Effect of spatial sampling from European flux towers for estimating carbon and water fluxes with artificial neural networks

Abstract

Empirical modeling 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. In this work, artificial neural networks (ANNs) were used for the prediction of gross primary production (GPP) and latent heat flux (LE) on local and European scales with the aim to assess the portion of uncertainties in extrapolation due to sample selection. ANNs were found to be a useful tool for GPP and LE prediction, in particular for extrapolation in time (mean absolute error MAE for GPP between 0.53 and 1.56 gC m−2 d−1). Extrapolation in space in similar climatic and vegetation conditions also gave good results (GPP MAE 0.7–1.41 gC m−2 d−1), while extrapolation in areas with different seasonal cycles and controlling factors (e.g., the tropical regions) showed noticeably higher errors (GPP MAE 0.8–2.09 gC m−2 d−1). 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, 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 to identify the meteorological drivers driving the uncertainty

Patterns and controls of inter-annual variability in the terrestrial carbon budget

“…The temporal dynamic of NEE has been addressed in numerous studies, based on either "top-down" approaches, which primarily focus on aircraft atmospheric budgets (Leuning et al, 2004), tower-based boundary layer observations (Bakwin et al, 2004) and tracer transport inversion (Baker et al, 2006;Gurney et al, 2002;Rödenbeck et al, 2003), or on "bottom-up" methods that rely on data-driven gridded products derived from the upscaling of flux data (Jung et al, 2011(Jung et al, , 2017Papale et al, 2015;Papale and Valentini, 2003) or process-based biogeochemical models that simulate regional carbon budgets (Desai et al, 2007(Desai et al, , 2008Mahadevan et al, 2008).…”

Grassland and Cropland Net Ecosystem Production of the U.S. Great Plains: Regression Tree Model Development and Comparative Analysis

“…Under sampling by reference or training data relative to the space and time to be mapped is an issue in both machine learning approaches and process-based mapping models used to map carbon fluxes [5,9,51,61], but the area to be mapped to number of flux tower ratios were similar in this study to other carbon flux mapping efforts [5,13,18].…”

Detecting impacts of extreme events with ecological in situ monitoring networks

“…These delineated ecoregions are taken to be representative of approximately homogeneous areas in the mean land-climate system state, and yield an equitable representation of land-surface processes in upscaling activities (e.g. the spatiotemporal inter-and extrapolation of land-atmosphere fluxes of CO 2 , H 2 O and others; Jung et al, 2011;Xiao et al, 2012;Papale et al, 2015) or model-data integration studies (sensu Williams et al, 2009). …”

“…In fact, ecological in situ networks play an increasingly important role in analysing ecological phenomena and often provide a complementary perspective on natural phenomena to EOs (Nasahara and Nagai, 2015;Papale et al, 2015;Wingate et al, 2015) and complement model analyses Sippel et al, 2017). One prominent example is FLUXNET, with its proven record of advancing our understanding of the functioning of terrestrial ecosystems (Balddocchi, 2014).…”