B24B-03
Do We Need Larger Eddy Covariance Site Networks? Sites Spatial Distribution and Network Size Effect on Carbon and Water Fluxes Upscaling Using Empirical Models

Tuesday, 15 December 2015: 16:30
2004 (Moscone West)
Dario Papale1, Thomas A Black2, Nuno Carvalhais3, Alessandro Cescatti4, Jiquan Chen5, Martin Jung3, Gerard Kiely6, Gitta Lasslop7, Miguel D Mahecha3, Hank A Margolis8, Lutz Merbold9, Leonardo Montagnani10, Moors Eddy11, Jørgen Olesen12, Markus Reichstein3, Gianluca Tramontana13, Eva Van Gorsel14, Georg Wohlfahrt15, Botond Ráduly16 and FLUXNET PIs in LaThuile, (1)Tuscia University, Department for Innovation in Biological, Agro-food and Forest systems (DIBAF), Viterbo, Italy, (2)University of British Columbia, Biometeorology Group, Faculty Land and Food Systems, Vancouver, BC, Canada, (3)Max Planck Institute for Biogeochemistry, Jena, Germany, (4)Joint Research Center Ispra, Ispra, Italy, (5)University of Toledo, Toledo, OH, United States, (6)Univ College Cork, Cork, Ireland, (7)Max Planck Institute for Meteorology, Hamburg, Germany, (8)Laval University, Centre d'étude de la forêt, Quebec City, QC, Canada, (9)ETH Swiss Federal Institute of Technology Zurich, Zurich, Switzerland, (10)Free University of Bolzano, Bolzano, Italy, (11)Alterra Wageningen, Wageningen, Netherlands, (12)Aarhus University, Tjele, Denmark, (13)Tuscia University, Viterbo, Italy, (14)CSIRO Ocean and Atmosphere Flagship, Yarralumla, Australia, (15)University of Innsbruck, Innsbruck, Austria, (16)Sapientia Hungarian University of Transylvania, Miercurea Ciuc, Romania
Abstract:
Empirical modelling approaches are frequently used to upscale local eddy-covariance observations of carbon, water and energy fluxes to regional and global scales. The predictive capacity of such models largely depends on the data used for parameterization and identification of input-output relationships, while prediction for conditions outside the training domain is generally uncertain. Artificial neural networks (ANNs) have been used in the past for empirical upscaling of eddy covariance measurements; here this modelling approach is used to predict gross primary production and latent heat flux on local and European scales with the aim to evaluate the effect of the sites network size and distribution on the total budgets and inter-annual variability predictions. ANNs were confirmed to be a valid tool for GPP and LE prediction, in particular for extrapolation in time. Extrapolation in space in similar climatic and vegetation conditions also gave good results, while extrapolation in areas with different seasonal cycles and controlling factors showed noticeably higher errors, confirming the importance of the training dataset representativeness. The distribution and the number of sites used for ANN training had a remarkable effect on prediction uncertainty in both, regional GPP and LE budgets and their interannual variability. Results obtained show that for ANN upscaling for continents with relatively small networks of sites (e.g. Africa), the error due to the sampling can be large and needs to be considered and quantified. The analysis of the spatial variability of the uncertainty helped also to identify which are the drivers less represented in the training dataset, contributing to the identification of hotspots where new towers would contribute more to the reduction of uncertainty in the empirical upscaling. Results are coherent with other studies on network representativeness although clearly related to the use of the data in these specific activities (upscaling).