Effects of scale and scaling in predictive modelling of forest site productivity

Abstract

Site productivity, commonly expressed by site index, is a key indicator of the potential of forested land to deliver ecosystem services like wood production and carbon sequestration. It is an important criterion for decision makers and managers of both production and multi-purpose forests. In many situations forest site index cannot be directly measured and must be estimated from site characteristics related to climate, topography and soil, using appropriate models. A major difficulty herewith is that the models must capture the spatial and temporal variability of the ecological processes, knowing that the magnitude and the variability of the driving forces and responses may show scale dependencies. Scale is therefore an important issue in successful forest site productivity modelling.In this study, empirical forest site productivity models are evaluated for their scale dependency whereby reference is made to the threefold concept of ‘scale’ (extent, support, coverage) as proposed by Bierkens et al. (2000). We also addressed the applicability of models at other extents or other supports than the one they were developed at, i.e. the effect of scaling. The results show that meaningful site index models for small extents require higher resolution support to catch the short-distance variability, whereas for larger extents a coarser support is sufficient to characterize the variability. Where it regards scaling, it is found that the validity of empirical site index models is restricted to the scale level for which they are calibrated. Also the application of site index models on an extent which is adjacent and not overlapping with the extent at which they were developed proved to result in inadequate predictions. Although the structure of site index models is scale-dependent and their applicability limited to the scale of development, it is beyond doubt that such models have the potential to provide good insight into the biophysical drivers of site productivity and can result in good predictions at unsampled locations whenever the scale of model establishment is adapted to the scale of the studied processes and predictions are restricted to the extent for which the model is calibrated.