A Comparative Analysis of Different Algorithms for Estimating Evapotranspiration with Limited Observation Variables: A Case Study in Beijing, China

Evapotranspiration (ET) plays a crucial role in the surface water cycle and energy balance, and accurate ET estimation is essential for study in various domains, including agricultural irrigation, drought monitoring, and water resource management. Remote sensing (RS) technology presents an efficient approach for estimating ET at regional scales; however, existing RS retrieval algorithms for ET are intricate and necessitate a multitude of parameters. The land surface temperature–vegetation index (LST-VI) space method and statistical regression by machine learning (ML) offer the benefits of simplicity and straightforward implementation. This study endeavors to identify the optimal long-term sequence LST-VI space method and ML for ET estimation under conditions of limited observed variables, (LST, VI, and near-surface air temperature). A comparative analysis of their performance is undertaken using ground-based flux observations and MOD16 ET data. The findings can be summarized as follows: (1) Long-term remote sensing data can furnish a more comprehensive background field for the LST-VI space, achieving superior fitting accuracy for wet and dry edges, thereby enabling precise ET estimation with the following metrics: correlation coefficient (r) = 0.68, root mean square error (RMSE) = 0.76 mm/d, mean absolute error (MAE) = 0.49 mm/d, and mean bias error (MBE) = −0.14 mm. (2) ML generally produces more accurate ET estimates, with the Random Forest Regressor (RFR) demonstrating the highest accuracy: r = 0.79, RMSE = 0.61 mm/d, MAE = 0.42 mm/d, and MBE = −0.02 mm. (3) Both ET estimates derived from the LST-VI space and ML exhibit spatial distribution characteristics comparable to those of MOD16 ET data, further attesting to the efficacy of these two algorithms. Nevertheless, when compared to MOD16 data, both approaches exhibit varying degrees of underestimation. The results of this study can contribute to water resource management and offer a fresh perspective on remote sensing estimation methods for ET.