Measuring Economic Development with Daytime Satellite Imagery

Klaus Ackermann

April 5, 2022

This paper proposes a methodology for predicting subnational economic development using high-resolution, daytime satellite imagery. We collected high-resolution satellite images and corresponding open-source ground truth data for over 28 million 1km×1km grid cells covering 25 European countries and 5 million 1km×1km grid cells covering 21 African countries. We first use a standard random forest model to identify a subset of features from the ground truth data that best predict fixed capital at the EU NUTSII level. We then train a convolutional neural network (CNN) to extract the relevant features from each daytime satellite image. The resulting measures are used to predict fixed capital at the EU-NUTSII level which then build the input in a standard development accounting framework to construct the regional GDP p.c. at the NUTSII level. The correlation between our constructed regional GDP and the official GDP figures is around 0.7. We then use this approach to construct a measure for local economic activity at the grid cell level for 21 African countries. We show that this new measure provides a more detailed picture of differences in subnational economic development in Africa compared to existing proxies. Our methodology can build the foundation to predict subnational indicators of economic development for areas where official data is either missing or unreliable.


Toxic Neighborhoods: The Effects of Concentrated Poverty and Environmental Lead Contamination on Early Childhood Development

Geoffrey Wodtke

April 19, 2022

Although socioeconomic disparities in cognitive ability emerge early in the life course, most research on the consequences of living in a disadvantaged neighborhood focuses on school-age children or adolescents. In this study, we outline and test a theoretical model of neighborhood effects on cognitive development during early childhood that highlights the mediating role of environmental health hazards, and in particular, exposure to neurotoxic lead. To evaluate this model, we follow 1,266 children in Chicago from birth through the time of school entry, tracking their areal risk of lead exposure and the socioeconomic composition of their neighborhoods over time. We then estimate the joint effects of neighborhood poverty and environmental lead contamination on receptive vocabulary ability. We find that sustained exposure to disadvantaged neighborhoods reduces vocabulary skills during early childhood and that this effect operates through a causal mechanism involving lead contamination.


Super-charging your (regression) models with space and geographical context

Daniel Arribas-Bel and Pedro Amaral

May 5, 2022

We're thrilled to host two old friends and long-time collaborators of the Center next week: Professors Daniel Arribas-Bel and Pedro Amaral. Dani is a Professor at the Department of Geography and Planning of the University of Liverpool (UK) and Pedro is a Professor in the Department of Economics at Universidade Federal de Minas Gerais in Brazil. Both have been working on spatial econometric modeling in Python (including the PySAL library) for many years.

They will do a guided walkthrough of two chapters of the forthcoming book Geographic Data Science with PySAL and the PyData Stack by Sergio J. Rey, Dani Arribas-Bel and Levi J. Wolf. This means that, if you have Python and data science experience, you should be able to follow along with your own laptop. If you don't, you can listen to the walkthrough as a lecture. Here's the summary:

Super-Charging your (Regression) Models with Space and Geographical Context

This talk will present an overview of different strategies to embed space and geographical context in (regression) models to improve their performance. Often, where a phenomenon takes place is relevant to both understand it and predict it, two of the main reasons for building empirical models. This talk will introduce you to two broad sets of techniques to do so in a systematic way: explicitly spatial regression, and spatial feature engineering. Spatial regression incorporates information about the spatial distribution of observations in the structural form of the model; spatial feature engineering does so in the “features” or variables that are then plugged into a model or algorithm. We will take a hands-on approach and illustrate several of these concepts using the modern Python stack for data science and rely on the upcoming book Geographic Data Science with Python. The only requirement for this session will be to bring an internet-connected laptop. Optionally, you may want to skim over chapters 11 and 12 of the book.


  Logistics: Thursday, May 5, 2022 
  • 1-2:30pm Chicago/Central Time: Spatial Regression (chapter 11)
  • 3-4:30pm Chicago/Central Time: Spatial Feature Engineering (chapter 12)
Recording of both talks