How do you test predictions of fire spread?

One way to test the accuracy of a predictive model of fire spread is to reconstruct a historical fire and compare the observed historical fire spread to the predicted fire spread. Data showing the fire's growth at a specific interval in time, oftentimes acquired using a helicopter equipped with a GPS, provides the requisite observations for comparison. A reconstruction of two historical Southern California wildfires, the 1996 Calabasas Fire and the 1998 Ogilvy Fire, was performed to assess the accuracy of predictions made using the FARSITE and HFire fire spread models. FARSITE, developed by Dr. Mark A. Finney, is the current US standard operational model of fire spread. HFire, developed by Marco Morais at UCSB, is an experimental model of surface fire spread through (Southern California) chaparral fuels.
Read More About HFire

1996 Calabasas Fire

The 1996 Calabasas Fire was a Santa Ana-driven fire which burned a total of 5159 hectares in the Santa Monica Mountains, California. The fire was actively spreading from the time it started along the 101 freeway on October 21 at 1100 hours until it was contained sometime during the late morning on October 22. The weather conditions under which the fire spread, an average windspeed of 14.3 m/s (32 miles/hr) was measured by a local RAWS at 1100 hours, and consequent fire behavior are representative of an extreme fire event.

1998 Ogilvy Fire

Unlike the Calabasas Fire, the 1998 Ogilvy Fire burned in the mountains of Ventura County, California under far more moderate weather conditions. The Ogilvy Fire was actively spreading for 161 hours, from the time of ignition on October 16 at 1500 hours until October 23 at 0800 hours. During this span, 1714 hectares, almost exclusively consisting of chaparral, were consumed by the fire. This represents less than half of the area consumed by the Calabasas Fire during the first 10 hours of spread. As a result, fire behavior during the Ogilvy Fire is representative of a more moderate fire event.