EnglishEspanolPortuguesItalianoFrancaisEllinika

WildEngine

WildEngine is the fire and smoke nowcasting core of the Wildcaster ecosystem. It computes a full 24-hour forecast of a wildfire's behaviour in one to three seconds, entirely on the device it runs on, with no connection to any network and no specialist equipment.

Historically, a wildfire behaviour forecast was produced in a command centre or a research institution, on powerful computers, by specialist analysts, and reached the field only if time and connectivity allowed. WildEngine removes that dependency. It is a nowcasting engine: where a conventional forecast is produced at a distance and describes the days ahead, a nowcast is an immediate, short-range projection built from the conditions happening right now, answering what this fire, in this wind and on this ground, will do over the coming hours, computed on the spot.

Two properties make this matter operationally. The first is raw speed: a full day of fire spread computed in seconds keeps the forecast in step with the fire rather than behind it. The second, and for an organisation arguably the more important, is that it runs fully offline. Because the engine needs no servers, no connectivity and no specialist infrastructure, the same simulation that once belonged to command centres can be placed in the hands of everyone at the same time, from the civil-protection director shaping the operation to the crew leader on the line. It democratises access to fire intelligence.

Simulate complex fire scenarios

Wildfire advancing through forest and olive groves toward a valley town, with a smoke column rising over the ridge
On device Fully offline 1 to 3 seconds
How WildEngine turns live inputs into a tactical nowcast, on the device, in seconds.

What WildEngine does

Every fire is a conversation between four things, and the engine follows all of them at once: the fuel on the ground, from cured grass that carries fire quickly to heavy timber and slash that burn long and hot; the land, since fire runs uphill far faster than it backs downhill and funnels through saddles and drainages; the weather, and above all the wind, followed hour by hour across the day ahead; and the fire itself, meaning where it is at this moment, which is the starting line from which the engine grows the fire forward in time.

The data that feeds these calculations comes from Europe's space and Earth-observation infrastructure. The fuels and land cover, the terrain, and the weather that drives the forecast are derived from Copernicus services and other European Space Agency resources, while active-fire detection draws on satellite hotspot data from instruments such as VIIRS and MODIS. This grounding in continuously updated space data lets the engine begin from a realistic picture of the landscape and the fire across its whole area of operation.

From these ingredients WildEngine advances the fire across the landscape and returns answers a tactical team can act on directly:

  • Where the fire will be, as a sequence of hourly perimeters that show the timing as well as the footprint: when the head reaches a road, when it threatens a settlement, and how much time you genuinely have.
  • How fast and how hard it is burning along each part of its edge, distinguishing a flank you can hold from a head you cannot.
  • When it climbs into the canopy and becomes a crown fire, the transition that changes everything about how the fire is fought.
  • Where it may jump, projecting the wind-driven embers that start new fires ahead of the front and outflank crews.

It is complete in the sense that counts operationally, modelling the full range of fire behaviour, surface, crown and spotting, across every fuel type, rather than a single simplified case. And because everything it needs is held on the device, it keeps working once connectivity is lost, which is the normal condition where fires happen.

WildEngine simulation of a wildfire perimeter with hourly spread contours and affected critical points over satellite terrain
Fire propagation: simulated perimeter and hourly spread contours, with affected critical points.

When the active phase eases, the same engine turns to a threat that reaches far beyond the fireline, into towns and along highways many kilometres away: the smoke. WildEngine simulates the plume drifting and spreading through the atmosphere and maps the resulting air quality on the ground, using the same colour-coded health categories the public already recognises from air-quality alerts. This is a deeper and heavier calculation than the fire forecast, and it is built for the command post and the duty office rather than the split-second call on the line, answering the questions that drive evacuations and public-health warnings: which neighbourhoods are about to be affected, which roads remain usable, and where and when people should be moved or told to shelter.

WildEngine smoke-dispersion simulation showing the plume drifting downwind from the fire over satellite terrain
Smoke dispersion: the modelled plume drifting downwind from the fire.

