Defining a Flood Event in the United States
Apr 18, 2014
Editor's Note: Defining what constitutes a flood event in reinsurance contracts presents many challenges. This article describes a new approach to event definition used in the upcoming AIR Inland Flood Model for the United States that can drive innovation in how flood risk is understood and transferred.
Inland flooding poses an extremely high risk to homes and businesses across the contiguous U.S. At any given time, a storm is occurring somewhere in the region that could produce flood damage. However, despite the frequency of flood, the complexity and ambiguity of conditions that contribute to inland flooding make it one of the least understood of all natural perils for insurance risk managers.
As with other natural hazards, flood insurance cover is applied based on individual flood occurrences. In the case of inland flooding, however, it can be extremely difficult to attribute losses to a particular occurrence because of the complexities in defining a flood event. Unlike damage from hurricanes, earthquakes, and severe thunderstorms, which typically can be linked to a single causative event that lasts from seconds to days at most, the cause of flooding and flood damage is not always clear. The main culprit is precipitation often combined with snowmelt, but the process of determining which low-pressure systems triggered a flood is no trivial task. Large areas containing multiple river basins often flood in sequence as the rivers are affected by any number of storms that may occur over the course of a month or more. A storm that saturates the ground may or may not be the same one that causes a river to burst its banks, and whether a flood occurs depends on many variables, including soil type, antecedent soil conditions, drainage conditions, land use and land cover, and flood defenses.
The fact remains that flood is an underinsured peril in the United States, with the risk borne to a large extent by the federal government. To increase private market participation, insurers will need a more detailed and robust understanding of the location, frequency, and severity of flooding, and this starts with an objective definition of what constitutes a flood event.
The Hours Clause
The ambiguities that surround flood event definition have prompted the insurance industry to manage flood risk using methods that are dissociated with the physical properties of the peril. One such method is the application of an hours clause, which specifies a limited time period during which insurers can aggregate claims for a reinsurance recovery. For inland floods in the contiguous U.S., many reinsurers use 168 hours (one week) to define a single flood occurrence.
While hours clauses provide some flexibility to insured parties (who can select the time at which to commence the claim period), they are prone to misinterpretation and inconsistencies in their application. They provide no insight into the actual underlying flood risk, which could help the industry manage losses from future floods. In some policies, any flooding that occurs within the same time window can be aggregated into one occurrence, even if it clearly involves separate weather systems thousands of miles apart.
In Europe there has been increasing interest in using more meteorologically-based methods for defining a flood occurrence. After the 2002 floods along Germany's Elbe River, it became clear that the 168-hour time window, commonly used at the time, should be extended. Germany, and many other European countries, now use a 504-hour (three-week) time window, which better reflects the observed duration of flood waters during a single event. The time window of 504 hours also serves well to separate storms that contribute to flooding in Europe. In Germany, this time window accommodates the length of time it usually takes a flood wave to propagate along the length of Elbe or Rhine rivers from their headwaters to the North Sea.
Understanding Flood Occurrence
Applying a predefined time window to define flood in the contiguous U.S. presents a challenge. Flooding of some nature occurs almost continually across the country, and floods contract and expand as they propagate through different regions of varying conditions. In addition, it can take a flood wave up to two months to propagate the length of certain rivers in the U.S. It is therefore quite difficult to define a flood using an hours clause that can encompass the geographical and meteorological complexities surrounding flood risk across the United States.
To accommodate current practices in the industry, the new AIR Inland Flood Model for the United States takes the 168-hour clause into account when defining a flood event. AIR's unique methodology, however, is based on a comprehensive and objective understanding of what causes flooding and how floods propagate through the country's extensive river networks.
To build the model's stochastic catalog, AIR simulated atmospheric activity and soil conditions on a continuous basis to model floods across the entire contiguous U.S (an area of nearly 7.6 million km2).1 This physical approach, which leverages a global climate model coupled with numerical weather prediction, is able to accurately capture the myriad environmental circumstances that contribute to flood, including antecedent conditions, precipitation, and snowmelt. The model then calculates the discharges at each river segment, which are also continuously monitored (at hourly time steps). The model thus registers all flood peaks that exceed the bankfull flow of the river (the discharge at which the river is full to the top of its banks) and could potentially cause a flood on the floodplain.
