No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Combining satellite imagery and machine learning to predict poverty
Abstract
Measuring consumption and wealth remotely
Nighttime lighting is a rough proxy for economic wealth, and nighttime maps of the world show that many developing countries are sparsely illuminated. Jean et al. combined nighttime maps with high-resolution daytime satellite images (see the Perspective by Blumenstock). With a bit of machine-learning wizardry, the combined images can be converted into accurate estimates of household consumption and assets, both of which are hard to measure in poorer countries. Furthermore, the night- and day-time data are publicly available and nonproprietary.
Science , this issue p. 790 ; see also p. 753
... Second, the current SDG index dataset for U.S. cities is meant for a single year and only focuses on the city level, which hinders monitoring of multi-scale and multi-year SDG progress 9 . Alternatively, built upon the rapid development of remote sensing and deep learning techniques, satellite imagery showing nearly real-time and bird's-eye view information in cities has been broadly investigated as a data source for SDG monitoring [10][11][12][13][14][15][16][17] . Therefore, monitoring SDGs in cities with satellite imagery is of great significance in promoting sustainable urban development. ...
... Satellite imagery provides a near real-time bird's-eye view of the earth's surface. Combined with machine learning techniques, satellite imagery has been widely used in predicting socioeconomic status, especially in urban research, which includes poverty/asset prediction 11,14,17 , urban pattern mining 15 , commercial activity prediction 16,35 , and population prediction 12,36 . Inspired by the interpretable feature generation from satellite imagery 14 , we provide satellite imagery visual attributes in our dataset to promote the Content courtesy of Springer Nature, terms of use apply. ...
... Specifically, GBDT has a prediction performance of R 2 reaching about 0.155 and 0.338 for median household income and POI density, respectively. These results are consistent with the findings in previous research 11,14,16,82 that socioeconomic status can be inferred from satellite imagery, confirming the validity of the provided dataset and demonstrating the potential to monitor SDGs from satellite imagery. While the education and health indicators are predicted with low precision, which encourages dataset users for future enhancement. ...
Cities play an important role in achieving sustainable development goals (SDGs) to promote economic growth and meet social needs. Especially satellite imagery is a potential data source for studying sustainable urban development. However, a comprehensive dataset in the United States (U.S.) covering multiple cities, multiple years, multiple scales, and multiple indicators for SDG monitoring is lacking. To support the research on SDGs in U.S. cities, we develop a satellite imagery dataset using deep learning models for five SDGs containing 25 sustainable development indicators. The proposed dataset covers the 100 most populated U.S. cities and corresponding Census Block Groups from 2014 to 2023. Specifically, we collect satellite imagery and identify objects with state-of-the-art object detection and semantic segmentation models to observe cities’ bird’s-eye view. We further gather population, nighttime light, survey, and built environment data to depict SDGs regarding poverty, health, education, inequality, and living environment. We anticipate the dataset to help urban policymakers and researchers to advance SDGs-related studies, especially applying satellite imagery to monitor long-term and multi-scale SDGs in cities.
... Los mapas de pobreza son proyectos informáticos creados por organizaciones como la Universidad de Stanford y el Qatar Computing Research Institute (QCRI), con apoyo del Banco Mundial, con el objetivo de medir y mapear de manera más precisa la pobreza para generar datos que puedan ser utilizados por los tomadores de decisiones. Para ello se utiliza principalmente información proveniente de los teléfonos celulares como el consumo de datos, la actividad en redes sociales y las transacciones financieras, en combinación con imágenes satelitales de los usos de tierra, asentamientos urbanos y sus materiales de construcción, automóviles, luces nocturnas, infraestructura urbana como caminos y carretera, etc. (Blumenstock et al., 2015;Jean et al., 2016). Esta información es procesada y relacionada mediante algoritmos de IA y aprendizaje automático que da como resultado un conjunto de indicadores socioeconómicos georreferenciados (Kellenberger et al., 2021;Xie et al., 2016). ...
... Esta información es procesada y relacionada mediante algoritmos de IA y aprendizaje automático que da como resultado un conjunto de indicadores socioeconómicos georreferenciados (Kellenberger et al., 2021;Xie et al., 2016). Estos datos son utilizados por los tomadores de decisiones, principalmente de países en desarrollo como Nigeria, Uganda, Tanzania, Malawi, Ruanda, Pakistán y Filipinas, para mejorar la formulación de las políticas públicas diseñadas para combatir la pobreza, así como para su monitoreo y evaluación (Agyemang et al., 2023;Jean et al., 2016;Tingzon et al., 2019). ...
... La principal ventaja de los mapas de pobreza es que se presentan como una alternativa más constante, precisa y de menor costo para estimar, medir y evaluar la pobreza en comparación con los métodos tradicionales que utilizan encuestas y censos que son levantados generalmente cada 5 o 10 años, los cuales resultan ser más costosos y muchas veces no tienen un nivel alto de representatividad (Agyemang et al., 2023). Adicionalmente, en países donde los censos y las encuestas son limitados o inexistentes, los mapas de pobreza se presentan como la mejor alternativa para medir y evaluar la pobreza (Blumenstock, et al., 2015;Jean et al., 2016). ...
En la última década se observa un creciente uso de la Inteligencia Artificial (IA) en la administración pública; sin embargo, el estudio científico del tema es relativamente incipiente pues las afirmaciones sobre sus ventajas y desventajas se basan en suposiciones y predicciones de los investigadores y carecen de suficiente evidencia empírica. El objetivo de este trabajo es analizar las ventajas y desventajas del uso de la IA en el ciclo de las políticas públicas para contribuir a la solución de este vacío. Para ello se realiza un estudio comparativo de ocho casos a nivel internacional. El análisis muestra que la principal ventaja del uso de la IA es que permite procesar y analizar gran cantidad y diversidad de información de forma inmediata para automatizar diversos procesos dentro del ciclo de las políticas públicas; sin embargo, existen desventajas como la exclusión, sesgos en las estimaciones, falta de privacidad y poca transparencia.
... One notable advancement was the utilisation of nighttime light (NTL) data, initially showing promise due to its strong correlation with traditional measures of economic growth, as demonstrated by (Keola et al., 2015). However, its application has faced limitations, particularly when striving for finer resolutions or in economically poor regions (Blumenstock, 2016;Jean et al., 2016a). ...
... Furthermore, other notable approaches include the utilisation of high-resolution daytime satellite data (Hall et al., 2023;Jean et al., 2016a) mobile phone metadata (Aiken et al., 2023;Blumenstock, 2016), internet search history, and social media activities (Fatehkia et al., 2020). Researchers have also explored various combinations of these data sources (Pokhriyal and Jacques, 2017). ...
... Machine learning approaches can also incorporate diverse data sources, including satellite imagery (Puttanapong et al., 2022), household surveys, and socioeconomic indicators, to enhance the predictive power. For example, a study conducted in five African countries utilised survey data and RS satellite information to demonstrate the effectiveness of training a convolutional neural network (Jean et al., 2016a). The network successfully identified visual characteristics in images that could account for approximately 75% of the differences observed in economic outcomes at the local level. ...
Energy poverty affects billions worldwide, including people in developed and developing countries. Identifying those living in energy poverty and implementing successful solutions require timely and detailed survey data, which can be costly, time-consuming, and difficult to obtain, particularly in rural areas. Through machine learning, this study investigates the possibility of identifying vulnerable households by combining satellite remote sensing with socioeconomic survey data in the UK. In doing so, this research develops a machine learning-based approach to predicting energy poverty in the UK using the low income low energy efficiency (LILEE) indicator derived from a combination of remote sensing and socioeconomic data. Data on energy consumption, building characteristics, household income, and other relevant variables at the local authority level are fused with geospatial satellite imagery. The findings indicate that a machine learning algorithm incorporating geographical and environmental information can predict approximately 83% of districts with significant energy poverty. This study contributes to the expanding body of research on energy poverty prediction and can help shape policy and decision-making for energy efficiency and social fairness in the UK and worldwide.
