Free Artificial Intelligence Competition to Tackle Environmental Pollution Challenges

Zindi hosts the Urban Air Pollution Challenge, an open learning competition for participants worldwide to predict daily PM2.5 concentrations in cities using weather and satellite observations. The challenge has no closing date, and participants may compete individually or in teams of up to four.

About the Urban Air Pollution Challenge

Participants use weather data and daily pollutant observations from the Sentinel 5P satellite to predict PM2.5 particulate matter concentration for each city and day. PM2.5 is a commonly used measure of air quality that normally requires ground-based sensors. The competition data covers the last three months represented in the dataset and spans hundreds of cities around the world. The task is intended to help track air quality and its changes, including in places without ground-based sensors.

Entries are evaluated using Root Mean Squared Error. Submission files must follow the supplied sample format, with one column named Place_ID X Date and another named target for the predictions. Participants can enter alone or form a team of no more than four people. When a team is created, its total submission count must be no greater than the maximum allowed at that time. The team receives the competition submission allowance minus the highest number of submissions already made by any member before the team was formed.

This is a learning competition, not a prize competition. There are no prizes beyond the knowledge gained, but participants receive 25 points for their first submission and 50 points for their first non-sample submission. A maximum of 10 submissions per day is allowed. A submission counts only after it passes processing; a failed submission does not receive a score or count against the daily limit.

Solutions must use publicly available open-source packages and only the datasets provided for the challenge. External data and automated machine-learning tools are not allowed. Pretrained models may be used if they are openly available to everyone. Competition data must not be redistributed or made available to people who are not participating. Work on a cloud platform is permitted only if the data remains private and the use complies with Zindi’s rules.

Multiple accounts are prohibited, as is collaboration or membership across multiple teams. Code must not be shared privately outside a team; code that is shared must be made available to all participants through the platform. The organizers may disqualify individuals, teams, accounts or submissions for rule violations. The challenge will not close, although its timeline may be updated. Explore more competitions and awards or browse related AI opportunities.

Who Should Apply

The challenge is suited to people interested in prediction, data analysis and applying computational methods to air-quality questions. It is open to participants in any country and welcomes both individual entries and teams of up to four. Applicants should be prepared to follow the dataset, submission and collaboration rules. The competition page also provides guidance on getting started with Zindi challenges, forming a team and running notebooks in Colab.

Key Highlights

• Online competition hosted on Zindi.
• Predict daily PM2.5 concentrations for cities around the world.
• Use the competition’s weather and Sentinel 5P satellite data.
• Scored with Root Mean Squared Error.
• No closing date and no prizes beyond learning and platform points.
• Individual participation or teams of up to four people.