The boreal forests of the Northern Hemisphere (i.e., covering the USA, Canada and Russia) are the grandest carbon sinks of the world. A significant increase in wildfires could cause disequilibrium in the Northern boreal
forest’s capacity as a carbon sink and cause significant impacts on wildlife and people worldwide. That is why the ability to forecast wildfires is essential in order to minimize all risks and vulnerabilities. We present a novel
methodology utilizing the Bayesian Machine Learning models to identify climatic variations that induce high and low wildfire activity cycles and forecast long-term occurrences of wildfires. The data analyzed are observed
records of wildfires, climate change and climate teleconnections, atmospheric, oceanographic, and environmental factors, starting from the first half of the 20th century. Our Bayesian machine learning models show that a
new phase of high wildfire activity in the USA, Canada and Russia began in 2020. While USA has a detectable, oscillation of 40 ±5 years; Russia and Canada have oscillatory patterns of 30 ±5 and 60 ±5 years, respectively.
Also, our Machine Learning model forecasts peak wildfire activity at around 2022 ±3,2035 ±3, and 2045 ±5 years for USA, Russia, and Canada, respectively. The new high wildfire activity phase will persist in Russia, USA,
and Canada, until 2045, 2030, and 2055, respectively.