... Many have used the transfer learning approach on high-resolution images and sometimes there have been attempts at interpreting the model, i.e. exploring what parts of the image that the model reacts to. Jean et al. [2] [3] show that the CNN filters react to structures in the landscape, with more reactions in urban areas than in rural areas. An analysis of if and how these filters contribute to the Wealth Index predicted value is not provided. ...
... Tang et al. [14] use the low-resolution normalized difference vegetation index (NDVI) as input. The CNN model is trained using the same transfer learning trick as used by Jean et al. [3]. They then use random forest for estimating the Wealth Index from the CNN features, concluding that NDVI-based predictions are as good as the predictions achieved with daytime satellite images. ...
... All this is in agreement with previously published findings. [3]. In 2017, Head et al. replicated some of their experiments, albeit not for Tanzania, and concluded that they got pretty much the same values [16], with variations in the second decimal. ...
... Survey data-based models, such as those published by Alsharkawi et al., offer significant interpretability but may sacrifice accuracy (6). In contrast, models like Engstrom et al. and Jean et al.'s, which use satellite imagery and Rolf et al.'s, which indirectly employs satellite imagery, achieve high accuracy but struggle with interpretability, thus hindering our understanding of poverty's causes (5,7,2). Alsharkawi et al.'s gradient boosting model also suffers from limited interpretability despite high accuracy. ...
... The high dimensionality and potential for nonlinear relationships between the 40 survey features seem to favor this more complex neural network model over simpler counterparts, such as the random forest and softmax classification models. These results align with existing literature that shows the proficiency of neural network models in managing high-dimensional, complex data (7). ...
... Additionally, while our study adds to the existing knowledge of using machine learning for poverty prediction, it stands out by exploring the trade-off between model complexity and interpretability in this specific context. Studies such as (7) have highlighted the importance of this balance, but few Taken together, our study demonstrated the effective use of machine learning models in predicting poverty levels and highlighted the challenges and potential improvements required for more accurate classifications. The findings offer insights not only on the technological front but also potentially inform policy decisions by identifying key areas of focus. ...
Addressing global poverty requires understanding of the most poverty-stricken regions. One approach towards achieving this is through poverty prediction, a task that entails classifying poverty levels of households using available data. While machine learning (ML) has been applied in numerous fields with considerable success, its application in poverty prediction using exclusively household survey data is yet to be thoroughly explored. Household survey data offers a detailed view into the living conditions, lifestyle, and socio-economic factors affecting households. Hence, we aim explore the use of this data type in predicting poverty levels. Our study primarily focuses on three ML models: softmax classification, random forest classification, and multilayer perceptron (MLP). We chose Cambodia for this study due to its unique socio-economic landscape and as a representative of developing nations struggling with poverty. This analysis will serve as the foundation for applying this approach to other nations. The analysis was based on a dataset consisting of 15,825 household samples and 1,873 features obtained through the Demographic and Health Surveys (DHS) program in Cambodia. The study's aim was to validate the effectiveness of ML in poverty prediction using household data and identify the best performing model among the selected three. We hypothesized that the MLP, due to its advanced neural network structure, would provide superior results compared to the softmax classification and random forest models. As anticipated, the multilayer perceptron outperformed the other models, achieving an accuracy of 87% against the 81% and 80% accuracy of the random forest and softmax classification models respectively.
... Indeed, the feature extraction property of convolutional layers, and the powerful training methods of the CNN, are reasons for the high performance of CNNs on image classification. CNNs have been widely used in computer vision tasks [16,33], object detection [35], bioinformatics [54], economics [33], and natural language processing [75]. ...
... Indeed, the feature extraction property of convolutional layers, and the powerful training methods of the CNN, are reasons for the high performance of CNNs on image classification. CNNs have been widely used in computer vision tasks [16,33], object detection [35], bioinformatics [54], economics [33], and natural language processing [75]. ...
Deep Convolutional Neural Networks (CNNs) have been successfully used in different applications, including image recognition. Time series data, which are generated in many applications, such as tasks using sensor data, have different characteristics compared to image data, and accordingly, there is a need for specific CNN structures to address their processing. This paper proposes a new CNN for classifying time series data. It is proposed to have new intermediate outputs extracted from different hidden layers instead of having a single output to control weight adjustment in the hidden layers during training. Intermediate targets are used to act as labels for the intermediate outputs to improve the performance of the method. The intermediate targets are different from the main target. Additionally, the proposed method artificially increases the number of training instances using the original training samples and the intermediate targets. The proposed approach converts a classification task with original training samples to a new (but equivalent) classification task that contains two classes with a high number of training instances. The proposed CNN for Time Series classification, called CNN-TS, extracts features depending the distance of two time series. CNN-TS was evaluated on various benchmark time series datasets. The proposed CNN-TS achieved 5.1% higher overall accuracy compared to the CNN base method (without an intermediate layer). Additionally, CNN-TS achieved 21.1% higher average accuracy compared to classical machine-learning methods, i.e., linear SVM, RBF SVM, and RF. Moreover, CNN-TS was on average 8.43 times faster in training time compared to the ResNet method.
... For these reasons, poverty mapping is sparking interest in welfare economics and geography studies (Jean et al., 2016;Hall et al., 2023), as an essential methodology to support the investigation of the spatial distribution and regional characteristics of poverty. In addition, if performed at a granular level, it enables a geographical targeting, namely the area-driven allocation of resources for poverty alleviation (Elbers et al., 2007;Liu et al., 2017). ...
... Survey estimates are increasingly unreliable when considering finer spatial levels, translating into smaller and smaller sample sizes (Pratesi and Salvati, 2016). In addition, remote sensing (RS) or mobile phone usage data are recently employed in poverty mapping, especially in developing countries (Jean et al., 2016;Schmid et al., 2017). As opposed to survey data, such alternative sources are highly informative at fine spatial levels, while losing power when aggregated. ...
... GDP per capita, population density, and urbanization rate were calculated according to the administrative units of districts and counties. Nighttime light data reflect the dynamic spatial distribution of human activities and the level of social and economic development (Feng et al., 2020;Jean et al., 2016;Yang et al., 2021a). ...
Urban agglomerations are important spatial carriers for regional economic development, but sustained urban expansion has triggered a series of issues such as conflicts between "people and land" and "built-up areas and green spaces". However, many studies have rarely incorporated spatiotemporal patterns and driving mechanism into scenario simulation of urban growth, and has also overlooked subsequent discussions on the specific integration of simulation results with spatial planning. Considering the case of Min Delta Region, this study analyzes the spatiotemporal evolution of urban expansion based on multi-source data and explore the influential driving mechanism using Random forest algorithm. A combination of Markov and FLUS models was used to dynamically simulate future urban growth patterns. The results showed that 1) From 1995 to 2015, the urban expansion increased by 2.26 times, showing a trend of first accelerating and then slowing down. It was also discovered that there is an anisotropy toward expansion in the trajectory of urban growth. 2) With marginal increase and enclave increase, general urban expansion was observed to be changing from diffusion to agglomeration, and the urban expansion hotspots mainly located in the southeast coastal area. 3) Humanistic links, geographic position, social and economic considerations all played a role in the development of the Min Delta region, and can be summarized as the "four forces" driving mechanism model. 4) By 2035, the trend of continuous development with the Min Delta built-up area as the main body will be basically formed, and incorporating ecological restrictions can effectively restrain urban growth's encroachment into ecological spaces. Based on the simulation results, Min Delta region gradually showing a spatial structure evolution trend from the single core to the dual core mode, and then to the multi-center and networked mode. This study would serve as a multi-angle decision-making reference for regional spatial and ecological protection planning and urban growth management, and as a scientific foundation for high-quality development planning in the region.
