Upscaling from the Tree to the Stand Level

2.1.2 Challenges

The construction of statistical biomass models encompasses several challenges. Applicable models have to be accurate, which means precise and unbiased and should provide confidence limits to assess their applicability. Other typical issues are clustered data structures, heteroskedasticity of data and the desired additivity of biomass components. If sufficient data is available nonparametric fc-NN approaches have also shown to provide good results in estimating tree biomass (Fehrmann et al. 2008) but a constant challenge remains the scarcity of biomass data. This frequently leads to models based on a small number of trees from a few stands only. Chave et al. (2004) showed the exponential decline of the error in the estimation of total aboveground biomass for tropical tree species. Their data provides evidence that a minimum of about 50 trees would be required to achieve an error of 10 % in the determination of aboveground biomass; 5 % were achieved with a minimum of about 150 sampled trees. A logical conclusion seems to be to pool all available data on a tree species, to increase sites and tree numbers and achieve a more generic model (e. g. Wirth et al. 2003). This mostly valid option is unfortunately often impaired by the manifold of different ways used to measure biomass (cut­off diameters, drying prescriptions, definition of components etc.), all of which complicates the compilation of coherent data sets.