Evaluating a multi-step collocation approach for an ensemble climatological dataset of actual evapotranspiration over Italy

Authors: CAMMALLERI, C - University Of Milan; Anderson, Martha; CORBARI, C - University Of Milan; YANG, YUN - Mississippi State University; HAIN, C - Nasa Marshall Space Flight Center; SALAMON, P - European Commission-Joint Research Centre (JRC); MANCINI, M - University Of Milan

Submitted to: Journal of Hydrology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/5/2024
Publication Date: 10/22/2024
Citation: Cammalleri, C., Anderson, M.C., Corbari, C., Yang, Y., Hain, C.R., Salamon, P., Mancini, M. 2024. Evaluating a multi-step collocation approach for an ensemble climatological dataset of actual evapotranspiration over Italy. Journal of Hydrology. https://doi.org/10.1016/j.jhydrol.2024.132209.
DOI: https://doi.org/10.1016/j.jhydrol.2024.132209

Interpretive Summary: Evapotranspiration (ET) is a crucial variable in the hydrological balance, representing the main loss of water from the land surface. Reliable estimates of actual ET play a major role in quantifying the water use and stress in the agricultural sector, and in informing water management at watershed and basin scale. To support these applications, there are several methods for mapping ET rates over landscapes, including water balance techniques and energy balance estimates using satellite imagery. While all ET methods will have biases and errors, an effective error mitigation approach is to use multiple models and compute an ensemble average. In this ways, biases and errors will often cancel leading to a superior combined product. This ensemble method was tested over Italy for 1991-2020, using three water balance models and three remote sensing models. The ensemble product did indeed perform better than any individual model in comparison with daily ET observations obtained at flux tower sites across the country. These results demonstrate a feasible approach for generating higher accuracy estimates of water use and crop stress to support agricultural water use decision making.

Technical Abstract: Accurate estimations of evapotranspiration (ET) are key in a variety of water balance applications, but divergent results can be obtained due to the large range of available methodologies. The use of an ensemble approach is a suitable alternative, as it summarizes multiple sources in an optimized strategy. In this study, an expert-based multiple collocation (MC) approach is tested to merge six actual ET datasets, with the aim of reconstructing a spatiotemporal-consistent monthly dataset for the climatological period 1991-2020 for Italy. The merged products are: three water balance datasets (BIG BANG, LSA SAF, and LISFLOOD), two residual surface energy balance models (SSEBop, and ALEXI) and the MODIS standard product. The results of the analysis highlight how the merged product outperforms each single base dataset, with good accuracy (MAD = 0.47 ± 0.17 mm/d, RD = 27.9 ± 7.5 %), limited bias (BIAS = -0.17 ± 0.26 mm/d,) and high correlation (r = 0.83 ± 0.10) against eddy covariance observations. The merged ET dataset is accompanied by an estimation of the spread of the ensemble, which highlights large differences in ET estimates in some areas and periods characterized by severe water stress, such as in southern Italy during the summer. This large spread seems to be mostly driven by systematic differences among datasets, which affect the estimation of the reference climatology, suggesting how inter-model spread can have a defining role in further improving the merging strategies.