ausblenden:
Schlagwörter:
Carbon stocks
Biomass functions
Expansion factor
Prediction error
Lucf
Biomass
Forest
Zusammenfassung:
Most industrial countries have implemented national forest inventories (NFIs) with systematic and permanent sample plots. These data are principally an excellent database for carbon accounting at tree-level. The purpose of this study was to establish tree-level functions for predicting aboveground biomass carbon of European beech. In a subsequent step these functions were applied to inventory data and the error of the predictions was assessed. The regression equations were based on a model data set of 116 trees of beech (Fagus sylvatica) sampled in four climatic regions of the western German state North Rhine-Westphalia (NRW). The tree parameters diameter at breast height (d), tree height (h) and as a proxy for climatic conditions the altitude (alt) were used as predictor variables. The estimated mean relative prediction error of the favoured d-h and d-h-alt models at tree-level was 16.7 and 15.7%, respectively. The carbon content of wood ranged between 48.9 and 50.7%. Beech forest in NRW stored on average 120 Mg C ha(-1) in the aboveground biomass (excluding understory and reserved beeches). The total prediction error was less than ±2%. Based on such a high precision and due to the fact that the identical inventory plots are remeasured, stock changes in the order of 1% can be detected. The main uncertainty arises from the sampling error of the inventory grid. Comparing the predicted increment and the potential harvest during the next decade suggests that the carbon stock stabilises within the range of uncertainty. Further, we predicted the biomass of the 116 sample trees with six published, mostly stand-specific functions. The relative mean prediction error of the calculated aboveground carbon content was about ±15-32% of the observed values at tree-level and ±10% over all sample trees. But most of the predictions were systematically biased. This illustrates the need to use regional functions based on a large model data set. (C) 2003 Elsevier B.V. All rights reserved. [References: 29]
