Time Series of Satellite Imagery Improve Deep Learning Estimates of Neighborhood-Level Poverty in Africa

These maps contain estimates of material wealth at the neighborhood-level across Africa. These estimates were made using a novel deep-learning model for predicting material wealth based on satellite images from Landsat, DMSP and VIIRS. The model was trained on DHS survey data as described in the corresponding paper (Pettersson et al. 2023).
The maps cover all populated areas according to the Global Human Settlement Lay…. The spatial resolution of the maps is 6.72 x 6.72 km and the unit of measurement is the International Wealth Index…, scaled from 0 to 1. Each map represents a three-year time span between 1990 to 2019. In the tif file these maps are stored as bands in the image, resulting in the following configuration:

• Band 1: IWI estimates for 1990-1992
• Band 2: IWI estimates for 1993-1995
• Band 3: IWI estimates for 1996-1998
• Band 4: IWI estimates for 1999-2001
• Band 5: IWI estimates for 2002-2004
• Band 6: IWI estimates for 2005-2007
• Band 7: IWI estimates for 2008-2010
• Band 8: IWI estimates for 2011-2013
• Band 9: IWI estimates for 2014-2016
• Band 10: IWI estimates for 2017-2019

For a full explanation of the map-generating process and estimates, see the corresponding paper.