Sample-Based Estimation of Greenhouse Gas Emissions From Forests-A New. Approach to Account for Both Sampling and Model Errors

摘要

The Good Practice Guidance (GPG) for reporting emissions and removals of greenhouse gases from the land use, land-use change, and forestry (LULLICF) sector of the United Nation's Framework Convention on Climate Change states that uncertainty estimates should always accompany the estimates of net emissions. Two basic procedures are suggested: simple error propagation and Monte-Carlo simulation. In this article, we argue that these methods are not very well-suited for uncertainty assessments in connection with sample-based surveys such as national forest inventories (NFIs), which provide a majority of the data for the LULUCF sector reporting in several countries. We suggest that a more straightforward approach would be to use standard sampling theory for assessing the sampling errors; however, it may be important to also include the error contribution from biomass and other models that are applied and this requires new methods for the variance estimation. In this article, a method for sample-based uncertainty assessment, including both model and sampling errors, is developed and applied using data from the NFIs of Finland and Sweden. The study revealed that the model error contribution to the combined sampling-model mean square error of ratio estimators of mean aboveground biomass on forestland amounted to about 10% in both countries. In estimating 5-year change of the corresponding biomass stocks, using permanent sampling units, the model error contribution was reduced to less than 1%. The smaller impact in the case of change estimation is due to the fact that any tendency of models to either over- or underestimate due to random parameter estimation errors will be the same both at the beginning and the end of a study period. The fairly small model error contributions in our study are due to the large number of sample trees used in the fitting of biomass models in Finland and Sweden; with less sample trees the model error contributions could be expected to be substantial. The proposed framework applies not only to greenhouse gas inventories but also to traditional NFI estimates of, e.g., growing stock in which uncertainties due to model errors typically are neglected in applications.