Publicação
Wildfire susceptibility modelling in mainland Portugal
| Resumo: | Wildfires are a recurrent phenomenon in mainland Portugal, mainly of a seasonal nature, during summer, even though, depending on favourable conditions for ignition and fire spreading, wildfires can occur at any time of the year. Wildfires represent a problem for mainland Portugal because they destroy significant areas of which some populations are dependent on, but also because wildfires lead to significant expenditures in suppression efforts. In this thesis, a low complexity model, integrating high spatial correlation layers has been tested with an updated land cover coverage, showing a good predictive capacity with only land cover, slope and historical data as layers. However, this model benefits from a double integration of historical data in that past wildfires not only enter the model as an independent layer but are also the basis for computing favourability scores for any other layer. For that reason, another method has been explored: Weights of Evidence. This second method is statistically robust and unbiased, and when put to the test with several evidence layers, such as land cover, slope, elevation, aspect, population density, population growth ratio and distance to roads, has shown that it has comparable results to those of the simpler, lower complexity model. It has also shown that adding more evidence layers does not necessarily lead to improved predictive performance. Wildfire susceptibility assessment at a regional level (NUTSII) has also been studied, showing that with the exception of the smaller NUTSII region (Lisboa), all other regions show worse predictive performance than the model run for the entire mainland, and frequency-area statistics show that while large wildfires are responsible for most of the total burnt area in mainland Portugal, they are generally smaller than what would be expected from a power-law of good fitting. On the other side of the spectrum, smaller wildfires are found to be larger in area than what they should be in regards to statistics. |
|---|---|
| Autores principais: | Verde, João |
| Assunto: | Teses de doutoramento - 2015 |
| Ano: | 2015 |
| País: | Portugal |
| Tipo de documento: | tese de doutoramento |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | Wildfires are a recurrent phenomenon in mainland Portugal, mainly of a seasonal nature, during summer, even though, depending on favourable conditions for ignition and fire spreading, wildfires can occur at any time of the year. Wildfires represent a problem for mainland Portugal because they destroy significant areas of which some populations are dependent on, but also because wildfires lead to significant expenditures in suppression efforts. In this thesis, a low complexity model, integrating high spatial correlation layers has been tested with an updated land cover coverage, showing a good predictive capacity with only land cover, slope and historical data as layers. However, this model benefits from a double integration of historical data in that past wildfires not only enter the model as an independent layer but are also the basis for computing favourability scores for any other layer. For that reason, another method has been explored: Weights of Evidence. This second method is statistically robust and unbiased, and when put to the test with several evidence layers, such as land cover, slope, elevation, aspect, population density, population growth ratio and distance to roads, has shown that it has comparable results to those of the simpler, lower complexity model. It has also shown that adding more evidence layers does not necessarily lead to improved predictive performance. Wildfire susceptibility assessment at a regional level (NUTSII) has also been studied, showing that with the exception of the smaller NUTSII region (Lisboa), all other regions show worse predictive performance than the model run for the entire mainland, and frequency-area statistics show that while large wildfires are responsible for most of the total burnt area in mainland Portugal, they are generally smaller than what would be expected from a power-law of good fitting. On the other side of the spectrum, smaller wildfires are found to be larger in area than what they should be in regards to statistics. |
|---|