... However, these data often fail to reveal the inner spatial differences of irregular administrative units, posing a challenge for understanding the complex interactions between human beings and the environment [1]. To overcome this limitation and obtain high-spatial-resolution socioeconomic data, spatialization studies have been developed to allocate such data from statistical units to regular grids [2][3][4]. Among the various socioeconomic parameters, gross domestic product (GDP) holds particular significance as the broadest measure of economic development and resource allocation on national and local scales [5]. ...
Accurate spatial distribution of gridded gross domestic product (GDP) data is crucial for revealing regional disparities within administrative units, thus facilitating a deeper understanding of regional economic dynamics, industrial distribution, and urbanization trends. The existing GDP spatial models often rely on prediction residuals for model evaluation or utilize residual distribution to improve the final accuracy, frequently overlooking the modifiable areal unit problem within residual distribution. This paper introduces a hybrid downscaling model that combines random forest and area-to-area kriging to map gridded GDP. Employing Thailand as a case study, GDP distribution maps were generated at a 1 km spatial resolution for the year 2015 and compared with five alternative downscaling methods and an existing GDP product. The results demonstrate that the proposed approach yields higher accuracy and greater precision in detailing GDP distribution, as evidenced by the smallest mean absolute error and root mean squared error values, which stand at USD 256.458 and 699.348 ten million, respectively. Among the four different sets of auxiliary variables considered, one consistently exhibited a higher prediction accuracy. This particular set of auxiliary variables integrated classification-based variables, illustrating the advantages of incorporating such integrated variables into modeling while accounting for classification characteristics.
... Our model performs comparatively well to state-of-theart machine learning methods that explained 56% to 75% of the variation in wealth using geospatial data at similar spatial resolutions (Chi et al., 2022;Jean et al., 2016;Yeh et al., 2020). These studies used between 1,400 to 66,000 samples and uninterpretable geospatial features extracted from satellite imagery using deep learning models. ...
Physical and economic access to food vary spatially. Methods to map that variability at high levels of spatial detail over large areas are scarce, even though suitable datasets and methods exist. Using open-access data for Ethiopia, we developed a method to map the disparities in physical and economic food access at 1-km resolution. We selected 25 access-related variables for 486 geo-located communities from the 2018 Ethiopian Living Standards Measurement Study to create a food access index (FAI). The index was based on a weighted summation of the 25 variables from a principal component analysis (PCA). We then extrapolated the FAI to the rest of Ethiopia using a generalized additive model (GAM) to produce a 1-km resolution FAI map and used that to describe the spatial variability of food access. Economic access had a heavier weight than physical access in the FAI reflecting the fact that proximity to food markets alone is insufficient if one cannot afford food. The GAM had an R ² of 0.57 and a normalized root mean square error of 22.2% which are comparable to measures of model performance in studies that provided micro-level estimates of relative wealth. Peri-urban areas, representing 67% of the population, had relatively low food access, suggesting that these areas should be a priority for infrastructure or economic intervention. The scarcity of detailed spatial information on food access may limit the effectiveness of targeted policymaking to improve food security. The methodology developed in this study uses widely available and carefully selected datasets and can contribute to more spatially detailed estimates of food access in other countries.
... As machine learning continues to advance, we see its rise in both the number of domains it is impacting and the overall complexity of the challenges it solves (Jean et al. 2016;Kube, Das, and Fowler 2019). As such, the training data that is fundamental to these advancements must similarly be able to keep up in both scope and complexity, requiring complex and domain-specific annotations to further this growth. ...
Video object tracking annotation tasks are a form of complex data labeling that is inherently tedious and time-consuming. Prior studies of these tasks focus primarily on quality of the provided data, leaving much to be learned about how the data was generated and the factors that influenced how it was generated. In this paper, we take steps toward this goal by examining how human annotators spend their time in the context of a video object tracking annotation task. We situate our study in the context of a standard vehicle tracking task with bounding box annotation. Within this setting, we study the role of task complexity by controlling two dimensions of task design -- label constraint and label granularity -- in conjunction with worker experience. Using telemetry and survey data collected from 40 full-time data annotators at a large technology corporation, we find that each dimension of task complexity uniquely affects how annotators spend their time not only during the task, but also before it begins. Furthermore, we find significant misalignment in how time-use was observed and how time-use was self-reported. We conclude by discussing the implications of our findings in the context of video object tracking and the need to better understand how productivity can be defined in data annotation.
... [18][19][20][21] It has also been applied in computational research and modeling of crystal structures. 22,23 Similar approaches were broadly used for a variety of applications including the search for exotic particles from the Large Hadron Collider data, 24 analysis of satellite images, 25 and cancer detection. 26 However, supervised ML methods explicitly rely on the availability of the labeled data. ...
Recent advances in scanning tunneling and transmission electron microscopies (STM and STEM) have allowed routine generation of large volumes of imaging data containing information on the structure and functionality of materials. The experimental data sets contain signatures of long-range phenomena such as physical order parameter fields, polarization, and strain gradients in STEM, or standing electronic waves and carrier-mediated exchange interactions in STM, all superimposed onto scanning system distortions and gradual changes of contrast due to drift and/or mis-tilt effects. Correspondingly, while the human eye can readily identify certain patterns in the images such as lattice periodicities, repeating structural elements, or microstructures, their automatic extraction and classification are highly non-trivial and universal pathways to accomplish such analyses are absent. We pose that the most distinctive elements of the patterns observed in STM and (S)TEM images are similarity and (almost-) periodicity, behaviors stemming directly from the parsimony of elementary atomic structures, superimposed on the gradual changes reflective of order parameter distributions. However, the discovery of these elements via global Fourier methods is non-trivial due to variability and lack of ideal discrete translation symmetry. To address this problem, we explore the shift-invariant variational autoencoders (shift-VAE) that allow disentangling characteristic repeating features in the images, their variations, and shifts that inevitably occur when randomly sampling the image space. Shift-VAEs balance the uncertainty in the position of the object of interest with the uncertainty in shape reconstruction. This approach is illustrated for model 1D data, and further extended to synthetic and experimental STM and STEM 2D data. We further introduce an approach for training shift-VAEs that allows finding the latent variables that comport to known physical behavior. In this specific case, the condition is that the latent variable maps should be smooth on the length scale of the atomic lattice (as expected for physical order parameters), but other conditions can be imposed. The opportunities and limitations of the shift VAE analysis for pattern discovery are elucidated.
... Previous research leveraged real-time or near-real-time information for enhancing responses to natural disasters using diverse data sources such as satellite imagery and communication data for the rapid identification of affected areas and damage assessment (Bagrow et al., 2011;Bengtsson et al., 2011;Tomaszewski, 2014). Big data has also been proven instrumental in tracking the propagation of infectious diseases (Peak et al., 2018;Wesolowski et al., 2012) and in offering insights into economic disparities (Decuyper et al., 2014;Jean et al., 2016;Pokhriyal & Jacques, 2017). Furthermore, big data analytics significantly contributed to informed decision-making in infrastructural development investments through the analysis of mobility patterns and urban data (Martinez-Cesena et al., 2015). ...
