# Satellite Image and Machine Learning based Knowledge Extraction in the Poverty and Welfare Domain

**Authors:** Ola Hall, Mattias Ohlsson, Thortseinn Rögnvaldsson  
**arXiv:** [2203.01068](https://arxiv.org/abs/2203.01068)  
**Date:** 2022-03-02  
**Summary page:** https://www.summarizepaper.com/en/paper/2203.01068v1/

## Key points

- Recent advances in AI and machine learning have revolutionized the measurement of human development indicators, particularly asset-based poverty.
- Satellite imagery and machine learning techniques can estimate poverty levels with similar accuracy as traditional methods.
- There is interest in using this technology for scientific discovery and generating new knowledge in the field of poverty and welfare.
- Transparency, interpretability, and explainability are three core elements needed to achieve scientific insights.
- Existing literature shows that these core elements vary across studies and do not fully meet the requirements for scientific insights and discoveries.
- Some papers demonstrate transparency by providing detailed information about their models, while others lack clarity in methodology.
- Interpretability is often limited as researchers do not thoroughly discuss how different features impact their results or what changes in an image would alter the output.
- Explainability plays a crucial role in supporting wider acceptance and dissemination of research findings.
- Ethical considerations may arise when sharing data or final predictions.
- Work needs to be done to ensure transparency, interpretability, and explainability within the field of estimating poverty using satellite imagery.

## Summary

Recent advances in artificial intelligence and machine learning have revolutionized the measurement of human development indicators, particularly asset-based poverty. By combining satellite imagery and machine learning techniques, researchers are now able to estimate poverty levels with a similar accuracy as traditional methods such as face-to-face interviews and household surveys. However, beyond static estimations, there is a growing interest in whether this technology can contribute to scientific discovery and generate new knowledge in the field of poverty and welfare. To achieve scientific insights, it is crucial to have domain knowledge that translates into explainability and scientific consistency. In this review, the focus is on three core elements: transparency, interpretability, and explainability. Transparency refers to the ability to understand the inner workings of the models used in machine learning algorithms. Interpretability involves understanding how different features affect the output of these models. Explainability goes beyond interpretability by considering how well the results can be communicated and disseminated. The review of existing literature reveals that the status of these three core elements varies across studies and does not fully meet the requirements for scientific insights and discoveries. While some papers demonstrate transparency by providing detailed information about their models and parameters, others lack clarity in their methodology. Similarly, interpretability is often limited as researchers do not thoroughly discuss how different features impact their results or what changes in an image would alter the output. It is argued that explainability plays a crucial role in supporting wider acceptance and dissemination of research findings. Explainability encompasses more than just interpretability; it involves making all aspects of a study transparent so that other researchers can reproduce and explore the models for further analysis. However, ethical considerations may arise when sharing data or final predictions. Overall, while recent advancements in AI and machine learning have opened up new possibilities for estimating poverty using satellite imagery, there is still work to be done to ensure transparency, interpretability, and explainability within this field. By addressing these challenges, researchers can enhance scientific discovery and contribute to a deeper understanding of poverty and welfare.