[[TOC]]

SPITFIRE stands for SPread and InTensity of FIRE and is regarded as a process-based fire model. It simulates the climatic risk, ignition, spread and effects of wildfires originally in natural vegetation. It has stronger feedbacks with the vegetation than the more general fire model Glob-FIRM. With SPITFIRE you can simulate climatic fire danger, number of fires, area burnt (in ha) and biomass burnt (in gC/m²) at the daily time step.
Simulating wildfires means, that they burn in natural vegetation, being forest, open woodlands or natural grasslands. But they can be ignited naturally or anthropogenically.

Climatic fire danger is described through the Nesterov index, which takes the difference between daily maximum temperature and dew-point temperature on precipitation-free days (P<3mm).
Lightning- and human-caused fire ignitions are calculated. Lightning-caused fires are derived from optical satellite-date, which assume a constant fraction being ground-strikes, of which it is assumed that 1/5 have enough energy to heat a wood particle to its ignition point. A 12-month climatology is used as an input to capture seasonal pattern for each 0.5° grid cell. Human-caused ignitions are approximated from human-population density, assuming that the potential number of ignitions is higher in rural areas than in urban areas. Please note that the parameterisation of human-caused ignitions is done for current (last 30-years) conditions, to apply the algorithm to other historic times requires re-parameterisation. Future simulations always imply that current relations in ignition pattern remain the same over the next decades.
Fire spread is calculated using a potential available fuel load depending on its fuel moisture the latter being derived from soil moisture content in the upper soil layer. This is used in the fire spread routine which is based on the BEHAVE fire spread model, a classic fire spread model. This fire spread model requires dividing the litter pool into fuel classes, as fine fuels dry out faster and play an important role in accelerating fire spread. The classification of the fuel classes combining leaves, sapwood and heartwood is done using fixed proportions.
The calculated rate of spread is then used to calculate surface fire intensity. This surface fire intensity is used in two ways, 1) to decide if a fire in that day developed enough energy to sustain burning (>50 kW/m) and 2) if the flame height is sufficient to scorch the tree crown. If surface fire intensity is <50 kW/m, then number of fires, area burnt and biomass burnt are set to zero, else it is summed up for monthly output variables.
Please note that SPITFIRE in its current form only calculates crown scorching, which is not the same as active crown fires. For simulating active crown fires, information such as density of the crown and live fuel moisture would have to be parameterised.
To calculate crown scorching, the scorch height is calculated from rate of spread and is compared with the length of the canopy. The smaller the tree the easier it is for the flame to reach the crown. Parameter in the scorch height equation, crown length and bark thickness are PFT parameter. This means that vegetation height and diameter of the average individual play a role in simulated fire effects.
SPITFIRE calculates two types of post-fire mortality: mortality from crown scorch and cambial damage are simulated. Trees with thin bark have a higher mortality in high-intensity fires than trees with a thick bark under low-intensity fires. The biomass of trees dying due to these two types of mortality is added to the litter pool, the carbon pools and number of plant individuals (nind) is updated. This influences the simulation of the fire processes in the next simulation day.
Biomass burning is the sum of biomass consumed in surface fire (scorch height of the flame is lower than the lower end of the canopy), i.e. litter and living grass, and the biomass lost in crown scorching.

Model behaviour is as such that if grass PFTs reach a critical FPC, fire spread increases 100-fold and accelerates tree loss. On the other hand, higher fuel bulk density of deciduous trees leads to small rates of spread. A close canopy reduces the influence of wind, thus rate of spread.
The effect of land-use on fire is simply that the crop fraction within a grid cell reduced the potential area burnt to be maximum the stand size of the natural vegetation. It can be regarded to be a approximation of large-scale fragmentation.
Fires do burn only one day because LPJmL does not simulate the distribution of stands spatially explicit and the difficulties to approximate landscape fragmentation at this scale. If climate conditions permit, “new” fires ignite on the next simulation day leading potentially to overestimation of fire ignitions but to the same monthly burnt area as daily burnt areas are accumulated. The latter is the important variable to connect to the vegetation part of LPJmL.

A detailed description of the model is published, including a list of variables, in the open-access journal Biogeosciences (see References).
For further information you can refer to the presentation of the SPITFIRE model in the LPJmL_seminar.

Spitfire is currently working on natural vegetation only. At the moment Kirsten and Susanne are implementing and testing a version with spitfire on managed grassland.

In SPITFIRE, fire disturbances are simulated as the fire processes risk, ignition, spread and effects separately. The climatic fire danger is based on the Nesterov index NI(N_d), which describes atmospheric conditions critical to fire risk for day N_d:

where T_max} and T_dew are the daily maximum and dew-point temperature, and d is a positive temperature day with a precipitation of less than 3 mm. The probability of fire spread P_spread decreases linearly as litter moisture ω₀ increases towards its moisture of extinction m_e:

