Predicting Wildfires with Neural Networks: An Approach to Preventing California Wildfires
CSEF · 2019 Earth & Environmental Sciences
Overview
Objectives In recent years, wildfires have caused billions of dollars worth of damage across the state of California. An increase in frequency and intensity of these wildfires has made it difficult for emergency managers and fire officials to plan for them. Scientists are currently looking for new ways to predict the occurrence of wildfires. The objective of this project was to develop a machine learning algorithm to predict the risk of a wildfire occurring within a given area. Methods Data of temperature, vegetation, evapotranspiration, and historical wildfire incidence was obtained from the MODIS satellite using the Google Earth Engine API. A deep neural network programmed using the Keras library in Python was trained on the data. The neural network consisted of three hidden layers, each with 128 neurons. The output layer had a sigmoid activation function. Results The neural network achieved up to 77.87% accuracy after 16 epochs (or passes over the dataset while training), and 65.99% validation accuracy after 20 epochs. The accuracy stayed approximately the same throughout the experiment, and the validation accuracy stayed in the range of about 55-65%. Conclusions The neural network showed signs of overfitting, a phenomenon where the neural network fits the training dataset too well. The difference in accuracy and validation accuracy is evidence that overfitting occurred. The overfitting was most likely caused by a lack of factors in the dataset. In a future experiment, if more factors were used than just temperature, vegetation, and evapotranspiration data, the accuracy and validation accuracy of the network might be closer to each other.
Summary statement
I developed and trained a neural network on environmental variables from a satellite, and it predicted wildfires with moderate accuracy.
Help received
I collected the satellite data and built the neural network myself. Nick Nauslar, a fire weather forecaster with NOAA, guided me on the MODIS satellite and the data it provides. Canyon Robins, a computer science major at Stanford University, told me about Google Earth Engine and gave me advice on how I
Competition history
- CSEF 2019
Resources
Related projects
ISEF · 2018
Satellite Modeling of Wildfire Susceptibility in California by Using Artificial Neural Networking
CSEF · 2019
Application of Meteorological Data to Predict the Chances of a Forest Fire Using Machine Learning and Neural Networks
ISEF · 2022
It's Flaming Out: Using Artificial Intelligence To Emulate Critical Aspects of Wildfire Growth
ISEF · 2021
An Accessible, Low Cost Tool for Citizen Scientists: Using Remote Sensing Techniques to Predict Fire Damage Propensity
ISEF · 2023
Developing a Deep Learning Model to Approximate the Convolutional Geometry of Wildfires
CSEF · 2023
A Novel AI Pipeline for Predicting Next Day Wildfire Spread
ISEF · 2021
FyreWatch: Deep Learning for Accurate Wildfire Environmental Conditions Detection
JSHS · 2024
Predicting Next-Day Wildfire Spread with Environmental Data and Machine Learning
Closest projects by meaning, across every fair and year in the corpus.
Browse more like this
Source: California Science & Engineering Fair public projects