BURNED AREA PREDICTION USING SMOKE PLUME DETECTION FROM HIGH SPATIAL RESOLUTION IMAGERY

Abstract

The fast-spreading wildfire engulfs the dense parched flora and all obstructions in its way, transforming a woodland into a volatile reservoir of combustible materials. Once ignited, fires can expand at a velocity of up to 23 km/h. As flames spread across vegetation and woodlands, they have the potential to become self-sustaining, propagating sparks and embers that can spawn smaller fires miles away. The proximity of the burning materials to the observer has a direct impact on the density of smoke produced by the fire. This relationship is crucial for fire management teams and emergency responders and helps them assess the severity of a fire, predict its behavior, and make informed decisions regarding evacuation measures, resource allocation, and the protection of affected communities and ecosystems. Drones are valuable tools in the fight against forest fires. They can capture high-resolution imagery, thermal imaging, and video footage, supplying insights into the properties, behavior, and direction of the fire. By employing classical image processing techniques, it is possible to analyze these images and promptly determine the extent of land cover affected. According to the Swedish Civil Contingencies Agency, more than 25000 ha of forest burned down during the period of 2012-2021, which resulted in severe damage costs. The presence of a reliable and easily accessible smoke detection and assessment tool could significantly reduce the impact of wildfires. This study utilizes low and mid-level image processing techniques to analyze the domain of wildfires, leveraging smoke properties to estimate the extent of land affected by the flames.