Evapotranspiration restores stormwater retention capacity of green roofs during inter-event dry periods and thus affects their stormwater retention function. In this study, a conceptual hydrological green roof model was developed based on the Penman-Monteith method and water balance theory. The model was calibrated and verified with continuously observed meteorological and hydrological data of a green roof in Beijing, and then applied to simulate evapotranspiration processes of green roofs located in various climatic zones. The results show that the hydrological model could accurately simulate evapotranspiration process of the green roof, with the Nash-Sutcliffe Efficiency Coefficient of 0.6385 and 0.6014 as well as the Determination Coefficient of 0.7191 and 0.6168 for calibration and verification, respectively. For green roofs with the same depth of substrate, daily average value of actual evapotranspiration (AET) increases as the climate changes from semi-arid (Lanzhou) to semi-humid (Beijing) and humid (Guangzhou and Wuhan). Increasing depth of substrate results in the increased AET from green roofs, but the daily average AET from green roofs changes marginally when the depth of substrate increases beyond thresholds. The thresholds of substrate depth of green roofs at Lanzhou, Beijing, Wuhan, and Guangzhou are 10 cm, 17 cm, 24 cm, and 25 cm, respectively. The daily average AET of green roofs located in different climate zones also have various thresholds, and the daily average AET threshold of green roofs at Guangzhou is in turn larger than that at Wuhan, Beijing, and Lanzhou. This study is expected to provide a scientific reference for estimation of green roof evapotranspiration and design of green roofs under different climatic zones in China.