Research papers

Estimates of the Priestley-Taylor coefficient based on FLUXNET data at multiple spatiotemporal scales

Evapotranspiration (ET) plays a vital role in the hydrological cycle and exerts significant impacts on the regional climate and water resources management (Xue et al., 2022). Accurate estimation of land ET is crucial for understanding hydrological fluctuations and water resource management in the warming climate due to its high influences on surface runoff (Li et al., 2023). Although local (ecosystem) scale ET measurements can be achieved via various techniques, such as the eddy covariance method (Lee et al., 2011, Xu and Singh, 2000), the regional and global ET estimation is still relatively limited (Castellvi et al., 2001).

Direct ET measurements over a large scale is difficult due to the scarcity of observations, necessitating the use of indirect estimation methods. Many indirect methods have been proposed to estimate ET (Gavin and Agnew, 2004, Martínez Pérez et al., 2017). Among them, the Priestley-Taylor (PT) method, with its straightforward formulas and minimal data requirements, is commonly adopted for estimating water demand by the atmosphere or reference evapotranspiration in hydrological and ecological models (Priestley and Taylor, 1972, Priestly and Taylor, 1972, Cristea et al., 2013, Li et al., 2007, Nikolaou et al., 2023). However, the method’s reliability is often contingent on the accuracy of the PT coefficient (α) (Eichinger et al., 1996a). The PT method presupposes an equilibrium state in the atmosphere during evaporation from a fully wet surface, a condition that is seldom satisfied in reality due to atmospheric advection, even over large wet expanses. Therefore, the PT coefficient (α) is multiplied to counteract this “nonequilibrium” effect with a default value of 1.26 (Yan et al., 2023). In theory, stronger advection leads to more rapid turbulence, resulting in a higher α. Generally, α values fluctuate between 1.08 and 1.34, with an average of 1.26 (Mauder et al., 2018, Zhou et al., 2023). Many studies indicated that the default value of $\mathrm{\alpha }$(1.26) may not be reliable in windy and dry regions, where aerodynamic components are not adequately accounted for (Gong et al., 2022, Li et al., 2016). Nevertheless, accurate estimates of ET are especially essential for water management in semiarid and arid regions. Therefore, it is crucial to determine the precise values of $\mathrm{\alpha }$in various climates. While some regional studies have been conducted (Lhomme and Moussa, 2016, Shu et al., 2022, Zhou et al., 2020), the spatial and temporal evolution of $\mathrm{\alpha }$at global scales and the impacts of meteorological and vegetable parameters on its variability have not been elucidated to date (Jung et al., 2010). Extensive research has been conducted on PT parameters (Izaurralde et al., 2006, Kim et al., 2008, Timsina and Humphreys, 2006). It is widely acknowledged that the physical significance of $\mathrm{\alpha }$is the degree of dryness of the air and the role of advection in the atmosphere (Ai and Yang, 2016). $\mathrm{\alpha }$reflects the distribution of the net radiation energy ${\mathrm{R}}_{\mathrm{n}}$between the latent heat flux ($\mathrm{L}\mathrm{E}$) and the explicit heat flux ($\mathrm{H}$), influenced by the complexity of the natural environment. It depends on a number of environmental factors, including vapor-pressure deficit (Jarvis and Mcnaughton, 1986, Jury and Tanner, 1975), leaf area index (Sumner and Jacobs, 2005), elevation (Cho et al., 2013), and solar radiation (Burakowski et al., 2018, Flint and Childs, 1991, Pereira and Nova, 1992). To date, numerous scholars have developed formulas for $\mathrm{\alpha }$(Yao et al., 2013, Yao et al., 2015). Some researchers separated the contributions of canopy and soil to $\mathrm{\alpha }$and developed a formula driven by the LAI and P to calculate $\mathrm{\alpha }$(Szilagyi et al., 2017). Nevertheless, most observation-based assessments have been confined to particular geographic region, with relatively few studies on global scales (Aschonitis et al., 2017). Because the equilibrium evapotranspiration conditions of PT equation are hard to achieve and the applicable area is restricted, notably in windy and dry regions. This limitation leads to variability in the parameter $\mathrm{\alpha }$, resulting in inaccurate evapotranspiration calculations derived from the PT equation. Change law and mechanism of $\mathrm{\alpha }$need further global analysis. Machine learning techniques have gained popularity due to the abundance of remote sensing data. Some researchers have establish regional parameter models using these techniques, but they lack a large-scale worldwide application, and the accuracy needs to be improved (Lin et al., 2023). Currently, machine learning methods have been increasingly developed for explicitly quantifying variables regionally and globally (Rosset et al., 1997, Su and Singh, 2023). Several approaches, including multiple linear regression and random forest (RF), have been employed to predict regional parameters accurately (Lhomme and Moussa, 2016, Shu et al., 2022). Relevant studies have calculated the theoretical $\mathrm{\alpha }$for many years without noticing obvious trend, indicating the utility of machine learning in determining the global annual average $\mathrm{\alpha }$through long-term analysis. (Wu et al., 2021).

In this study, we calculated the Priestley-Taylor parameter ($\mathrm{\alpha }$) using FLUXNET site data and analyzed the influencing mechanisms of meteorological and vegetation factors for $\mathrm{\alpha }$. We trained the RF model with these FLUXNET measurements, and then we utilized the RF model to make global predictions. This work provides a foundation for understanding the significant role of vegetation in land–atmosphere interactions and climate through a focus on energy partitioning.