This study introduces the firebrand spotting parameterization implemented in WRF-Fire and applies it to the Marshall Fire, Colorado (2021) to demonstrate that, without fire spotting, wind-driven fire simulations cannot accurately represent the fire behavior. Spotting can be a dominant fire spread mechanism in wind-driven events, particularly those that occur in the wildland-urban interface (WUI), such as the Marshall Fire. To simulate these fires, the model’s ability to spot is critical, in that it accelerates the rate of spread and enables the fire to spread over streams and urban features such as highways. The firebrand spotting parameterization was implemented in WRF-Fire to improve simulations of wind-driven fires in a fire-atmosphere coupled system. In the parameterization, particles are generated with a set of fixed firebrand properties, from locations vertically aligned with the fire front. Firebrands are transported using a Lagrangian framework and firebrand physics is represented by a burnout (combustion) parameterization. Fire spots may occur when firebrands land on unburned grid points. The parameterization components are illustrated through idealized simulations and its application is demonstrated through simulations of a devastating real case - the Marshall Fire (Colorado, 2021). The simulations were verified using time of arrival and contingency table metrics. Our metrics show that when fire spots were included in the simulations, fire rate of spread and burn area consistently improved.