Abstract: In recent years, the Three Gorges reservoir has implemented drought-relief water replenishing operation several times in emergency response to drought events in the middle and lower Yangtze River. However, it is not yet clear how an increase in the discharge replenishment influences the water levels at the main cross sections downstream of the reservoir, and an index system has not been established yet for evaluating the drought-relief water replenishment benefits of a reservoir. This paper develops a water level simulation model for a river section based on long short-term memory neural networks, and simulates the time variations in water level at the typical sections of the study reaches. And an index system for the reservoir’s benefits evaluation is constructed considering comprehensively water supply, ecology, shipping, and power generation. The results show that under the reservoir’s three replenishing schemes of 500, 1000, and 2000 m3/s in August, 2022, the water level and discharge at the main control stations in the study reaches would be able to rise to a certain extent. The restored irrigation area would be 2.9 thousand ha, and the economic benefit of agricultural irrigation be 13.55 million yuan. The increased power generation benefits would be 14.00, 27.25, and 32.25 million yuan under the three schemes, respectively. The ecological and shipping benefits would not change considerably, because of only a small increase in water level and discharge. This study helps formulate drought-relief water replenishing plans for the Three Gorges reservoir.

Key words: Three Gorges Reservoir, emergency drought-relief water replenishment operation, section water level simulation, long short-term memory neural networks, benefit evaluation