Satellite Remote Sensing and U-Net for Predicting Population Density in Kenya

Fundamental institutions rely on the work and results of empirical economic research whose main purpose is to answer why economic systems move in the way they do. This is the cardinal question for informing effective and efficient policy. In recent years, with the growing availability of datasets from increasingly powerful satellite sensors and This paper is broken down into three main sections:

1. Background: to provide the reader with a knowledge base to interpret view the project in. I’ll discuss the neural networks including the U-net architecture used in this project as well as introduce Google Earth Engine (GEE).

2. Project Report: to walk through the unique workflow as well as the modeling (including model inputs and response, sampling, training, and prediction results), and discussion on experience and limitations.

3. Implications: in a conclusion of sorts, to further talk about the promise of this sort of data analysis and it’s use to economic research. Through slight training set and model adjustments, I was able to improve a standard U-net regression model by 72%, leaving still lots of room for further improvement and tuning.