Evaluation of vegetation around roads by LiDAR data processing and multispectral images

Resumo

Nowadays, forest fires are of great importance to society because of the social and economic impacts which they entail. Although the area burned on a global scale has decreased over the last decades, many scientists have predicted that this situation will be reversed due to extreme fires. Among the reasons for these predictions are the changes in social-ecological systems as well as in the climate, therefore two aspects of global changes are carrier for uncertain consequences. A high percentage of fires take place on the surrounding roads, which represent the escape route for the population in case of wildfire. On the contrary, in case of collapse, roads also entail a high risk to people which, unfortunately, the 2017 events located in Galicia and Portugal are an example of this. Therefore, the management of vegetation around roads is very important in terms of wildfire prevention. Fuels are broadly defined as any combustible material which properties are highly variable in space and time and which can be described by their composition and structure. Fuel is the only element of the fire behaviour triangle that can be directly manipulated. Consequently, their characterization helps to represent and model fire risk and its effects. The use of new technologies and data analytics provides collaboration among scientists and managers transforming fire science and management. These technologies allow for generating current predictions about wildland fires, measure, or monitoring fires. In this thesis, automatic methodologies for LIDAR data and multispectral images are presented, focusing specifically on the vegetation placed around roads. The main purpose is to propose solutions and different approaches to the detection, analysis, modelling, and evaluation of vegetation in order to prevent fire around roads through forest management. This objective is aborded by six main aspects: the detection of vegetation, the modelling of its structure both vertical and horizontal as well as its changes over time, the type of species linked to the wildfire events, and the risk of forest fire in a concrete area due to the interaction of many factors. The geometric information of LiDAR and multispectral images allowed the development of algorithms for the detection and evaluation of vegetation around roads and the design of methodologies for the structure characterization of vegetation. Moreover, a method to evaluate the vegetation in places where there were registered wildfires was carried out by multispectral images. The work of this thesis has been tested in real case studies, obtaining results that contribute to the solution of the above-mentioned objectives. This thesis contains eight scientific publications, four of them published in different scientific journals, indexed on the Journal Citation Report (JCR), and four are double-blind peer-reviewed conference proceedings.