a resolution of 60 cm per pixel, which enables a more precise understanding of intra-city disparities. To support policymaking and humanitarian efforts, we make our data and code publicly available.

Recent advances in deep learning have enabled the inference of urban socioeconomic characteristics from satellite imagery. However, models relying solely on urbanization traits often show weak correlations with poverty indicators, as unplanned urban growth can obscure economic disparities and spatial inequalities. To address this limitation, we introduce a novel representation learning framework that captures multidimensional deprivation-related traits from very high-resolution satellite imagery for precise urban poverty mapping. Our approach integrates three complementary traits: (1) accessibility traits, learned via contrastive learning to encode proximity to essential infrastructure; (2) morphological traits, derived from building footprints to reflect housing conditions in informal settlements; and (3) economic traits, inferred from nightlight intensity as a proxy for economic activity. To mitigate spurious correlations - such as those from non-residential nightlight sources that misrepresent poverty conditions - we incorporate a backdoor adjustment mechanism that leverages morphological traits during training of the economic module. By fusing these complementary features into a unified representation, our framework captures the complex nature of poverty, which often diverges from economic development trends. Evaluations across three capital cities - Cape Town, Dhaka, and Phnom Penh - show that our model significantly outperforms existing baselines, offering a robust tool for poverty mapping and policy support in data-scarce regions.

Lack of access to Water, Sanitation, and Hygiene (WASH) services is a major public health concern in refugee camps, where extreme crowding accelerates the spread of communicable diseases. The Rohingya settlements in Cox's Bazar, Bangladesh, exemplify these conditions, with large populations living under severe spatial constraints. We develop a semi-supervised segmentation framework using the Segment Anything Model (SAM) to map shelters from multi-temporal sub-meter remote sensing imagery (2017-2025), improving detection in complex camp environments by 4.9

Mongolia is among the countries undergoing rapid urbanization, and its temporary nomadic dwellings—known as Ger—have expanded into urban areas. Newly formed ger communities in cities are potentially recognized as informal settlements, or slums. The distinctive circular, tent-like shape of gers enables their detection through very-high-resolution satellite imagery. We develop a computer vision algorithm to detect gers in Ulaanbaatar, the capital of Mongolia, utilizing satellite images collected from 2015 to 2025. Results reveal that ger settlements have relocated towards the capital’s peripheral areas. The predicted ger household ratio based on our results exhibits a significant correlation (r = 0.85) with the World Bank’s district-level poverty data. Our nationwide extrapolation suggests that housing improvements in informal settlements have fallen short of official projections by an average of 2.8% since the COVID-19 pandemic. We discuss the potential of machine learning on satellite imagery in providing insights into urbanization patterns and sustainable development.

Mongolia is among the countries undergoing rapid urbanization, and its temporary nomadic dwellings-known as Ger-have expanded into urban areas. Ger settlements in cities are increasingly recognized as slums by their socio-economic deprivation. The distinctive circular, tent-like shape of gers enables their detection through very-high-resolution satellite imagery. We develop a computer vision algorithm to detect gers in Ulaanbaatar, the capital of Mongolia, utilizing satellite images collected from 2015 to 2023. Results reveal that ger settlements have been displaced towards the capital's peripheral areas. The predicted slum ratio based on our results exhibits a significant correlation (r = 0.84) with the World Bank's district-level poverty data. Our nationwide extrapolation suggests that slums may continue to take up one-fifth of the population after the COVID-19 pandemic, contrary to other official predictions that anticipated a decline. We discuss the potential of machine learning on satellite imagery in providing insights into urbanization patterns and monitoring the Sustainable Development Goals.

Consistent and timely assessment of climate risks is crucial for planning disaster mitigation and adaptation to climate change at the local community level. This article presents an automatized method for monitoring climate risks with machine learning on satellite imagery, specially targeting riverine and coastal floods. Our research demonstrates that disaster-related risk measurement becomes more comprehensive and multi-faceted by including the following components: hazards, exposure, and vulnerability. Our model first maps hazard-related risks with geo-spatial data, then extends the model to incorporate exposure and vulnerability. In doing so, we adopt a clustering-based supervised algorithm to sort satellite images to produce the climate risk scores at a grid-level. The developed model was tested over multiple ground-truth datasets on flood risks in the region of Jakarta, Indonesia. Results confirm that our model can assess climate risks in a granular scale and further capture potential risks in the marginalized areas (e.g., slums) that were previously hard to predict. We discuss how computational methods like ours can support humanitarian projects for developing countries.

