The present study established a fixed-point lightning warning model based on multi-source data for point targets or small-scale targets, such as scenic spots, mines, oil depots, and industrial parks, etc. By analyzing the lightning positioning data, atmospheric electric field data, Doppler radar data, and other data from the Jinfo Mountain lightning experiment field during the period of 2020 to 2022, the data quality control was performed, and a representative sample set of thunderstorm and non-thunderstorm events were sorted out. Decision tree models employing two algorithms: Classification and Regression Tree (CART) as well as Chi-Square Automatic Interaction Detection (CHAID) were established using SPSS Statistical software. The performance of these models was compared and analyzed when the lead time ranged from 0 to 54 minutes, with evaluation based on the F1 score metric. The results indicated that the two models performed differently, and the CART model demonstrated a simpler topology and superior classification accuracy compared to the CHAID model. The research findings offered valuable insights for lightning warning work in scenic areas, mines, oil depots, industrial parks and other relevant objects.