Comparison between the trapezoid method and two energy balance models (TSEB and 3SEB) to estimate evapotranspiration of a tree-grass ecosystem

Tree-grass ecosystems (TGEs) comprise nearly 1/6th of Earth's surface in many climates while being biodiversity hotspots. These transitory landscapes dominate global biogeochemical cycles and are one of the most sensitive to global climate change. Indeed, these issues, combined with increasing pressures from agricultural land conversion, livestock grazing, and wildfires, require better characterization of these ecosystems.  Actually, the performance of evapotranspiration (ET) remote sensing algorithms tends to have more significant uncertainties in these landscapes due to the poor representation of both (i) the vertical multiple-layered vegetation strata (i.e., overstory with tree/shrub canopies over a herbaceous understory) having distinct phenological variations and bare soil, and (ii) the openness of the horizontally distributed high vegetation, causing inherent pixel heterogeneity at the conventional satellite scale. This study assessed and inter-compared remote sensing-based ET models having different modelling assumptions and data requirements. In this case, we applied an empirical and analytical vegetation index-temperature trapezoid method (VITT) and two different surface energy balance models: the two-source energy balance (TSEB) and three-source energy balance (3SEB). TSEB decouples the energy balance between vegetation and soil, while 3SEB incorporates an extra vegetation layer within the TSEB model structure to better depict ecosystems with multiple vegetation layers, such as TGEs. The VITT method considers as TSEB the decoupling of soil and vegetation, but the latter only in its photosynthetically active state.  The study sites are a grass-oak-pine savanna and grassland, two experimental core sites from the Ameriflux network, Tonzi and Vaira sites, located in California, USA. The dataset comprises flux tower data, meteorological data, land cover data, and airborne images from Aviris-Classic (reflectance) and MASTER (temperature) sensors downsampled to 35m spatial resolution. We evaluated the robustness of the methods to estimate ET through key phenological stages (e.g., drying of the grass layer, biomass peaks, and inter-intra annual variations). We analysed how well each method portrays vegetation water stress. The simpler the vegetation structure of the ecosystem, the more similar methods' behaviors and capabilities were. Methods that separate the ET from the different layers were more suitable for assessing the different layer influences for this open and partially covered system. The VITT method raised some limitations as used in a nonconventional way by accounting for two vegetation layers. One may expect better results to be achieved when at least one of the vegetation layers is senescent. Finally, our results can help us understand the possible constraints to face when applying these types of ET algorithms with future satellite missions (TRISHNA, SBG).