- This topic is empty.
March 31, 2015 at 6:57 pm #1703meganParticipant
I had a couple of questions about fire modeling in state-and-transition vegetation models and I thought some of the users on this forum might have some good insights. There are two related questions:
One question is about fire return interval versus fire rotation. Several people have mentioned to me that they can be very different things, and as I’ve delved into this topic, it appears that the fire rotation can be quite a bit longer than the fire return interval. I have recently become concerned that the method we are using to compare historic and current vegetation conditions may be confounded by differences between fire return interval and rotation. For historic conditions we have been using LANDFIRE fire probabilities (the inverse of which is the fire return interval, according to their documentation). For current conditions we have been using recent fire size data, calculating a fire probability by dividing the total number of acres burned in a vegetation type by the total area of that vegetation type by the number of years in the record. From the research I’ve done, I think the inverse of this number is the rotation. If the return interval and rotation are expected to be different, we may be confounding our historic vs current comparison by an inconsistent fire regime characterization between the two. If anyone has any thoughts about whether this is a problem or not, whether my interpretation is correct, or other approaches they have used, it would be much appreciated.
Another question is about a discrepancy between the expected number of acres burned vs actual modeled fire in some models we are running in Colorado. These models are for forested vegetation types, and each state class in the model has the same overall fire probability, whether it is a single transition or several transitions (that all add up to the same overall fire probability when summing the probability x proportion for each). I was calculating the expected number of acres burned as the area of the vegetation type x the probability of fire in each state class. However, in the model output, the number of acres burned in the model is always lower than the estimate based on area and probability. There are other probabilistic transitions that are also occurring, but I hoped that the simple estimate would provide a ballpark of acres burned. Perhaps this assumption is incorrect, but if so, we may be modeling less fire than we are intending.
Thanks to anyone who may have any insights!
Megan CreutzburgApril 3, 2015 at 3:08 pm #1872Leonardo FridKeymaster
Here are a couple of answers to your questions.
- You may want to take a look at an article by William Reed (2006) entitled "A note on fire frequency concepts and definitions". In this article Reed shows mathematically that in almost all cases the fire cycle (syn. with fire rotation as you describe it) will be greater than the fire interval and recommends against the use of the fire cycle as a method to estimate fire probabilities in simulation models.
- I can think of a couple of reasons why your model is not modeling the expected amount of fire based on its probability and area available:
- If you are running a spatial model and have defined a size distribution that expects large fires, your model may not be able to satisfy that distribution because of fragmentation in the burnable area. If this is the case you will see less fire than in a landscape where the burnable area has no barriers to fire spread.
- In a non-spatial model, if the sum of all probabilities for a cell (fire and non-fire) exceeds one, you will see a reduction in the realized probability of transitions that you see in your outputs.
Let me know if either of these are possible reasons for lower than expected fire amounts. If not, perhaps you could share your library and I can have a look.
LeonardoMay 11, 2015 at 4:59 pm #1873Leonardo FridKeymaster
Here are a couple of more questions from an ST-Sim user:
As a fire spreads one pixel to another, does what information about the state class of the recipient affects whether or not fire spread is successful? Is it a binary function (i.e., if fire is wired to the state class, then the cell will burn)? Or, is the spread probability modified by the fire probabilities associated with the target state class?
Fire spread in ST-Sim is a stochastic process that incorporates information provided by the user about topography "slope multipliers", wind "direction multipliers" and fire probability. The fire grows to its target size by adding cells at the perimeter according to relative travel time based on topography and wind. Cells to be added are then selected randomly with those that have lower travel times and higher fire probabilities being more likely to be added to the fire than those with longer travel time and lower probabilities. We are currently working on a publication that will include the full details of the transition spread algorithm.
One other fire-spread-related question that I have is whether there’s a way to tune fire spread to the various fire severity transitions we have wired in the ILAP models. Seems like the probability of a stand replacing wildfire spreading to the adjacent cell ought to be higher than that for a nonlethal wildfire, especially in the context of suppression. However, I am currently driving fire spread with an allWildFire transition group so that nonlethal wildfires can spawn stand replacing fires in adjacent cells, and vice versa. Will I get wonky results if I add size distributions for stand replacing, mixed severity and nonlethal wildfire transitions on top of the existing allWildFire size distributions?
Good question… You can either specify fire size distributions to fire as a whole, or individually by type. You can’t do both. If you do the latter, individual fires will always be of only one type. Also, make sure that in this case you set the primary group to be for the individual fire types not for fire as a whole. Note that when a fire is spreading, if you are using the AllFire group as the primary group, the model selects which fire type to apply to a cell based on the relative probabilites of the differetn fire types that are possible.
- You must be logged in to reply to this topic.