Dataset

Wildfire Susceptibility Mapping for South-eastern Australia by Evolutionary Algorithms and Statistical Methods

University of New South Wales
Lim, Samsung ; Jia, Xiuping
Viewed: [[ro.stat.viewed]] Cited: [[ro.stat.cited]] Accessed: [[ro.stat.accessed]]
ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Adc&rfr_id=info%3Asid%2FANDS&rft_id=info:doi10.26190/bqpn-8b33&rft.title=Wildfire Susceptibility Mapping for South-eastern Australia by Evolutionary Algorithms and Statistical Methods&rft.identifier=https://doi.org/10.26190/bqpn-8b33&rft.publisher=UNSW&rft.description=Australia is one of the most flammable counties due to fuel accumulation and frequent droughts. The number and size of wildfire incidents have increased during the last decades. Global warming, industrialisation and extensive human activities played an important role in the increase of wildfire incidents. Wildfires are a considerable threat to human lives and properties, especially in populated areas. In addition, wildfires will negatively impact the components of our ecosystem such as vegetation, soil, water and forests.Wildfire susceptibility maps show the areas with different probabilities of fire occurrence. These maps help managers and policymakers to act efficiently and reduce the negative impacts of wildfires. Many models were created by Geospatial Information System (GIS) and Remote Sensing (RS) to predict wildfires.This thesis aims to investigate wildfire susceptibility in Victoria located in south-eastern Australia with an area of 227,444 km2. The elevation in this area ranged between -76 m to 1,986 m. More than a million hectares burned in Victoria in the last bushfire season in 2019-2020. In addition, more than 110 homes or businesses were destroyed during this period. A wildfire susceptibility model could be a useful tool to control and manage the future wildfires by predicting vulnerable areas.This study aims to generate wildfire susceptibility maps for the south-eastern part of Australia. The main research objectives are as follows.1. To generate a wildfire inventory map from the Moderate Resolution Imaging Spectroradiometer (MODIS) data.2. To develop the conditioning factors and map layers.3. To generate wildfire susceptibility maps using statistical methods e.g., Frequency Ratio (FR) and Logistic Regression (LR) and evolutionary algorithms separately.4. To apply ensemble techniques (statistical methods combined with evolutionary algorithms) to generate wildfire susceptibility maps.5. To evaluate the performance of the proposed methods by using the Receiver Operating Characteristics (ROC) curve.   Methodology Topography, meteorology, hydrology and vegetation data will be prepared prior to the creation of the model. Digital Elevation Model (DEM) will be provided from the National Aeronautics and Space Administration (NASA) (ASTER 30-m GDEM). Elevation, slope, aspect and topographic wetness index will be extracted from DEM. The distance to rivers and roads data will be collected from Open Street Map website by mean Euclidean distance. Temperature and rainfall data will be collected from the Bureau of Meteorology for the study area. Land cover data, NDVI will be collected from NASA (MODIS 1-km MYD13A3 NDVI) and wildfire inventory map also obtained from NASA (MODIS 500-m MCD 64 Monthly).Meteorologically related factors will be mapped with the inverse distance weighted (IDW) interpolation method. All applied conditioning factors will be prepared as a separate map layer with the same resolution and will be categorised based on their attributes and relationships with wildfire susceptibility.The correlation between conditioning factors will be calculated using multicollinearity analysis. The inflation factors (VIF) and tolerances (TOL) will be used to evaluate degree of multicollinearities of the conditioning factors . The factors with TOL < 0.1 or VIF > 10 will be categorised as multicollinear.We will use 70% of wildfires randomly for calibration and 30% for validation of fire susceptibility models based on Majority of the previous research.Statistical methods such as logistic regression will be used to determine the wildfire susceptibility mapping. Then GP and GEP will be used to generate the wildfire susceptibility as well. GA is one of the first population-based algorithms inspired by the biological evolution and Darwin’s natural selection theory. Briefly, GA algorithm starts with a random population which represent chromosomes of individuals. Individuals that are most fit is more likely to survive and have more chance to reproduce. Operations have been conducted on individuals to create the next generation. Genetic operators included reproduction, crossover and mutation. Reproduction copies the individual in the next generation. In crossover, two solutions (parent solution) combined to produce two new solutions (children solution). In mutation, one or multiple genes are altered in chromosome and create new solution. The rate of mutation in GA are low. The mutation operator introduces another level of randomness and maintain the diversity of the solutions.Finally, the hybrid technique (combinations of methods) will be used to increase the chance of obtaining a more accurate model for wildfire probabilities. The ROC will be used to evaluate the accuracy of the generated models.&rft.creator=Lim, Samsung &rft.creator=Jia, Xiuping &rft.date=2021&rft_rights=University of New South Wales 2021 &rft_rights=Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International &rft_subject=Wildfire&rft_subject=Gene Expression Programming&rft_subject=ArcGIS&rft_subject=Susceptibility&rft_subject=CIVIL ENGINEERING&rft_subject=ENGINEERING&rft.type=dataset&rft.language=English Access the data

