An international panel of experts in forest assessment met in March 2000 to review progress and plans for completing FRA 2000, and to make recommendations for how the final results could be completed by the deadline of December 31 2000. The group reviewed the available sources of data and the proposed methods for combining the data, proposed additional data sources and analysis methods, and evaluated several alternatives for both scientific rigor and for practicality within the time and resource constraints. Criteria used to evaluate the alternatives included (1) statistical accuracy and precision of estimates, (2) ability to be used within available time, resource, and data availability constraints, (3) degree of general scientific acceptance of methods, (4) degree of transparency in methods, (5) ability to yield estimates for all countries of interest regardless of data constraints, and (6) simplicity.
After significant debate, the Panel recommended a combination of the Delphi Technique (an iterative process using blind expert opinion to refine estimates) and a Convergence of Evidence (a method for weighing and combining different sources of information to arrive at a single country level estimate). The recommended approach uses a small group of experts to weigh the different available data sources for each country and attempt to arrive at final estimates. The experts are led in each region by the FRA Regional Coordinator. The Coordinator summarizes all available information (from TM image analysis, country reports, relevant models, other satellite imagery, etc.) and derives an initial estimate or range of estimates for each country, explaining their decision process and documenting the weight that they assigned to the various available sources of information. They then share the results with the small set of regional or country level experts to validate the results, asking the other experts to critique the reasoning or assignation of weights in the production of country level estimates and attempting to seek some consensus on final estimates. The process is illustrated schematically in Figure 1.
All available data will be used in the process. However, there were some observations and recommendations regarding the reliability of data sources. The use of AVHRR data was to be carefully considered, since the observed low correlation between FRA TM data and AVHRR map, especially on dry and semi-dry lands in the Tropics. Therefore, the panel recommended FRA2000 to consider an additional sample of TM images to complete FRA2000 change assessment in Africa where other sources of data are very limited. The change model based on population, due to its limitations mentioned earlier, was recommended to use only as ancillary data in cased of very poor quality data from other sources.
The primary strength of this combined approach is that it allows the flexibility to include a variety of information sources assessed by people familiar with the situations in the countries, while mitigating some of the disadvantages inherent in relying on a single expert to integrate the data sources. This approach makes maximal use of available information, which varies from country to country. It allows incorporation of experience and qualitative information which is relevant but is normally not considered by more objective, purely quantitative analyses. The validation procedure adds a second set of information useful for comparative purposes or for calibration.
The primary weakness of this approach is the reliance on sufficiently diverse group of experts to add value in each country, and the unavoidable reliance on the availability of expertise and data. Time and resource constraints may limit the ability of FRA staff to seek sufficient input for each country. This difficulty could be partly mitigated by conducting a triage, whereby certain countries with known strong data were given less discussion, and countries with weaker data had more attention. Triage might also include stratifying countries by amount of forest cover to ensure that countries with the most forest are examined more closely, in order to minimise the chances of making relatively larger estimation errors.