The science underneath

Beneath the speed sits a well-established body of fire and atmospheric science, implemented in full and checked against the original studies. What follows names the specific models WildEngine uses and links to the source literature, so the science can be verified rather than taken on trust.

Surface fire spread

The rate at which a fire advances through a given fuel under a given wind and slope follows the semi-empirical spread equation of Rothermel (1972), with the computational refinements of Frank Albini (1976) that made it practical to evaluate. Its accuracy depends on an honest description of the fuel, so WildEngine drives it with the two internationally standard fuel-model sets: the original thirteen of Anderson (1982) and the expanded forty of Scott and Burgan (2005), which between them describe grass, shrub, timber-litter and slash fuels, and the distinct way each ignites, carries fire and burns out.

Crown fire

The passage from a surface fire to a fire running through the canopy is among the most dangerous transitions on any incident, and WildEngine models it explicitly. It uses the initiation criteria of Van Wagner (1977) to determine when a surface fire has enough intensity to ignite the crowns, the active-crown rate-of-spread model of Cruz, Alexander and Wakimoto (2005) to predict how fast it then travels, and the surface-to-crown linkage of Scott and Reinhardt (2001) to handle the transition between the two regimes.

Ember spotting

Long-range spotting, in which embers are lofted downwind to start new fires ahead of the front, is treated as what it physically is: a matter of probability resolved many thousands of times over. WildEngine builds on the pioneering firebrand-transport model of Frank Albini (1979), together with the crosswind-scatter formulation of Himoto and Tanaka (2005), the short-grass ignition-probability curves of Schroeder (1969) and the flight-survival modelling of Perryman (2012). It launches a large population of virtual embers, carries each on the wind, and tests whether it survives its flight and finds receptive fuel where it lands.

Fire-front propagation

To turn this physics into a clean, advancing perimeter, WildEngine represents the fire's edge with the level-set method introduced by Osher and Sethian (1988) and adapted to wildland fire by Mallet, Keyes and Fendell (2009). This front-based approach to fire simulation has a rich research lineage, including the influential front-tracking and coupled fire-atmosphere modelling of Filippi and colleagues (2009); WildEngine's implementation is its own, but it shares that tradition's central insight of focusing computation on the moving fire front. A narrow-band scheme confines the calculation to the immediate vicinity of the live edge, which is a large part of why the perimeters behave like real fires and compute so quickly.

Fuel moisture

All of the above depends on knowing how dry the fine fuels are, since the moisture in dead grass and twigs decides whether an ember dies or takes hold. WildEngine estimates this continuously using the dead-fuel-moisture science of Fosberg and Deeming (1971), calibrated for the hot, dry, wind-driven conditions of the Mediterranean fire season.

Smoke dispersion

The smoke simulation is a substantial piece of atmospheric physics. Instead of smearing smoke across a map, WildEngine follows it as a swarm of individual particles, each carried by the wind and scattered by turbulence, an approach known as Lagrangian stochastic dispersion that is also used to forecast volcanic ash and radiological plumes. Every particle obeys the well-mixed condition of Thomson (1987), the theoretical requirement that keeps the modelled smoke physically consistent with the turbulence it moves through, with the turbulence statistics themselves taken from Hanna (1982). Because the atmosphere behaves very differently across the day, the model treats the daytime convective boundary layer with the skewed-turbulence scheme of Cassiani, Stohl and Brioude (2015), and grounds the near-surface wind and stability in Monin-Obukhov similarity theory. The height to which the fire's own heat drives the plume is set by the injection-height model of Sofiev, Ermakova and Vankevich (2012), building on the classical buoyant-plume work of Briggs (1975, 1984). Smoke is removed from the air realistically as particles settle under gravity and as rain scavenges them below cloud, following the size-resolved scavenging parameterisation of Wang, Zhang and Moran (2014). The result is a ground-level map of fine-particle pollution, the PM2.5 fraction that matters most for human health, sorted into the standard air-quality bands and driven directly by the live fire through the biomass-burning emission factors of Andreae (2019).

