Seasonal trends of satellite-based evapotranspiration algorithms over a complex ecosystem in East Asia

Accurate land surface evapotranspiration (ET) estimations over a heterogeneous ecosystem are important to understand the interaction between the land surface and atmosphere including practical applications in integrated water resources management. Though numerous studies have been adopted and developed based on remote sensing technology to make a more accurate prediction of regional ET distribution, there has been still degree of uncertainty due to high spatio-temporal variability of the ecohydrologic parameter. This study suggested the revised remote sensing-based Penman-Monteith (Revised RS-PM) model and Trapezoid Interpolation Model (TIM) using only remotely sensed data as input data to assess applicability over complex topography in East Asia. Ground observations at the two flux sites having different land surface conditions were used to evaluate intra-annual seasonality in 2008. Both models represented temporal compatibility yielding biases of − 74.25-38.79 W m− 2 and Root Mean Square Error (RMSE) values of 68.96-90.90 W m− 2 while the models consistently overestimated ET at the forests due to large amount of interception relatively restraining sufficient water supply to the plants. The revised RS-PM showed slight overestimation due to the overvalued Leaf Area Index (LAI) as an input parameter and classified parameterization in transpiration calculation. This algorithm was developed for global mapping of ET so that errors occurring from vegetation parameterization are inevitable. TIM reproduced higher ET than the measurements in a non-growing season since remotely sensed Normalized Difference Vegetation Index (NDVI) as an input parameter could be affected by cloud contamination. In contrast to the dormant season, the revised RS-PM estimated a larger amount of ET distribution than TIM in the growing season. A conservative estimation of TIM was mainly caused by the structural characteristics. Relationships between the land surface temperature and NDVI were contextually used to determine both maximum and minimum limits of the Priestley-Taylor parameters in the image data. Determination of the contextual relationship should be carefully conducted to achieve reliable estimations. Results of a sensitivity analysis shows that the net radiation (RN) plays the most significant role in both models (± 17-20% for ET by ± 20% change of RN). Variation of LAI impacted ET mostly in the revised RS-PM in a dormant season due to a logarithmic relationship between canopy conductance and LAI. TIM, on the other hand, was barely affected by LAI because of the simple structure of the algorithm including the Priestley-Taylor equation. The results suggest that the models can be applied to a regional scale with heterogeneous topography over long term periods if the input data handling is carefully conducted. In particular, the models can be usefully applied where ground ancillary data are not readily available.