With the increasing interest in leveraging mobile robotics for fire detection and monitoring arises the need to design recognition technology systems for these extreme environments. This work focuses on evaluating the sensing capabilities and image processing pipeline of thermal imaging sensors for fire detection applications, paving the way for the development of autonomous systems for early warning and monitoring of fire events. The contributions of this work are threefold. First, we overview image processing algorithms used in thermal imaging regarding data compression and image enhancement. Second, we present a method for data-driven thermal imaging analysis designed for fire situation awareness in robotic perception. A study is undertaken to test the behavior of the thermal cameras in controlled fire scenarios, followed by an in-depth analysis of the experimental data, which reveals the inner workings of these sensors. Third, we discuss key takeaways for the integration of thermal cameras in robotic perception pipelines for autonomous unmanned aerial vehicle (UAV)-based fire surveillance.

The influence of a mesoscale convective system (MCS) in the evolution of two wildfires that started during the afternoon of 17 June 2017 in Pedrógão Grande, Central Portugal is discussed and analysed using weather radar data, weather stations, video images and fire spread analysis. As the MCS approached the region, its convectively driven flows started to influence the fires. The overturning flows were formed by two main limbs: one organised as front-to-rear deep layer inflows that propagated over the convective region of the MCS and the other as rear-to-front mid-level inflows that descended below the anvil structure of the MCS. The rear-to-front inflows, while accelerating and descending to lower levels, contributed to modify the fires’ intensity and plume characteristics. After the two fires merged, the resulting junction fire became very intense and impossible to control. Then, a firestorm was generated, causing the deaths of 66 people. The main goal of this study is to detail the influence of the MCS in the fire spread, thus contributing to the general knowledge of outstanding fire behaviour modifications due to the influence of atmospheric convective processes.

Mathematical models and numerical solution procedures for predicting the trajectory, oscillation, possible rotation, and mass and size time-evolution of cylindrical wind-driven firebrands are described and discussed. Two test problems and the results, used for validating the mathematical models, are presented. In one, experimental measurements of non-burning cylindrical particles falling in still air are compared to numerical predictions and in the other, predictions of time-evolution of mass and size of stationary burning particles in air flows are compared with experimental results reported in the literature. Results yielded by the proposed models for a demonstration problem involving cylindrical wind-driven firebrands, with the same initial volume, mass and position, but different initial aspect ratios and distinct initial orientations relative to the wind velocity, are then presented. These results show the following: the horizontal distance travelled by the firebrand from release to landing locations is an increasing function of its initial aspect ratio; and the initial orientation of the firebrand, and its subsequent oscillations including possible rotation, have a significant influence on its trajectory, thus it is important to account for them in mathematical models formulated for predicting the spread of fires by spotting.

The number, dimensions, and initial velocity of the firebrands released from burning Quercus suber, Eucalyptus globulus, Quercus robur, and Pinus pinaster trees were analyzed in laboratory experiments using a particle image velocimetry system. Additionally, the flame height, tree mass decay, vertical flow velocity, and temperature at the top of the trees were measured during the experiments. The relationship between the various parameters was analyzed and a good connection was found. The specimens burnt were mostly young trees, so large particles (e.g., pine cones, thick trunk barks, branches) were not included in this study as they were not present. Actually, the firebrands produced in the laboratory tests, mainly burning leaves, had a cross-sectional area of <1,600 mm², having the potential to cause short distance spotting (up to tens of meters). Quercus trees are often considered to have a lower fire risk than eucalyptus or pine trees. However, in this study, Quercus suber and Quercus robur were the species that produced more firebrands, both in terms of number and total volume. The tests with Quercus suber were the only ones using specimens from an adult tree, confirming the great importance of the age of trees in the propensity to release firebrands. The results obtained with Quercus robur confirmed the high tendency of this species to originate spot fires at a short distance. Thus, these results are of great relevance to afforestation plans and to evaluating the risk of the presence of these species in wildland–urban interface areas.

Spotting is a very important mechanism of forest fire spread. Its negative effect increases in extreme fire danger conditions. In order to predict the maximum spotting distance, the duration of the combustion reaction of potential firebrands should be evaluated. This paper reports the results of an experimental laboratory study of the combustibility of firebrand material (pine cones and scales and pieces of bark of eucalypt) of two representative species in Portugal, Pinus pinaster Ait. and Eucalyptus globulus Labill. The main purpose was to assess the role on the burning conditions of the firebrand particle orientation angle θ in relation to the airflow and of the flow velocity (U) around the particle. Tests were made varying the angle of orientation of the main axis of the particle in relation to incident flow in the range of ±90°; flow velocities were tested from 0 to 6.5 m s–1. After ignition, particle mass loss owing to flaming or glowing combustion of the particle was continuously measured. Residual mass, duration of the flaming phase and the burnout times are reported for each case. Empirical models to estimate the trends of variation of some of these parameters with θ and U are proposed to illustrate their importance for the range of test conditions.