Forest fire patterns are likely to be altered by climate change. We used boosted regression trees modelling and the MODIS Global Fire Atlas dataset (2003–15) to characterise relative influences of nine natural and human variables on fire patterns across five forest zones in China. The same modelling approach was used to project fire patterns for 2041–60 and 2061–80 based on two general circulation models for two representative concentration pathways scenarios. The results showed that, for the baseline period (2003–15) and across the five forest zones, climate variables explained 37.4–43.5% of the variability in fire occurrence and human activities were responsible for explaining an additional 27.0–36.5% of variability. The fire frequency was highest in the subtropical evergreen broadleaf forests zone in southern China, and lowest in the warm temperate deciduous broadleaved mixed-forests zone in northern China. Projection results showed an increasing trend in fire occurrence probability ranging from 43.3 to 99.9% and 41.4 to 99.3% across forest zones under the two climate models and two representative concentration pathways scenarios relative to the current climate (2003–15). Increased fire occurrence is projected to shift from southern to central-northern China for both 2041–60 and 2061–80.

Fire-induced forest loss has substantially increased worldwide over the last decade. In China, the connection between forest loss and frequent fires on a national scale remains largely unexplored. In this study, we used a data set for a time-series of forest loss from the Global Forest Watch and for a MODIS-derived burned area for 2003–2015 to ascertain variations in forest loss and to explore its relationship with forest fires (represented by burned areas) at the country- and forest-zone levels. We quantified trends in forest loss during 2003–2015 using linear regression analysis and assessed the relation between forest loss and burned areas using Spearman’s correlation. Forest loss increased significantly (264.8 km² a⁻¹; R² = 0.54, p < 0.01) throughout China, with an average annual increase of 11.4% during 2003–2015. However, the forest loss trend had extensive spatial heterogeneity. Forest loss increased mainly in the subtropical evergreen broadleaf forest zone (315.0 km² a⁻¹; R² = 0.69, p < 0.01) and tropical rainforest zone (38.8 km² a⁻¹; R² = 0.66, p < 0.01), but the loss of forest decreased in the cold temperate deciduous coniferous forest zone (− 70.8 km² year⁻¹; R² = 0.75, p < 0.01) and the temperate deciduous mixed broadleaf and coniferous forest zone (− 14.4 km² a⁻¹; R² = 0.45, p < 0.05). We found that 1.0% of China’s area had a significant positive correlation (r ≥ 0.55, p < 0.05) with burned areas and 0.3% had a significant negative correlation (r ≤ − 0.55, p < 0.05). In particular, forest loss had a significant positive relationship with the burned area in the cold temperate deciduous coniferous forest zone (16.9% of the lands) and the subtropical evergreen broadleaf forest zone (7.8%). These results provide a basis for future predictions of fire-induced forest loss in China.

A sub-surface forest fire is a type of smoldering combustion with a slower spread rate, longer combustion time, and lower combustion temperature compared with flame combustion. Sub-surface fires are usually accompanied by surface fires, and the surface fires’ conversion from sub-surface fires has great uncertainty. Therefore, there are considerable difficulties in monitoring and fighting sub-surface fires. However, there are few studies on the conversion from sub-surface fires to surface fires, and the mechanism and influencing factors of the conversion remain unclear. This study focuses on Larix gmelinii forests, which are representative of the boreal forest of China and hot spots of sub-surface fires, studies the moisture content limit of sub-surface fires’ conversion to surface fires by simulating a smoldering experiment, and establishes a monitoring model of sub-surface fires and an occurrence probability prediction model of sub-surface fires’ conversion to surface fires. The results showed that the moisture content limit of the conversion was 25% in the grass–Larix gmelinii forest and Ledum palustre–Larix gmelinii forest and 20% in Rhododendron dauricum–Larix gmelinii forest. There was a significant positive correlation between the time and temperature caused by the smoldering. The monitoring model of the sub-surface fires based on the surface temperature and moisture content had a good fitting effect (p < 0.01). The occurrence probability prediction model of the sub-surface fires’ conversion to surface fires, based on a logistic regression model, had high prediction accuracy (AUC = 0.987). The lower the moisture content of the humus, the closer the smoldering came to the surface and the higher the probability of conversion. This research could contribute to the study of the mechanism of sub-surface fires’ conversion into surface fires.