Urban slums in the Global South have a formidable challenge of mismanaged waste. Behind this challenge lies urban politics, creating disproportionate exposure to waste in marginalized settlements. This paper articulates an urban political ecology of uneven waste accumulation in slums with a case study of Jakarta, Indonesia. I apply a mixed-methods approach, by integrating a spatial regression model with a critical qualitative analysis, to draw the connection between the slums’ waste crisis and neoliberal waste infrastructure. Since the mid-2010s, Jakarta’s waste governance has shifted from a conventional collect-transport-dispose model to a circular economy model operating under a technocratic and neoliberal economy. Under the shifted governance, the informal sector has been strengthened through the integration of high-tech infrastructure to reduce waste and extract new profits from materials. However, the sector has been unevenly strengthened across the city, creating two flows of waste accumulation activities. First, the strengthening has been accelerated in low-income residential areas by introducing financial incentives for the informal sector to make up the shortfall created by public services. Consequently, the informal sector in slums—with its limited focus on recyclable materials—has led to the accumulation of mismanaged non-recyclable waste in unregulated dumpsites of slums. On the other hand, public services are relatively prevalent in capital-intensive areas to promote the economic growth. However, due to its compact land-uses in those areas, the services utilize the unregulated dumpsites in slums as semi-permanent landfills for storing the collected waste.

Machine learning approaches using satellite imagery are providing accessible ways to infer socioeconomic measures without visiting a region. However, many algorithms require integration of ground-truth data, while regional data are scarce or even absent in many countries. Here we present our human-machine collaborative model which predicts grid-level economic development using publicly available satellite imagery and lightweight subjective ranking annotation without any ground data. We applied the model to North Korea and produced fine-grained predictions of economic development for the nation where data is not readily available. Our model suggests substantial development in the country’s capital and areas with state-led development projects in recent years. We showed the broad applicability of our model by examining five of the least developed countries in Asia, covering 400,000 grids. Our method can both yield highly granular economic information on hard-to-visit and low-resource regions and can potentially guide sustainable development programs.

High-resolution daytime satellite imagery has become a promising source to study economic activities. These images display detailed terrain over large areas and allow zooming into smaller neighborhoods. Existing methods, however, have utilized images only in a single-level geographical unit. This research presents a deep learning model to predict economic indicators via aggregating traits observed from multiple levels of geographical units. The model first measures hyperlocal economy over small communities via ordinal regression. The next step extracts district-level features by summarizing interconnection among hyperlocal economies. In the final step, the model estimates economic indicators of districts via aggregating the hyperlocal and district information. Our new multi-level learning model substantially outperforms strong baselines in predicting key indicators such as population, purchasing power, and energy consumption. The model is also robust against data shortage; the trained features from one country can generalize to other countries when evaluated with data gathered from Malaysia, the Philippines, Thailand, and Vietnam. We discuss the multi-level model's implications for measuring inequality, which is the essential first step in policy and social science research on inequality and poverty.

As coastal communities across the Global South confront the multiple challenges of climate change, overfishing, poverty and other socio-environmental pressures, there is an increasing need to understand diverse coastal governance responses and livelihood trajectories from a comparative perspective. This paper presents a holistic investigation of the pressures coastal communities face in four countries and examines possible meeting points between bottom-up initiatives and top-down policies. We compare the experiences of eight fishing areas in Ghana, Tanzania, Thailand and the Philippines and ask how small-scale fishing communities perceive overfishing and other socio-environmental pressures; what factors determine the success and failure of coastal governance initiatives; and how different initiatives can be made congruent to improve coastal, rural development outcomes. Results from an extensive survey of 835 fisherfolk and semi-structured interviews with 196 key informants show that overfishing remains a significant driver of livelihood trajectories in the communities and that fisherfolk respond through informal mechanisms of collective action. Drawing from these diverse experiences, we propose viewing coastal livelihood trajectories through the integrated dimensions of socio-environmental relationships and coastal governance options and discuss implications that address institutional scalar flexibility, illegal fishing, and persistent marginalisation.