Bu çalışma, göç çalışmaları alanında gerçekleştirilen büyük veri araştırmalarında karşılaşılan etik zorluklara odaklanmakta ve verinin sorumlu bir şekilde kullanımı için izlenmesi gereken ilkeler hakkında bir diyaloğa katkı sunmayı amaçlamaktadır. Çalışmada göç araştırmacılarının çalışmalarını tasarlama, yürütme ve bulgularını yayma süreçlerinde karşılaştıkları zorluklara odaklanılmaktadır. Bu bağlamda, çalışmada araştırmacıların birey ve grup gizliliği ile güç dengesizliklerinin arttırımı ve yeni eşitsizliklerin yaratılmasına ilişkin etik sorumlukları olmalıdır. Çalışmada ayrıca göç araştırmacılarının büyük veri araştırmaları için etik standartların belirlenmesindeki kritik rolü vurgulanmaktadır.
... Cluster II has 13 references, the joint maximum of all the clusters, and Vinuesa et al. (2020) is the most cited reference in this cluster. Cluster III has ten references, with Jean et al. (2016) as the most cited reference. The focus of this paper was the effective use of satellite imagery and machine learning techniques to predict poverty in five different African countries. ...
Abstract
This paper reviews the application of AI & ML techniques in achieving the UN Sustainable
Development Goals, as documented in various studies during 2017–2022. A systematic
bibliometric review of a sample of 250 peer-reviewed journal articles selected from two
scientific databases, Scopus and Web of Science, was undertaken (i) to gauge the trend in
publications on the application of specific innovative technologies, especially AI and ML,
for achieving the SDGs; (ii) to analyze the blind spots of AI adversely affecting sustainability,
which are derived from the literature review and to examine the solutions offered in
the literature to counter the adverse effects of AI, and (iii) to gauge the future direction of
research. The highest number of studies originated from China, the USA, Spain, the UK,
and Australia. Evident collaborations between countries and universities are also discernible.
The study identified the journals, Sustainability, Remote Sensing, IEEE Access, and
Journal of Cleaner Production as core sources through Bradford’s law. The findings show
that AI holds promise, but there is overexuberance about its positive outcome. The study
shows a need to impose regulatory requirements and enforce regular verification to ensure
that AI remains a subject of constant scrutiny for trust, transparency, and adherence to universal
ethical standards for SDG achievement. The findings could also provide researchers
with a direction for integrating AI/ML in achieving the SDGs.
Keywords Artificial intelligence (AI) · Bibliometric coupling · Co-citation ·
Co-occurrence · Sustainable development goals (SDG)
... Cluster II has 13 references, the joint maximum of all the clusters, and Vinuesa et al. (2020) is the most cited reference in this cluster. Cluster III has ten references, with Jean et al. (2016) as the most cited reference. The focus of this paper was the effective use of satellite imagery and machine learning techniques to predict poverty in five different African countries. ...
This paper reviews the application of AI & ML techniques in achieving the UN Sustainable Development Goals, as documented in various studies during 2017–2022. A systematic bibliometric review of a sample of 250 peer-reviewed journal articles selected from two scientific databases, Scopus and Web of Science, was undertaken (i) to gauge the trend in publications on the application of specific innovative technologies, especially AI and ML, for achieving the SDGs; (ii) to analyze the blind spots of AI adversely affecting sustainability, which are derived from the literature review and to examine the solutions offered in the literature to counter the adverse effects of AI, and (iii) to gauge the future direction of research. The highest number of studies originated from China, the USA, Spain, the UK, and Australia. Evident collaborations between countries and universities are also discernible. The study identified the journals, Sustainability, Remote Sensing, IEEE Access, and Journal of Cleaner Production as core sources through Bradford’s law. The findings show that AI holds promise, but there is overexuberance about its positive outcome. The study shows a need to impose regulatory requirements and enforce regular verification to ensure that AI remains a subject of constant scrutiny for trust, transparency, and adherence to universal ethical standards for SDG achievement. The findings could also provide researchers with a direction for integrating AI/ML in achieving the SDGs.
... Most prominent are applications where GDP growth has been proxied by night light emissions (e.g. Hu and Yao, 2022;Jean et al., 2016;Mellander et al., 2015), as in this study. 3 For example, remote sensing data has been used to delineate economically strong regions (Florida et al., 2008;Taubenböck et al., 2017;Georg et al., 2018) or with the underlying aim of analyzing real regional GDP without measurement error (Gennaioli et al., 2014). ...
We present a novel approach to analyze the effects of EU cohesion policy on local economic activity. For all municipalities in the border area of the Czech Republic, Germany, and Poland, we collect project-level data on EU funding in the period between 2007 and 2013. Using night light emission data as a proxy for economic development, we show that receiving a higher amount of EU funding is associated with increased economic activity at the municipal level. Our paper demonstrates that remote sensing data can provide an effective way to model local economic development also in Europe, where comprehensive cross-border data are not available at such a spatially granular level.
... Work in multisystem dynamics bridges nexus approaches with advances in formal modeling techniques (7). Advances in machine learning can enable using remote sensing information in regions where on-the-ground data are sparser (27). Advances in defining purpose. ...
This Perspective evaluates recent progress in modeling nature–society systems to inform sustainable development. We argue that recent work has begun to address longstanding and often-cited challenges in bringing modeling to bear on problems of sustainable development. For each of four stages of modeling practice—defining purpose, selecting components, analyzing interactions, and assessing interventions—we highlight examples of dynamical modeling methods and advances in their application that have improved understanding and begun to inform action. Because many of these methods and associated advances have focused on particular sectors and places, their potential to inform key open questions in the field of sustainability science is often underappreciated. We discuss how application of such methods helps researchers interested in harnessing insights into specific sectors and locations to address human well-being, focus on sustainability-relevant timescales, and attend to power differentials among actors. In parallel, application of these modeling methods is helping to advance theory of nature–society systems by enhancing the uptake and utility of frameworks, clarifying key concepts through more rigorous definitions, and informing development of archetypes that can assist hypothesis development and testing. We conclude by suggesting ways to further leverage emerging modeling methods in the context of sustainability science.
... For example, the complete statistical data of 2023 will be available only one or two years later. In addition, there exist some inconsistencies caused by the change in administrative units [26][27][28][29][30][31]. ...
Poverty is a social issue of global concern. Although socioeconomic indicators can easily reflect poverty status, the coarse statistical scales and poor timeliness have limited their applications. While spatial big data with reasonable timeliness, easy access, and wide coverage can overcome such limitations, the integration of high-resolution nighttime light and spatial big data for assessing relative poverty is still limited. More importantly, few studies have provided poverty assessment results at a grid scale. Therefore, this study takes the Pearl River Delta, where there is a large disparity between the rich and the poor, as an example. We integrated Luojia 1-01, points of interest, and housing prices to construct a big data poverty index (BDPI). To evaluate the performance of the BDPI, we compared this new index with the traditional multidimensional poverty index (MPI), which builds upon socioeconomic indicators. The results show that the impoverished counties identified by the BDPI are highly similar to those identified by the MPI. In addition, both the BDPI and MPI gradually decrease from the center to the fringe of the study area. These two methods indicate that impoverished counties were mainly distributed in ZhaoQing, JiangMen and HuiZhou Cities, while there were also several impoverished parts in rapidly developing cities, such as CongHua and HuaDu Counties in GuangZhou City. The difference between the two poverty assessment results suggests that the MPI can effectively reveal the poverty status in old urban areas with convenient but obsolete infrastructures, whereas the BDPI is suitable for emerging-development areas that are rapidly developing but still lagging behind. Although BDPI and MPI share similar calculation procedures, there are substantial differences in the meaning and suitability of the methodology. Therefore, in areas lacking accurate socioeconomic statistics, the BDPI can effectively replace the MPI to achieve timely and fine-scale poverty assessment. Our proposed method could provide a reliable reference for formulating targeted poverty-alleviation policies.