The AIR model captures off-floodplain flooding as well, which occurs when heavy precipitation falls on saturated soil or paved urbanized areas, resulting in extreme runoff. As the runoff streams down slopes, it can create temporary ponding or dangerous ephemeral brooks and rills that are capable of transporting large amounts of sediment and debris.
In the AIR model, excess runoff generated at each small area flows into the nearest stream and propagates through the river network. During each storm, multiple small streams, which are shallower, reach flood stage first. The flow is then funneled into larger streams that have more capacity, and eventually a single large river conveys the flow from the entire basin. This process creates complex patterns of on- and off-floodplain events in the catalog that, if considered over a relatively small area, can be associated with a particular storm system.
However, for very large basins such as that of the Mississippi River, a flood wave can be caused by runoff from multiple storm systems and may take months to propagate through the entire basin (from Minnesota to New Orleans, for example).2 Defining such a flood event in the context of an hours clause, and based on proximity to a particular storm system, is therefore a highly complex task.
A New Approach to Defining Events
To solve this problem, AIR takes a new approach for flood event definition. This approach clusters extreme runoff and river flows into flood events according to their spatial and temporal proximity. AIR's clustering algorithm is based on a hierarchical process. Each registered extreme flow or runoff (approximately 300,000 occur annually on average across the country) is initially considered an individual event centered at its spatial and temporal coordinates. An iterative process allows the clusters to expand as individual extreme occurrences are aggregated around the locations of highest densities of excess flow. This continues until certain criteria are met concerning cluster size, while simultaneously considering separation distance between each pair of excess flows—restricted to approximately 3,300 km (2,050 miles)—and the elapsed time.
The clustering process allows the model to separate, both spatially and temporally, the extreme river flows and excess runoff into events that conform to the 168 hours clause. An example of two separate events is illustrated in Figure 1. Both events last four days (96 hours) and they each have multiple flood extents that are within 3,300 km of one another. The only flood occurrence on Day 1 is Event 1 (green) while Event 2 (blue) starts on Day 2. Event 2 continues to strengthen as Event 1 subsides although by Day 4 both events are subsiding. On Day 5, only Event 2 is still occurring. The separation in space and time prevents the two events from overlapping, and also accommodates their different intensities, which strengthen and ebb on different days.
The Future of Flood Risk Management
Flood risk management has always been a challenge in the U.S., and one of the main reasons is the shortage of effective tools to help the insurance industry understand, price, and transfer the risk. For the first time, a fully probabilistic and physically-based model is coming to market that provides a scientifically defensible simulation of flooding and results in realistically modeled flood extent footprints and inundation depths. By simulating flood events on both a temporal and spatial basis, the AIR Inland Flood Model for the United States offers an objective starting point for defining—and perhaps redefining—flood occurrence.
Stakeholders in the industry can use this model to revisit the ways that flood events are defined in reinsurance contracts. Doing so may, as it did in Europe, result in more comprehensive and effective management of inland flood risk in the U.S. The hours clause commonly in use falls short by many measures: its time span of 168 hours does not adequately reflect the reality of flood duration across the U.S; its application is open to inconsistency and misinterpretation; and the counterparties may not have a common understanding of what constitutes one event. Because AIR's flood event definition is a robust process, it can be adapted to different time durations as the industry evolves its approach to managing flood risk.
The AIR model, scheduled for release this summer, is the industry's most detailed and comprehensive probabilistic model, and companies can use it not only to price risk and enter new markets, but to drive innovation in how this risk is understood and transferred.
1 This is a different approach than the one used for the AIR inland flood models for Europe, which separate storm systems as part of the process for defining separate flood events.
2 A good example of this is the Great Flood of 1993 along the eastern Mississippi River basin. At that time, a series of weather systems formed an "Omega Block." Omega Blocks are very stable weather systems in which a large high pressure system is flanked by two low pressure systems, forming a pattern similar to the Greek letter Omega. The precipitation pattern from these systems therefore followed the same track (from southwest to northeast) resulting in relentless rain over the same area. The first storms saturated the soil and subsequent storms produced excess runoff, forming floods.