The WACMOS-ET project – Part 1: Tower-scale evaluation of four remote-sensing-based evapotranspiration algorithms
D. Michel, C. Jimenez, D.G. Miralles, M. Jung, M. Hirschi, A. Ershadi, B. Martens, M.F. McCabe, J.B. Fisher, Q. Mu, S.I. Seneviratne, E.F. Wood, D. Fernandez-Prieto
Hydrology and Earth System Sciences 20, pp. 803-822, (2016)
WACMOS-ET, Global, Evaporation, Climate, HESS, PT-JPL, GLEAM, SEBS, Hydrology
The Water Cycle Multi-mission Observation Strategy – EvapoTranspiration (WACMOS-ET) project has compiled a forcing data set covering the period 2005–2007 that aims to maximize the exploitation of European Earth Observations data sets for evapotranspiration (ET) estimation. The data set was used to run four established ET algorithms: the Priestley–Taylor Jet Propulsion Laboratory model (PT-JPL), the Penman–Monteith algorithm from the MODerate resolution Imaging Spectroradiometer (MODIS) evaporation product (PM-MOD), the Surface Energy Balance System (SEBS) and the Global Land Evaporation Amsterdam Model (GLEAM). In addition, in situ meteorological data from 24 FLUXNET towers were used to force the models, with results from both forcing sets compared to tower-based flux observations. Model performance was assessed on several timescales using both sub-daily and daily forcings. The PT-JPL model and GLEAM provide the best performance for both satellite- and tower-based forcing as well as for the considered temporal resolutions. Simulations using the PM-MOD were mostly underestimated, while the SEBS performance was characterized by a systematic overestimation. In general, all four algorithms produce the best results in wet and moderately wet climate regimes. In dry regimes, the correlation and the absolute agreement with the reference tower ET observations were consistently lower. While ET derived with in situ forcing data agrees best with the tower measurements (R2 = 0.67), the agreement of the satellite-based ET estimates is only marginally lower (R2 = 0.58). Results also show similar model performance at daily and sub-daily (3-hourly) resolutions. Overall, our validation experiments against in situ measurements indicate that there is no single best-performing algorithm across all biome and forcing types. An extension of the evaluation to a larger selection of 85 towers (model inputs resampled to a common grid to facilitate global estimates) confirmed the original findings.