Despite the world making remarkable progress in combating poverty through the United Nations Millennium Development and Sustainable Development Goals, over 300 million people in Africa still live in extreme poverty.
African countries such as Nigeria and Madagascar had similar poverty levels to those of China and Vietnam in the early 1990s. Yet, the Asia Pacific-based countries have been successful in reducing poverty, while the African countries have not.
Researchers lacked the information and data required to determine the causes and consequences of this poverty - in particular, the reason why poverty reduction strategies and anti-poverty programmes haven’t been as effective in Africa as they have been in other countries.
To bridge the gap with African poverty data accessibility, the ‘Multidimensional Poverty Data - Africa’ project was created by the University of Gothenburg and Harvard University. Together, they are working towards providing past poverty data to help determine future poverty trends within African communities using Google Earth Engine (GEE).
Multidimensional poverty identifies the various deprivations experienced by individuals; such as poor health, lack of education and inadequate living standards. By measuring more than one factor - usually income, researchers can better capture and understand the true reality of poverty in differing communities.
This project is one of the 32 projects selected as part of the GEO - Google Earth Engine Program that provides funding to tackle environmental and social challenges using open Earth data.
We spoke with Adel Daoud, the project’s Principal Investigator from the University of Gothenburg, to learn more about the project and how his team will be using GEE to provide multidimensional poverty data for all African communities.
Adel, can you describe the challenge that you want to tackle with this project?
Past poverty data can assist us with predicting future poverty trends. However, policymakers and researchers currently lack sufficient data to conduct such predictions.
In the coming two years, our goal is to produce historical poverty data for Africa dating from 1984 to 2020. The key challenge we face is to identify what resources we can use in GEE, and how to set up an optimal technical pipeline to assemble data, train the algorithms, and wrap our code in a reproducible format.
How will you address this challenge with your project?
Our project aims to combine artificial intelligence methods and publicly available satellite images to produce multidimensional poverty data for all African communities from 1984 to 2020.
To do this we will:
(i) Train and adapt image recognition algorithms (CNNs) to identify multidimensional poverty from Landsat satellite images of African communities over time and space, quarterly, from 1984 to 2020.
(ii) Use these CNNs to compare the quality of poverty data produced by:
- Landsat images
- Google Maps (2005-)
- Sentinel 2 (2015-)
- RapidEye (2009-2020)
- PlanetScope (2009-)
(iii) Create statistical software that enables us and others to access the current poverty estimates to produce future estimates using different satellite technologies.
(iv) We will, based on our methodological advances, forecast the levels and speed that African communities are emerging from multidimensional poverty and their likelihood to achieve the relevant Sustainable Development Goals by 2030.
How has the challenge impacted your community?
My community consists of global poverty researchers working in sociology, political science and economics. The lack of historical poverty data limits the scope of analyses that my community is able to do.
With the technical competence required to produce such data from satellite images, very few social-scientists attempt to tackle this data limitation. The research conducted so far in this area has been conducted by Earth and computer scientists combining machine learning and satellite images to predict poverty levels. Our aim is to build on this research and make these data and models more accessible to a wider range of researchers.
How does Google Earth Engine help you achieve your project-related goals?
Downloading all the images of Africa for the past 40 years and training the image-recognition algorithms on our local machines requires a lot of computing power. GEE provides an integrated portal to access both satellite image data and image-recognition algorithms for experimentation.
How will the GEO-GEE funding help your project?
The GEO-GEE funding will help us:
- Easily access a wide range of satellite technologies and images
- Conduct experiments by combining a variety of image-recognition algorithms and satellite imagery parameters
- Provide mentorship for both the technical and social networking aspects of marketing our research findings
What does success look like to you?
Hopefully by the end of this project we have created a pilot dataset of historical poverty trends that we can release to the wider research community.
EO Data Science’s role in the GEO - GEE Program
EO Data Science partnered with Google Earth Engine and the Group on Earth Observations to launch the GEO-GEE Program, which supports GEO member countries to operationalise their science as they strive to tackle the world’s biggest sustainable development challenges.
In July 2020, 32 projects across 22 countries were selected into the program which offers $3 million USD towards product licenses and $1 million USD in technical support from EO Data Science. This funding and support will help these projects tackle global challenges using open Earth data. Read the announcement and list of winners here.
Follow us on Twitter, LinkedIn or Facebook to hear about more GEO-GEE winning projects!
Back To News Stories