Page 147 - AI for Good-Innovate for Impact Final Report 2024

AI for Good-Innovate for Impact

Use case – 33: Embrace the Forest 33-umgrauemeio

Country: Brazil

Organization: umgrauemeio

Contact person: Osmar Bambini, osmar.bambini@ umgrauemeio .com

33�1� Use case summary table

Domain Management of wildfires

Problem to be A holistic, multi-stakeholder approach, adding high-end tools and
addressed technologies, respecting local knowledge, and cultural and biological
diversity.
• Currently the fire management approach is reactive.
• Loses time, effort.
• Emission reduction from wetland.
Key aspects of the Predictive approach by managing fire, avoiding fire,
solution Prevention, response.
In line with local knowledge.
Community empowering technologies
Early detection is key to extinguishing fires.
Used LabVIEW in the past (until 2020) earlier / switched to Python

Technology NASA, Geographic positioning of communities,
keywords Power lines (are shut many times due to wildfire, 30% of the cause is
wildfire), this affects the energy utilities availability KPIs.
Computer vision (cameras based on python, identify smoke, 20 km
range) and satellites.
Satellite detection (has limitations, latency between detection and
action, satellite have diverse capabilities, coordination of the satellite
orbit is not always possible).
Satellites can help in fire propagation model. (French collaboration)
Post fire analysis.

Data availability Most of the image data is private. Link
Satellite detection data can be shared.
Metadata (type of Streamed videos, detect smoke and light, Satellite images are used
data) only for post-fires, Sensors from satellite which can detect heat (diverse
generation of sensor).

Model Training Anomaly detection for smoke and light.
and fine tuning No opensource algorithms.

131