Description and design

Purpose:

The MedSpread model is a spatially explicit raster-based model designed to mimic fire spread of a given set of fires. The model simulates fire spreading from an ignition until the target area is reached. It can be applied to mimic the spread and burning of an observed real-life fire perimeter from its known ignition point. MedSpread can be also used to simulate fire scars from estimated ignitions points. The purpose of the MedSpread model is to examine the spatial interactions between vegetation (i.e. fuel load and forest composition), topography, and wind forces when determining fire spread.

Architecture:

The model simulates fire spread from an ignition until the target area is reached (i.e. when the fire has completely burnt the total area to be burnt), and an early fire extinction only happens if all active fronts arrive at non-burnable areas (e.g. water, urban settlements or rocks/bare soil). MedSpread is implemented on the SELES platform and it requires two types of input data: (1) a set of ignitions characterized by spatial location with the associated required metadata; and (2) a set of raster layers representing the potential landscape drivers of fire spread. In its current version, spatial resolution is 1 ha.

Design:

MedSpread simulates fire spreading (SR) based on a polynomial algorithm formulation, where explanatory factors are species flammability (SppFlam), fuel load (Fuel), aspect (Aspect), slope in relation to fire front (Slope), and wind effect in relation to dominant wind direction (Wind). These explanatory variables are multiplied by weight-parameters representing the relative influence of each factor on fire front progression (Eq.1).

SR = wW*Wind + wS*Slope + wA*Aspect + wF*Fuel + wSpp*SppFlam (Eq. 1)

When an ignition is activated, SR is calculated for its 8 neighbors (queen’s case). The order in which these evaluated cells will spread and burn depends on the speed at which the fire is spread from the source cell to them. SpeedTime for each neighboring cell is then calculated as a negative exponential of the accelerated SR (Eq.2). First, the SR value is multiplied by an acceleration value Acc which determines the strength of the driving factors in accelerating fronts. In addition, SpeedTime is multiplied by a random value that will eventually provide some stochasticity to the fire spread process.

SpeedTime = -exp(-SR*Acc) * U(StochasticSpread,1) (Eq. 2)

The evaluated cells are thus entered into a priority queue according to their SpeedTime value. The first cell in the queue is activated and then its neighbors are evaluated by calculating SR and SpeedTime for all its burnable neighbors. All the evaluated cells will either way be added to the priority queue according to their currently calculated SpeedTime and respecting the time-ascending order of the queue. However, a cell will only be activated from the fire front that has reached that cell fastest, discarding other burning attempts.

After selecting the activated cell, it is time to decide if it has to burn. The probability of burning (pBurn) is a function of the SR and dictates whether a cell burns or not, proxying fire intensity (Eq.3). Cells will effectively burn if pBurn is greater than a selected value from a uniform random distribution. The rPb parameter allows to modulate fire intensity, namely the relation between fire spread and probability of burning.

pBurn = (1-exp(-SR))^rPb (Eq. 3)

MedSpread is a widely applicable model, but in order to initialize the spatial factors describing landscape characteristics in Eq.1, analysts should consult bibliographic research for the region where it is applied. Any values assigned to the factors have to be constrained between 0 and 1. Since MedSpread is an empirical model based on the relative effect of factors on the fire spread of actual fires, weights of the factors in the polynomial formulation are to be calibrated (wW, wS, wA, wF, and wSpp). Each weight-parameter can range from 0 to 1, and all the weights must sum to 1 to explain 100% of relative factors effects.