A Novel AI Pipeline for Predicting Next Day Wildfire Spread

Overview

Wildfires cause significant damage, resulting in over 33,000 fatalities and destruction of more than 10.1 million acres annually. 40% of wildfire damage occurs in California. Unfortunately, firefighters often face challenges in containing the fires due to limited information and difficulty in predicting their direction. To address this issue, I developed a novel AI pipeline that can predict the direction of wildfires over a 24-hour period using a combination of current fire information and meteorological data. This approach enables firefighters to focus their resources on specific targets, leading to more efficient resource allocation and a higher likelihood of containment. To develop and test the pipeline, I utilized a publicly available dataset containing information on 18,445 grids of 64kmx64km per grid, including a mask describing fires at 1kmx1km resolution. The dataset also included a variety of relevant factors such as elevation, wind direction, wind speed, temperature, humidity, precipitation, drought index, vegetation, population density, energy release component, and previous fire mask. Using K Nearest Neighbors and Random Forest algorithms, I hyper-parameter tuned each algorithm on an 80/20 split of training and validation data. Afterward, I assessed the accuracy and created a confusion matrix based on the data. The results were promising, with Random Forest algorithm performing the best, achieving up to 95% accuracy when using only previous day fire status. The accuracy increased to 97% when accounting for other parameters and neighboring grid square status. These findings demonstrate that machine learning can effectively predict the direction of wildfires.

A Novel AI Pipeline for Predicting Next Day Wildfire Spread

Overview

Source coverage

Awards (1)

Competition history

Resources

Related projects

Predicting Next-Day Wildfire Spread with Environmental Data and Machine Learning

It's Flaming Out: Using Artificial Intelligence To Emulate Critical Aspects of Wildfire Growth

A Multi-Modal Wildfire Prediction and Personalized Early-Warning System

Predicting Wildfires with Neural Networks: An Approach to Preventing California Wildfires

Developing a Novel Wildfire Spread Prediction Model Using Spatio-Temporal Analysis and Time-series Forecasting

Developing a Deep Learning Model to Approximate the Convolutional Geometry of Wildfires

BlazeAlert: A Novel Spaciotemporal Deep Convolutional Neural Network Model for Real Time Wildfire Risk Assessment

Application of Meteorological Data to Predict the Chances of a Forest Fire Using Machine Learning and Neural Networks

Browse more like this