The summer of 2026 has seen extreme weather events emerge with unprecedented frequency.
A typhoon severely impacted Guangxi, affecting 375,000 people, while a subsequent storm with Category 13 winds made landfall in Zhejiang, forcing the emergency relocation of 1.716 million individuals. Since the start of this year's flood season, meteorological authorities have issued warnings for 609 rivers exceeding alert levels, with 14 rivers surpassing recorded historical water level peaks. Concurrently, intense heatwaves have been prevalent. In just the first week of July, areas in China experiencing temperatures of 35°C or higher exceeded 4 million square kilometers, with 15 monitoring stations in Sichuan and Chongqing breaking historical July high-temperature records.
Data released by the National Disaster Prevention, Mitigation, and Relief Commission Office indicates that in the first half of 2026, various natural disasters caused varying degrees of impact to 17.276 million people nationwide, resulting in 155 deaths and missing persons. Emergency relocations and urgent life assistance were required for 587,000 people, with direct economic losses reaching 42.14 billion yuan.
Insurance industry payouts have risen alongside economic losses. As of July 13, 2026, due to recent heavy rain, floods, and typhoon disasters across 20 provinces including Guangxi, Hubei, and Zhejiang, the insurance sector has received nearly 380,000 claims, with estimated losses of 6.38 billion yuan and 2.89 billion yuan already paid out.
As "once-in-a-century" extreme weather events repeatedly strike, becoming "once-a-year" occurrences, and as their frequency increases, historical data is proving increasingly inadequate in the face of climate warming. The fundamental assumptions upon which the insurance industry bases its pricing are being shaken.
When discussing the role of the insurance industry within the disaster prevention and loss mitigation system, one topic is always central: catastrophe models. These models are described as the "core chip" of risk management and form the foundational technology for the insurance sector's core function of risk reduction. As technology advances, catastrophe models are becoming a critical variable in how the insurance industry confronts the challenges of climate change.
How do these catastrophe models function, and how will they, through their own development, enhance the insurance industry's capacity for risk quantification? To explore these questions, an interview was conducted with Fang Jing, Deputy General Manager (Acting Head) and Deputy Party Secretary of China Re Catastrophe Risk Management Co., Ltd.
Backed by China Reinsurance (Group) Corporation, China Re Catastrophe Management, as the country's first fintech company dedicated to catastrophe risk management, has completed the development of three major models: the China Earthquake Model, the China Typhoon Model, and the China Flood Model.
As a leader in model development, Fang Jing believes the primary challenge in catastrophe risk management is the difficulty in quantifying risk, as traditional actuarial methods cannot accurately assess its tail risk. Catastrophe models serve as an effective risk management tool that addresses this challenge.
Understanding the Core Function of Catastrophe Models
To understand catastrophe models, one must first grasp the unique characteristics of catastrophe risk. It features "low frequency but high severity"—major earthquakes and typhoons do not occur often, but when they do, the losses are immense. We often refer to "once-in-a-century" catastrophes, but in reality, China has only several decades of systematic and complete meteorological and disaster records, with precise seismic observation records being even shorter. Traditional insurance actuarial methods struggle to use just a few decades of historical data to project extreme losses over a century or longer cycles.
Catastrophe models solve this problem. They leverage computer simulation technology, integrating multidisciplinary achievements from geophysics, meteorology, hydrology, civil engineering, and insurance actuarial science to generate sets of catastrophe events that could potentially occur over the next several hundred or even tens of thousands of years.
Taking the earthquake model as an example, based on a 100,000-year time window, it simulates millions of seismic events—encompassing both those that have historically occurred and those that, while not recorded, are geologically plausible. The typhoon model operates on a 10,000-year scale, simulating 200,000 to 300,000 typhoon tracks. This data is then combined with information on building distribution, structural types, population distribution, and economic value within a region to conduct a probabilistic quantitative assessment of potential disaster losses.
The Implications of Extreme Weather as the "New Normal" for Insurance
According to the climate trend forecast for the main flood season of 2026 released by the National Climate Center, it is expected that overall extreme weather and climate events will be more frequent during this year's main flood season (June to August). This assessment has been incorporated into the company's risk outlook report.