... Mobile phone usage data and machine learning methods has been used to predict spatial distribution of wealth and poverty (Blumenstock et al. 2016;Steele et al. 2017). Satellite imagery has been used to successfully estimate both average household consumption expenditure and asset wealth as measured at the cluster level (Jean et al. 2016). While these "modern methods" and data has had success with aggregate wealth and poverty prediction, to be best of knowledge the feasibility of using these methods and data for understanding household seasonal consumption behaviour is yet to be explored. ...
There is little discussion, let alone systemic evidence, on how droughts affect seasonal/intra-annual consumption within a drought year and subsequent years. This paper reviews the evidence on whether households can smooth food and non-food consumption and how factors constraining consumption smoothing are experienced in the context of drought, focusing primarily on empirical evidence from Sub-Saharan Africa. Households are unable to smooth consumption for various reasons including income seasonality, imperfect credit markets, incomplete insurance, storage constraints, and liquidity constraints. They reduce non-food consumption to partly smooth food consumption during lean season, and there is an overwhelming amount of evidence that households reduce food consumption to cope with drought. Droughts not only exacerbate seasonality in income and price through crop losses and changes in prices and wages, which are triggered by the initial losses, but also extend the lean season. Access to formal or informal credit markets is further constrained for covariate shocks such as drought.
This paper recommends further research into the following areas: i) What is the timing of any reduction in food and non-food consumption in response to drought? ii) How do the patterns and timing of consumption reduction vary by household? What are the characteristics that that make some households more vulnerable to drought? iii) Which type of households can afford to smooth consumption by destabilising assets instead of dangerously reducing consumption and smoothing assets? iv) Does access to internet and telecommunications reduce the incompleteness of informal insurance by facilitating information sharing and reducing transaction costs. The empirics of understanding drought's effects on seasonal consumption are limited by costly and infrequent data. “Big data” and “modern methods of data collection” could be explored and leveraged-either on their own or linked with traditional surveys- to obtain more frequent and disaggregated observations of household consumption.
... No Poverty Predicting poverty regions [31,32], optimizing social security payments [33], improving microfinance services [34,35] ...
Artificial intelligence (AI) and deep learning (DL) have shown tremendous potential in driving sustainability across various sectors. This paper reviews recent advancements in AI and DL and explores their applications in achieving sustainable development goals (SDGs), renewable energy, environmental health, and smart building energy management. AI has the potential to contribute to 134 of the 169 targets across all SDGs, but the rapid development of these technologies necessitates comprehensive regulatory oversight to ensure transparency, safety, and ethical standards. In the renewable energy sector, AI and DL have been effectively utilized in optimizing energy management, fault detection, and power grid stability. They have also demonstrated promise in enhancing waste management and predictive analysis in photovoltaic power plants. In the field of environmental health, the integration of AI and DL has facilitated the analysis of complex spatial data, improving exposure modeling and disease prediction. However, challenges such as the explainability and transparency of AI and DL models, the scalability and high dimensionality of data, the integration with next-generation wireless networks, and ethics and privacy concerns need to be addressed. Future research should focus on enhancing the explainability and transparency of AI and DL models, developing scalable algorithms for processing large datasets, exploring the integration of AI with next-generation wireless networks, and addressing ethical and privacy considerations. Additionally, improving the energy efficiency of AI and DL models is crucial to ensure the sustainable use of these technologies. By addressing these challenges and fostering responsible and innovative use, AI and DL can significantly contribute to a more sustainable future.
... Machine learning techniques hold promise for predicting water quality using data from household surveys given the wealth of information on household socio-economic conditions and a range of geospatial information from global datasets that can be integrated using the household survey cluster locations (where available). Prior research drawing on data from the World Bank's Living Standard Measurement Study (LSMS) surveys has applied machine learning to a wide range of topics covered in these surveys including poverty 6 , housing rental value 7 , food security 8 , crop type mapping 9 , crop yield 10 , and fertilizer pricing 11 . Recent studies have investigated the ability to predict microbial drinking water quality on smaller scales for specific water source types including piped water in the Democratic Republic of Congo 12 and groundwater in Uganda and Bangladesh 13 . ...
Monitoring access to safely managed drinking water services requires information on water quality. An increasing number of countries have integrated water quality testing in household surveys however it is not anticipated that such tests will be included in all future surveys. Using water testing data from the 2016 Ethiopia Socio-Economic Survey (ESS) we developed predictive models to identify households using contaminated (≥1 E. coli per 100 mL) drinking water sources based on common machine learning classification algorithms. These models were then applied to the 2013–2014 and 2018–2019 waves of the ESS that did not include water testing. The highest performing model achieved good accuracy (88.5%; 95% CI 86.3%, 90.6%) and discrimination (AUC 0.91; 95% CI 0.89, 0.94). The use of demographic, socioeconomic, and geospatial variables provided comparable results to that of the full features model whereas a model based exclusively on water source type performed poorly. Drinking water quality at the point of collection can be predicted from demographic, socioeconomic, and geospatial variables that are often available in household surveys.
Machine learning (ML) is increasingly being used in official statistics with a range of different applications. The main focus of ML models is to accurately predict attributes of new, unlabeled cases whereas the focus of classical statistical models is to describe the relations between independent and dependent variables. There is already a lot of experience in the sound use of classical statistical models in official statistics, but for ML models this is still under development. Recent discussions concerning the quality aspects of using ML in official statistics have concentrated on its implications for existing quality frameworks. We are in favor of the use of ML in official statistics, but the main question remains as to what factors need to be considered when using ML models in official statistics. As a means of raising awareness regarding these factors, we pose ten propositions regarding the (sensible) use of ML in official statistics.
Purpose
This study aims to examine the current state of academic research pertaining to the role played by artificial intelligence (AI) in the achievement of a critical sustainable development goal (SDG) – poverty alleviation and describe the field’s development by identifying themes, trends, roadblocks and promising areas for the future.
Design/methodology/approach
The authors analysed a corpus of 253 studies collected from the Scopus database to examine the current state of the academic literature using bibliometric methods.
Findings
This paper identifies and analyses key trends in the evolution of this domain. Further, the paper distils the extant literature to unpack the intermediary mechanisms through which AI and related technologies help tackle the critical global issue of poverty.
Research limitations/implications
The corpus of literature used for the analysis is limited to English language studies from the Scopus database. The paper contributes to the extant research on AI for social good, and more broadly to the research on the value of emerging technologies such as AI.
Practical implications
Policymakers and government agencies will get an understanding of how technological interventions such as AI can help achieve critical SDGs such as poverty alleviation (SDG-1).
Social implications
The primary focus of this paper is on the role of AI-related technological interventions to achieve a significant social objective – poverty alleviation.
Originality/value
To the best of the authors’ knowledge, this is the first study to conduct a comprehensive bibliometric analysis of a critical research domain such as AI and poverty alleviation.
The chapter provides an overview of the role of information technology in development, examining its functions, applications, and impacts. This chapter reviews the ways in which information technology has “digitized development,” discussing the rollout of information technology by the private sector, as well as development-driven interventions that use digital technologies. It provides a framework for thinking about the role of digital in development in five key areas: health, agriculture, education, financial services, and social protection programs.
This article introduces machine learning techniques to identify politically connected firms. By assembling information from publicly available sources and the Orbis company database, we constructed a novel firm population dataset from Czechia in which various forms of political connections can be determined. The data about firms' connections are unique and comprehensive. They include political donations by the firm, having members of managerial boards who donated to a political party, and having members of boards who ran for political office. The results indicate that over 85% of firms with political connections can be accurately identified by the proposed algorithms. The model obtains this high accuracy by using only firm‐level financial and industry indicators that are widely available in most countries. These findings suggest that machine learning algorithms could be used by public institutions to improve the identification of politically connected firms with potentially large conflicts of interest.
