Adversarial AI Model for Fact-Checking Wildfire Chatbot Answers, Reliability, and Accuracy

Authors

  • Ayan Patel Center for Spatial Information Science and Systems, Department of Geography and Geoinformation Science, George Mason University, Fairfax, VA
  • Rohit Kamath Center for Spatial Information Science and Systems, Department of Geography and Geoinformation Science, George Mason University, Fairfax, VA
  • Nabiha Sharif Center for Spatial Information Science and Systems, Department of Geography and Geoinformation Science, George Mason University, Fairfax, VA
  • Abigail Lei Center for Spatial Information Science and Systems, Department of Geography and Geoinformation Science, George Mason University, Fairfax, VA
  • Meghana Koramutla Ramesh Center for Spatial Information Science and Systems, Department of Geography and Geoinformation Science, George Mason University, Fairfax, VA
  • Ziheng Sun Center for Spatial Information Science and Systems, Department of Geography and Geoinformation Science, George Mason University, Fairfax, VA

Abstract

Wildfires pose a severe threat to both the environment and human health, causing over one
billion dollars in infrastructure damage annually. Wildfire smoke contaminates the air with
hazardous pollutants such as lead, exacerbating the risk of cardiovascular and respiratory
diseases. This study compares the accuracy of our wildfire prediction model with that of the
existing WildfireGPT, using actual values from the NASA Fire Radiative Power (FRP) and
MODIS MOD14 Wildfire datasets. A comprehensive dataset, including features such as the
Fire Weather Index (FWI), Vapor Pressure Deficit (VPD), temperature (T), and pressure (P),
was utilized to evaluate the accuracy of our prediction model. The analysis specifically
compares the predicted FRP values against observed data to assess the model’s performance.
This research aims to significantly impact wildfire prediction by providing a detailed
comparative analysis that can guide future improvements. Accurate wildfire prediction is
crucial for saving lives and protecting natural habitats. The study highlights the ongoing need
to develop and refine predictive technologies, aiming for enhanced accuracy and usability in
future applications. It also contributes to a deeper understanding of how improved wildfire
prediction can advance management practices.

Published

2024-10-13

Issue

Section

College of Science: Department of Geography and Geoinformation Science