Before analyzing a variety of simulated fire regimes, we sought to validate that HFire could recreate fire histories similar to those observed in reality. Preliminary validation of HFire in the Santa Monica Mountains (SAMO) therefore required identification of baseline parameter settings that generated realistic fire patterns. Because there are many possible combinations of input parameters that could result in reasonable output, the baseline values must also be realistic for this region and the size of the simulation domain.

Derivation of baseline parameter settings is described in the sections below, and baseline output is discussed at the end of this page.

Ignitions

Long-term simulation of fire regimes requires that the user specify an annual rate of ignitions on the landscape. An "ignition" can be any source of fire, regardless of whether or not it propagates beyond its initial location. The vast majority of ignitions do not become "fires," either because they are quickly extinguished or because they go out before igniting something that will allow further fire spread. The mapped fire history for SAMO includes only the largest fires, so it is not possible to estimate the average number of ignitions per year from this dataset. However, Los Padres National Forest (LPNF) is a nearby shrubland-dominated region with an excellent record of ignitions, allowing one to calculate a rough estimate of ignition rates for a region the size of SAMO.

On average, the shrublands of LPNF have experienced a total of ~0.37 ignitions per square km over the period 1911-1995. For a region the size of SAMO (i.e., ~900 square km), this equates to an average ignition rate of ~4/yr. However, it is likely that some fraction of ignitions in LPNF have gone unreported, leading to annual ignition rates that are probably somewhat higher. In addition, large expanses of LPNF are unroaded wilderness, while SAMO is more heavily roaded and embedded in the densely populated Los Angeles area. For these reasons, we estimate that the average rate of ignitions for SAMO is substantially higher than 4 per year, and we settled on a value of 8 for the baseline simulation runs.

Weather

Suppression

Identifying a baseline value for the extinction rate of spread threshold is difficult, because there is little published information about what this rate should be. In addition, the baseline rate must be one that generates realistic fire patterns under a variety of weather conditions and in different vegetation types, despite the fact that such an extinction threshold should probably vary.

After preliminary tests of the model, we have chosen a baseline spread threshold of 0.033 m/s, which is ~6 chains/hr. This is a reasonable estimate, based on discussions with various fire behavior personnel, other fire simulation work in shrublands (e.g., Davis and Burrows 1993, Davis and Michaelsen 1995), and comparison of model output with mapped fire histories.

Davis, F.W. and D.A. Burrows. 1993. Modeling fire regime in Mediterranean landscapes. Pages 247-259 in S. Levin, T. Powell, and J. Steele, editors, Patch Dynamics. Springer-Verlag Lecture Notes in Biomathematics, New York.

Davis, F.W. and J. Michaelsen. 1995. Sensitivity of fire regime in chaparral ecosystems to climate change. Pages 435-456 in J.M. Moreno and W.C. Oechel, editors, Global change and Mediterranean-type ecosystems. Springer-Verlag Ecological Studies v. 117, New York, New York, U.S.A.

Baseline Output

As a validation of our parameter choices, simulated fire regimes should:
1) consist of realistic fire patterns on the landscape,
2) generate reasonable fire return intervals, and
3) approximate historical fire size distributions.

Fire patterns:
Select the figure below to see a comparison between real (top) and simulated (bottom) fire histories.

The actual SAMO fire history is for the period 1925-1999, and the last 10 years of fires are shown in red.   For comparison, a random 75-year period of simulated fires is also shown with the last 10 years of fires in red.

Both real and simulated fires are relatively compact and space-filling (as opposed to fractal). Both also show some of the same areas that tend not to burn very often. The real fire history for SAMO only records a few fires smaller than 100 ha, while the simulated output includes fires of all sizes.

Fire return intervals:
Shrubland vegetation is relatively sensitive to fire return intervals that are too short, which can result in type-conversion of native plant species. It is therefore important that simulated fire intervals be within a range that allows persistence of natural vegetation.

"Fire cycle" is a measure often used to approximate how long average fire intervals are across the landscape. Fire cycle is calculated as the number of years it would take to burn an area equivalent to the size of the simulation domain (i.e., ~900 square km), based on average annual burning rates. Using the baseline parameter settings discussed above, average fire cycles are ~33 yr. This interval length is close to that observed for some shrubland ecosystems (e.g., Los Padres National Forest), although it is not far from intervals that could risk type conversion (e.g., < 25 yr).

Fire size distributions:
One of our goals has been to analyze the sensitivity of fire size distributions to changes in various parameters; therefore, it is important that baseline settings generate fire regimes that approximate historical distributions.

Because the mapped fire history for SAMO is quite short and incomplete, it was not useful for this step in the validation process. As was the case for ignitions (above), the mapped fire history of LPNF was used in assessing the baseline simulation settings. Select the figure below to see a comparison between simulated fire size distributions and those characterizing historical data. Simulated fire regimes approximate those observed in real shrublands to a remarkable degree.

The LPNF distribution includes fires only for the shrubland portions of the fire history, 1911-1995; the simulated fire regime is for 200 yr in SAMO. Both distributions show fires > 1 ha in size. Differences in record length do not effect relative frequencies of events.

Back to Simulation
Overview