Tree Species Impact on Forest Fire Spread Susceptibility in Sweden
(2023) In Master Thesis in Geographical Information Science GISM01 20232Dept of Physical Geography and Ecosystem Science
- Abstract (Swedish)
- Klimatförändringar förväntas leda till längre brandperioder och ökad brandrisk. I Sverige, där homogen produktionsskog utgör största delen av landskapet, finns ett behov av bättre förståelse för skogens mottaglighet för bränder för att kunna mildra risker. I denna avhandling avser skogsbrandmottaglighet sannolikheten för att skogen brinner efter att antändning har ägt rum och inledande brandspridning har skett. Skogsbrandmottaglighet bidrar direkt till den risk som bränder utgör för miljö, samhälle och ekonomi. Det är väl fastställt att de huvudsakliga faktorerna som styr brandrisk är klimat och tillgängligt bränsle; av dessa kan endast bränsle manipuleras och hanteras. Det är därför viktigt att undersöka skogsbrandmottaglighet i... (More)
- Klimatförändringar förväntas leda till längre brandperioder och ökad brandrisk. I Sverige, där homogen produktionsskog utgör största delen av landskapet, finns ett behov av bättre förståelse för skogens mottaglighet för bränder för att kunna mildra risker. I denna avhandling avser skogsbrandmottaglighet sannolikheten för att skogen brinner efter att antändning har ägt rum och inledande brandspridning har skett. Skogsbrandmottaglighet bidrar direkt till den risk som bränder utgör för miljö, samhälle och ekonomi. Det är väl fastställt att de huvudsakliga faktorerna som styr brandrisk är klimat och tillgängligt bränsle; av dessa kan endast bränsle manipuleras och hanteras. Det är därför viktigt att undersöka skogsbrandmottaglighet i sammanhanget av skogsparametrar.
I denna studie utvärderades effekten av flera skogsparametrar på skogsbrandmottaglighet med hjälp av hypotesprövning och generaliserade linjära modeller (logistiska regressionsmodeller). Data bestod av avgränsade bränder från 2018, 2019 och 2020 samt SLU Skogskartans raster.
Utförd hypotesprövning identifierade signifikanta skillnader mellan brända och obrända klasser av trädålder (år), trädhöjd (m) och stående biomassa av tall (m³sk ha¯¹), björk (Betula spp.) (m³sk ha¯¹) och total lövskog (m³sk ha¯¹). Efter korrigering för multipel testning behöll endast trädålder och stående biomassa av tall signifikans, medan björk och total lövskog visade marginell signifikans.
Två logistiska regressionsmodeller tränades med 5-faldig korsvalidering, med 20% av datamängden undanhållen för noggrannhetstestning. Modell 1 tränades med ålder samt stående biomassa av total lövskog, tall och gran. Modell 2 tränades med stående biomassa av tall och gran. Noggrannheten för båda modellerna var låg, med AUC-värden på 0,556 för modell 1 och 0,551 för modell 2. I modell 1 hade total lövskogs stående biomassa och ålder inget signifikant bidrag till skogsbrandmottaglighet, stående biomassa av tall och gran hade vid bästa marginal en bidragande betydelse till skogsbrandmottaglighet (p-värden på 0,078 respektive 0,060), där tall ökade och gran minskade skogsbrandmottagligheten. Efter justering av p-värden för multipel testning bidrog varken tall eller gran signifikant (justerade p-värden på 0,312 respektive 0,240). I modell 2 visade sig stående biomassa av gran marginellt minska skogsbrandomottagligheten (p-värde på 0,034 och justerat p-värde på 0,067), och stående biomassa av tall hade ingen signifikans (justerat p-värde på 0,121) för skogsbrandmottaglighet.
Hypotesprövning framhöll olika variabler av betydelse jämfört med de som framkom i de logistiska regressionsmodellerna. Stående biomassa av gran var den enda variabeln med något signifikant bidrag till skogsbrandomottaglighet efter justering av p-värden för multipel testning (modell 2), vilken inte hade någon signifikant skillnad mellan brända och obrända klasser i hypotesprövning.