Thunderstorms can be a large source of disruption for European air-traffic management causing a chaotic state of operation within the airspace system. In current practice, air-traffic managers are provided with imprecise forecasts which limit their ability to plan strategically. As a result, weather mitigation is performed using tactical measures with a time horizon of three hours. Increasing the lead time of thunderstorm predictions to the day before operations could help air-traffic managers plan around weather and improve the efficiency of air-traffic-management operations. Emerging techniques based on machine learning have provided promising results, partly attributed to reduced human bias and improved capacity in predicting thunderstorms purely from numerical weather prediction data. In this paper, we expand on our previous work on thunderstorm forecasting, by applying convolutional neural networks (CNNs) to exploit the spatial characteristics embedded in the weather data. The learning task of predicting convection is formulated as a binary-classification problem based on satellite data. The performance of multiple CNN-based architectures, including a fully-convolutional neural network (FCN), a CNN-based encoder–decoder, a U-Net, and a pyramid-scene parsing network (PSPNet) are compared against a multi-layer-perceptron (MLP) network. Our work indicates that CNN-based architectures improve the performance of point-prediction models, with a fully-convolutional neural-network architecture having the best performance. Results show that CNN-based architectures can be used to increase the prediction lead time of thunderstorms. Lastly, a case study illustrating the applications of convection-prediction models in an air-traffic-management setting is presented.
This article explores the potential ways in which artificial intelligence (AI) can support veganism, a lifestyle that aims to promote the protection of animals and also avoids the consumption of animal products for environmental and health reasons. The first part of the article discusses the technical requirements for utilizing AI technologies in the mentioned field. The second part provides an overview of potential use cases, including facilitating consumer change with the help of AI, technologically augmenting undercover investigations in factory farms, raising the efficiency of nongovernment organizations promoting plant-based lifestyles, and so forth. The article acknowledges that the deployment of AI should not happen in a “solutionist” manner, meaning to always consider nontechnical means for achieving desired outcomes. However, it is important for organizations promoting veganism to realize the potential of modern data-driven tools and to merge and share their data to reach common goals.
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.
Scaling up case-based knowledge to understand human-environmental interactions is key to addressing the sustainability challenges we face in the Anthropocene. Mapping social-ecological system (SES) archetypes has emerged as a promising tool in identifying these interactions in a spatially explicit manner and developing transferrable sustainability solutions based on context-specific challenges. However, state-of-the-art SES archetype mapping is mainly bound to administrative borders. Lack of spatial details in SES maps does not always capture the panorama of SES configurations and hidden mechanisms. Empowered by multiple open-access geospatial big data streams, this study mapped two-tier nested SES archetypes at 1 km fine resolution in the Beijing-Tianjin-Hebei (BTH) urban agglomeration with high validation accuracy. Global SES archetypes derived from top-down rule-based classification revealed broad social-ecological configurations and social-ecological gradients across SESs. Meanwhile, regional SES sub-archetypes generated by bottom-up hierarchical clustering presented detailed SES characteristics that helped to identify location-specific sustainability challenges. Additionally, by scrutinizing the relationships among SES variables, we identified empirical evidence supporting the existence of four archetypical human-environmental interactions in an urban agglomeration area, manifesting as telecoupling between human and environment, human dependence on the environment, human adaption to the environment, and human modification and conservation of the environment. Generally, fine-resolution SES archetypes demonstrate great potential in supporting the recognition of local needs and gaps in existing policies, and further promoting sustainable landscape management in the BTH region and similar regions.
Data is one of the most crucial issues in achieving sustainable development in rural areas. Hitherto, emerging
techniques, like remote sensing and machine learning, provide a promising foundation aiming at the data acquisition,
calculation, and analysis, especially for vast, scattered, and full of territory heterogeneity villages. Specifically, the
methodology of the Computable Village establishes the imagery-figure relationship via deriving systematically the multisourced data of rural natural and human settlement subjects; besides describing various rural elements quantificationally
levering index calculating, data mining, and Artificial Intelligence methods, which could further unveil the spatial
discrepancy and dynamic tendency of rural development in the total territory, and thus provides a possibility for rural
researches striding from local to global. In this paper, a technical framework of the Computable Village is proposed
including: (1) the categories and features of multi-sourced rural data such as remote sensing; (2) the computing
approaches for rural data and the intelligent interpretation techniques for imagery-figure relationship; (3) the supports
from the computed results to rural sustainable developments. Finally, the current challenges as well the further research
directions are discussed and envisaged aiming at the Computable Village.
Modern cities meet many of the criteria for a comfortable and safe life for their citizens. However, the rapid researches and developments in the fields of information technology and telecommunications stimulates further modernization of urban environment in order to provide maximum possible conveniences to everyone - from a resident to an urban planner and manager. The work of the latter largely determines the comfort and safety of urban residents. In this regard, it is necessary that urban planners and managers can make decisions based on objective factual data, which needs to be easily and quickly obtainable without spending a lot of time on it. For this reason, there is a need to develop IT solutions aimed at the optimization of the work of planners and managers. In order to justify solutions for transforming the urban environment to make it more comfortable, safe and harmonious, sustainable and integrated development, methods for analyzing and assessing the current state and monitoring changes are needed. This paper considers the problem of collecting data on the state of urban infrastructure. The application of spatial data analysis to solve the problems of urban territories management and urban planning are investigated. A method for estimating the area of the city, occupied by the objects of different categories, is presented. Implementation of the proposed approach is shown on the example of calculating the area occupied by industrial objects.
For a growing economy like India, most of its energy resources are obtained through extractive processes such as mining of coal and other minerals. Mining can however have many negative social and ecological impacts if it is not well regulated. Illegal mining or inadequate reclamation of abandoned mines can amplify these impacts, emphasizing the need to develop methods that can monitor changes in the land-use patterns in and around mining sites. We develop a method using machine learning on freely available satellite data to monitor the extent of mines, and augment it with outputs from land use and land cover classification, deforestation detection, and PM2.5 particulate matter estimation from remote sensing data to track land-use and ecological changes taking place in the proximity of mining sites. We provide evaluation results of our mining delineation classifier, a feasibility check of this suite of tools to monitor mining areas over a period of four years, and a temporal characterization study over 628 mines in India that were granted a clearance for operations during the period 2006 to 2012. We further use this suite of monitoring tools to compare socio-economic development and health indicators across mining and non-mining areas, across various states in India, to study whether extractive processes of mining benefit the immediate population in their neighbourhood.
The degeneration of the intervertebral discs in the lumbar spine is the common cause of neurological and physical dysfunctions and chronic disability of patients, which can be stratified into single- (e.g., disc herniation, disc prolapse, or disc bulge) and comorbidity-type degeneration (i.e., simultaneous presence of two or more conditions of disc herniation, prolapse, and bulge in a patient) respectively. The degenerative changes of the lumbar spine differentiate in the level of severity and type. A sample of lumbar magnetic resonance images from multiple clinical hospitals in China was collected and used in the proposal assessment. Theoretically, we devised a novel transfer learning framework VIRD by ensembling four pre-trained models including Densenet169, ResNet101, InceptionResNetV2, and VGG16. Thereafter, the proposed approach was applied to the clinical data and achieved 99% accuracy versus 94%, 96%, 96%, 96%, and 98% for compositional benchmark models of VGG16, InceptionResNetV2, DenseNet169, ResNet101, and other ensemble deep learning respectively. Furthermore, improved performance was observed as well for the metric of the area under the curve, producing a 1% increase relative to other ensemble learning, a 2% increase relative to most-studied models, and a 3% increase relative to the baseline models. The novel methodology can be used as a guide in the initial and efficient screening of complex degeneration of lumbar intervertebral discs and assist in the early-stage selection of clinically differentiated treatment options.
