Chesapeake Conservancy is an environmental group that fights for the conservation of the entire basin of the bay with the same name. Two years ago, featured one of the highest resolution maps ever made in the US, of a territory that covered 259,000 square kilometers, with 207 cities and rural counties.
This map was based on updated satellite images and had a resolution of one meter, compared to the previous one it was from 2011 and its resolution was 30 meters: this meant that a single pixel of this map no longer represented an entire farm, but allowed to identify individual crops.
The map consisted of 80,000 cells, each equivalent to about 33.7 square kilometers, which after being analyzed by a computer, were reviewed one by one by humans to classify ground cover (vegetation, water, concrete, etc.), a task that required a team of 30 people working for 10 months.
Algorithms to identify crops, lakes and roads
The computerized analysis of the map was carried out using the open source tool Microsoft Cognitive Toolkit 2.0 and the Azure cloud, thanks to the collaboration between environmentalists and the AI for Earth program launched by Microsoft.
This has enabled the Chesapeake Conservancy to use software capable of analyzing and identify each element of the map thanks to machine learning: “Your brain contains an algorithm that has been trained to identify images: ‘That’s a tree, that’s a car, that’s a boat,” explains Microsoft’s environmental manager Lucas Joppa.
In the Microsoft Environment blog, they summarize the challenges they face when trying to make these kinds of maps:
“High-resolution, real-time land cover maps can guide conservation efforts, but creating these maps using available imagery and tracking changes over time requires complex algorithms and computing resources.[…] able to keep pace with the increasing speed of data collection. “
Thanks to what they learned working with the map of the bay, Microsoft was able to use its Brainwave Project to process 200 million satellite images in just 10 minutes (and for a cost of less than € 40) and immediately generate a draft of the land cover map of the entire United States.
These high-resolution maps will now allow environmental groups to assess changes in the environment at a much higher frequency, and focus where changes are fastest and most aggressive. That’s why the Chesapeake Conservancy is now collaborating with groups in other regions of the US (Iowa, Arizona, the Great Lakes) to generate new maps, thanks to the processing power that Microsoft technology gives them.
However, not everything learned in the Chesapeake is exportable to other areas: for example, the machine learning algorithm, trained in analyzing the blue and green tones of the Chesapeake basin, did not show good performance when facing beige and sandstone colors that dominate the arid landscape of Arizona.