Meteorological forecasts answer the question of "how much rain will fall," while catastrophe models answer the question of "if this much rain falls, what will the economic losses be?" The catastrophe model itself does not perform rainfall forecasting; it only quantifies losses after inputting extreme precipitation scenarios. It is noteworthy that even if the model's event set includes precipitation scenarios of equal intensity, it is difficult to match the actual spatial distribution of rainfall beforehand due to constraints like local micro-topography and sudden convective activity, leading to potential deviations in loss assessment.
If the judgment that extreme weather is the "new normal" holds true, it implies that the entire logic of risk management within the insurance industry needs to be rewritten.
For the socio-economy, frequent extreme weather means: infrastructure defense standards require re-evaluation, as previous "once-in-a-century" events may become "once-in-fifty" or even "once-in-twenty" year events, rendering original standards insufficient against future climate risk baselines; supply chain resilience faces testing, as a single extreme weather event could disrupt production in a region, transmitting impacts nationwide or even globally; the logic of asset pricing is also changing, with the long-term value of coastal real estate and agricultural production areas needing to factor in climate risk.
For the insurance industry, the changes are even more comprehensive. First, the pricing logic for insurance products has changed. We can no longer use historical average losses as the pricing benchmark; we must incorporate forward-looking judgments on climate trends. Second, product forms need innovation. Traditional annual policies may be unable to cope with the continuous upward shift in the risk baseline, necessitating accelerated promotion of innovative coverage products like index insurance. Third, reinsurance arrangements face restructuring. Pricing discrepancies for climate change risks are widening in the global reinsurance market, requiring more refined ceding strategies from primary insurers. Fourth, insurance companies need to genuinely transform from "post-event compensators" to "pre-event risk managers," which is also the most crucial factor in addressing climate change.
Practical Application and Decision-Making Conversion of Catastrophe Models
China Re Catastrophe Management plays the role of a technology enabler. Its client base covers primary insurers, reinsurers, brokerage firms, government departments, and large enterprises. The core value it provides is "infrastructural capability for risk quantification."
Take a major coastal city's catastrophe insurance project as an example. The project required designing a catastrophe insurance scheme covering typhoon and flood disasters for the city. The company first used the typhoon and flood catastrophe models to conduct detailed calculations of the annual expected loss and probable maximum loss for various return periods (e.g., 50-year, 100-year, 200-year) for different areas of the city, achieving a spatial resolution down to the street level.
Based on these quantitative results, in terms of risk assessment, high-risk areas for urban waterlogging and typhoon winds can be identified, which would require higher premiums or stricter risk control measures. In terms of actuarial assumptions, exceedance probability curves replace the simple historical averages of traditional actuarial methods, providing pure premium references under different coverage levels. In product design, the model tests the impact of different combinations of parameters like deductibles, coverage limits, and co-insurance ratios on premiums and loss ratios to find the optimal solution.
In terms of risk reduction, the model can, before a typhoon arrives and in conjunction with meteorological bureau typhoon track forecasts, push risk alerts and disaster prevention advice to policyholders in specific areas up to 72 hours in advance. Behind these seemingly simple actions lie economic cost considerations—making contingency plans and evacuating people incur costs. Decisions on whether to evacuate or relocate can be informed by a quantitative judgment from the model.
Impact of Climate Change on Model Components and Technical Challenges
Catastrophe models generally consist of four core modules: hazard, vulnerability, exposure, and financial. When climate changes, the hazard module is the first to be impacted. Climate change systematically alters the frequency, intensity, and spatial distribution characteristics of hazard events by modifying ocean, atmospheric, and land surface water cycle conditions—for example, northward shifts in typhoon tracks, increased precipitation intensity, and prolonged drought durations. The core task of the hazard module is to "generate" vast numbers of hazard events that conform to physical laws. If the probability distribution at this "input" end is itself changing, the entire model's output will be affected.
However, the vulnerability module also faces challenges. Urban infrastructure and building standards are constantly evolving; the seismic and flood resistance capabilities of old residential areas from a decade ago are completely different from newly constructed buildings today. The value of new types of exposed assets, like underground spaces and subways, is rising rapidly. This means the "hazard intensity—economic loss" mapping relationship is also shifting. In the development of the vulnerability module, there is an increasing focus on incorporating the latest building design codes, urban infrastructure census data, and high-resolution topographic information.
