Can satellite derived insights be a credible compliment to survey design, monitoring and evaluation efforts in data sparse environments? At Atlas AI, we are determined to find out.

Anaïs Tadlaoui

Any operator looking to better understand ground conditions in data scarce environments faces a multitude of challenges that include — but are hardly limited to — the limitations of official statistics, wrangling disparate data sources that are insufficient in scope and often out of date, and the difficulty of collecting new data — an expensive, time-consuming and resource intense endeavor.

Over the last several years, the remarkable increase in satellite-derived information and advancements in machine learning techniques have begun to significantly shift how we create new data and make inferences accordingly. It’s an exciting time for the monitoring and evaluation (M&E) space but with excitement comes a host of important questions that the research community is actively exploring:

  • Can we trust the new data we are creating?
  • Do we need to adjust our current methodologies?
  • How can we collectively advance our work to ensure equitable access to these new opportunities?

The founders of Atlas AI, three Stanford scientists — Marshall Burke, David Lobell and Stefano Ermon — recently released an article in Science that reviewed and synthesized the rapidly growing scientific literature that seeks to use satellite imagery to measure and understand various human outcomes, including many outcomes directly related to the Sustainable Development Goals (SDGs).

Their work made four main points:

  1. Satellite-based outcomes predictions work reasonably well and the methods are improving.
  2. Training data — and not imagery — is now the biggest rate limiter to model development.
  3. Satellite-derived insights are most likely to strengthen — and not replace — well-developed and trusted ground-based data collection efforts.
  4. Limited uptake of this emergent technology is driven by a host of factors including the newness of the science, hesitancy in adopting measures that cannot fully be understood, the challenges with interpretability and maintenance of the status quo.

Over the next six months, we will unpack each of these points, discussing more about the work we are doing with trusted partners, and sharing exciting findings emerging from new initiatives.

Read the full article in Science here.

Learn more about Atlas AI’s work here.

Stay tuned!

Sign up for our newsletter to get the latest news on investing and growing your business

world map illustration