Change Detection for Land Cover Mapping in the areas surrounding the Murchison Falls National Park

Background: 

This project is one of the 11 selected worldwide under National Geographic and Microsoft’s AI for Earth RFP for 2019. Our project seeks to provide information that will help answer some of the questions about why and how oil development has and will affect ecologically sensitive areas like the Murchison Falls National Park and Lake Albert Delta. We are interested in the latter park because it’s the oldest, largest and most ecologically biodiverse in the country.
Moreover, only a few months into this project, the government of Uganda publicly announced its plans to build a new dam over the Nile, which is partly housed in the park. The public and related institutions strongly objected to this new development, sighting a range of counterproductive reasons. This makes our work in the study site increasingly relevant.
This project therefore demonstrates how AI, applied to geography and remote sensing data, can be used to extract critical information from a changing and dynamic system to support conservation, Agricultural and Biodiversity management in ecologically rich and environmentally sensitive areas, towards a more sustainable planet.

Goals: 

Based on contextual and qualitative information, Our project goals are

  1. Detect a decade of land cover change in the area surrounding the Murchison National Park and Lake Albert delta.
  2. Quantify a decade of land cover change in the area surrounding the Murchison National Park and Lake Albert delta and
  3. Monitor land cover change over the next decade.
Facts/Method: 

We used satellite imagery from two sensors of the Landsat imaging program. Going forward into the next decade, we’ll lookout for Landsat 9, planned for 2023. However, due to cloud cover, two smaller cloud free areas around the study site were selected. The bulk of training data was obtained from the satellite images from the two selected sites. In addition, ground truth data was used as test data for model tuning. From the AI side of things, we used a supervised decision tree classifier and documented the data and model on Microsoft Azure’s AI + Machine Learning workspace. The maps below show a subset of team efforts over the last few months.

Notes: 

Note: North is up for all the maps and images

Comment: 

This page briefly documents team efforts on the project over the last couple of months. Project results at this point may contain uncertainty resulting from automated imagery analysis. You shouldn’t rely upon the information on the website as a basis for making any business, legal or any other decisions.

Acknowledgement: 

This work was supported by National Geographic Society and Microsoft’s AI for Earth Program. The brief account of our work is written by Ketty Adoch.

Publications: 

Coming soon.