The wildfires
package provides easy access to data on the intersection
between social vulnerability and wildfire risk for the UK's nations:
- England: Middle Layer Super Output Area (MSOA)
- Wales: MSOA
- Scotland: Intermediate Zones
- Northern Ireland: Super Data Zones
This package is based on Hasan Guler's thesis "Spatial Assessment of Wildfire Vulnerability in England and Wales: Coupling Social Vulnerability with Predicted Wildfire Susceptibility," aiming to spatially assess wildfire vulnerability in England and Wales by coupling social and biophysical vulnerability.
The impact of wildfire widely vary based on the resilience of exposed communities, influenced by the time and location of the incident in conjunction with socioeconomic variables. It is therefore necessary to gain a deeper understanding of how socioeconomic disparities intersect with environmental hazards, to enable targeted interventions and informed decision-making in measures to decrease hazard exposure.
Install the development version from GitHub with:
# install.packages("devtools")
devtools::install_github("humaniverse/wildfires")
The package provides a comprehensive toolkit for analyzing social vulnerability and wildfire risks, including:
- Calculation of the Social Vulnerability Index (SoVI) using PCA.
- Wildfire risk prediction using Random Forest models.
- Combined datasets highlighting areas at high risk due to both social vulnerability and wildfire susceptibility.
- England:
wildfires::sovi_england
- Wales:
wildfires::sovi_wales
- Nothern Ireland:
wildfires::sovi_ni
- Scotland:
wildfires::sovi_scotland
Indicators of social vulnerability, which have been used to create the SoVI are also available for all UK nations.
- England & Wales:
wildfires::indic_msoa_eng_wales
- Nothern Ireland:
wildfires::indic_sdz_ni
- Scotland:
wildfires::indic_msoa_scotland
Summer
- England:
wildfires::wildfire_risk_summer_england
- Wales:
wildfires::wildfire_risk_summer_wales
- Nothern Ireland:
wildfires::wildfire_risk_summer_ni
- Scotland:
wildfires::wildfire_risk_summer_scotland
Spring
- England:
wildfires::wildfire_risk_spring_england
- Wales:
wildfires::wildfire_risk_spring_wales
- Nothern Ireland:
wildfires::wildfire_risk_spring_ni
- Scotland:
wildfires::wildfire_risk_spring_scotland
The datasets including indicators of social vulnerability at the MSOA level or equivalent for all UK nations were standardised using z-score normalization to ensure comparability. Then, correlation among the variables was assessed to ensure the suitability of PCA. The Kaiser-Meyer-Olkin (KMO) measure and Bartlett's test were applied, with high KMO scores and significant Bartlett's test p-values, justifying the PCA approach for all nations.
PCA was conducted using the psych
package's principal
function with
varimax rotation to enhance interpretability. Five components were
extracted based on eigenvalues, representing the underlying structure of
the data. The variance explained by each component was calculated, and
factor loadings were examined to understand the variable contributions.
The factor loadings shed light on the dimensions of variation within the data:
- Factor 1: Represents urban vulnerability, highlighted by housing conditions and ethnic compositions.
- Factor 2: Captures socio-economic and health-related vulnerabilities, particularly around disability and education levels.
- Factor 3: Emphasizes linguistic and ethnic diversity, alongside housing conditions.
- Factor 4: Reflects geographic mobility and housing market dynamics, indicating areas with high internal migration and private rental activity.
- Factor 5: Distinguishes demographic structures, particularly the balance between younger and older populations.
Finally, the Social Vulnerability Index (SoVI) was calculated as a weighted sum of the component scores, with weights based on the variance explained by each component. The SoVI scores were then standardized within country.
Data Collection
Spatial data on fire occurrences (dependent variable) and various independent variables was collected from a variety of sources. Independent variables were categorized into topographical, climatological, vegetation cover, and anthropogenic factors.
Preprocessing and Data Preparation
Independent variables were organized into raster stacks for both spring and summer seasons, and standardised using z-score normalization.
Pseudo-background points, serving as absence data, was generated in a quantity equal to that of the recorded fire points for the season. These points were randomly distributed across the UK within the country's boundaries.
Model Training and Validation
Two random forest models were trained on a subset of the data - one for summer data, the other for spring.
Performance Metrics & Feature Importance
Using another subset of the data, the models' performances were assessed using a confusion matrix from which accuracy, precision, recall, and F1 score were derived.
The significance of each independent variable in predicting wildfires was also evaluated through feature importance analysis, identifying the most influential factors for fire occurrence.
Wildfire Risk Index Creation
Utilizing both Random Forest models, distinct probabilities of wildfire occurrences were predicted across the UK for summer and spring. They were then aggregated to the MSOA level, resulting in the Wildfire Risk Index. Missing values were imputed using higher geography averages when necessary. Similarly to the Social Vulnerability Index, the Wildfire Risk Index was standardized separately for England, Wales, Scotland, and Northern Ireland to enable comparison within each nation.
If you encounter a clear bug, please file an issue with a minimal reproducible example on GitHub.
Please note that this project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms.