WG on Landcover Classification

WG on Landcover Classification

Starts: Thursday August 11 at 6 - 7 PM EDT
Satellites can provide detailed snapshots of the Earth’s surface. Here we will take what we learnt in the previous iterations of this Working Group & build an improved landcover classification model to identify the different land classes in Canada.
Prior experience using Python is required for this project. We recommend that user’s are also familiar with scikit-learn, Tensorflow and Numpy for ML-related modelling tasks. No previous remote sensing knowledge is needed for this project.


Top performers of previous cohort!
notion image
Fynn Davis | Lead ~ BSc. (Ecology), Dip. (Data Science)
notion image
Leah Lourenco | Lead ~ BSc., MSc., Dip (Data Science) & Entrepreneur

Based on the works of … in the first Remote Sensing Working Group

notion image
~ Fraser King | Lead PhD.c @ University of Waterloo, Canada
notion image
Dr Karen Smith | Advisor ~ Assistant Professor, U of Toronto
notion image
Dr Andre Erler | Advisor ~ Senior Climate Scientist at Aquanty
❗ Meeting Time: 7:30 PM EDT on Thursdays (starting the 11th of August)
❗ Slack channel: Join here [If issues, drop ammar@ai.science an email]

Project Overview

  • Learn how to access, download and align remotely sensed Earth imagery (Sentinel-2, Sentinel-1 etc.) for industry ML applications.
  • Stretch goal: Creation of sufficiently accurate classifier for industry application

Work Architecture

notion image

Tentative Timeline

Major Milestones
Expected Finish
Get familiar with domain and previous work. Core Team skimmed out.
3 weeks
Download sample data, perform feature extraction and run baseline model using existing code
1 week
Collaborative optimization [playing with classes, features and models for improvement]
2 weeks
[Industry application]
2 weeks
  • To join the Core Team at the end of session #3 [Be Active to get shortlisted!]: Apply here
  • Time commitment: 3-5 hrs a week.
  • Even if you don’t make the Core Team, you can still stay involved as a spectator.
  • But only the Core Team will be spotlighted at the group’s conclusion.


  • This working group will build on skills and techniques created during the first land cover classification working group to develop an accurate and robust classifier that takes as remote sensing data as well as other data sources participants are able to integrate.
  • Classification uses NRCan 2015 land cover classes, at a resolution of 60 m. Initial data is based on Sentinel-2 satellite imagery however other sources are encouraged. We have included an example of what the downloaded S2 data will look like from Sentinel-Hub here: https://i.imgur.com/t6blupG.png
  • Unlike the previous efforts, this working group is less of a competition and more focused on collaboration. We encourage participants to share code, discoveries and notebooks throughout the project to maximize the progress we can all make in a short period of time.
  • For testing purposes, the expected input for your models will be a dataset where each column is a feature and each row is a pixel. In the cases of where the model requires the pixels in the original structured image shape code should be included to reshape from the dataset.
  • If you have any questions about the competition, please let us know. Good luck everyone!

Why join?

Aggregate Intellect hosts one of the most diverse ML communities in the world. Over the course of the working group
  • You’ll get an immersion into that community & walk out with some cool new friends.
  • Advance your ML skills by working on real world problems with classification algorithms of increasing sophistication
  • Contribute to a study area (land cover classification) which has major impacts to resource management practices, wildlife habitat protection and in advancing our understanding of the Earth’s biophysical systems
Looking forward to meeting everyone in our working group! Please feel free to reach out if you have any questions about the planned project.

This effort is being sponsored by our friends at

notion image