Urban green space is thought to contribute to citizen happiness by promoting physical and mental health. Nevertheless, how urban green space and happiness are related across many countries with different socioeconomic conditions has not been explored. By measuring the urban green space score (UGS) from high-resolution satellite imagery of 90 global cities covering 179,168 km 2 and 230 million people in 60 developed countries, we find that the amount of urban green space and GDP are correlated with a nation’s happiness level. More specifically, urban green space and GDP are each individually associated with happiness. Yet, only urban green space is related to happiness in the 30 wealthiest countries, whereas GDP alone can explain happiness in the subsequent 30 countries in terms of wealth. We further show that the relationship between urban green space and happiness is mediated by social support and that GDP moderates this relationship. These findings corroborate the importance of maintaining urban green space as a place for social cohesion to support people’s happiness.

Adaptation and coping have been frequently compared. However, their relationship is still in dispute. So far, three approaches have been suggested: interchangeable, distinct and interrelated. We argue that the third is the most useful as it provides insights into how long-term adaptation can be achieved by a series of short-term coping mechanisms. Within this focus, we interpret adaptation in a novel way: as a complex cumulative result based on the interaction between multiple coping mechanisms and vulnerability dynamics. As such we reorient Smit et al.’s work on “cumulative adaptation”. Our empirical case is slum households affected by floods from the Mekong River in Phnom Penh, Cambodia. The inquiry is based on 119 surveys and 25 semi-structured interviews in nine slum communities. The results capture new trajectories of adaptation (or maladaptation), livelihoods and local collective action. The article explores implications for local development in slum communities in the global South.

How can we teach machines to quantify economic activities from satellite images? In this paper, we share the research progress in answering the question. We document what we have learned so far – characteristics of geospatial data including satellite images and recent developments in computer vision and image processing. We then identify some challenges in adopting the machine learning techniques to address the question. We present two of our proposed deep learning models that address some of the challenges: the first model predicts economic indicators from a satellite image by resolving the mismatch in data representation, and the second model learns to score the level of economic development of a satellite image even without ground-truth data. We also talk about our future research agenda to improve the models and to apply them for economic research and policymaking in practice.

Reliable and timely measurements of economic activities are fundamental for understanding economic development and designing government policies. However, many developing countries still lack reliable data. In this paper, we introduce a novel approach for measuring economic development from high-resolution satellite images in the absence of ground truth statistics. Our method consists of three steps. First, we run a clustering algorithm on satellite images that distinguishes artifacts from nature (siCluster). Second, we generate a partial order graph of the identified clusters based on the level of economic development, either by human guidance or by low-resolution statistics (siPog). Third, we use a CNN-based sorter that assigns differentiable scores to each satellite grid based on the relative ranks of clusters (siScore). The novelty of our method is that we break down a computationally hard problem into sub-tasks, which involves a human-in-the-loop solution. With the combination of unsupervised learning and the partial orders of dozens of urban vs. rural clusters, our method can estimate the economic development scores of over 10,000 satellite grids consistently with other baseline development proxies (Spearman correlation of 0.851). This efficient method is interpretable and robust; we demonstrate how to apply our method to both developed (e.g., South Korea) and developing economies (e.g., Vietnam and Malawi).

Previous research has reported a connection between urban green space and public health that ultimately contributes to happiness. Existing studies have mainly investigated green space over small areas. This paper revisits this significant correlation by examining the relationship between country-level happiness and the amount of urban green space as measured systematically from satellite images. Based on 2018 and 2013 data from 30 developed countries, we found that there is a correlation between urban green space and happiness, and this relationship becomes stronger among countries with higher GDP. We also found that the relationship between happiness and green space has grown stronger over time.

Reliable and timely measurement of economic activities is fundamental for understanding economic development and for delivering humanitarian aid and disaster relief where needed. However, many developing countries still lack reliable data. This paper introduces a novel approach for measuring economic development from high-resolution satellite images in the absence of ground truth statistics. Our method’s novelty is at breaking down a computationally challenging problem into sub-tasks, which involves a human-in-the-loop solution. With the combination of unsupervised learning and the partial orders of dozens of urban versus rural clusters, our method can estimate the economic development scores of over 10,000 satellite grids with less human labor than other baseline approaches (Spearman correlation of 0.851). We demonstrate how to apply our method to both developed and developing economies.