Introduction The Canadian Forest Fire Danger Rating System (CFFDRS) is a globally known wildland fire risk assessment system, and two major components, the fire weather index system and the fire behavior prediction system, have been extensively used both nationally and internationally to aid operational wildland fire decision making. Methods In this paper, we present an overview of an R package cffdrs, which is developed to calculate components of the CFFDRS, and highlight some of its functionality. In particular, we demonstrate how these functions could be used for large data analysis. Results and Discussion With this cffdrs package, we provide a portal for not only a collection of R functions dealing with all available components in CFFDRS but also a platform for various additional developments that are useful for the understanding of fire occurrence and behavior. This is the first time that all relevant CFFDRS methods are incorporated into the same platform, which can be accessed by both the management and research communities.

A 3.6 ha experimental fire was conducted in a black spruce peatland forest that had undergone thinning the year prior. After 50 m of spread in a natural stand at 35–60 m min⁻¹, the crown fire (43,000 kW m⁻¹ intensity using Byram’s method) encountered the 50% stem removal treatment; spread rates in the treatment were 50–60 m min⁻¹. Fuel consumption in the control (2.75 kg m⁻²) was comparable to the treatment (2.35 kg m⁻²). Proxy measurements of fire intensity using in-stand heat flux sensors as well as photogrammetric flame heights had detected intensity reductions to 30–40% of the control. Crown fuel load reductions (compensated by higher surface fuel load) appear to be the most significant contributor to the decline in intensity, despite drier surface fuels in the treatment. The burn depth of 5 cm in moss and organic soil did not differ between control and treatment. These observations point to the limited effectiveness (likely reductions in crown fire intensity but not spread rate) of stem removal in boreal black spruce fuel types with high stem density, low crown base height and high surface fuel load. The observed fire behaviour impacts differ from drier conifer forests across North America.

An experimental fire was conducted in one-year-old mulched (masticated) boreal fuels, where all aboveground biomass was mulched with no stems removed or left standing. Typical mulching practices remove remnant biomass; leaving biomass in situ reduces overall management input. While fuel quantities were not explicitly reduced, availability of fuels to fire was reduced. Infrared imagery was obtained to quantify rate of spread and intensity to a 1 m resolution. In-stand totalizing heat flux sensors allowed for the observation of energy release near the surface. When compared with the pre-treatment fuel-type M-2 (mixedwood, 50% conifer), rates of spread were reduced 87% from an expected 8 m min⁻¹ to observed values 1.2 m min⁻¹. Intensity was also reduced from 5000 kWm⁻¹ to 650kWm⁻¹ on average. Intermittent gusts caused surges of fire intensity upwards of 5000 kW m⁻¹ as captured by the infrared imagery. With reference to a logging slash fuel type, observed spread rates declined by 87% and intensity 98%. Independent observations of energy release rates from the radiometers showed similar declines. As mulching is a prevalent fuel management technique in Alberta, Canada, future studies will contribute to the development of a fire behaviour prediction model.

Background: Wildfire behavior models such as the Canadian Fire Behavior Prediction (FBP) system predict wildfire characteristics given fuel, weather and topography data. The Wildfire Intelligence and Simulation Engine (W.I.S.E.) builds upon this system to predict fire spread, supporting wildfire preplanning and response. Despite their usefulness, assessment of fire spread models remains difficult due to the scarcity of real fire spread data. Aims: A fire spread model assessment procedure is presented comparing fire spread predictions to thousands of satellite-derived daily wildfire boundaries. Methods: W.I.S.E. is used to generate spread predictions for 19,848 individual days of 2210 wildfires using historical data. Predictions are compared to the Canadian Fire Spread Dataset (CFSDS) using normalized area difference, precision, recall, F1 score, Intersection-over-Union (IoU) and Hausdorff distance. Key results: Using default model parameters, fire spread predictions achieve an average F1 score of 0.259. Diagnostic scenarios are also performed to understand model sensitivity: scenario 2 optimizes burn duration, reaching an average F1 score of 0.498. Scenario 3 optimizes both wind direction and burn duration; F1 score rises to 0.539. Conclusions: The assessment method provides an effective, logical and reproducible measure of perimeter prediction accuracy. Implications: These findings suggest that the assessment method may be used to assess and improve wildfire behavior prediction models.