A carregar...
Publicação
Improving the estimation of fire danger, fire propagation and fire monitoring : new insights using remote sensing data and statistical methods
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 7.01 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
This thesis covers three major topics related to wildfires, remote sensing and meteorology: (i) quantifying and forecasting fire danger combining numerical weather forecasts and satellite observations of fire intensity; (ii) mapping burned areas from satellite observations with multiple spatial and spectral resolution; and (iii) modelling fire progression taking into account weather conditions and fuel (vegetation) availability. Regarding the first topic, an enhanced Fire Weather Index (FWI) is proposed by using statistical methods to combine the classical FWI with an atmospheric instability index with the aim of better forecasting the fire danger conditions favourable to the development of convective fires. Furthermore, the daily definition of the classical FWI was extended to an hourly timescale, allowing for assessment of the variability of the fire danger conditions throughout the day. For the second topic, a method is proposed to map and date burned areas using sequences of daily satellite data. This method, tested over several regions around the globe, provide burned area maps that outperform other existing methods for the task, particularly regarding the consistency and accuracy of the date of burning. Furthermore, a method is proposed for fast assessment of burned areas using 10-meter resolution satellite data and making use of Google Earth Engine (GEE) as a tool for preprocessing and downloading of data that is then used as input to a deep learning model that combines a coarse burned area map with the medium resolution data to provide a refined burned area map with 10-meter resolution at event level and with low computational requirements.
Finally, for the third topic, a method is proposed to estimate the fire progression over a 12-hour period with resource to an ensemble of models trained based on the reconstruction of past events. Overall, I am confident that the results obtained and presented in this thesis provide a significant contribution to the remote sensing and wildfires scientific community while opening interesting paths for future research on the topics described.
Descrição
Palavras-chave
índice de perigo de incêndio áreas queimadas propagação do fogo deteção remota aprendizagem automática fire weather index burned areas fire progression remote sensing machine learning