Modeling Fire Following Earthquake for the 2013 Japan Model Update

Improved Resolution: Capturing the Details

The degree of risk associated with fire following earthquakes is intrinsically linked to the characteristics of the buildings where fires might ignite and through which they can spread. To model fire following earthquake risk at high resolution, a detailed understanding of the layout and distribution of buildings throughout all of Japan is required. To accomplish this, AIR developed an innovative approach to modeling the spatial distribution of buildings. It is an approach that provides a significantly more detailed view of how buildings are arranged within blocks, from block to block, in cities or elsewhere.

AIR scientists analyzed hundreds of city blocks throughout Japan and identified 20 representative block types from which the vast majority of observed building configurations can be constructed. AIR recreated the details of each block type digitally—and added an open-space block that contains no buildings at all but may represent possible parks, broad thoroughfares, open squares, etc.—to create a set of 21 "characteristic blocks." Figure 2 illustrates one step in the process, in this case identifying building footprints by height in one city block using satellite imagery.

This set of characteristic block configurations reflect building occupancies, heights and sizes, and layouts and is used in turn to construct the entire built environment of Japan. Land use and building density data at 1 km resolution are combined with the characteristic block parameters to produce a modeled distribution of 170 characteristic blocks of buildings or open space for every 1 km grid cell in the country. This approach to capturing building distributions at high resolution enables AIR to employ fire ignition and fire spread models at similar high resolution when estimating risk.

A New Approach to Ignition Modeling

The updated model applies the latest ignition rate research to historical ignition data and Japanese construction practices to create an ignition rate function that captures fire risk in Japan. The new ignition function is based on 1.0 second spectral acceleration (a measure of ground shaking correlated to building damage), which has been found to be more closely correlated with fire ignitions and building damage than peak ground acceleration, the variable traditionally used.

Another important consideration in ignition modeling is human activity and interactions with the surrounding environment. For example, research by the Tokyo Fire Department found that ignitions in Japan are correlated with seasonality, as more heating and lighting appliances are in use in winter than in summer. The 2013 AIR fire following earthquake module explicitly accounts for this seasonal variation in ignition rates.

A similar ignition phenomenon became apparent during the Tohoku earthquake. Fires can ignite hours or even days after the earthquake has struck, as electricity is gradually restored to damaged buildings. However, the immense scale of the Tohoku earthquake slowed the power restoration process for millions of customers throughout the affected region. This delay likely reduced fire ignitions, as it allowed more time for damaged buildings to be inspected and for hazardous situations (toppled appliances, frayed wiring, etc.) to be corrected. This understanding of the impact of power system infrastructure damage on ignitions from large earthquakes is also explicitly incorporated into the new AIR ignition model.

Innovative Approach to Modeling Fire Spread

Over the past decade, fire spread modeling has evolved from empirically based equations—which simulate a fire as a growing ellipse (as in the widely adopted Hamada model)—to cellular automata and physics-based models. These cutting-edge models explicitly take into account the dynamics of fire spread. To date, most of these models are applied only at the scale of a city or a section of a city because they are computationally intensive, requiring extremely high resolution building data.

However, using the latest research in the field, AIR scientists have developed a technique that simulates fire following risk countrywide. The technique makes use of a cellular automata model that can capture the detailed components of fire spread while remaining less computationally and data intensive than physics-based fire spread models.

In this approach, a region is modeled as a uniform grid in which the grid cells are assigned properties that represent the underlying building information (combustible, noncombustible, etc.), as illustrated in a simplified schematic in Figure 3. The status of a fire—whether newly ignited or actively burning—is indicated in each cell, as shown in the bottom panel of Figure 3. Fire growth is simulated as the result of interaction between the cells, which is governed by a physically derived set of rules.