Combining NI and P_spread, we can calculate the fire danger index FDI:

to interpret the qualitative fire risk in quantitative terms. The value of α_p defines the slope of the probability risk function given as the average PFT parameter for all existing PFTs (n). SPITFIRE considers human-caused and lightning-caused fires as sources for fire ignition. Lightning-caused ignition rates are prescribed from the OTD/LIS satellite product Christian et al. 2003). Since it quantifies total flash rate, we assume that 20% of these are cloud-to-ground flashes Latham et al. 2001) and that, under favourable burning conditions, their effectiveness to start fires is 0.04 (Latham et al. 2001,1989). Human-caused ignitions are modelled as a function of human population density assuming that ignition rates are higher in remote regions and declines with increasing level of urbanisation and associated effects of landscape fragmentation, infrastructure and improved fire monitoring. The function is:

where

P_D is the population density [individuals km^-2], and a(N_D [ignitions individual^-1 day^-1] is a parameter describing the inclination of humans to use fire and cause fire ignitions. In absence of further information a(N_D) can be calculated from fire statistics using the following approach

where N_h,obs is the average number of human-caused fires observed during the observation years t_obs in a region with the average length of fire season (LFS) and the mean human population density. Assuming that all fires ignited in one day have the same burning conditions in a 0.5° grid cell with the grid cell size A, we combine fire danger, potential ignitions and the mean fire area A_f to obtain daily total burnt area with:

We calculate E(n_ig) with the sum of independent estimates of numbers of lightning (n_l,ig) and human-caused ignition events (n_h,ig), disregarding stochastic variations. A_f is calculated from forward and backward rate of spread which depends on the dead fuel characteristics, fuel load in the respective dead fuel classes and wind speed. Dead plant material entering the litter pool is subdivided into 1-, 10-, 100- and 1000 hour fuel classes, describing the amount of time to dry a fuel particle of a specific size (1-hour fuel refers to leaves and twigs and 1000 hour fuel to tree boles). As described by Thonicke et al. 2010): "The forward rate of spread ROS_f,surface [m min^-1] is given by:

where I_R is the reaction intensity, i.e. the energy release rate per unit area of fire front [kJ,m^-2 min^-1; ζ is the propagating flux ratio, i.e. the proportion of I_R that heats adjacent fuel particles to ignition; Φ_w is a multiplier that accounts for the effect of wind in increasing the effective value of ζ; ρ_b is the fuel bulk density [kg m^-3}], assigned by PFT and weighted over the 1-, 10- and 100-hour dead fuel classes; ε is the effective heating number, i.e. the proportion of a fuel particle that is heated to ignition temperature at the time flaming combustion starts; and Q_ig is the heat of pre-ignition, i.e. the amount of heat required to ignite a given mass of fuel [kJ kg^-1]. With fuel bulk density ρ_b defined as a PFT parameter, surface-area-to-volume ratios change with fuel load." Assuming that fires burn longer under high fire danger, we define fire duration (t_fire) [min] as

In the absence of topographic influence and changing wind directions during one fire event or discontinuities of the fuel bed, fires burn an elliptical shape. Thus, the mean fire area [in ha] is defined as follows:

with L_B is length to breadth ratio of elliptical fire, and D_T is the length of major axis with:

with ROS_b,surface, surface as the backward rate of spread. L_B for grass and trees, respectively, is weighted depending on the foliage projective cover of grasses relative to woody PFTs in each grid cell.
SPITFIRE differentiates fire effects depending on burning conditions (intra- and interannual). If fires have developed insufficient surface fire intensity (< 50 kW m^-1), ignitions are extinguished (and not counted in the model output). If the surface fire intensity has supported high enough scorch height of the flames, resulting scorching of the crown is simulated. Here, the tree architecture through the crown length, height of the tree determines fire effects and describes an important feedback between vegetation and fire in the model. PFT-specific parameters describe the trees sensitivity to or influence on scorch height and crown scorch. Surface fires consume dead fuel and live grass as a function of their fuel moisture content. The amount of biomass burnt results from crown scorch and surface fuel consumption.
Post-fire mortality is modelled as a result of two fire mortality causes: crown and cambial damage. The latter occurs when insufficient bark thickness allows the heat of the fire to damage the cambium. It is defined as the ratio of the residence time of the fire to the critical time for cambial damage. The probability of mortality due to crown damage (CK) is:

where r_CK is a resistance factor between 0 and 1, and p is in the range of 3 to 4. The biomass of trees which die from either mortality cause is added to the respective dead fuel classes.
In summary, the PFT composition and productivity strongly influences fire risk through the moisture of extinction, fire spread through composition of fuel classes (fine vs. coarse fuel), openness of the canopy and fuel moisture, fire effects through stem diameter, crown length and bark thickness of the average tree individual. The higher the proportion of grasses in a grid cell the faster fires can spread, the smaller the trees and/or the thinner their bark the higher the proportion of the crown scorched and the higher their mortality.

A detailed technical description, containing reasoning of individual equations and PFT parameterisation is found in SPITFIRE_technical_document_freeze.pdf (see link to pdf below).

To activate the SPITFIRE model in the lpjml.conf file you need to set the flag #WITH_SPITFIRE.
The lpjml.conf and input.file are written as such that during the spinup with natural vegetation, only lightning-caused ignitions are simulated (ifndef FROM_RESTART NO_POPULATION #ELSE POPULATION). If you start from a restart file, human-caused igntions are automatically simulated. If you do not want to do this, you need 1) to set POPULATION to NO_POPULATION in the #else statement and 2) comment out the directory and file name of the human population density file in the input.conf.

Specific inputs required are: lightning ignitions, potential human_caused ignitions, wind speed and human population density, see Input.

SPITFIRE has been developed by Kirsten Thonicke, Allan Spessa and Colin I. Prentice.

Fire