Direct foreign aid faster with AI

Deployed computer vision model on whole country to find vulnerable locations

Project context: External blogpost

For my contributions, I was privileged to be granted primary authorship of the following 2 papers, which got accepted into the globally recognized conference NeurIPS.

  • 2021 NeurIPS Primary author for Mapping Access to Water and Sanitation using Publicly Accessible Satellite Imagery;
  • 2020 NeurIPS Primary author for Mapping New Informal Settlements Using Machine Learning and Satellite Images;

Summary

Deployed computer vision model on whole country to find vulnerable locations and guide humanitarian efforts.

This project harnessed the power of satellite imagery and advanced Computer Vision AI to address a critical challenge: identifying vulnerable locations and guiding foreign aid distribution in crisis-stricken regions. Focusing on Colombia, this initiative made previously impossible large-scale satellite analysis feasible, demonstrating the profound impact of AI in humanitarian efforts.

Situation

Colombia, a country facing various crises, required precise, large-scale data to identify vulnerable areas and effectively direct foreign aid. Traditional ground-based data collection was often too slow, dangerous, or resource-intensive to provide timely insights over vast and often inaccessible regions. This created significant challenges for international aid organizations needing to make rapid, informed decisions about resource allocation.

Task

My core task was to pioneer the use of satellite imagery and AI to overcome these data collection hurdles, enabling a comprehensive and rapid assessment of vulnerability across an entire country. This involved:

  • Data Feasibility: Making previously impossible satellite analysis feasible for humanitarian applications.
  • Model Development: Developing and deploying a high-resolution prediction model for a vast geographical area.
  • Client Engagement: Leading interactions with international clients to ensure alignment and impact.
  • Academic Contribution: Contributing to the broader scientific community through research publications.

Action

I took on a pivotal role spanning data engineering, model development, and client engagement, driving the project from conception to published results:

A. Pioneering Geospatial Data Engineering

  • I pioneered the use of Google Earth Engine for data extraction, ingestion, and processing. This was a critical step, as GEE’s capabilities allowed for efficient handling of petabytes of satellite imagery, making the nationwide analysis feasible where it previously wasn’t.

B. Advanced Satellite Image Prediction Model

  • I owned the development of a satellite image prediction model designed to identify key indicators of vulnerability. This model produced high-resolution 250-meter resolution maps covering an expansive 300,000 square kilometer land area, providing granular insights that were previously unavailable.
  • My technical contributions in this area were significant enough to be recognized in two co-authored NeurIPS papers (one on detection, one on prediction), demonstrating rigorous scientific validation and novel approaches.

C. International Client & Stakeholder Management

  • I led meetings with international clients across different time zones (such as ET and UTC+8), effectively communicating technical progress, gathering requirements, and ensuring the AI solutions directly addressed their humanitarian aid objectives. This cross-cultural and time-zone management was crucial for project alignment and successful deployment.

Result

My efforts resulted in a groundbreaking capability to direct foreign aid more effectively and rapidly, directly impacting vulnerable populations in Colombia:

  • Directing Foreign Aid: By making previously impossible satellite analysis feasible with AI, I helped direct foreign aid in crisis-stricken Colombia, enabling more targeted and efficient allocation of resources to vulnerable locations.
  • Nationwide Vulnerability Maps: The development of the satellite image prediction model produced detailed 250-meter resolution maps over a 300,000 sq km land area, providing actionable intelligence for aid organizations.
  • Scientific Recognition: The innovative approaches and results were validated and disseminated through two co-authored NeurIPS papers, solidifying the project’s impact both in the humanitarian field and the scientific community.

This project stands as a testament to the power of AI to drive positive social impact, demonstrating how advanced technology can bridge critical information gaps and accelerate humanitarian efforts.