Assessment of Different Complementary-Relationship-Based Models for Estimating Actual Terrestrial Evapotranspiration in the Frozen Ground Regions of the Qinghai-Tibet Plateau

Abstract

Actual evapotranspiration (ETa) is important since it is an important link to water, energy, and carbon cycles. Approximately 96% of the Qinghai-Tibet Plateau (QTP) is underlain by frozen ground, however, the ground observations of ETa are particularly sparse–which is especially true in the permafrost regions–leading to great challenge for the accurate estimation of ETa. Due to the impacts of freeze-thaw cycles and permafrost degradation on the regional ET process, it is therefore urgent and important to find a reasonable approach for ETa estimation in the regions. The complementary relationship (CR) approach is a potential method since it needs only routine meteorological variables to estimate ETa. The CR approach, including the modified advection-aridity model by Kahler (K2006), polynomial generalized complementary function by Brutsaert (B2015) and its improved versions by Szilagyi (S2017) and Crago (C2018), and sigmoid generalized complementary function by Han (H2018) in the present study, were assessed against in situ measured ETa at four observation sites in the frozen ground regions. The results indicate that five CR-based models are generally capable of simulating variations in ETa, whether default and calibrated parameter values are employed during the warm season compared with those of the cold season. On a daily basis, the C2018 model performed better than other CR-based models, as indicated by the highest Nash-Sutcliffe efficiency (NSE) and lowest root mean square error (RMSE) values at each site. On a monthly basis, no model uniformly performed best in a specific month. On an annual basis, CR-based models estimating ETa with biases ranging from −94.2 to 28.3 mm year−1, and the H2018 model overall performed best with the smallest bias within 15 mm year−1. Parameter sensitivity analysis demonstrated the relatively small influence of each parameter varying within regular fluctuation magnitude on the accuracy of the corresponding model.

Trade‐Offs Between Spatial and Temporal Accuracy of Complementary Relationship Models for Evaporation in an Ungauged Basin

“…This is difficult to do in most regions, and even more impossible in ungauged basins. For convenience, one of the two parameters of CR models is often fixed first by default values, empirical formulas, transplanting from elsewhere, or estimation under extreme wet conditions, while optimizing the other parameter (Brutsaert et al., 2020; Liu, Liu, & Brutsaert, 2016; Shang et al., 2022; Wang et al., 2020). As mentioned above, the sensitivity of the parameter α e was much greater than that of the other parameters of CR models, and the determination of α e directly affected the temporal and spatial accuracy of CR models.…”

“…However, because of the nature of the CR principle, CR models only focus on the atmospheric status and ignore the land surface status. It is difficult to capture rapid changes in land surface status solely through changes in atmospheric status, especially at small spatiotemporal scales (Shang et al., 2022) such as irrigation and crop harvesting. As mentioned above, a more reasonable spatial distribution is obtained by obtaining spatially distributed dynamic parameters.…”

“…To improve estimation accuracy, two parameters of CR models are often calibrated simultaneously, but the risks of unreasonable spatial distribution may increase concurrently. Most current studies only focus on the accuracy of CR models in the time series or the spatial pattern (e.g., Brutsaert et al., 2020; Crago et al., 2022; Ma, Jin, et al., 2021; Ma, Szilagyi, et al., 2021; Shang et al., 2022), while fewer studies consider both the temporal and spatial accuracy. Moreover, systematic comparisons of the performance of CR models with spatially and temporally fixed and distributed parameters are not yet available.…”

“…Yang et al (2021) found that five CR models were only able to estimate daily E during the thawed period but not during the thawing, freezing, or frozen periods. Shang et al (2022) assessed the performance of different CR models at four flux sites in frozen ground regions of the Tibetan Plateau and found that five CR models are generally able to simulate E changes during the warm season. However, current studies in cold alpine regions are mainly based on in situ data, while analysis based on basin long-term data is rare.…”

Land-Atmosphere Interactions and Effects on the Climate of the Tibetan Plateau and Surrounding Regions

“…More than 70% of the models overestimated the LAI during the 1981-2014 growing seasons, indicating that the greening of grassland in the TP was greatly overestimated. Moreover, five complementary relationship-based models, requiring only routine meteorological variables to estimate actual terrestrial evapotranspiration, were evaluated by Shang et al [24], with biases ranging from −94.2 to 28.3 mm/year. These models provide a simple and convenient approach for evapotranspiration estimates.…”