Licence & Rights:

Non-Derivative Licence view details
CC-BY-NC-ND

Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International

University of New South Wales 2021

Access:

Open

Contact Information

Laptop, RDS Archive

Full description

Australia is one of the most flammable counties due to fuel accumulation and frequent droughts. The number and size of wildfire incidents have increased during the last decades. Global warming, industrialisation and extensive human activities played an important role in the increase of wildfire incidents. Wildfires are a considerable threat to human lives and properties, especially in populated areas. In addition, wildfires will negatively impact the components of our ecosystem such as vegetation, soil, water and forests.Wildfire susceptibility maps show the areas with different probabilities of fire occurrence. These maps help managers and policymakers to act efficiently and reduce the negative impacts of wildfires. Many models were created by Geospatial Information System (GIS) and Remote Sensing (RS) to predict wildfires.This thesis aims to investigate wildfire susceptibility in Victoria located in south-eastern Australia with an area of 227,444 km2. The elevation in this area ranged between -76 m to 1,986 m. More than a million hectares burned in Victoria in the last bushfire season in 2019-2020. In addition, more than 110 homes or businesses were destroyed during this period. A wildfire susceptibility model could be a useful tool to control and manage the future wildfires by predicting vulnerable areas.This study aims to generate wildfire susceptibility maps for the south-eastern part of Australia. The main research objectives are as follows.1. To generate a wildfire inventory map from the Moderate Resolution Imaging Spectroradiometer (MODIS) data.2. To develop the conditioning factors and map layers.3. To generate wildfire susceptibility maps using statistical methods e.g., Frequency Ratio (FR) and Logistic Regression (LR) and evolutionary algorithms separately.4. To apply ensemble techniques (statistical methods combined with evolutionary algorithms) to generate wildfire susceptibility maps.5. To evaluate the performance of the proposed methods by using the Receiver Operating Characteristics (ROC) curve.

 

Methodology

Topography, meteorology, hydrology and vegetation data will be prepared prior to the creation of the model. Digital Elevation Model (DEM) will be provided from the National Aeronautics and Space Administration (NASA) (ASTER 30-m GDEM). Elevation, slope, aspect and topographic wetness index will be extracted from DEM. The distance to rivers and roads data will be collected from Open Street Map website by mean Euclidean distance. Temperature and rainfall data will be collected from the Bureau of Meteorology for the study area. Land cover data, NDVI will be collected from NASA (MODIS 1-km MYD13A3 NDVI) and wildfire inventory map also obtained from NASA (MODIS 500-m MCD 64 Monthly).Meteorologically related factors will be mapped with the inverse distance weighted (IDW) interpolation method. All applied conditioning factors will be prepared as a separate map layer with the same resolution and will be categorised based on their attributes and relationships with wildfire susceptibility.The correlation between conditioning factors will be calculated using multicollinearity analysis. The inflation factors (VIF) and tolerances (TOL) will be used to evaluate degree of multicollinearities of the conditioning factors . The factors with TOL < 0.1 or VIF > 10 will be categorised as multicollinear.We will use 70% of wildfires randomly for calibration and 30% for validation of fire susceptibility models based on Majority of the previous research.Statistical methods such as logistic regression will be used to determine the wildfire susceptibility mapping. Then GP and GEP will be used to generate the wildfire susceptibility as well. GA is one of the first population-based algorithms inspired by the biological evolution and Darwin’s natural selection theory. Briefly, GA algorithm starts with a random population which represent chromosomes of individuals. Individuals that are most fit is more likely to survive and have more chance to reproduce. Operations have been conducted on individuals to create the next generation. Genetic operators included reproduction, crossover and mutation. Reproduction copies the individual in the next generation. In crossover, two solutions (parent solution) combined to produce two new solutions (children solution). In mutation, one or multiple genes are altered in chromosome and create new solution. The rate of mutation in GA are low. The mutation operator introduces another level of randomness and maintain the diversity of the solutions.Finally, the hybrid technique (combinations of methods) will be used to increase the chance of obtaining a more accurate model for wildfire probabilities. The ROC will be used to evaluate the accuracy of the generated models.

Valid: 14 09 2020 to 14 09 2024

Click to explore relationships graph
Subjects

User Contributed Tags    

Login to tag this record with meaningful keywords to make it easier to discover

Identifiers