Transforming How We Prevent and React to Natural Disasters

Bridging the Gap Between Science and Response

Wildfires, floods, hurricanes — the cost of natural disasters is rising, and so is their complexity. Yet across the globe, our response remains largely reactive. We wait for imminent risk and consequences to act on. We analyze the aftermath instead of anticipating the threat. In too many cases, our tools are built to understand what happened — not what’s about to. We believe it’s time to shift from reaction to prediction and prevention, from delayed awareness to real-time situational intelligence. 

At SkyTL, we’re closing the gap between cutting-edge science and real-world decision-making. We bring research-grade prediction models into the hands of those who need them most — firefighters, dispatchers, and emergency planners — through intuitive, real-time, and field-ready technology. Disasters may be inevitable — but their worst outcomes don’t have to be. With SkyTL, we’re not just observing and analyzing. We’re staying ahead of it.

The Problem: Outdated Tools in an Escalating Crisis

In the chaos of a natural disaster, every second counts — yet responders are often forced to make critical decisions without real situational awareness. Most tools available today operate reactively: they analyze what’s already happened instead of predicting what’s about to unfold.

This reactive approach stems from a fundamental disconnect between science and real-world operations. While advanced research has produced powerful modeling tools, these tools rarely make it into the field due to three systemic barriers:

📊Lack of Accurate and Timely Input Data

The behavior of natural disasters is inherently sensitive to local conditions — wind speed, humidity, vegetation , and terrain — but most existing models rely on coarse or outdated data. Weather stations are too sparse, satellite revisit times introduce lag, and manual data entry slows everything down. Without real-time, hyperlocal data, predictions lose their edge.

💻 Dependence on High-Performance Computing

Many of the most accurate simulation tools are computationally intensive, requiring HPC clusters to run. That makes them impractical for field use, where insights are needed in minutes, not hours. As a result, even well-funded agencies often resort to simplified models or wait on slow processing cycles that can’t keep up with a fast-moving threat.

📱Lack of Intuitive User Interfaces

Even when good data and models exist, they are often locked behind complex interfaces designed for researchers, not responders. Field personnel need tools that work under pressure — that can be used with a touchscreen in a fire truck or a tablet in the field. Yet, most academic tools require technical expertise to operate, making them unusable at the edge.

❗The consequence:  Emergency managers operate with incomplete information, juggling multiple disconnected systems and relying on intuition rather than intelligence. Critical threats — like embers crossing containment lines or shifting wind conditions — are missed until it’s too late.

The SkyTL Solution: Turning Data into Decisive Action

SkyTL is built from the ground up to solve this. It’s not just a prediction model — it’s a modular risk platform that fuses real-time data, physics, and artificial intelligence to deliver predictive situational awareness.

1. Real-Time Data Ingestion and Fusion

SkyTL continuously ingests high-resolution data from:

• Satellites: MODIS, VIIRS, GOES for thermal anomaly detection and fuel state.

• IoT Sensors and Weather Stations: To localize wind, humidity, and fuel moisture levels.

• UAS: For low-altitude reconnaissance and thermal imaging.

• User Inputs: Allowing users to input data that is fed into the predictions models.

• Historical Data Pipelines: Built using LLMs to extract insights from fragmented datasets, helping train AI models on fire patterns, behavior, and ignition triggers.

This ingestion system is built on a cloud-native, scalable architecture using services like Google Cloud Pub/Sub, Dataflow, and BigQuery for real-time transformation, storage, and analytics.

2. AI Modeling: The Power of Physics + Data

At the core of SkyTL is a hybrid modeling architecture that blends:

• Physics-Informed Neural Networks (PINNs): These deep learning models embed the laws of fluid dynamics and thermodynamics into the learning process, approximating the solutions of high computing requirements software, such as CFD, with accurate answers in real-time.

• Research-grade Open Source Simulators: Academic models (e.g., FARSITE, FlamMap) are integrated into SkyTL but optimized through AI-based calibration.

• Custom ML Algorithms: A combination of proprietary physics and AI models that enhance the latest academic research with information that is most relevant in the field.

• LLM-powered data extractors: Used to compile, clean, and structure vast amounts of unstructured fire event data to continuously improve model accuracy.

This hybrid approach allows SkyTL to:

• Predict disaster spread across natural and man-made barriers.

• Simulate how disasters evolve hour-by-hour under dynamically changing conditions.

• Provide “what-if” scenario generation for proactive decision-making.

3. Cloud Infrastructure and Scalability

SkyTL is powered by a containerized, serverless infrastructure that offers:

• Elastic Scaling with Kubernetes (GKE): Automatically adjusts compute based on active fires or surge in users.

