TRANSFORMATIVE AI FRAMEWORK FOR WILDFIRE MANAGEMENT

Abstract

The purpose of this study is to investigate the methods for AI transformation in the Fire industry, focusing on predicting wildfire characteristics such as wildfire spread and contextual information related to the fire scene. To achieve this, the study utilizes numerical data from various satellite data sources, as this data platform enables scalability for AI-integrated real-world applications on a global scale. The research employs various unsupervised machine learning algorithms on unlabeled data. It proposes new clustering algorithms that predict wildfire characteristics, including contextual information like the threat level to the nearest residence.
This study addresses the challenges associated with integrating AI predictions into real-world applications on time. It proposes a system software architecture designed to effectively schedule GPUs by logically grouping data points from various sources, ensuring that accuracy is maintained. The application is divided into these groupings, enabling the computation of AI predictions from a regional to a national level.
The approach involves continuously acquiring real-time raw fire event data from various sources, including MODIS and VIIRS, and utilizing a deployable cloud platform. This platform is designed to constantly preprocess real-time data and compute predictions using the selected unsupervised algorithms. The study then illustrates the method for exposing the algorithm's predictions and integrating them into a visualization system within a real-world application, ensuring that the information is readily accessible and usable by the firefighting community for effective decision-making and management during wildfire incidents. This research makes a substantial contribution to the business sector by enhancing the management capabilities of firefighting organizations.