The lack of reliable data in developing countries is a major obstacle towards
sustainable development, food security, and disaster relief. Poverty data, for
example, is typically scarce, sparse in coverage, and labor-intensive to
obtain. Remote sensing data such as high-resolution satellite imagery, on the
other hand, is becoming increasingly available and inexpensive. Unfortunately,
such data is highly unstructured and we lack techniques to automatically
extract useful insights to inform policy decisions and help direct humanitarian
efforts. We propose a novel machine learning approach to extract large-scale
socioeconomic indicators from high-resolution satellite imagery. The main
challenge is that training data is very scarce, making it difficult to apply
modern techniques such as Convolutional Neural Networks (CNN). We therefore
propose a transfer learning approach where nighttime light intensities are used
as a data-rich proxy. We train a fully convolutional CNN model to predict
nighttime lights from daytime imagery, simultaneously learning features that
are useful for poverty prediction. The model learns filters identifying
different terrains and man-made structures, including roads, buildings, and
farmlands, without any supervision beyond nighttime lights. We demonstrate that
these learned features are highly informative for poverty mapping, even
approaching the predictive performance of survey data collected in the field.
Deep learning allows computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction. These methods have dramatically improved the state-of-the-art in speech recognition, visual object recognition, object detection and many other domains such as drug discovery and genomics. Deep learning discovers intricate structure in large data sets by using the backpropagation algorithm to indicate how a machine should change its internal parameters that are used to compute the representation in each layer from the representation in the previous layer. Deep convolutional nets have brought about breakthroughs in processing images, video, speech and audio, whereas recurrent nets have shone light on sequential data such as text and speech.
The ImageNet Large Scale Visual Recognition Challenge is a benchmark in
object category classification and detection on hundreds of object categories
and millions of images. The challenge has been run annually from 2010 to
present, attracting participation from more than fifty institutions.
This paper describes the creation of this benchmark dataset and the advances
in object recognition that have been possible as a result. We discuss the
challenges of collecting large-scale ground truth annotation, highlight key
breakthroughs in categorical object recognition, provide detailed a analysis of
the current state of the field of large-scale image classification and object
detection, and compare the state-of-the-art computer vision accuracy with human
accuracy. We conclude with lessons learned in the five years of the challenge,
and propose future directions and improvements.
Recent results indicate that the generic descriptors extracted from the
convolutional neural networks are very powerful. This paper adds to the
mounting evidence that this is indeed the case. We report on a series of
experiments conducted for different recognition tasks using the publicly
available code and model of the OverFeat network which was trained to perform
object classification on ILSVRC13. We use features extracted from the OverFeat
network as a generic image representation to tackle the diverse range of
recognition tasks of object image classification, scene recognition, fine
grained recognition, attribute detection and image retrieval applied to a
diverse set of datasets. We selected these tasks and datasets as they gradually
move further away from the original task and data the OverFeat network was
trained to solve. Remarkably we report better or competitive results compared
to the state-of-the-art in all the tasks on various datasets. The results are
achieved using a linear SVM classifier applied to a feature representation of
size 4096 extracted from a layer in the net. The results strongly suggest that
features obtained from deep learning with convolutional nets should be the
primary candidate in most visual classification tasks.
We develop a statistical framework to use satellite data on night lights to augment official income growth measures. For countries with poor national income accounts, the optimal estimate of growth is a composite with roughly equal weights on conventionally measured growth and growth predicted from lights. Our estimates differ from official data by up to three percentage points annually. Using lights, empirical analyses of growth need no longer use countries as the unit of analysis; we can measure growth for sub- and supranational regions. We show, for example, that coastal areas in sub-Saharan Africa are growing slower than the hinterland. (JEL E01, E23, O11, 047, 057)
Evaluating the total numbers of people at risk from infectious disease in the world requires not just tabular population data, but data that are spatially explicit and global in extent at a moderate resolution. This review describes the basic methods for constructing estimates of global population distribution with attention to recent advances in improving both spatial and temporal resolution. To evaluate the optimal resolution for the study of disease, the native resolution of the data inputs as well as that of the resulting outputs are discussed. Assumptions used to produce different population data sets are also described, with their implications for the study of infectious disease. Lastly, the application of these population data sets in studies to assess disease distribution and health impacts is reviewed. The data described in this review are distributed in the accompanying DVD.
Socio-economic maps contain important information regarding the population of a country. Computing these maps is critical given that policy makers often times make important decisions based upon such information. However, the compilation of socio-economic maps requires extensive resources and becomes highly expensive. On the other hand, the ubiquitous presence of cell phones, is generating large amounts of spatiotemporal data that can reveal human behavioral traits related to specific socio-economic characteristics. Traditional inference approaches have taken advantage of these datasets to infer regional socio-economic characteristics. In this paper, we propose a novel approach whereby topic models are used to infer socio-economic levels from large-scale spatio-temporal data. Instead of using a pre-determined set of features, we use latent Dirichlet Allocation (LDA) to extract latent recurring patterns of co-occurring behaviors across regions, which are then used in the prediction of socio-economic levels. We show that our approach improves state of the art prediction results by 9%.
We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 dif- ferent classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implemen- tation of the convolution operation. To reduce overfitting in the fully-connected layers we employed a recently-developed regularization method called dropout that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry
GDP per capita and household survey means present conflicting pictures of the rate of economic development in emerging countries.
One of the areas in which the national accounts-household surveys debate is key is the measurement of developing world poverty.
We propose a data-driven method to assess the relative quality of GDP per capita and survey means by comparing them to the
evolution of satellite-recorded nighttime lights. Our main assumption, which is robust to a variety of specification checks,
is that the measurement error in nighttime lights is unrelated to the measurement errors in either national accounts or survey
means. We obtain estimates of weights on national accounts and survey means in an optimal proxy for true income; these weights
are very large for national accounts and very modest for survey means. We conclusively reject the null hypothesis that the
optimal weight on surveys is greater than the optimal weight on national accounts, and we generally fail to reject the null
hypothesis that the optimal weight on surveys is zero. Additionally, we provide evidence that national accounts are good indicators
of desirable outcomes for the poor (such as longer life expectancy, better education and access to safe water), and we show
that surveys appear to perform worse in developing countries that are richer and that are growing faster. Therefore, we interpret
our results as providing support for estimates of world poverty that are based on national accounts.
Accurate and timely estimates of population characteristics are a critical input to social and economic research and policy. In industrialized economies, novel sources of data are enabling new approaches to demographic profiling, but in developing countries, fewer sources of big data exist.We show that an individual's past history of mobile phone use can be used to infer his or her socioeconomic status. Furthermore, we demonstrate that the predicted attributes of millions of individuals can, in turn, accurately reconstruct the distribution of wealth of an entire nation or to infer the asset distribution of microregions composed of just a few households. In resource-constrained environments where censuses and household surveys are rare, this approach creates an option for gathering localized and timely information at a fraction of the cost of traditional methods.
How flocks of small, cheap satellites, hatched in Silicon Valley, will constantly monitor a changing Earth.
We investigate the role of national institutions on subnational African development in a novel framework that accounts for
both local geography and cultural-genetic traits. We exploit the fact that the political boundaries on the eve of African
independence partitioned more than 200 ethnic groups across adjacent countries subjecting similar cultures, residing in homogeneous
geographic areas, to different formal institutions. Using both a matching type and a spatial regression discontinuity approach
we show that differences in countrywide institutional structures across the national border do not explain within-ethnicity
differences in economic performance, as captured by satellite images of light density. The average noneffect of national institutions
on ethnic development masks considerable heterogeneity partially driven by the diminishing role of national institutions in
areas further from the capital cities. JEL Codes: O10, O40, O43, N17, Z10.