The implementation

Building on this leading science, WildEngine contributes a tactical, robust implementation of it: one that runs the complete chain, surface fire, crown fire, spotting and smoke, in seconds, offline, on ordinary handheld hardware. That is achieved through a set of proprietary optimisations that concentrate the computation only on the ground the fire can actually reach and follow only its live edge, delivering research-grade fire and smoke modelling at field speed with no supporting infrastructure. This is what places a genuine nowcast in the hands of every responder, from the operations room to the fireline, wherever the fire happens to be.

References

  • Cassiani, M., Stohl, A., Brioude, J. (2015). Lagrangian stochastic modelling of dispersion in the convective boundary layer with skewed turbulence conditions and a vertical density gradient. Boundary-Layer Meteorology 154, 367-390. doi:10.1007/s10546-014-9976-5
  • Andreae, M. O. (2019). Emission of trace gases and aerosols from biomass burning - an updated assessment. Atmos. Chem. Phys. 19, 8523-8546. doi:10.5194/acp-19-8523-2019
  • Sofiev, M., Ermakova, T., Vankevich, R. (2012). Evaluation of the smoke-injection height from wild-land fires using remote-sensing data. Atmos. Chem. Phys. 12, 1995-2006. doi:10.5194/acp-12-1995-2012
  • Wang, X., Zhang, L., Moran, M. D. (2014). Bulk or modal parameterizations for below-cloud scavenging of fine, coarse, and giant particles. J. Adv. Model. Earth Syst. 6, 1301-1310. doi:10.1002/2014MS000392
  • Cruz, M. G., Alexander, M. E., Wakimoto, R. H. (2005). Development and testing of models for predicting crown fire rate of spread in conifer forest stands. Can. J. For. Res. 35, 1626-1639. doi:10.1139/x05-085
  • Mallet, V., Keyes, D. E., Fendell, F. E. (2009). Modeling wildland fire propagation with level set methods. Computers & Mathematics with Applications 57, 1089-1101. doi:10.1016/j.camwa.2008.10.089
  • Filippi, J.-B., Bosseur, F., Mari, C., et al. (2009). Coupled atmosphere-wildland fire modelling. J. Adv. Model. Earth Syst. 1, 11. doi:10.3894/JAMES.2009.1.11
  • Rothermel, R. C. (1972). A mathematical model for predicting fire spread in wildland fuels. USDA Forest Service Res. Pap. INT-115. treesearch/32533
  • Van Wagner, C. E. (1977). Conditions for the start and spread of crown fire. Can. J. For. Res. 7, 23-34. doi:10.1139/x77-004
  • Scott, J. H., Reinhardt, E. D. (2001). Assessing crown fire potential. USDA Forest Service RMRS-RP-29. treesearch/4623
  • Scott, J. H., Burgan, R. E. (2005). Standard fire behavior fuel models. USDA Forest Service RMRS-GTR-153. treesearch/9521
  • Osher, S., Sethian, J. A. (1988). Fronts propagating with curvature-dependent speed. J. Comput. Phys. 79, 12-49. doi:10.1016/0021-9991(88)90002-2
  • Thomson, D. J. (1987). Criteria for the selection of stochastic models of particle trajectories in turbulent flows. J. Fluid Mech. 180, 529-556. doi:10.1017/S0022112087001940
  • Also cited: Anderson (1982) INT-122; Albini (1976) INT-30 and Albini (1979) INT-56; Schroeder (1969); Himoto & Tanaka (2005); Perryman (2012); Fosberg & Deeming (1971) RM-207; Hanna (1982); Briggs (1975, 1984).

Available for integration

WildEngine is available as a licensed component for civil-protection platforms, infrastructure and transport operators, and research institutions that need fast, complete, on-device fire and smoke nowcasting within their own systems.