Temporal Poverty Prediction using Satellite Imagery

In order to effectively alleviate poverty, the measurement and tracking of support initiatives over time is a necessary step for targeting aid efforts and guiding policy decisions. However, obtaining such data is time and labor intensive, so coverage of poverty stricken areas is often sparse or nonexistent. Previous research has shown that a viable alternative for measuring poverty levels can be achieved through remote sensing methods. Specifically, satellite images processed through convolutional neural networks have shown promise in predicting the intensity of nighttime lights, which can then be used to gauge the underlying poverty level. This paper attempts to extend on past work by finding methods for measuring the change in poverty levels across different years using the same type of publicly available data. We are able to verify the original results of predicting poverty at a single point in time. More work is still needed to produce meaningful results in predicting temporal poverty.