Background

Machine learning helps researchers solve geospatial challenges related to agriculture, natural disasters, and land use. However, geospatial machine learning models must be trained on well-curated data to be effective. Training a model on global data does not necessarily translate to a model that performs well for specific regional areas, which is the ultimate goal when building a geospatial model that can solve real-world issues. 

The Competition

The Grow by Numbers challenge, run by Arizona State University under a NASA grant, invites participants to curate a training sample for a geospatial machine learning model that can classify cropland in a specific geographical area. Participants will be given a pool of candidate training samples from the CropHarvest benchmark dataset, from which they will create their own training sample.

The submitted list of training samples will be used to train multiple machine learning models and evaluated based on their combined performance across test crop classification tasks in four geographies: China, Togo, Tigray (Ethiopia), and Senegal.

Winners

Teams with winning or compelling solutions will be invited to collaborate on a workshop or conference paper submission with the competition organizers. The results of the competition will help address the important challenge of how to select training samples from a global pool to train geospatial machine learning models that excel at a local level. This will contribute directly to NASA Harvest’s operational machine learning pipeline for global agriculture mapping and monitoring.

While the Grow by Numbers challenge is based around agriculture, the results will influence a wide array of geospatial machine learning applications, such as disaster damage assessment, poverty mapping, and flood mapping.

Learn More

The challenge will run through August 3rd. To learn more about the competition, read the rules, and participate, visit the Grow by Numbers Challenge website.