While Africa may have overcome its growth tragedy, it is facing a statistical tragedy, in that the statistical foundations of the recent growth in per-capita GDP and reduction in poverty are quite weak. In many countries, GDP accounts use old methods, population censuses are out of date, and poverty estimates are infrequent and often not comparable over time. The proximate reasons have to do with weak capacity, inadequate funding, and a lack of coordination of statistical activities. But the underlying cause may be the political sensitivity of these statistics, and some donors' tendency to go around countries' own National Statistical Development Strategies (NSDS). Greater openness and transparency of statistics, and a higher profile for the NSDS, possibly with “naming and shaming” of those who try to circumvent it, may help Africans turn around their statistical tragedy.
The latest generation of Convolutional Neural Networks (CNN) have achieved
impressive results in challenging benchmarks on image recognition and object
detection, significantly raising the interest of the community in these
methods. Nevertheless, it is still unclear how different CNN methods compare
with each other and with previous state-of-the-art shallow representations such
as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts
a rigorous evaluation of these new techniques, exploring different deep
architectures and comparing them on a common ground, identifying and disclosing
important implementation details. We identify several useful properties of
CNN-based representations, including the fact that the dimensionality of the
CNN output layer can be reduced significantly without having an adverse effect
on performance. We also identify aspects of deep and shallow methods that can
be successfully shared. A particularly significant one is data augmentation,
which achieves a boost in performance in shallow methods analogous to that
observed with CNN-based methods. Finally, we are planning to provide the
configurations and code that achieve the state-of-the-art performance on the
PASCAL VOC Classification challenge, along with alternative configurations
trading-off performance, computation speed and compactness.
A major assumption in many machine learning and data mining algorithms is that the training and future data must be in the same feature space and have the same distribution. However, in many real-world applications, this assumption may not hold. For example, we sometimes have a classification task in one domain of interest, but we only have sufficient training data in another domain of interest, where the latter data may be in a different feature space or follow a different data distribution. In such cases, knowledge transfer, if done successfully, would greatly improve the performance of learning by avoiding much expensive data-labeling efforts. In recent years, transfer learning has emerged as a new learning framework to address this problem. This survey focuses on categorizing and reviewing the current progress on transfer learning for classification, regression, and clustering problems. In this survey, we discuss the relationship between transfer learning and other related machine learning techniques such as domain adaptation, multitask learning and sample selection bias, as well as covariate shift. We also explore some potential future issues in transfer learning research.
Reliably extracting information from aerial imagery is a difficult problem with many practical applications. One specific
case of this problem is the task of automatically detecting roads. This task is a difficult vision problem because of occlusions,
shadows, and a wide variety of non-road objects. Despite 30 years of work on automatic road detection, no automatic or semi-automatic
road detection system is currently on the market and no published method has been shown to work reliably on large datasets
of urban imagery. We propose detecting roads using a neural network with millions of trainable weights which looks at a much
larger context than was used in previous attempts at learning the task. The network is trained on massive amounts of data
using a consumer GPU. We demonstrate that predictive performance can be substantially improved by initializing the feature
detectors using recently developed unsupervised learning methods as well as by taking advantage of the local spatial coherence
of the output labels. We show that our method works reliably on two challenging urban datasets that are an order of magnitude
larger than what was used to evaluate previous approaches.
A pervasive issue in social and environmental research has been how to improve the quality of socioeconomic data in developing countries. Given the shortcomings of standard sources, the present study examines luminosity (measures of nighttime lights visible from space) as a proxy for standard measures of output (gross domestic product). We compare output and luminosity at the country level and at the 1° latitude × 1° longitude grid-cell level for the period 1992-2008. We find that luminosity has informational value for countries with low-quality statistical systems, particularly for those countries with no recent population or economic censuses.
This paper has an empirical and overtly methodological goal. The authors propose and defend a method for estimating the effect of household economic status on educational outcomes without direct survey information on income or expenditures. They construct an index based on indicators of household assets, solving the vexing problem of choosing the appropriate weights by allowing them to be determined by the statistical procedure of principal components. While the data for India cannot be used to compare alternative approaches they use data from Indonesia, Nepal, and Pakistan which have both expenditures and asset variables for the same households. With these data the authors show that not only is there a correspondence between a classification of households based on the asset index and consumption expenditures but also that the evidence is consistent with the asset index being a better proxy for predicting enrollments--apparently less subject to measurement error for this purpose--than consumption expenditures. The relationship between household wealth and educational enrollment of children can be estimated without expenditure data. A method for doing so - which uses an index based on household asset ownership indicators- is proposed and defended in this paper. In India, children from the wealthiest households are over 30 percentage points more likely to be in school than those from the poorest households.
Using data from India, we estimate the relationship between household wealth and children's school enrollment. We proxy wealth by constructing a linear index from asset ownership indicators, using principal-components analysis to derive weights. In Indian data this index is robust to the assets included, and produces internally coherent results. State-level results correspond well to independent data on per capita output and poverty. To validate the method and to show that the asset index predicts enrollments as accurately as expenditures, or more so, we use data sets from Indonesia, Pakistan, and Nepal that contain information on both expenditures and assets. The results show large, variable wealth gaps in children's enrollment across Indian states. On average a "rich" child is 31 percentage points more likely to be enrolled than a "poor" child, but this gap varies from only 4.6 percentage points in Kerala to 38.2 in Uttar Pradesh and 42.6 in Bihar.
We consider an asset-based alternative to the standard use of expenditures in defining well-being and poverty. Our motivation is to see if there exist simpler and less demanding ways to collect data to measure economic welfare and rank households. This is particularly important in poor regions where there is limited capacity to collect consumption, expenditure and price data. We evaluate an index derived from a factor analysis on household assets using multipurpose surveys from several countries, We find that the asset index is a valid predictor of a crucial manifestation of poverty - child health and nutrition. Indicators of relative measurement error show that the asset index is measured as a proxy for long-term wealth with less error than expenditures. Analysts may thus prefer to use the asset index as an explanatory variable or as a means of mapping economic welfare to other living standards and capabilities such as health and nutrition.
- S Devarajan
S. Devarajan, Rev. Income Wealth 59, S9-S15 (2013).
Benefits and costs of the data for development targets for the Post-2015 Development Agenda
- M Jerven
M. Jerven, "Benefits and costs of the data for development
targets for the Post-2015 Development Agenda," Data for
Development Assessment Paper Working Paper, September
(Copenhagen Consensus Center, Copenhagen, 2014).
- J Sandefur
- A Glassman
J. Sandefur, A. Glassman, J. Dev. Stud. 51, 116-132 (2015).
- J V Henderson
- A Storeygard
- D N Weil
J. V. Henderson, A. Storeygard, D. N. Weil, Am. Econ. Rev. 102,
994-1028 (2012).
- X Chen
- W D Nordhaus
X. Chen, W. D. Nordhaus, Proc. Natl. Acad. Sci. U.S.A. 108,
8589-8594 (2011).
- M Pinkovskiy
- X Sala-I-Martin
M. Pinkovskiy, X. Sala-i-Martin, Q. J. Econ. 131, 579-631 (2016).
- J Blumenstock
- G Cadamuro
- R On
J. Blumenstock, G. Cadamuro, R. On, Science 350, 1073-1076 (2015).
- S J Pan
- Q Yang
S. J. Pan, Q. Yang, IEEE Trans. Knowl. Data Eng. 22, 1345-1359 (2010).
- D E Sahn
- D Stifel
D. E. Sahn, D. Stifel, Rev. Income Wealth 49, 463-489 (2003).
- O Russakovsky
O. Russakovsky et al., Int. J. Comput. Vis. 115, 211-252 (2014).
- E Hand
E. Hand, Science 348, 172-177 (2015).