Spatio-temporal dynamics of forest and grassland fire risk in Xinjiang and quantification of variable influence

• Step_RF model is the optimal model for simulating forest and grassland fire risk in Xinjiang. • BEAST breakpoint analysis was innovatively employed to detect trend changes of fire risk. • In 2015-–2019, nearly half of the region experienced conversion of trend from increasing to decreasing. • Aboveground biomass had a significant influence on the change of fire risk. • The contributions of different driving factors exhibit spatial heterogeneity. Under the dual pressures of increasing extreme climate events and the expanding scope of human activities, forest and grassland fires are becoming more frequent and severe. Developing high-precision fire risk assessment models and identifying key driving factors are essential for effective fire prevention and control. However, selecting appropriate modeling variables and revealing the spatial heterogeneity of driving forces remain insufficient in current research. To address these issues, this study compared and analyzed multiple variable selection methods (PCA, LASSO, Stepwise Regression) coupled with machine learning algorithm (RF, SVM, BP-ANN) to identify the optimal modeling framework for simulating the spatial distribution of forest and grassland fire risk in Xinjiang from 2001 to 2022. Furthermore, we detected temporal breakpoints and trend conversions, and quantified the contribution of driving factors. The results indicate that: (1) Step_RF model is the optimal model (Acc = 89.31 %, F1 = 88.19 %, AUC = 0.9546) for simulating forest and grassland fire risk in Xinjiang; (2) Spatially, fire risk exhibited a decreasing gradient from northwest to southeast, with higher risk in regions of high altitude and high vegetation coverage; (3) Most of the region showed a significant increasing trend, nearly half of the region experienced a trend shift from increasing to decreasing in 2015–2019; (4) All variables exhibited bidirectional effects, with aboveground biomass showed a particularly pronounced positive influence in shrublands, open forest land, medium-coverage and low-coverage grasslands. This study provides scientific support for the precise prevention and control of forest and grassland fires under the context of climate change.