Modeling fire severity in California, USA

Understanding the conditions that determine fire severity is important for predicting potential fire effects. The ability to predict and forecast risk of high severity fires would allow land managers to prioritize fuel treatments. The probability of large fire occurrence has been successfully predicted and forecast using hydro-climate variables and statistical modeling techniques. We are employing similar statistical techniques to test the potential for predicting and, given success, forecasting fire severity for California, USA. We have a mapped fire severity dataset for 243 fires spanning seventeen years (1984-2001) for California. Our hydro-climate dataset was developed using the VIC hydrologic model with the LDAS parameterization; we thus used the LDAS 1/8º grid to sample the fire severity dataset.

Initial exploratory analyses show that low fractional fire severity is positively correlated to cumulative precipitation for the 12-month prior period and the 12-month period ending 6-month prior to the month of the fire, while moderate to high fractional severity is negatively correlated. Opposite in sign, but comparative in magnitude, relationships were found for cumulative adjusted moisture deficit for 12 months prior to the month of fire and the 12 month period ending 6 months prior to the fire. Additionally, fire area was positively correlated to both cumulative moisture deficit variables. Current year March and April snow water equivalent were negatively correlated to moderate and high fractional fire severity.

We were able to predict high severity fraction (fraction of the fire in the high severity category relative to fire size) using a third order polynomial regression with the following independent variables: one year moisture deficit, average relative humidity, monthly average temperature, mean elevation, the latter three for month of fire. These relationships are expected as we know that fuel moisture is an important control on potential fire severity and all will affect fuel moisture content. We also know that total fuel quantity and type affect potential fire severity. We will explore fuel quantity and type in addition to the aforementioned hydro-climatic variables to determine how the predictive capability might be affected. Our initial results indicate that prediction of fire severity could be feasible. We will present full prediction/forecast results.