Discussion & Conclusion
Throughout the duration of this project a number of errors quickly became apparent, which reduced the accuracy and precision of the final vulnerability raster. To begin, there are only 14 weather stations in Mackenzie County, most of which are located in the south east. The interpolation tool used to create the output for the likelihood model predicts values for raster cells based on these data points, but assumes that these data points are all spatially correlated. Due to the small number of weather stations distributed across Mackenzie County however, this is not the case. As a result the humidity, temperature, wind, and precipitation variables used during the analysis are stretched and fail to reflect reality. For the purposes of this study however, the values generated by the interpolation tool are acceptable. This is mainly the result of the scale. Currently the final vulnerability raster is represented by cells with a scale of 1 kilometer by 1 kilometer - very large overall. This allows room for some level of error or generalization.
In terms of the lightning data, this information is at a fairly coarse resolution. The data used in order to produce the likelihood raster is derived from a map produced by the Alberta government, rather than from original source data alone. Although the map generated as a result of digitization mirrors the original map fairly well, it is important to recognize that this is not the most reliable way to utilize data for the purpose of creating a model and correponding map. There is a risk of introducing human error to the study. People often fail to capture the physical features on a map accurately; this is known as positional error.
In order for the models within this study to identify the areas that are more likely to experience a forest fire or be increasingly susceptible to a forest fire, it is essential that weights are selected that accurately protray this. Unfortunately selecting weights is largely dependent on the user, meaning that bias and inaccuracy can easily be introduced into a study. Although the weights used for this project are reflective of scientific literature, some level of inaccuracy remains as a result of bias.
Lastly, before running the model, some input data was clipped to the study area, which can cause errors. For instance, a road ouside the study area may still have an effect on the study area. To mitigate this error, the data should have been clipped to a widely buffered study area (at least 100km on either side), and then only clipped to the study area for the final result.
Limitations & Improvements
Although this study provides meaningful and applicable results that can be used to adequately predict the areas of highest vulnerability to forest fires in Mackenzie County, Alberta, there are a number of improvements that could potentially be applied to this study if research was to be continued in the future. These could further improve the validity of the study, as well as the results generated as a component of both the likelihood and susceptibility models.
Many of the suggested improvements for this study are related to the expansion of the source data and using data that is of a higher resolution. This could include finding data that more accurately represents human land use in Mackenzie County, rather than using data that only represents roads and powerlines. This would be more effective, as no bias or assumptions regarding data are mistakenly intrduced into the study. This would improve the likelihood of the model immensely.
Local landuse data was difficult to obtain in Mackenzie County. In fact, for the purposes of this study it could not be acquired at all. This is primarily why the data used to determine ignition likelihood largely revolves around relatively basic human activities such as proximity to roads, towns, and parks. If this study was conducted at a smaller scale it would be possible to invest more time in determining whether other local trails, ATV tracks, and other recreational areas in this county exist, especially in small, isolated communities that may be significantly impacted as a result of forest fire.
If this research project is conducted again, it is also possible to use data containing a record of previous lightning-caused fires. This would be beneficial as a model could be created using historical data that accurately protrays where forest fires initiated by lightning occur, therefore indicating what areas may be more likely to have lightning-based fires in the future. This could be accomplished by creating a density raster of the historical fires. This would act to supplement the likelihood model, as the data that is currently being used in the likelihood and susceptible models for Mackenzie County is severely lacking in complete accuracy as it was digitized and adapted to protray a smaller scale. Additionally a study such as this could be used to inform further study, pinpointing regions in the county that may be more vulnerable or susceptible to fire. This may aid mitigation efforts and attempts to reduce the likelihood of fire.
Using higher resolution lighting strike data would be beneficial as well. As with the previous suggested lightning data improvement, this would help better pinpoint areas that are at high risk of frequent lightning strikes. In addition it is also important to recognize that lightning is random and unpredictable, as a result, it is very challenging to create a model with results that adequately predict areas of extreme risk or vulnerabilty with perfect accuracy or certainty.
High quality weather data for the region would greatly improve the accuracy of the model. Mackenzie County has a total of 14 weather stations, which do not offer very good coverage of such a large area. It is difficult to assume that merely 14 points of data concentrated primarily in the southeast have the ability to accurately represent the weather patterns that occur for over 80,458.19 square kilometers of land (Statistics Canada, 2019). It is clear that there is a high risk of error from interpolation as a result of this. By possibly supplimenting the station data with satellite-based imagery and information, the resolution and accuracy of this component of the research project can be greatly increased. Additionally, since many of these stations are remote, automatic units, they sometimes go down for a month or more at a time leading to gaps in weather data within a season.
Implementing any of the improvements above and re-running the model would be a good next step to improve the current results. More importantly however, it is possible to use this study as an initial starting point and focus on the areas of extreme vulnerability in Mackenzie County. By doing this it would be possible to create a new study area at a larger scale that is significantly more accurate.
The study area used for the purposes of this project is currently represented by pixels at a scale of 1 kilometer by 1 kilometer. Despite the fact that this does successfully generate models that provide a general overview of vulnerability, very little is actually accomplished in terms of representing those specific regions that may be severely impacted by forest fire. If this analysis is to be conducted once again it is important that the focus of the study is on a much smaller region, perhaps a small city or town that is highly vulnerable or at a high level of risk for fire in the county. As the literature suggests, this would allow planners, developers, and community members to have a better understanding of the risks associated with fire, as well as understand what areas require the extensive use of mitigation activities and technologies such as fire resistant building materials, fire buffers, and the removal of flammable materials in forested areas (ISL Engineering and Land Service, 2009; Alberta Agriculture and Rural Development, 2015). Additionally, a study such as this may inform legislation or act to improve the current laws or protocols in place that alleviate the likelihood of fire, one example being bylaw 811-11 that defines public responsibility in terms of preventing fire in Mackenzie County.
At the conclusion of this initial study it may also be possible to conduct a hotspot analysis in order to determine what specific areas within Mackenzie County are statistically significant in terms their vulnerability to forest fire. Individual cells with similar values will be clustered together, creating areas in the final map that either protray a very high or low level of vulnerability overall. This will lend valuable insight into what areas of the county should be focused on in the event that the scale of this project is reduced and this study is conducted again in the future.
The purpose of this research project was to use GIS and remote sensing to identify the areas of Mackenzie County, Alberta that are highly vulnerable to forest fire. This was accomplished by creating a susceptibility and likelihood model. When combined together, these models create a vulnerability raster, highlighting the regions in the county that are more vulnerable to flame. Based on the results of this study, areas that are in close proximity to roads and settlements are at the highest level of vulnerability. The regions of lowest vulnerability are primarily in the northeast due to their decreased proximity to human activity, as well as their close proximity to the conditions that act to mitigate fire such as an increased likelihood of rain. In the future it is possible for this study to be used to guide planning and development decisions which may act to combat vulnerability in communities or settlements that are at a high risk for experiencing forest fires.