5.2.5 National choices in emissions and removals factor estimation
The decision tree in Figure 13 is intended to guide countries through the choices likely to arise in practice when considering the use of data to estimate biomass emissions and removals factors, taking account of the preceding discussion. The numbers in the decision point boxes refer to explanations following the tree. Non-biomass pools and other GHG are covered more generically in Section 5.2.4.
Considerations at the decision points in the tree are as follows:
Decision Points 1 and 2. Use ground reference data(1) to estimate emissions and removals factors
The answer to the question in decision point 1 will usually be YES because countries for which REDD+ activities are key, and/or for which significant amounts of ground-based data exist, will generally make use of the data for REDD+ estimation. If ground reference data are not to be used (answer NO at decision point 1 leading to decision point 2) countries can use IPCC default values in Tier 1 estimation. This should not be done for key categories or where adequate ground data are available. Countries may also want to consider the IPCC Emission Factor Database as a possible source of emission/removal factors. The database contains factors that have undergone an editorial review process though not the full IPCC review so they do not have the same status as defaults contained in the IPCC’s methodological reports, and their use is a matter of scientific judgement by technical experts responsible for the estimates. Uncertainty ranges associated with the use of defaults should be taken from the IPCC guidance and guidelines.
Decision Point 3. Do you have data that can be harmonized?
Harmonization entails putting data on the same basis – for example by use of consistent measurement thresholds, consistent definitions, and common assumptions regarding species wood density or carbon conversion factors. The NFMS should check that this is the case with the collated data, leading to the response YES at this decision point. In the absence of ground data that can be harmonized (leading to the response NO) the NFMS should initiate the collection of the ground data needed.
Decision Point 4. Use auxiliary data?
Auxiliary data refers to the information used in stratification to increase sampling efficiency or as an input to models. Increased efficiency implies lower costs for given precision so in general, where available, auxiliary data should be used for stratification purposes. The answer to this question will generally be YES unless the forests in the country are so uniform as to render stratification unnecessary.
Decision Points 5a, 6a and 7a Decisions related to sample size in the presence of auxiliary data
The data sample size is sufficient if the confidence intervals associated with the emission/removal factors estimated for the strata defined using auxiliary data meet the specified precision criterion. If this is not already known to be the case (which would already lead to a YES at point 5a) then with a probability sample (which will lead to a YES at decision point 7a that immediately follows it can be determined in the first instance by reconnaissance calculations of the type described under sample size in the account of sampling in Appendix B. In the absence of a probability sample (leading to a NO at decision point 7a) then, Monte-Carlo or other uncertainty analysis being used with model-based inference will be needed. If the reconnaissance estimates or Monte Carlo analysis indicate a NO at point 5a then at 6a the sampling will need augmentation as described under supplementary sampling in Appendix B.
Decision Point 5b.6b, 7b Decisions related to sample size in the absence of auxiliary data
Considerations apply as for decision points 5a, 6a and 7a except that in the absence of auxiliary data there will be no basis for stratification. Unless the forests in question are a statistically uniform population this will increase the amount of sampling needed to satisfy precision requirements, and hence increase sampling costs. If this is an issue the NFMS should consider obtaining the auxiliary data needed for stratification, so that the left hand branch can be executed following a YES at decision point 4.
Decision Point 8. Use auxiliary data without harmonized data already available?
Having answered NO at decision point 3 the assumption is that the NFMS will make arrangements to gather the data needed for estimating emissions/removals factors to satisfy precision requirements. On this side of the decision tree there is no need for consideration of augmentation of an existing dataset because the sampling is designed from the beginning. In most cases auxiliary data (collated by the NFMS) will be used for stratification because of the need to increase sampling efficiency and reduce costs, and therefore the answer at decision point 8 will be YES. If auxiliary data are not being used there will be no basis for stratification and the NO branch should be followed.
Decision Point 9a. Select probability sample of ground plots with auxiliary data?
In case of a probability sample (leading to the left-hand branch below 9a), the sample will need to be sufficient if the confidence intervals associated with the emission/removal factors estimated the strata defined using auxiliary data meets the specified precision criterion. This can be determined in the first instance by reconnaissance surveys of the type described under sample size in the account of sampling provided in Appendix B. The individuals to be sampled will depend on the purpose of the sampling as described in Appendix B under selecting which individuals to sample. If the sampling is in conjunction with model-based inference (leading to the right hand branch below 9a) the sampling will be used to establish model parameters and needs to be sufficient that the confidence interval for model outputs of interest (e.g. carbon densities) meets criteria set out by the NFMS for the policy purpose intended. The model sensitivity analysis and exploratory runs are the equivalent of reconnaissance surveys to establish what is needed.
Decision Point 9b. Select probability sample of ground plots without auxiliary data?
Considerations apply as for decision point 9a except that in the absence of auxiliary data there will be no basis for stratification. Unless the forest population in question is statististically relatively homogeneous with respect to the target variable(s) the amount of sampling needed to achieve target precision will increase, and hence increase sampling costs. If this is an issue the NFMS should consider obtaining the auxiliary data needed for stratification, so that the left hand branch can be executed following a YES at decision point 8.
Decision Point, final: design- or model-based inference?
As set out in the box above, design-based , inference is based on sample points distributed according to probabilistic rules across the forest landscape - whereas in the case of model-based inference the sampling is used to establish model parameters and need not follow the same probabilistic rules, though to be effective it should cover the range of forest types and circumstances likely to be encountered in practice. Model-based inference relies on correct model specification as the basis for valid inference and to minimize bias, rather than a probability sampling design. The advantages are that a model offers opportunities for incorporating scientific understanding e.g. on the relationship between carbon pools, and this may increase predictive power. Model-based inference can also accommodate sample data that may not have been gathered according to a particular sampling design. The disadvantages are that there is no general agreement on what model to use, and the analysis of uncertainties is more complicated because sampling theory applied to models does not yield relatively simple formulae. For this reason Monte-Carlo analysis is often used to generate uncertainty estimates, though this relies on sufficient understanding of the correlations that may exist between different parameters.
As noted in the glossary reference data are the best available assessment of conditions on the ground for a given location or spatial unit. Reference data can be used to estimate areas or carbon densities and associated standard errors based on sampling. Reference data are generally collected according to probabilistic sampling design.