Evaluation of evapotranspiration partitioning models in the Amazon forest

Abstract. Although models that simulate actual ET have been widely used globally, their performance in tropical forests is unsatisfactory. The distribution of ET components is one of the key reasons. In this study, we evaluated the ability of three ET models (Forest-CEW, PML-V2, and PT-JPL) in a complex forest by analyzing their components. The data comes from seven ground-based eddy covariance flux towers in Brazil, which are part of the "Large Scale Biosphere- Atmosphere Experiment in Amazonia" (LBA) project. Our study found that the R² of Forest-CEW was 0.64, that of PT-JPL was 0.43, and that of PML-V2 was only 0.29. The average results of the model show that T/ET=63.2 %±16 %, Ei/ET=32.3 %±16 %, and Es/ET=6 %±5 %. The model simulates better results in Savanna (RMSE=10.4 mm/month) than in the rainforest (RMSE=17.6 mm/month). Rn is the main driving variable of the model ET and T, with a sensitivity of 20 %, temperature is the main driver of Ei, accounting for 17 %, and LAI is the main driver of Es, but it produces a negative effect (-22.5 %). Our analysis emphasizes the differences in the ability of existing models to simulate ET dynamics in complex forests. Improving the formulation of ET components, particularly the canopy interception part, holds significant potential for substantially enhancing the accuracy and reliability of these ET models.