Mer forskning krävs med en större datamängd och förbättrade urvalsmetoder för att fullständigt förstå sambanden mellan skogsparametrar och skogsbrandmottaglighet i Sverige. (Less) - Abstract
- Climate change is expected to result in longer fire seasons and increased fire risk. In Sweden, where homogenous production forest makes up the majority of the landscape, there is a need for better understanding of forest fire susceptibility in order to mitigate risk. In the context of this thesis forest fire susceptibility is the likelihood of forest burning after ignition and initial fire propagation has taken place. Forest fire susceptibility contributes directly to the risk posed to the environment, society and economy by fire.It is well established that the main factors governing fire risk are climate and available fuel, of these, only fuel can be feasibly manipulated and managed. It is therefore important to investigate forest fire... (More)
- Climate change is expected to result in longer fire seasons and increased fire risk. In Sweden, where homogenous production forest makes up the majority of the landscape, there is a need for better understanding of forest fire susceptibility in order to mitigate risk. In the context of this thesis forest fire susceptibility is the likelihood of forest burning after ignition and initial fire propagation has taken place. Forest fire susceptibility contributes directly to the risk posed to the environment, society and economy by fire.It is well established that the main factors governing fire risk are climate and available fuel, of these, only fuel can be feasibly manipulated and managed. It is therefore important to investigate forest fire susceptibility within the context of forest parameters.
This study assessed the impact of several forest parameters on forest fire susceptibility, using hypothesis testing and Generalised Linear Models (logistic regression models). Data consisted of delineated fires from 2018, 2019 and 2020 and the SLU Forest Map rasters.
Hypothesis testing identified significant differences between the burned and unburned classes of tree age (yrs), tree height (m) and standing biomass volume of scots pine (m³sk ha¯¹), birch (Betula spp.) (m³sk ha¯¹) and total deciduous (m³sk ha¯¹). After correction for multiple testing only tree age and standing biomass volume of scots pine retained significance, with birch and total deciduous showing marginal significance.
Two logistic regression models were trained with 5-fold cross validation, while also holding back 20% of the dataset for accuracy testing. Model 1 was trained with age, and standing biomass volume of total deciduous, pine and spruce. Model 2 was trained with and standing biomass volume of pine and spruce. Accuracies for both models were low, with AUC values of 0.556 for model 1 and 0.551 for model 2. In model 1 total deciduous standing biomass volume and age had no significant contribution to forest fire susceptibility, standing biomass volume of pine and spruce had at best a marginal contribution to forest fire susceptibility (p values of 0.078 and 0.060 respectively), where pine increased and spruce decreased forest fire susceptibility. After adjusting p values for multiple testing neither pine nor spruce contributed significantly (adjusted p values of 0.312 and 0.240 respectively). In model 2 spruce standing biomass volume was found to marginally decrease forest fire susceptibility (p value of 0.034 and adjusted p value of 0.067) and pine standing biomass volume had no significance (adjusted p value of 0.121) on forest fire susceptibility.
Hypothesis testing highlighted different variables of importance than those of the logistic regression models. Spruce standing biomass volume was the only variable with any significant contribution to forest fire susceptibility after adjusting p-values for multiple testing (model 2), which had no significant difference between burned and unburned classes in hypothesis testing.
More research is needed with a larger dataset and improved sampling methods to fully understand the relationships between forest parameters and forest fire susceptibility in Sweden. (Less) - Popular Abstract
- Climate change is expected to result increased fire activity. In Sweden most of the landscape is made up of production forest, and we need to understand how susceptible this production forest is to fire, so that we can protect our environment, society and economy.
Forest fire susceptibility is how likely it is for forests to catch fire and for the fire to spread after it has started. Climate and fuel are the main factors that determine forest fire susceptibility. While we can’t feasibly control the climate we can manage the fuel available by changing what species of trees are planted.
I studied fires in Sweden from 2018, 2019 and 2020 using maps of the fires and forest maps. I wanted to identify if any specific tree species make a... (More) - Climate change is expected to result increased fire activity. In Sweden most of the landscape is made up of production forest, and we need to understand how susceptible this production forest is to fire, so that we can protect our environment, society and economy.
