Improving Evapotranspiration Model Performance by Treating Energy Imbalance and Interaction

Abstract

Input uncertainty and deficiencies in the key elements of hydrologic models are fundamental challenges to improving model performance. Evapotranspiration (ET) models are sensitive to energy imbalance and energy interaction between the canopy and surface, and these errors can bias model simulation. This paper presents a Bayesian framework that accounts for energy imbalance and energy interaction uncertainty in calibration of the Shuttleworth-Wallace (SW) model and discusses whether these efforts can improve the SW model performance. Specifically, the DiffeRential Evolution Adaptive Metropolis (DREAM) algorithm is used to analyze the energy imbalance error and energy interaction correction during SW model calibration, and Bayesian model evidence (BME) is used to scrutinize multiple hypotheses about alternative representations of energy. The results indicate that explicit treatment of the energy imbalance and conceptual corrections of the energy interaction significantly reduce the uncertainty of model parameters and provide energy corrections that are more appropriate and largely improve the SW model performance. Moreover, the optimal model selected by BME is mostly consistent with that evaluated using traditional statistical criteria (TSC) during a validation period. This result shows that the improved SW model also provides the best predictions, and thus consistent model selection should also be asymptotically efficient. The findings may allow improvement of ET model performance, and hence the accuracy of hydrologic and climatic simulations, toward fully understanding the key process underlying hydrologic and climatic variability and change over land.

Advancing Process Representation in Hydrological Models: Integrating New Concepts, Knowledge, and Data

“…From a model perspective, Wei et al. (2020) refined estimates of evapotranspiration using the Shuttleworth‐Wallace model and Bayesian model evaluation to scrutinize multiple hypotheses about alternative representations of energy exchange, specifically how to account for the energy imbalance in flux observations and energy interaction between the canopy and the surface. Finally, evapotranspiration estimates at 1 resolution were also used to constrain irrigation amounts applied in the Haihe River Basin.…”

“…In summary, the studies in this special section range from a pure physical description of the bare soil evaporation process (Novak, 2019; Trautz et al., 2019) to estimates of evaporation and evapotranspiration at the field (Cui et al., 2020; Wei et al., 2020), catchment (Koch et al., 2020), and global scale (Or & Lehmann, 2019). While diverse in methodology and scope, all the studies highlight how evapotranspiration remains central to the hydrological discussion and how important it is to separate the abiotic evaporation from the biotic transpiration flux as well as to link ET with water sources in the soil column.…”

Two Alternatives to the Two‐Source Energy Balance Evapotranspiration Model

“…Another challenge is how to diagnostically evaluate the ET models with different structures during the model development and refinement stages. In current model evaluation studies, it is difficult to attribute the performance differences between any two models to individual components and, or hypotheses (Koster & Milly, 1997; Wei et al., 2020). It is increasingly recognized that quantitative evaluations of model behavior using statistical difference measures cannot effectively discover inadequacies in a model.…”

Construction of an evapotranspiration model and analysis of spatiotemporal variation in Xilin River Basin, China

“…Therefore, these models are relatively less practical for calculating the actual ET in arid and semi-arid regions owing to the low vegetation coverage. In recent years, the research on ET models has mainly focused on the following aspects: the application of ET models [10][11][12], the comparison and evaluation of models [13][14][15][16][17][18][19][20][21], the improvement and optimization of existing ET models [22][23][24][25][26], and the construction of new ET models [10,16,20]. However, there are few studies on model construction, some of which use multiple linear regression [16], and some of which use machine learning algorithms such as random forest to estimate ET [10,20].…”