Catastrophe modeling for a resilient future—powered by AI

Around the world, extreme weather events and natural catastrophes are reshaping the risk landscape, with stakeholders facing mounting pressure to respond more rapidly and accurately to protect communities and maintain financial stability.

To meet the growing need for faster, more precise risk insights, catastrophe models continue to serve as the scientific backbone of risk assessment, with artificial intelligence (AI) emerging as a powerful accelerator to enhance their speed and impact.

The role of catastrophe models

Catastrophe models are science‑based frameworks designed to quantify the severity, frequency, and potential financial impact of natural disasters. They enable a spectrum of stakeholders, including insurers and reinsurers, to make informed decisions about underwriting, portfolio management, and reinsurance strategies.

At their core, these models combine the following five key components:

• Event set: A library of simulated and historical catastrophe scenarios that defines the location, frequency, and severity of events, providing a foundation for deriving hazard, vulnerability, and loss calculations.
• Hazard: The probability and intensity of natural catastrophe events such as hurricanes, floods, or wildfires.
• Vulnerability: How susceptible structures are to damage.
• Exposure: Assets at risk, including properties and infrastructure.
• Financial module: Translates physical damage into financial outcomes, with metrics such as probable maximum loss (PML), average annual loss (AAL), and return‑period losses, providing stakeholders with a financial perspective on risk.

By simulating thousands of plausible scenarios, catastrophe models provide critical insights ranging from pricing risk, setting reserves, and managing capital.

Model accuracy depends heavily on the quality of input data and the efficiency of processing workflows, and in both these areas, AI is making a transformative impact.

Figure 1: Catastrophe modeling general framework

Where AI adds value

1. Event response, post-event reconnaissance, and validation data

When a catastrophic event occurs, post-event surveys have traditionally required on-the-ground teams, a process that can often be slow and hazardous. Using post-event satellite and aerial imagery, AI-driven algorithms can rapidly detect and quantify structural damage, revolutionizing how we assess damage after catastrophic events.

Example 1: AI-enhanced structure verification and classification after the Los Angeles County 2025 wildfires

In the aftermath of the Los Angeles County wildfires, AI played a pivotal role in identifying not just the extent of the damage, but the type of structures affected; a critical distinction for insurers when responding to early claims signals.

Moody’s AI-powered image analysis compared pre-event satellite and aerial imagery with post-event images to determine whether a structure was fully destroyed, partially damaged, or even untouched.

Beyond detecting damage levels, the models were trained to distinguish between primary residences and appurtenant structures such as sheds and garages.

This distinction between structure types proved essential for analyzing wildfire footprints where multiple buildings occupied a single parcel.

In many cases, outbuildings were lost while the main home remained intact, a scenario that, without AI, could lead to overstated early loss estimates or misalignment with eventual claims.

The LA event demonstrated three high-value outcomes for our clients of AI-enabled structure typing:

• More accurate representation of insurable losses to inform the claims response, avoiding the double-counting of non-residential structures.
• Faster reconciliation between modeled losses and real claims activity, improving event response activities.
• Clearer granularity, enabling an understanding of which structures were truly impacted to focus event response activities on the right areas.

By combining damage detection with structure classification, AI enables a more reliable view of wildfire impacts, strengthening the accuracy of early loss guidance and model validation.

Figure 2: Los Angeles 2025 wildfires structure damage identification from pre‑ and post-event imagery.

Example 2: Hurricane Ian (2022)

Hurricane Ian made landfall in southwest Florida in September 2022 as a powerful Category 4 storm, bringing catastrophic storm surge, destructive winds, and extensive structural damage across Lee County and surrounding areas.

The scale of destruction, combined with widespread power outages and flooding, made traditional on-the-ground assessments difficult in the early days following landfall, creating an urgent need for rapid, remote damage intelligence.

The image below (see Figure 3) illustrates how AI applied to post-event satellite imagery enabled property-level damage assessment in affected neighborhoods.

Each building footprint is outlined and classified using computer‑vision models trained to detect structural loss patterns: red indicates visible damage, while green indicates structures with no observable damage.

This approach, coupled with catastrophe modeling, provides stakeholders with a fast, consistent view of damage severity across entire communities, supporting more accurate early loss estimates and claims planning.

Figure 3: Imagery showing Hurricane Ian wind damage detection

Example 3: Water and wind damage caused by Hurricanes Helene and Milton (2024)

With Hurricane Helene making landfall as a Category 4 hurricane on Florida’s Gulf Coast on September 27, 2024, and Hurricane Milton, a Category 3 storm making landfall further north up the coast two weeks later, AI was used to classify damage from water versus wind, and distinguish partial from total structural loss, enabling informed decision-making and faster insured loss estimates. 

Figure 4: Satellite-based AI classification of damaged and undamaged structures

2. Enhancing model input data

Accurate inputs are the foundation of reliable catastrophe models. AI enables the use of enhanced digital terrain models (DTMs) and land-use/land-cover data. These DTMs are critical for natural catastrophe modeling, for extracting elevation, surface roughness, vegetation, and patterns in the built environment with far greater precision. These richer inputs reduce uncertainty and help ensure that modeled losses more closely reflect real-world behavior during extreme events.

Building on these enhanced hazard and terrain inputs, AI also strengthens the exposure side of the model through high‑resolution property intelligence. Moody’s use of CAPE Property Intelligence deploys computer vision and machine learning to help deliver property-level risk attributes, roof condition, vegetation coverage, and defensible space at scale.

These attributes correlate strongly with insurance losses, enabling address-level risk scoring and more precise vulnerability assessments. By integrating Moody’s catastrophe risk modeling with geospatial AI through our new Enhanced Exposure Enrichment capabilities, insurers gain more granular insights that improve underwriting accuracy and portfolio resilience.

3. Efficiency gains

Workflows that had traditionally consumed significant time and resources are now being streamlined using AI:

• Automating repetitive tasks: Data cleaning, feature extraction, and parameter tuning.
• Optimizing algorithms: Identifying key variables to provide faster simulations.
• Real-time data processing: Updating models during live events for immediate decision support.

These efficiencies reduce computational overhead and accelerate the delivery of actionable insights.

Why it matters

AI doesn’t rewrite the science behind catastrophe models, but it amplifies their impact. By improving input fidelity, automating processes, and enabling rapid post-event analysis, AI helps insurers respond faster, manage risk more effectively, and develop policy solutions that help close the protection gap.

In an era of escalating natural catastrophe risk, speed to insight is not a luxury; it’s a necessity.

Limitations and reality check

Catastrophe models remain the cornerstone of risk quantification, with AI serving as a critical accelerator. While in these examples, AI is enhancing workflows, it does not directly price risk.

Risk pricing will always involve additional factors such as administrative costs, profit margins, and regulatory considerations.

Closing thoughts

AI is not a silver bullet, but it is a powerful accelerant. By combining the rigor of proven catastrophe science with AI‑driven data enrichment, automation, and event intelligence, Moody’s helps insurers make faster, smarter decisions in a world facing climate risk uncertainty.

At the same time, it’s important to remember that the insurance industry is fundamentally concerned with extremes; the rare, high‑impact events that sit in the tails of the distribution.

Most AI methods aren’t designed for these edge cases by default, so careful evaluation and validation are essential to ensure meaningful, reliable outcomes.

Used responsibly and in partnership with scientific modeling, AI can enhance, not replace, the discipline required to understand and manage catastrophic risk. Together, they enable a more resilient future.