Abstract: Evapotranspiration and soil infiltration, as the basic links in the rainfall water cycle, play an extremely important role in regional hydrological characteristics and water resource utilization. This paper takes Zhaoshu Island as the study area, estimates the evapotranspiration based on the PM(penman-monteith) formula and the SEBAL(surface energy balance algorithm for land) formula, constructs a BP(back propagation) neural network model to simulate the evapotranspiration in the vegetated and non-vegetated areas of the island, and simulates rainfall infiltration in beaches and bare soil based on the Kostiakov infiltration model and characteristics of precipitation infiltration process. The results show that the determination coefficients(R²) of the two evapotranspiration simulation models are higher than 0.85, and the model effectiveness index (EF) are higher than 0.85. The average daily evapotranspiration of each land cover types on Zhaoshu Island is the highest in May and the lowest in December of a year. The monthly change of evapotranspiration in the vegetated area is the most obvious. In areas with high soil sand content, the R² of the Kostiakov infiltration model is generally greater than 0.85. The steady infiltration rate of soil in Zhaoshu Island is about 0.33 cm·min-1. Multi-year rainfall data shows that the rainfall intensity in the study area is basically less than the steady infiltration rate, so the rainfall to the ground basically all infiltrates, and the rainfall infiltration in the dry season and the rainy season is significantly different. This study provides a feasible solution for the simulation of evapotranspiration and rainfall infiltration in the South China Sea islands where there is a lack of measured data, and provides data support for the construction of the island precipitation cycle.

Key words: evapotranspiration, BP neural network, infiltration model, rainfall infiltration