Projects / Estimating Poverty Incidence in the Philippines Using Satellite Imagery and CNN–LSTM Architectures

Estimating Poverty Incidence in the Philippines Using Satellite Imagery and CNN–LSTM Architectures

2026 · Thesis · data pipeline

r = 0.73 in cross-validation · 1,970 areas mapped

Project TALA dashboard: a dark map of Central Luzon with each community shaded from poor (orange) to wealthy (blue)

The problem

Official Philippine poverty statistics come out infrequently and only to the province, so policymakers cannot see which communities inside a province are poorest.

What I built

With three teammates I built a model that reads a year of Sentinel-2 satellite images alongside road, building and night-light data to estimate a wealth index for each community, and a dashboard for policymakers. I built the data pipeline in Google Earth Engine, worked on the diagrams and ran the user survey.

What happened

It reached a correlation of 0.73 with survey-measured wealth in cross-validation, in line with published studies, and mapped 1,970 communities across 16 provinces. It shows wealth gaps inside provinces that official averages hide.

Built with

PythonTensorFlowGoogle Earth EngineNext.jsMapbox