WP178: Viewing society from space: Image-based sociocultural prediction models

Bienvenue à la section des publications d’Afrobaromètre. Pour des brèves analyses d’actualité, veuillez voir nos notes informatives (pour les séries d’enquêtes 1-5) et dépêches (à partir de la Série 6). Pour des analyses plus longues et techniques, se focalisant sur des questions de politique, regardez nos documents de politique. Nos documents de travail sont des analyses approfondies destinées à la publication dans des revues académiques ou des livres. Vous pouvez aussi rechercher dans toute la base des publications à partir des mots-clés, la langue, le pays, et/ou l’auteur.

Filter content by:

Documents de travail
John M. Irvine, Richard J. Wood, and Payden McBee

Applying new analytic methods to imagery data offers the potential to dramatically expand the information available for human geography. Satellite imagery can yield detailed local information about physical infrastructure, which we exploit for analysis of local socioeconomic conditions. Combining automated processing of satellite imagery with advanced modeling techniques provides a method for inferring measures of well-being, governance, and related sociocultural attributes from satellite imagery. This research represents a new approach to human geography by explicitly analyzing the relationship between observable physical attributes and societal characteristics and institutions of the region. Through analysis of commercial satellite imagery and Afrobarometer survey data, we have developed and demonstrated models for selected countries in sub-Saharan Africa (Botswana, Kenya, Zimbabwe).

The findings show the potential for predicting people’s attitudes about the economy, security, leadership, social involvement, and related questions, based only on imagery-derived information. The approach pursued here builds on earlier work in Afghanistan. Models for predicting economic attributes (presence of key infrastructure, attitudes about the economy, perceptions of crime, and outlook toward the future) all exhibit statistically significant performance. Although these results are encouraging, several avenues for advancement and improvement are proposed. Initial analysis of new methods for image processing and feature extraction have identified several avenues for promising enhancements.