The country data, available in the form of statistical tables or maps, have been the main source of information for the global forest resources assessment and the direct determinant of its quality and content. So far a statistically designed pan-tropical forest survey has not been attempted. Therefore, a review of the current state of the country forest inventories (see Annex 1) is an appropriate starting point for discussion of the Project methodology.
Table 1
State of forest inventory in the tropics at end 1990
| Region | Number of countries under assessment | Number of countries with forest resources data at national level | |||||||
| Forest area information (number of assessments and reference years) | Other topics covered | ||||||||
| No assessment | One assessment | More than one assessment | Forest conservation and management | Forest plantations | Volume and biomass | Forest harvesting and utilisation | |||
| before 1981 | 1981–90 | ||||||||
| Africa | 40 | 3 | 23 | 12 | 2 | 4 | 8 | 2 | 4 |
| Asia & Pacific | 17 | 0 | 1 | 6 | 10 | 10 | 8 | 7 | 7 |
| L. America & Caribbean | 33 | 0 | 15 | 9 | 9 | 11 | 8 | 9 | 4 |
| Total | 90 | 3 | 39 | 27 | 21 | 25 | 24 | 18 | 15 |
Source: FORIS database of the Project
The findings of the Project on the current state of the country forest inventories can be summarised as follows.
There is considerable variation among regions with respect to completeness and quality of the information, with Asia faring better than tropical America and the latter better than tropical Africa.
There is considerable variation in the timeliness of the information. The data is about ten years old, on average, which could be a potential source of bias in the assessment of change.
There are some countries which have carried out more than one assessment. These countries, however, have not used appropriate techniques, such as Continuous Forest Inventory (CFI) design, for change assessment.
Only a few countries have reliable estimates of actual plantations, harvest and utilisation although such estimates are essential for national forestry planning and policy-making.
No country has carried out a national forest inventory containing information that can be used to generate reliable estimates of the total woody biomass volume and change.
It is unlikely that the state and change information on forest cover area and biomass could be made available on a statistically reliable basis at the regional or global levels within the next ten or twenty years unless a concerted effort is made to enhance the country capacity in forest inventory and monitoring.
The above findings establish that forest resource assessments are among the most neglected aspects of forest resource management, conservation and development in the tropics.
Two main techniques of global assessments have been tried so far based on the country data: (i) a questionnaire approach and (ii) a centralised assessment approach. The questionnaire approach was started in 1948 and discontinued in 1965 because of many non- or incorrect responses. A centralised assessment was started by FAO/UNEP Tropical Forest Resources Assessment Project (1980) by collecting the existing country data from a wide variety of sources at FAO HQ, Rome, bringing them to a common conceptual and mensurational standard and adjusting results to the common reference year 1980 using an empirical approach.
The expert consultation held at Kotka Finland in 1987 recommended that the present Project follow the approach of the FAO/UNEP 1980 assessment for estimating the forest cover area at end 1990 and then estimate the change by taking the difference between the 1990 and 1980 figures. However, the first results of the Project demonstrated clearly that this approach is not suitable for the change estimation because the resulting estimate of change had high variance, being the sum of the variance of 1980 and 1990 estimates according to the law of propagation of errors. The following remarks clarify this point further.
It may be noted that at the time of the FAO/UNEP 1980 assessment:
forest cover information was available for only 23 countries. To alleviate the situation, the FAO/UNEP Project carried out area assessment for 13 countries by interpreting Landsat MSS images at 1:1 million scale;
the average date of the country information lagged substantially behind the global assessment year being 1980, requiring extrapolation to bring country data to a common reference date;
very few countries had more than one assessment of forest cover. Therefore, the only way to adjust results to the standard year was through an empirical process.
The situation for the FAO 1990 assessment was as follows.
All countries (except three in Africa) possessed one estimate of forest cover made during 1970 to 1990, mostly based on remote sensing.
The variations in the date of country data were still a major problem. It has been found that the average age of country data available for the 1990 assessment is, in fact, close to 1980.
Twenty-one countries had carried out more than one forest cover assessment.
There had been substantial developments in technology, in particular geographic information system (GIS), remote sensing and modelling techniques.
Another limitation associated with the recommended approach lies in the fact that the least precise of the two constituent assessments is the determinant of the variance of the estimate. Therefore, improved coverage, timeliness and reliability of the country data during 1980 to 1990 make little or no impact on the precision of change estimate.
| Error Propagation in Change Estimation from Independent Surveys |
| Let f1 and f2 denote estimates of forest cover areas for two points of time, here 1980 and 1990 respectively; V(f1) and V(f2) denote the respective variance. If the two assessments are independently conducted, then the change is estimated by (f1 - f2) and its variance by: |
| V(f1 - f2) = V(f1) + V(f2) |
| The above formulation clearly shows that the variance of change estimate is jointly determined by the variance of the two contributing assessments. If one of them has low value and the other high value, then the variance of difference will be high, irrespective of the low value of one of them. |
In Chapter 1 mention was made of the high expectations of the world community from the FRA 1990, whereas the review of the current state of forest inventory revealed major gaps in conducting a global assessment based on country data. The challenge was to produce the expected results making the best use of available data, appropriate techniques and new technology (remote sensing, GIS, computerised database management and modelling techniques). The Project, with the support of the cooperating institutions, set towards developing an approach which could fulfill many of the requirements listed earlier. The following are highlights of the tools and techniques investigated and applied.
Use of a database management system for easy storage, retrieval, analysis and updating of information.
Introduction of a modelling technique to estimate deforestation objectively.
Use of auxiliary variables, dynamic in nature, such as population density and population growth, for which data is readily available and which are among the important driving forces behind deforestation.
Reduction in size of the assessment unit from national to subnational level which is ecologically and demographically more homogeneous.
Use of variance-reducing techniques while making estimation, such as stratification of the subnational units by ecological criteria.
Use of multi-date high resolution satellite data to estimate changes on an objective and statistically sound basis.
| Continuous Forest Inventory Design |
| Using the continuous forest inventory (CFI) concept, it is possible to significantly reduce the error term. In this class of designs, the source data of surveys at two points of time are not independently but inter-dependently analysed. Assuming a design in which all units are measured on two dates, the error term associated with change would have the following value: |
| V(f1 - f2) = V(f1) + V(f2) - 2 Cov(f1, f2) |
| where V(f1) and V(f2) are variances of the forest cover observations on the same area unit on two dates and Cov(f1,f2) is the co-variance between them. If V(f1) = V(f2) and Cov(f1,f2) = 0.8 V(f1), which are reasonable assumptions, the variance of the difference in the case of inter-dependent assessment will be only one-fifth of the that associated with independent assessments. |
Working on the above ideas, the Project developed and put into operation a design which consisted of the following two phases:
A “model” approach using the existing country information, both in tabular and spatial form (i.e. in the form of maps). This low-budget technique is cost-effective and provides efficient country estimates, but is not able to satisfy completely all the five concerns listed in the introductory chapter, particularly at the regional and global levels.
A “survey” approach using multi-date high resolution satellite data. This technique satisfies rather well all the five requirements listed in Section 1 of Chapter I with reference to regional and global levels.
To add to the benefit, the two phases were integrated in the framework of a statistical design. The following chapter will report on the methodology and main results of the two phases.
