Abstract: Clarifying the driving forces of groundwater level and achieving quantitative control of water supply are crucial ways to ensure dynamic balanced management of groundwater and enhance the strategic reserve capacity of water resources. Based on the daily and monthly monitoring data of groundwater depth over the years at the Zhangfang large karstic groundwater source area adjacent to a river, this paper employs stepwise regression analysis to quantitatively determine the driving factors influencing the dynamics of groundwater depth at the water source across multiple time scales. The regression model is then used for predictive control of water supply safety. On a monthly scale, the most significant factor influencing groundwater depth is the depth from the previous month, followed by the current month's extraction volume, precipitation from the previous month, and precipitation from two months ago. On a daily scale, the most significant factor is the depth from the previous day, followed by the current day's extraction volume, the previous day's extraction volume, the current day's flow at the Zhangfang station, and the current day's precipitation. Across different time scales, human extraction is the direct driving factor causing dynamic changes in groundwater depth. Combining the aquifer characteristics of the water source, the groundwater depth responds rapidly to surface water and precipitation on the same day, benefiting from the close hydraulic connection between the surface and groundwater facilitated by surface karst fissures. The delayed response to precipitation from one to two months ago indicates that groundwater recharge originates from water flow in deep karst fissures. The results of the water supply safety regulation prediction show under the conditions of a maximum allowable groundwater depth of 60 meters and three consecutive years of low-water periods, extraction can be appropriately increased during the dry season from April to July, with the maximum extraction volume of 1000×10⁴ m³/month.