Wildfire-Smoke Forecasting: A Methodology for Local-Scale Dynamics from Wildfire-Spread to Atmospheric Dispersion

Wildfires have a significant impact on the human health of populations on the path of the generated smoke plumes. They emit high amounts of air-pollutants resulting in abnormally high concentrations of harmful particles and gases. These lead to increased diseases associated with bad air quality that are clearly perceived in hospital admissions. Wildfire smoke also impacts society on other levels, such as tourism and airplane routes or the firefighting operations themselves.

The focus of this work is on the forecasting of wildfire smoke. Such forecasts serve essentially two purposes. First the ability to quickly assess potential impacts of wildfires on air quality. Second, to decide on actions that mitigate those impacts. Examples of their usefulness are the FireSmoke platform used in North America, or the European CAMS air quality forecast.

This work presents a new methodology for forecasting wildfire smoke at local-scale.  Firstly, the emissions of wildfires are estimated using a fire-progression model. Then the dispersion of smoke at local-scale is estimated using a computationally efficient lagrangean model.

The implementation of this methodology was carried out for the region of Sardinia, Italy. Disperfire was used as the dispersion model, while the Sardinian Wildfire Simulator (SWS) was used in the estimation of fire-progression. A case-study of a past wildfire in the region was chosen to evaluate the developed methodology.

The SWS is a fire-growth model that, based on vegetation characteristics and state, topography and meteorology, is able to estimate how the fire-front will change over time. This type of models have been widely used to support firefighting operations.

Disperfire is a lagrangean dispersion model that works with a numerical grid at resolutions of hundreds of meters. In a first step the emissions at each grid-cell are calculated based on emission-factors and the intensity of the fire, previously estimated by SWS. Then, smoke is modelled as particles, each representing a given mass of smoke, that are advected along the wind velocity vectors. The diffusion phenomena are modelled by moving those particles according to a random normal distribution.

Several runs were carried out using different levels of discretization, by varying the time-step, the grid-resolution and the number of particles used in Disperfire. The influence of the different levels of discretization was assessed.

The newly developed methodology fills a gap by explicitly modelling phenomena such as local-scale dispersion and fire-progression which are often simplified or absent in mesoscale wildfire-smoke forecast systems. This approach provides a foundation for improving the accuracy of mesoscale smoke dispersion models in the future.