Human bias and CNNs' superior insights in satellite based poverty mapping.

Satellite imagery is a potent tool for estimating human wealth and poverty, especially in regions lacking reliable data. This study compares a range of poverty estimation approaches from satellite images, spanning from expert-based to fully machine learning-based methodologies. Human experts ranked clusters from the Tanzania DHS survey using high-resolution satellite images. Then expert-defined features were utilized in a machine learning algorithm to estimate poverty. An explainability method was applied to assess the importance and interaction of these features in poverty prediction. Additionally, a convolutional neural network (CNN) was employed to estimate poverty from medium-resolution satellite images of the same locations. Our analysis indicates that increased human involvement in poverty estimation diminishes accuracy compared to machine learning involvement, exemplified with the case of Tanzania. Expert defined features exhibited significant overlap and poor interaction when used together in a classifier. Conversely, the CNN-based approach outperformed human experts, demonstrating superior predictive capability with medium-resolution images. These findings highlight the importance of leveraging machine learning explainability methods to identify predictive elements that may be overlooked by human experts. This study advocates for the integration of emerging technologies with traditional methodologies to optimize data collection and analysis of poverty and welfare.

A review of explainable AI in the satellite data, deep machine learning, and human poverty domain.

Recent advances in artificial intelligence and deep machine learning have created a step change in how to measure human development indicators, in particular asset-based poverty. The combination of satellite imagery and deep machine learning now has the capability to estimate some types of poverty at a level close to what is achieved with traditional household surveys. An increasingly important issue beyond static estimations is whether this technology can contribute to scientific discovery and, consequently, new knowledge in the poverty and welfare domain. A foundation for achieving scientific insights is domain knowledge, which in turn translates into explainability and scientific consistency. We perform an integrative literature review focusing on three core elements relevant in this context-transparency, interpretability, and explainability-and investigate how they relate to the poverty, machine learning, and satellite imagery nexus. Our inclusion criteria for papers are that they cover poverty/wealth prediction, using survey data as the basis for the ground truth poverty/wealth estimates, be applicable to both urban and rural settings, use satellite images as the basis for at least some of the inputs (features), and the method should include deep neural networks. Our review of 32 papers shows that the status of the three core elements of explainable machine learning (transparency, interpretability, and domain knowledge) is varied and does not completely fulfill the requirements set up for scientific insights and discoveries. We argue that explainability is essential to support wider dissemination and acceptance of this research in the development community and that explainability means more than just interpretability.

Population centroids of the world administrative units from nighttime lights 1992-2013.

Knowledge about the past, current and future distribution of the human population is fundamental for tackling many global challenges. Censuses are used to collect information about population within a specified spatial unit. The spatial units are usually arbitrarily defined and their numbers, size and shape tend to change over time. These issues make comparisons between areas and countries difficult. We have in related work proposed that the shape of the lit area derived from nighttime lights, weighted by its intensity can be used to analyse characteristics of the population distribution, such as the mean centre of population. We have processed global nighttime lights data for the period 1992-2013 and derived centroids for administrative levels 0-2 of the Database of Global Administrative Areas, corresponding to nations and two levels of sub-divisions, that can be used to analyse patterns of global or local population changes. The consistency of the produced dataset was investigated and distance between true population centres and derived centres are compared using Swedish census data as a benchmark.

High-resolution African population projections from radiative forcing and socio-economic models, 2000 to 2100.

For its fifth assessment report, the Intergovernmental Panel on Climate Change divided future scenario projections (2005-2100) into two groups: Socio-Economic Pathways (SSPs) and Representative Concentration Pathways (RCPs). Each SSP has country-level urban and rural population projections, while the RCPs are based on radiative forcing caused by greenhouse gases, aerosols and associated land-use change. In order for these projections to be applicable in earth system models, SSP and RCP population projections must be at the same spatial scale. Thus, a gridded population dataset that takes into account both RCP-based urban fractions and SSP-based population projection is needed. To support this need, an annual (2000-2100) high resolution (approximately 1km at the equator) gridded population dataset conforming to both RCPs (urban land use) and SSPs (population) country level scenario data were created.

Nighttime lights and population changes in Europe 1992-2012.

Nighttime satellite photographs of Earth reveal the location of lighting and provide a unique view of the extent of human settlement. Nighttime lights have been shown to correlate with economic development and population but little research has been done on the link between nighttime lights and population change over time. We explore whether population decline is coupled with decline in lighted area and how the age structure of the population and GDP are reflected in nighttime lights. We examine Europe between the period of 1992 and 2012 using a Geographic Information System and regression analysis. The results suggest that population decline is not coupled with decline in lighted area. Instead, human settlement extent is more closely related to the age structure of the population and to GDP. We conclude that declining populations will not necessarily lead to reductions in the extent of land development.