4.3 The performance comparison of all the models
By comparing the three type models, we found that the radiation-based models display better performance(small RMSE) when compared with the Penman family models and temperature-based models, average RMSE is 1.09 mm d−1, 1.26 mm d−1 and 1.29 mm d−1 for radiation-based models, combination models and temperature-based models, respectively (Table 2). The better performance of the radiation-based models compared with other two type models might be attributed to the incorporation of important input parameter such as net radiation. Although the combination models also include net radiation parameter, the combination models incorporate much other meteorological factors that not mainly controlling ET except for net radiation, which lead to a poor performance for combination models. In addition, the radiation-based models require less meteorological input compared with combination models, and the net radiation was the dominated factor that controlling the ET for this humid region. Overall, most radiation-based models generally underestimated the measured ET during the whole study period, whereas the temperature-based models tended to overestimate ET. This is consistent with previous studies where the Makkink and Priestley-Taylor models generally underestimated ET (Fooladmand & Haghighat, 2010; Priestley & Taylor, 1972; Xu & Singh, 2002), while the Hargreaves equations often overestimate ET in cold-humid conditions and requires a local calibration (Berti, Tardivo, Chiaudani, Rech, & Borin, 2014). Given that the study region in our study belongs to humid alpine meadow, thus ET tended to be overestimated. An alternative explanation for the poor performance of the Hargreaves model in humid regions may also relate to the Ra parameter used in the Hargreaves model (Fontenot, 2004), which is based on the maximum possible radiation value and does not take the atmospheric transmissivity into account. However, the atmosphere transmissivity in humid regions is affected by many factors, such as atmospheric moisture; thus, the solar radiation reaching the surface is significantly reduced due to the high atmospheric moisture content (Temesgen, Allen, & Jensen, 1999), resulting in the overestimation of solar radiation, ultimately leading to an overestimation of ET by the Hargreaves method.
Furthermore, there were also common features of all three groups of models. All the models tended to underestimate the measured ET during the growing season (with larger evaporative demand), and overestimated ET during the non-growing season (with reduced evaporative demand), which was consistent with a previous study conducted in a semi-arid region (Liu et al., 2017b). Furthermore, we found that the measured ET and calculated ET0 were less correlated during non-growing season than during growing season. These discrepancies may relate to the dominant component between transpiration and evaporation. The transpiration was the dominant during growing season, almost account for 75% of evapotranspiration, whereas the evaporation was the dominant component during non-growing season in the same study site (Zhang et al., 2018). Considering the evaporation process was much complex and affected by many environmental factors compared with transpiration process, ultimately lead to a poor correlation between measured ET and calculated ET0 during non-growing season. Therefore, both Hargreave’s equations and other models need further local or region calibration before being applied to a given region (Xu & Singh, 2002). Besides, it should be noted that the data used in this study just obtained from a single lysimeter and a single weather station, which may insufficient to represent the whole humid climate or the alpine ecosystem but represent a specific site. Thus, more lysimeter systems should be used in the alpine ecosystem in the future to obtain more accurate estimates of evapotranspiration over the northeastern Qinghai-Tibetan Plateau.