The group was divided into three sub-groups to discuss how data might be synthesized into estimates. Modelling, convergence of evidence and remote sensing were three techniques reported at the end of the day.
The sub-group agreed that a model:
The group reviewed the 1990 model:
The use of this model in FRA estimations, could be an option as:
When the potential of the models was evaluated, experts agreed that:
Time series analysis could be used with a clear categorization of:
A formal approach was recommended, meaning that mathematical precision should be indicated when possible, and that the final estimate should carry a precision measure when possible.
The following table illustrates the recommended approach:
Table 4. Recommended approach.
Dataset |
Utility(Group 1) |
Estimateha/year |
Precision (s/range/class/rank) |
Systematic
errors |
USE OR NOT 0/1 |
Country data |
|||||
A |
|||||
B |
|||||
Remote sensing |
|||||
C |
|||||
D |
|||||
Models |
|||||
E |
|||||
F |
|||||
Other |
|||||
G |
Suggested alternative uses:
a) If statistical precision measurement or range with confidence level is available for the data set(s) to be used: The weights are calculated based on precision measure and use-or-not and final estimate plus estimated error are derived.
b) If statistical precision measurement or range with confidence level is not available: Only one data set can be used and no precision will be given for the final estimate. Ranks are used to create weights according to predefined rules.
This procedure will allow several data sets to be used. The final estimate does not have any precision measure.
It was recommended that all available data sets and how they were used be published.
In view of the recommendation that the number of sample units in Africa be increased, the remote sensing group was assigned the task of providing advice on how to best carry out the supplemental sampling, image interpretation and data analysis. It was first suggested that stratification by ecological zones (EZs) should be employed, jointly with information about population density, climate and socio-economic parameters. This additional information could help to cluster countries with similar characteristics, bearing in mind the need to make the sampling procedures as efficient as possible due to time, financial and logistic constraints.
One of the issues raised concerned the possible stratification of forest cover by high/ moderate/low rate of change to ensure that the sampling adequately covers these categories.
To identify these areas, continent-wide, several possibilities were discussed:
The remote sensing group found that due to the low correlation presented for Africa for AVHRR and thematic mapper forest cover, AVHRR should not be used for Africa, especially due to presence of dry forest.
The hot spot areas presented by the TREES Project are valuable for the humid forest. However, for TREES standards, the dry forest is not considered forest, so the expert consultation did not include identification of hot spots in that region. And such stratification would have to be done carefully as results are very sensitive to strata size for under-sampled cases.
This sampling approach, first taking into consideration the ecological zones, would allow values within the same ecological zone strata to be averaged out to provide an estimate (regional) for the entire strata.
It was proposed to first group countries with “similar” socio-economic, climatological and forest cover conditions and consider these country clusters as strata where samples would be allocated. The estimate for each stratum would be based on the actual values observed within that stratum. This approach does not eliminate the consideration of ecological zones, but employs the use of samples from countries belonging to the same ecological zones.
The use of a more systematic pattern to fit any future FRA plans was noted as an issue to consider.
The number of additional samples to supplement the existing 47 would be determined as a result of the final stratification adopted. It was recommended that a statistician be consulted to advise FAO in this respect. It must be recognized, however, that for country numbers these images will be used in a rather pragmatic fashion. The implications of the different stratification procedures should also be discussed with an expert. The remote sensing group left the issue open.
It was agreed that the sampling should not only cover those countries in the low-utility information areas of Africa, but should be distributed so as to cover the entire continent. This would ensure that a consistent methodology is applied to the entire region.
Regarding analysis of the supplemental sampling, the remote sensing group provided the following recommendations:
Ř The analysis should still consider the classes “closed forest” and “open forest” as distinct.
Ř Interpretation of the images should be carried out using digital data by analysis on the screen.
Ř Map change interpretation would need to be done carefully.
Ř Do as much as possible ground survey.
Ř Use Africa cover images if possible/available.
One of the tasks assigned to the remote sensing group concerned the possibility of using spatial statistics tools to generate a change gradient map based on the present 117 samples (and possibly other information). It was generally agreed that at this point it would be premature to develop such a map or use the information for country statistics. It could be a “lab” exercise. Since Brazil has annual data on forest cover change, the lab exercise could start from these.
Regarding the use of information from the present 117 samples, the group felt that the present procedure, using the transition matrix, bar charts and division by ecological zones, is very adequate and did not recommend any immediate change. It was suggested that some examples of images, image interpretation and data analysis procedures could be included on the web site.
As to the future, the remote sensing group recommended the following:
Ř Promote further discussions regarding a change from visual interpretation to a semi-digital method.
Ř Improve “ground truth” verification through use of ground inspectors, aerial observation and aerial photography.
Ř Explore medium-resolution sensor data, such as MODIS and CBERS (China-Brazil Earth Resource Satellite), of appropriate spatial resolution of 250 meters.
Ř Further explore the possibility of developing a geo-referenced database for each region. This would considerably simplify the analysis of cover change at any time interval.
Ř Use a GIS map approach to area determination.
Ř Digitize old and new polygons and do geospatial analyses of deforestation (e.g. size, number and proximity).
Ř Consider using a system of permanent satellite sample plots where forest state and change are monitored for many decades to come.
Ř Consider adopting monitoring of change events rather than (or in addition to) overall changes in state. For example, a clearing in closed forest would be identified and documented as a “deforestation event”. At present it would only contribute to area determination of deforestation (closed to open) or would be missed entirely as deforestation or missed entirely if it goes from closed to young closed in 10 years. Selected cuts could be a recorded event. This has interesting benefits and design implications.
Ř Country-based programs should be considered in high priority areas.
Criteria for proposed methods were discussed and several points mentioned:
Ř Consider compatibility with FRA 2010 or future FRA.
Ř FRA should determine if the programme objective is to be “the” world forest inventory or a useful source of information on inventories of the world .
Ř The results should stand the scrutiny of others in the field.
Ř The results should be defensible. Determine what weight should be put on defensibility.
