The Loess Plateau, a typical ecologically fragile region in China, is increasingly facing complex and intensified wildfire risks due to the combined impacts of climate change and human activities. Identifying the spatiotemporal patterns of wildfire hazard and assessing its potential ecological impacts are crucial for regional ecological security and fire prevention. Based on multi-period wildfire data from 2001 to 2024, the temporal evolution patterns of wildfire activity were systematically analyzed, a wildfire hazard model was constructed using machine learning algorithms, and it was coupled with ESV(ecosystem service value ) to conduct an ecological risk assessment.The results show that: wildfire hazard is characterized by significant spatiotemporal heterogeneity. A significant increasing trend is observed in wildfire activity during summer and autumn as well as during the non-fire-prevention period, revealing a new feature of increasing fire risk outside the traditional fire-prevention period. During spring and winter, as well as during the fire-prevention period, the high-hazard areas are more extensively distributed, exhibiting a continuous zonal pattern. The driving mechanisms of hazard factors are found to differ across periods, and the seasonal variations in the contributions of meteorology, vegetation, and human activities are identified. The spatial distribution of ESV is characterized by a gradient pattern, with higher values in the southeast and lower values in the northwest. Spring and winter, as well as the fire-prevention period, are identified as the critical periods for ecological risk management and control, during which the high-risk areas are characterized by strong continuity, and the expected ESV losses are significantly greater than those in other periods. The findings provide scientific support for developing spatiotemporally differentiated fire prevention strategies and for maintaining ecological security in the Loess Plateau.