Nowcasting Wildfire Ember Risk in the WUI with WindTL

Nowcasting Ember Spread in the Wildland–Urban Interface with WindTL

The increasing frequency and intensity of wildfires in the Wildland–Urban Interface (WUI) have highlighted the need for more advanced predictive tools to manage and mitigate fire risks effectively. The WindTL tool represents a significant breakthrough in operational wildfire ember spread prediction, offering unprecedented accuracy in forecasting the ignition and spread of spot fires—a critical factor in wildfire severity and responsible for 90% of structural losses during such events. This model leverages cutting-edge technologies, combining physics-based and machine learning techniques to deliver real-time risk assessments crucial for fire management and emergency response.

WindTL operates by aggregating in-situ, real-time data from a diverse array of sensors and sources, including Commercial Off-The-Shelf (COTS) sensors, Internet of Things (IoT) devices, Unmanned Aerial Systems (UAS), satellite imagery, and third-party meteorological data. By integrating these inputs, WindTL creates a comprehensive picture of the environmental conditions that influence fire behavior. The model's core functionality is divided into two components: an ignition model that identifies regions vulnerable to fire initiation and a fire propagation model that predicts the spread of spot fires across the WUI.

At its inception, WindTL initializes its coupled wildfire–atmospheric model using the latest satellite imagery from NASA's Fire Information for Resource Management System (FIRMS), atmospheric forecasts from the High-Resolution Rapid Refresh (HRRR) model, and in-situ observations collected from UAS and ground sensors. WindTL further integrates topographical and fuel data from LANDFIRE to account for the terrain's influence on fire spread. WindTL utilizes WindNinja's physics-based mass-conserving approach to estimate high-resolution wind speed and direction according to the terrain topography. This refinement is crucial for accurately predicting ember transport and the potential for creating new spot fires.

The fire propagation model includes Rothermel's surface fire spread equations via FlamMap to determine fire behavior attributes (rate of spread, flame length, crown fire spread) and to estimate ground fire perimeter progression. These data are used as inputs into Albini's spotting model to identify areas with a higher likelihood of ember transport and landing. For regions with potential ember landings, a neural network is employed to assess the likelihood of ignition. This machine learning component of WindTL continuously improves through real-time training using heat signature data obtained from UAS, enhancing the model's predictive accuracy over time. As the wildfire evolves, WindTL reruns its predictions, refining estimates for fire perimeter progression, rate of spread, and flame length based on the latest UAS observations.

WindTL's predictive capabilities are rooted in extensive training on field data, such as that from Page et al. (2019), which encompasses a wide range of variables, including meteorological conditions (wind speed, humidity, temperature, shading), fuel models (fuel load,, fuel moisture, canopy height, canopy bulk density), and topographical features (slope, aspect, elevation). The model's output is a dynamic risk map that captures real-time changes in fire behavior, such as wind gusts and shifts in wind direction. This risk map is vital for effective wildfire management and emergency planning, providing actionable insights into ember dispersal and potential ignition points.

The risk map generated by WindTL can be visualized through Improving Aviation's web-based interface, the Android Tactical Awareness Kit (ATAK) application, or shared with third-party systems in Geographic JavaScript Object Notation (GeoJSON) format. This standardized format ensures compatibility with other Geographic Information System (GIS) tools, facilitating seamless integration into existing emergency management frameworks.

Looking ahead, WindTL's development is focused on enhancing its risk maps by incorporating severity assessments. By integrating large repositories of geographical vectored data, the model will be able to account for critical infrastructure, vulnerable populations, and other key factors in its risk predictions. This evolution will enable WindTL to not only predict the likelihood of ember ignition with unparalleled accuracy but also to assess the potential severity of fire impacts. The result will be a comprehensive risk management system that provides emergency responders and decision-makers with the tools they need to protect lives and property in the WUI.

The WindTL tool has been developed under multiple SBIRs from NASA and NOAA and we now hold numerous partnerships established with providers of wildfire services that are conducting testing and validation of our model, evaluating WindTL in historic and active fires. Through these collaborations, we aim to earn the trust of operational partners and demonstrate the effectiveness of the ember spread model, ultimately equipping wildland firefighters and similar stakeholders with the WindTL tool in active, operational environments.

In conclusion, the WindTL tool stands at the forefront of wildfire prediction technology, offering a robust, data-driven approach to managing fire risk in the WUI. Its combination of physics-based models and machine learning techniques enables it to deliver highly accurate predictions that are crucial for real-time wildfire management and emergency planning. As the model continues to evolve, it promises to become an even more indispensable tool in the fight against wildfires, providing a comprehensive understanding of both the likelihood and severity of fire events.