Forest fire susceptibility is how likely it is for forests to catch fire and for the fire to spread after it has started. Climate and fuel are the main factors that determine forest fire susceptibility. While we can’t feasibly control the climate we can manage the fuel available by changing what species of trees are planted.
I studied fires in Sweden from 2018, 2019 and 2020 using maps of the fires and forest maps. I wanted to identify if any specific tree species make a forest more or less likely to burn, using statistical tests and models.
The statistical tests showed that tree age, and the volume of scots pine growing in the forest may have an impact on forest fire susceptibility. The volume of deciduous trees in the forest may also play a role, but this was inconclusive. The models that were trained were not very accurate and were also inconclusive.
More research is needed with more fire data, better quality forest maps and better methods to really understand how different tree species may affect forest fire susceptibility in Sweden. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9139934
- author
- Jones, Sara Sharon LU
- supervisor
-
- Igor Drobyshev LU
- Veiko Lehsten LU
- organization
- course
- GISM01 20232
- year
- 2023
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Geography, GIS, Physical Geography, Forest Fire, Forest Fire Susceptibility, Sweden, logistic regression modelling
- publication/series
- Master Thesis in Geographical Information Science
- report number
- 166
- language
- English
- id
- 9139934
- date added to LUP
- 2023-10-12 11:16:09
- date last changed
- 2023-10-12 11:16:09
@misc{9139934,
abstract = {{Climate change is expected to result in longer fire seasons and increased fire risk. In Sweden, where homogenous production forest makes up the majority of the landscape, there is a need for better understanding of forest fire susceptibility in order to mitigate risk. In the context of this thesis forest fire susceptibility is the likelihood of forest burning after ignition and initial fire propagation has taken place. Forest fire susceptibility contributes directly to the risk posed to the environment, society and economy by fire.It is well established that the main factors governing fire risk are climate and available fuel, of these, only fuel can be feasibly manipulated and managed. It is therefore important to investigate forest fire susceptibility within the context of forest parameters.
This study assessed the impact of several forest parameters on forest fire susceptibility, using hypothesis testing and Generalised Linear Models (logistic regression models). Data consisted of delineated fires from 2018, 2019 and 2020 and the SLU Forest Map rasters.
Hypothesis testing identified significant differences between the burned and unburned classes of tree age (yrs), tree height (m) and standing biomass volume of scots pine (m³sk ha¯¹), birch (Betula spp.) (m³sk ha¯¹) and total deciduous (m³sk ha¯¹). After correction for multiple testing only tree age and standing biomass volume of scots pine retained significance, with birch and total deciduous showing marginal significance.
Two logistic regression models were trained with 5-fold cross validation, while also holding back 20% of the dataset for accuracy testing. Model 1 was trained with age, and standing biomass volume of total deciduous, pine and spruce. Model 2 was trained with and standing biomass volume of pine and spruce. Accuracies for both models were low, with AUC values of 0.556 for model 1 and 0.551 for model 2. In model 1 total deciduous standing biomass volume and age had no significant contribution to forest fire susceptibility, standing biomass volume of pine and spruce had at best a marginal contribution to forest fire susceptibility (p values of 0.078 and 0.060 respectively), where pine increased and spruce decreased forest fire susceptibility. After adjusting p values for multiple testing neither pine nor spruce contributed significantly (adjusted p values of 0.312 and 0.240 respectively). In model 2 spruce standing biomass volume was found to marginally decrease forest fire susceptibility (p value of 0.034 and adjusted p value of 0.067) and pine standing biomass volume had no significance (adjusted p value of 0.121) on forest fire susceptibility.
Hypothesis testing highlighted different variables of importance than those of the logistic regression models. Spruce standing biomass volume was the only variable with any significant contribution to forest fire susceptibility after adjusting p-values for multiple testing (model 2), which had no significant difference between burned and unburned classes in hypothesis testing.
More research is needed with a larger dataset and improved sampling methods to fully understand the relationships between forest parameters and forest fire susceptibility in Sweden.}},
author = {{Jones, Sara Sharon}},
language = {{eng}},
note = {{Student Paper}},
series = {{Master Thesis in Geographical Information Science}},
title = {{Tree Species Impact on Forest Fire Spread Susceptibility in Sweden}},
year = {{2023}},
}