Boreal forests are globally extensive and store large amounts of carbon, but recent climate change has led to drier conditions and increasing fire activity. The objective of this study is to quantify trends in fire size and frequency using data spanning multiple scales in space and time. We use multi-temporal Landsat image compositing on Google Earth Engine and validate results with reference fire maps from the Canadian Park Service. We also interpret general fire trends through the concept of Self-Organized Criticality (SOC). Our study site is Wood Buffalo National Park, which is a fire hot spot in Canada due to frequent lightning ignitions. The relativize differenced normalized burn ratio (RdNBR) was the most accurate Landsat-based burn severity metric we evaluated (52.2% producer's accuracy, 87.6% user's accuracy). The Landsat-based burn severity maps provided a better fit for a linear relationship on the log-log scale of fire size and frequency than a manually drawn fire map. Landsat-based fire trends since 1990 conformed to a power-law distribution with a slope of 1.9, which is related to fractal dimensions of the satellite-based fire perimeter shapes. The unburned and low-severity patches within the burn severity mosaic influenced the power-law slope and associated fractal dimensionality. This study demonstrates a multi-scale and multi-dataset technique to quantify general fire trends and changing fire cycles in remote locations and establishes a baseline database for assessing future fire activity. Testing criticality by power laws helps to quantify emergent trends of contemporary fire regimes, which could inform the strategic application of prescribed fire and other management activities. Natural resource managers can utilize information from this study to understand local ecosystem adaptability to large fire events and ecosystem stability in the context of recent increasing fire activity.