How machine learning helps students fight wildfires
Wildfires destroy property and claim hundreds of lives every year, costing Americans billions of dollars in damages.
Two high school students from Monta Vista High School in Cupertino, California, are predicting wildfires using machine learning. They recognized that current systems used by the fire department are inefficient and time-consuming.
With support from California Department of Forestry and Fire Protection (Cal Fire), the students have successfully predicted wildfires with 89 percent accuracy.
Fire departments have ready tools that measure factors affecting wildfires, such as wind speed, wind direction, humidity, and temperature.
However, measuring biomass was more challenging. Fire prevention crews need to physically visit forest sites to collect samples to classify manually what they call dead fuel.
Dead fuels are branches and leaves that have 0 percent moisture content. This makes fires start easily and spread rapidly, as there is no water to slow down the burning process.
Aditya Shah and Sanjana Shah relied on a network of sensors to collect these key data. The Smart Wildfire Sensor devices were used to identify and capture images of biomass accumulated on the forest ground.
Using Google’s TensorFlow, the students were able to estimate the moisture content and size of biomass from the images. This allowed them to predict the amount of dead fuel present, and thus calculate the likelihood of wildfire in a forest.
With the Smart Wildfire Sensor devices, fire departments can spend less time manually collecting samples. Machine learning provides a more accurate estimation of biomass build-up, allowing fire departments to spend more time taking preventative measures.
This means fires will not spread as quickly or as aggressively, thus reducing costs needed to fight fires. Most importantly, it will help protect homes and lives.
The students have worked with Cal Fire to implement the devices this in three counties, covering major areas prone to wildfires. The devices have since captured enough images to train the machine learning model, with predictions boasting accuracy up to 100 square meters.
They will be improving prediction accuracy further using ground and aerial drones to capture more images of biomass.
If their system is widely rolled out across the US, this can save Americans billions of dollars, and protect more lives.
26 February 2024
26 February 2024
22 February 2024