华北地区地形复杂，太行山、燕山山脉植被覆盖度高，冬春季林火灾害频发，针对中小尺度下林火蔓延模拟预报需求，本文采用大气-野火耦合模式WRF-SFIRE，对2021年“2·20”晋冀交界林火过程进行高分辨率模拟研究。通过引入 GDEMV3 30 m 地形数据、GLC_FCS30_2020 30 m 土地利用数据、ESA_WorldCover 10 m地表覆盖度数据和CLDAS高精度土壤温湿度初始场，开展包括地形、土地利用、可燃物、陆面同化驱动、大涡模拟等七组敏感性试验。结果表明：WRF-SFIRE模式能较好地再现风速、风向和温度的日变化过程及概率分布特征，模拟气象场与观测具有良好一致性。林火过程清晰表现出启动、平稳、加速与衰阶的阶段性演变特征，过火区的空间形态与时间演变序列较好地反映了实际火势发展的规律，且模拟的过火范围与遥感反演结果基本吻合。不同场景的敏感性试验表明，各类高分辨率静态数据可显著改善风场结构与火行为的响应，其中土地利用与燃料精度对火势扩展尤为敏感，地形高程则在复杂地形下显著调控林火行为反馈过程。另外，CLDAS土壤温湿度作为驱动场有效提升了 2 m温度的模拟精度，验证了其在中小尺度林火预报中的应用价值。相比之下，在 1 km分辨率下直接开启大涡模拟并未改善模拟精度，反而引入风向扰动增强等不稳定因素，建议在中尺度下谨慎使用。综上所述，本研究聚焦中小尺度林区，评估了大气-野火耦合模型在林火数值模拟中的应用潜力，并解析了林火蔓延过程中的气象演变与火场响应机制。研究结果强调，合理整合多源高分辨率数据集和真实的下垫面条件是提升林火预报能力的关键。本研究不仅丰富了复杂地形下气火相互作用的理论认识，也为森林草原防灭火工作及华北地区生态系统的长期保护提供科学支撑。

North China is characterized by complex topography，featuring dense vegetation in the Taihang Mountains and Yanshan Mountains. Forest fires frequently occur during winter and spring. To address the need for simulating and predicting forest fire spread at meso-and micro-scale，this study employs the coupled atmo‐ spheric-wildfire model WRF-SFIRE to conduct high-resolution simulations of a forest fire event that occurred at the Jin-Ji junction during the“2-20”period in 2021. Seven sensitivity experiments are designed to assess the ef‐ fects of topography，land use，fuel type，land surface assimilation forcing，and large eddy simulation. The exper‐ iments incorporate GDEMV3 30 m topography，GLC_FCS30_2020 30 m land use，ESA_WorldCover 10 m land cover，and CLDAS high-precision soil temperature and moisture initial fields. The results demonstrate that the WRF-SFIRE model effectively reproduces the diurnal variations and probability distribution of wind speed，wind direction，and temperature，demonstrating strong agreement between simulated and observed meteorological fields. The simulated fire evolution exhibits distinct stages of ignition，stabilization，acceleration，and decay. The spatial and temporal characteristics of the simulated burn area closely match satellite-based fire detection re‐ sults，accurately reflecting the real fire development process. The sensitivity experiments under different configu‐ rations reveal that incorporating high-resolution static datasets substantially improves the model’s representation of wind field structure and fire behavior. Notably，land use and fuel accuracy have the most pronounced influ‐ ence on fire spread，while topographic elevation significantly modulates the feedback process of forest fire be‐ havior in complex terrain. Moreover，introducing CLDAS soil temperature and moisture fields as driving inputs notably enhances the simulation accuracy of 2 m temperature，which verified its value in meso-and micro-scale forest fire forecasting. Conversely，enabling large eddy simulation（LES）at 1 km resolution did not improve t performance and instead introduced wind-related instabilities，suggesting that LES should be applied cautiously at mesoscale resolutions. In summary，this study evaluates the applicability of the WRF-SFIRE model in meso‐ scale and microscale forest regions，analyzes the meteorological evolution and fire response mechanisms during the spread of forest fires. The findings highlight that integrating multi-source high-resolution datasets with authen‐ tic land surface conditions is crucial for enhancing wildfire forecasting capabilities. This research not only ad‐ vances the theoretical understanding of atmosphere-fire interactions in complex topography，but also provides scientific support for forest and grassland fire prevention and suppression efforts，as well as long-term ecosystem conservation in North China.