Using random forest model to analyze driving factors of extreme rainstorm peak discharge: taking the “23·7” rainstorm in Beijing as an example

Objective To clarify the driving factors of peak discharge formation under extreme rainstorms, an investigation of the extreme rainstorm peak discharge in Beijing’s “23·7” event was conducted, thereby revealing the patterns of peak discharge variation, and providing a scientific basis for regional post-disaster reconstruction planning and disaster prevention and mitigation work.

Method To analyze the drivers of peak discharge under extreme rainfall conditions, after the extreme rainfall event in Beijing in July, 2023, a research team surveyed 140 flood mark points across 56 gullies in the Mentougou District. Then, the random forest model was used to analyze the factors driving peak discharge based on the spatial changes.

Result (1) In the Mentougou District, the peak discharge of gullies increased from north to south, with values ranging from 27.5 to 1072.0 m³/s. Gullies exceeding the peak discharge levels for 10-year, 20-year, and 50-year events accounted for 82%, 71%, and 42% of the total, respectively. (2) The main factors influencing peak discharge differences in Mentougou District under rainfall conditions were basin area, gully length, mean relief amplitude, runoff coefficient and sediment accumulation, with importance values of 1.000, 0.524, 0.471, 0.382, 0.346, respectively. The topographic factors had the greatest influence on peak discharge in the gullies of Mentougou District under current heavy rainfall conditions. (3) These factors exhibited a complex nonlinear relationship with peak discharge.

Conclusion Based on the five key driving factors, it is recommended to implement a “One Gully, One Strategy” management approach for post-disaster recovery and reconstruction, using gullies as the basic unit. The strategy should include rational flood control measures and higher engineering standards, focusing on key controllable factors such as sediment accumulation and runoff coefficient. Additionally, emergency plans should be improved, and actions should be taken to address encroachments on gully flow areas to enhance flood resistance. Future research should expand to include human factors such as land use patterns and water infrastructure construction, analyzing the impact mechanisms of peak discharge in different regions to provide more precise guidance for disaster prevention and reduction efforts.