• Low-Latency Prediction: GPU-powered inference pipelines for real-time modeling outputs.

• API-first Architecture: Enables integrations with tools like Team Awareness Kit (TAK), emergency alert systems, or insurance dashboards.

• Infrastructure-as-Code (IaC): Using Terraform to deploy production, staging, and regional instances with consistency.

• Zero Downtime Deployments: Blue-green strategies ensure reliability in real-time deployments when it matters most.

By leveraging the latest Google Cloud technologies, we ensure that emergency responders get high-fidelity predictions without waiting hours for simulation results — even during high-traffic events like mass evacuations.

4. Designed for the Field

Technology only matters if it gets used. That’s why SkyTL was co-designed with dispatchers, firefighters, and utility risk managers over a 2-year pilot with NASA and NOAA funding. The result:

• Web-based UI: Clean, mobile-accessible, intuitive.

• No Technical Training Required: Visual overlays, intuitive toggles, and risk color-coding.

• Live alerts for wind shifts, real-time risk, or proximity to critical infrastructure.

Real-World Impact: Case Studies in Action

WindTL has evolved through rigorous field testing and real-world deployments, where it has repeatedly demonstrated the operational value of combining physics-based modeling with AI and real-time data fusion. Below are two key advancements that showcase the power and practical impact of our technology:

1. Predicting Ember Crossings During the 2024 Thompson Fire

In mid-2024, the Thompson Fire ignited near Oroville, California. Traditional wildfire prediction models projected that the fire would be contained by the river acting as a natural firebreak. Responders, trusting these legacy predictions, initially deployed suppression teams away from the riverbank.

However, a local firefighting unit using WindTL requested an updated risk map. Within minutes, our model simulated a hyperlocal wind pattern and predicted that embers would cross the river within two hours due to strong gusts and topographical funneling effects. WindTL flagged this as a high-risk ember zone.

Thanks to our real-time ember modeling, incident command was able to reassign suppression teams preemptively. The result: the embers did cross the river, but responders were already in position and successfully contained the fire before it reached homes.

✅ Impact: The event validated WindTL’s predictive accuracy in dynamic ember scenarios, and reinforced the importance of real-time, terrain-aware wind modeling — a capability few systems can offer with this level of precision and speed.

2. Reconstructing Ignition Origins in the 2019 Tennis Club Fire

WindTL also proved invaluable in forensic fire analysis. During a test with the Contra Costa County Fire Department, our team simulated the Tennis Club Fire, which originated from a utility-caused ignition. A second fire ignited several miles away, under disputed circumstances. The utility company denied responsibility, asserting that the grid was off in that zone.

Using WindTL, we reconstructed historical wind patterns and ember trajectories. Our model accurately simulated ember spread from the original fire and predicted a spot fire ignition precisely at the second location, matching the known fire perimeter.

✅ Impact: This retrospective validation showed that WindTL can reconstruct ember dynamics with scientific fidelity, providing evidence for risk attribution, liability assessment, and utility defense planning. It’s a breakthrough in using operational tools for post-event analysis, not just forecasting.

These advancements are more than technical milestones — they demonstrate that:

• High-resolution, real-time modeling is not only possible, it’s operationally deployable.

• AI-driven physics models like PINNs can outperform traditional systems in both speed and accuracy.

• Field personnel can use intuitive interfaces to generate actionable predictions within minutes — no technical background required.

What’s Next: Building for the Future

Disaster modeling shouldn’t live in academic papers — it should live in the field. SkyTL continues to evolve, not just because of breakthroughs in AI or cloud, but because we built it in collaboration with the people who rely on it to save lives and property.

Next steps include:

• Expanding our integrations with 3rd party incident management applications.

• Integrating additional risk layers for utility infrastructure, insurance zones, and vulnerable populations.

• Gathering more real-time data in areas without connectivity.

• Deploying a machine learning model that categorizes terrain features using high-resolution imagery.

Final Thought: From Lab to Line of Fire

At SkyTL, we’re redefining what it means to be prepared. Predicting disasters isn’t just about better science — it’s about delivering that science in the moments and places where it matters most. That’s why we’ve engineered WindTL to do more than model risk — we’ve made it intuitive, responsive, and mission-ready.

We don’t just simulate outcomes; we empower decisions. We don’t wait for catastrophe; we work to prevent it.

Whether it’s an ember crossing a fireline, a wind shift threatening a neighborhood, or a utility defending its grid, SkyTL brings real-time intelligence to the frontline — closing the loop between prediction and action. As climate-driven disasters grow in frequency and complexity, our mission is clear:

To give responders, planners, and communities the tools they need to act faster, smarter, and ahead of the threat.

Thanks for reading!