In reality, hazards often do not occur in isolation—a typhoon brings not only strong winds but also triggers heavy rain, urban waterlogging, river overflows, and even induces landslides and debris flows, forming a "disaster chain." However, at the model level, achieving full-chain joint simulation of disaster chains is a recognized technical challenge in the industry.
Moving from a typhoon to waterlogging or a debris flow is essentially a step-by-step derivation process. Like solving an equation, the first step yields a result, which then becomes the input for the second step, and so on. This process faces three major challenges: First, the complexity of physical mechanisms. Each step involves complex physical processes: typhoon wind fields and rainfall distribution are driven by meteorological dynamics, flooding by hydrology and hydrodynamics, landslides and debris flows by geotechnical mechanics. They differ in time scales and physical mechanisms, making cross-disciplinary coupled modeling exponentially difficult. Second, data dispersion. Multi-hazard coupling requires multi-dimensional data from meteorology, hydrology, geology, etc., and this data is scattered across different government departments and research institutions. Third, exponential growth in computational resource requirements. Single-hazard models already have high computational demands; multi-hazard coupling represents an order-of-magnitude increase, posing immense challenges for computing power scheduling and algorithm optimization.
The Role of AI in Catastrophe Models
The penetration of AI into the catastrophe models is a gradual process, with varying depths of integration and application forms across different modules.
In data preprocessing, AI has become a standardized tool, capable of performing tasks like automatic classification of satellite imagery, information extraction from historical disasters, and quality control and fusion of multi-source data. The pace of urban construction in China is very rapid, requiring automated identification of new buildings via AI analysis of remote sensing imagery to update the exposure database.
However, in the short term, AI cannot be used independently, divorced from traditional physics-driven models. The scientific community currently holds a cautious view regarding AI completely replacing physical models. AI models are often seen as "black boxes," lacking explicit physical equation constraints, making it difficult to trace and correct prediction biases. AI models also exhibit clear limitations in their predictive capability when faced with extreme hazard events not present in their training data.
Learning from Past Disasters and Future Climate Change Scenarios
Indeed, each major disaster event provides an impetus for model improvement. Taking the typhoon model as an example, in its upgrade in October 2025, data from 36 typhoons occurring between 2020 and 2024 was added, with some typhoons offering new insights. For instance, Typhoon In-Fa in 2021 lingered over land for an extended period after landfall, producing massive cumulative precipitation, highlighting the importance of cumulative precipitation effects in loss assessment. If a model only focused on the wind field at the moment of landfall and ignored the sustained precipitation during the lingering period, the final loss estimate would be severely understated. Another example is Typhoon Doksuri in 2023. After making landfall in Fujian and weakening into a tropical depression, its remnant circulation transported vast amounts of moisture northward, ultimately causing historically rare extreme heavy rainfall in Beijing, Hebei, and other areas. This underscored the understanding that typhoon models must not only focus on the landfall point but also track the remote moisture transport of remnant systems to capture a complete disaster chain. Therefore, the models require continuous refinement based on weather events.
Based on conclusions from sources like the IPCC AR6 (the Intergovernmental Panel on Climate Change's Sixth Assessment Report) and China's Blue Book on Climate Change, there is clear scientific consensus on the long-term evolution trend of climate change. An upward trend in the frequency and intensity of extreme events is discernible in most regions globally. China, as a monsoon climate zone, shows particularly significant warming amplitude and extreme precipitation responses. If global greenhouse gas emissions are not effectively controlled, the long-term risk projection results are indeed pessimistic—extreme disaster frequency and intensity will further increase, exposing society to higher and higher risks. When loss expectations exceed what society can accept, commercial insurance alone will not be able to bear the burden.
It should also be noted that the global energy transition is accelerating. China's new energy industry has already achieved large-scale, high-speed development. In the future, as carbon emissions are gradually reduced globally (especially in developing southern countries), the rate of temperature increase is expected to slow. However, climate is a long-term trend with immense inertia. How it will ultimately transform still requires time to tell.