Assessment of a bespoken remote-sensing Evapotranspiration model for Seasonally Dry Tropical Forests

Evapotranspiration (ET) plays an essential role in the water cycle, particularly in biodiverse environments with pronounced seasonal variability, such as the Caatinga. Accurately representing ET spatially and temporally in these ecosystems is indispensable, not only for understanding hydrological dynamics but also for natural resource management. However, the intricate nature of these environments poses significant challenges in modelling ET, which demands adaptive, site-specific approaches to capture their complex spatial and temporal variations. In this context, the STEEP (Seasonal Tropical Surface Energy Balance) model has been developed with the objective of capturing intrinsicalities of the dynamics and energy balance of the Caatinga forest. Despite its relatively good performance when compared to ground-based and global ET products, STEEP has not been extensively compared to other RS-based ET models. In this study, we used daily data from 2014 to revisit STEEP modelling outputs by comparing them to eddy covariance data and against five ET remote sensing models: PT-JPL, GLEAM, PM-MOD, SEBAL, and S-SEBI. We used the following statistics for performance evaluation: root mean squared error (RMSE), percent bias (PBIAS), and concordance correlation coefficient (CCC). Evaluation metrics for all models varied as follows: 0.69–1.31 mm day^-1 (RMSE), -13.54–41.13% (PBIAS), and 0.53–0.85 (CCC). STEEP overperformed four out of five models (i.e. SEBAL, S-SEBI, PT-JPL, and GLEAM), with RMSE = 0.80 mm/day, PBIAS = 11%, and CCC = 0.80. PM-MOD model exhibited the best performance metrics when driven with ground-truth data. We ascribe the best results of this model to its complex algorithm, which makes use of a wide range of spectral responses and environmental variables. Overall, all models exhibit some degree of ET overestimation during the dry season. This study highlights the ongoing need for precise model evaluation and adaptation to environmental nuances for improved ET estimation in biodiverse ecosystems like the Caatinga

Research Funding: National Council for Scientific and Technological Development (CNPq): grants nº 409341/2021–5 and 442799/2023-3