Modeling the Perfect Storm: How Antecedent Conditions Can Compound the Severity of Flood Tail Risk Events

What material difference do antecedent hydrologic conditions make to the severity and losses from flood events? During my recent RMS® Season of Flood webinar, I presented the results of a Florida-landfalling hurricane simulation and associated tropical cyclone precipitation of up to 10 inches (impacting states including Virginia, Pennsylvania, the Carolinas, New Jersey, and New York) from the RMS stochastic catalog.

Using our advanced high definition (HD) simulation framework, we ran this scenario against nine events, which included a wide range of antecedent conditions. The analysis showed that antecedent conditions could significantly impact loss variability – so much so that 75 percent of webinar attendees underestimated its role. We encourage you to watch the webinar, where we reveal the results (in minute 18).

Modeling Antecedent Conditions: Understanding the Challenge

In flood modeling, the initial soil moisture conditions dictate the partitioning of precipitation into infiltration and surface runoff, and capturing this interaction is complex. It requires factoring space and time processes into the initial state of the system, as well as understanding how antecedent rainfall events, snowmelt, baseflow conditions, and watershed characteristics impact the severity of flood risk. A model must also include event sets with multi-temporal scales ranging from weeks to months, depending on climate variability and watershed.

Elevated levels of antecedent soil moisture have compounded flood impacts and, in certain cases, contributed to pushing a flood into a tail risk event. For example, soil moisture levels in northwestern Europe were very high before the severe flooding during mid-July this year, with the top meter of soil completely saturated before the intense rainfall. RMS issued an insured loss estimate of between US$6 billion and US$7.7 billion.  

Similarly, regions of New York and New Jersey received more than six inches of rainfall on August 22 during Tropical Storm Henri, followed by more than seven inches of rainfall on September 1 from the remnants of Hurricane Ida. As a result, water levels in the Raritan River in central New Jersey crested at 27.7 feet during Ida.

This broke all historical flood records including water levels recorded during Hurricanes Floyd in 1999 (27.1 feet) and Irene in 2011 (26.24 feet). In this context, reliably representing the initial state of the system and antecedent conditions is important to accurately model the view of flood risk.

Hurricane Harvey and the Role of Antecedent Conditions in Event Losses

During the webinar, we looked at the role of antecedent conditions in 2017’s Hurricane Harvey, which ranks as the second most-costly hurricane to hit the U.S., with US$125 billion of damage. RMS estimated the insured losses as between US$30–$50 billion, much of this from tropical-cyclone-induced flooding as parts of Houston received more than 50 inches of rainfall. What role did antecedent conditions play in exacerbating the severity of the flooding?

Using data from NASA’s Soil Moisture Active Passive (SMAP) mission, which measures the amount of water in the surface soil anywhere on Earth, we observed that three weeks before Harvey’s landfall, soil moisture levels were in the 20–40 percent range in southern and eastern Texas due to a previous severe storm. These saturated conditions likely prevented rainfall from Harvey from infiltrating more deeply in soils. Studies showed that antecedent conditions had enhanced the flood peak and total flow volume during the hurricane.

At the same time, certain regions, such as west of Houston, had very dry conditions in which compacted or parched soils prevented rainfall absorption into the ground and were prone to runoff, hence increasing the likelihood of flooding. If the Texas soil was like a sponge, a dried-out sponge takes a while to loosen up and soak water, while a saturated sponge is little help with a new spill.

How RMS Incorporates Antecedent Conditions Into Flood Models

To address the complexity of antecedent conditions in our simulation-based, HD modeling for the U.S., Europe, and Asia-Pacific, we had to think holistically and innovate to ensure that our clients could fully capture all sources of flood risk. There are both meteorological and physical factors that affect runoff, from the intensity, frequency, and duration of rainfall through to the type of soil and vegetation, and much more.

Looking at the RMS® U.S. Inland Flood Model, antecedent conditions are built into the modeling approach, starting with precipitation modeling consisting of 50,000 years of continuous stochastic rainfall separated by tropical cyclone and non-tropical cyclone precipitation. A continuous hydrologic model simulates inputs for major and minor river networks – the river network across the U.S. covers 370,000 miles (600,000 kilometers) in length.

A hydrodynamic model is fed by simulated river discharge that, along with the runoff, determines inundation depths and extents. The modeling also accounts for snow accumulations and melt, evapotranspiration, and infiltration. In total, we simulate approximately 1.1 million flood events at different recurrence intervals.

Uncertainty in Hazard Data Driven by Shortcuts

We have seen a trend in recent years with several model vendors that only offer hazard data, relying mainly on event-based (non-continuous) simplified hydrologic models to simulate the rainfall-runoff processes. While such models are less complex to implement, require less input data, and run relatively fast to save on computational cost, they are not suitable to accurately represent the initial state of the system before a flood event.

These models can misrepresent the view of flood risk unless observed initial conditions are used, which is not possible in the context of cat modeling or climate change projections. Caution should be exercised, as hazard data produced by models that lack accurate representation of important hydrologic processes will come with higher, unavoidable associated uncertainty that will propagate into the final product.

Antecedent Conditions: Critical for a Robust Model

Looking at our modeling approach, you can see that the analysis of antecedent conditions forms just one integral part – but a crucial one of a robust modeling framework. As seen in the examples from the recent Europe and U.S. floods this year, and with Hurricane Harvey, where the ground is already saturated or is very dry, rainfall and runoff will respond differently.

A more realistic view of risk, using time-based simulation, also makes it easier to distinguish between successive flood events, which is essential for hours-clause contracts. It will help your business delve further into tail risk using a wide base of simulated events and establish the full potential of extreme rainfall events – and where events could be exacerbated by antecedent conditions.

Please watch my webinar, Modeling the Perfect Storm, and other webinars in our Season of Flood series.