Reduce preparedness and suppression costs by an estimated 3% ($750K annually).
Enhance wildfire spread projections and improve emergency response decisions, increasing public and firefighter safety.
Identify high-risk tree stands near communities and other important areas.
Following an extreme 2019 wildfire season, Alberta’s Ministry of Agriculture and Forestry commissioned a report to understand how wildfire operational processes and practices are performed, especially in extreme circumstances. This review highlighted that a modernized preparedness planning framework that balanced risk, values, hazard and cost will improve overall wildfire outcomes.
Following this, the Government of Alberta began working with AltaML to explore how state-of-the-art Artificial Intelligence/Machine Learning (AI/ML) techniques could support the Wildfire Management Branch in their efforts to allocate firefighting resources more effectively.
This project’s goal was to provide duty officers with an improved ability to predict the spread of a wildfire outbreak, conduct wildfire mitigation activities, and ultimately increase public and firefighter safety.
Duty Officers of the Wildfire Management Branch (WMB) of the Government of Alberta (GoA) use fuel grid maps to manage wildfire risk throughout the province. These maps primarily include classification of tree species, including identification of dead trees. As these are the most potent natural fuel for wildfires, they are a crucial part of directing fire prevention activities as well as prioritizing emergency response to wildfire outbreaks.
Currently, discrete regions of the provincial fuel grid map are updated using forest health survey flights and historical wildfire data. However, this approach only covers a small portion of the province every year and, consequently, many parts of the map are out of date. An out-of-date fuel grid map means projections on wildfire spread can be inaccurate and endanger the lives of firefighters and the public in surrounding communities. Thus, the objective of this use case was to find a way to generate and maintain a more accurate provincial fuel grid map.
Satellite imaging can be used to identify dead trees province-wide. However, the sheer size of Alberta makes it impractical for people to manually identify dead trees from satellite data (over 2.6 billion pixels at the resolution necessary to identify dead tree stands). Thus, AltaML developed an ML model that can identify dead trees in satellite imaging data, thereby informing updates to the fuel grid.
In the initial stages of the engagement, AltaML relied on research showing that dead vegetation can be differentiated from green vegetation with spectral indices. Based on this principle, the team built a machine learning model to classify forested areas, first into dead or alive vegetation, then into nine Fire Behaviour Prediction (FBP) fuel classes.
The model outputs shape files that can be input directly into the forest health’s ArcGIS mapping software. This supports enhanced decision-making when responding to fire behaviour and when planning prevention and presupression activities. In addition to this improvement in public safety, this solution also reduces time spent manually identifying changes to fuel types (that are not automatically captured to date). Regarding accuracy, our partners estimate that the high spatial accuracy of the ML-based polygons could outperform polygons manually acquired on fixed wing aircraft flights.
AltaML is planning to conduct a full pilot in the spring of 2023 with the wildfire management team. In addition, we are exploring the possibility of bringing these benefits, along with our growing portfolio of wildfire solutions, to other regions. If you’d like to connect us with potential customers, discuss opportunities or problems that could benefit from advanced analytics, please give us a shout at email@example.com.