During the past decade concern for the environment and in particular for the consequences of the depletion of natural resources has been rapidly growing in the conscience of people around the world. The fate of the forests in general, and tropical forests in particular has been receiving increasing attention owing to the far reaching consequences that deforestation and degradation have on the environment, the well-being of mankind and biological diversity.
This concern has been expressed by national and international institutions, governmental and non-governmental organizations (NGOs), opinion groups, etc.. and has resulted in innumerable resolutions, some of which have global relevance. A high point in international interest was reached at the United Nations Conference on Environment and Development (UNCED) in Rio de Janeiro (1992) which devoted a full chapter of its Agenda 21 (entitled “Combating Deforestation”) to forest conservation and development and adopted the “non-legally binding, authoritative statement of principles for a global consensus on the management, conservation and sustainable development of all types of forests”.
A major contributory factor to this concern for the current state of tropical forests has been the large degree of uncertainty associated with the available information on the rate of global deforestation and the total lack of quantitative information on the associated change processes.
Several related questions have been posed by policy-makers, the scientific community and the public at large. What is the area of the remaining tropical forests? At what rate are they being depleted? Is the rate of deforestation accelerating or slowing down? What are the causes? What are the ecological, economic and social effects of forest change?
Questions on global climatic change, loss of biodiversity, etc. demand both the increased precision of estimates and new types of information. Forests could be a source or a sink of global carbon depending on whether forest biomass increases or decreases; a question related closely to tropical forest depletion.
These issues called for a global deforestation assessment of a new generation. Five main requirements for the assessment technique to be developed were identified:
The database used should be verifiable and the procedures of assessment objective.
The technique should have a statistical basis and provide not only mean values but also their confidence limits (e.g., mean deforestation rate and its standard error).
The procedure must provide consistent and comparable estimates over time, because deforestation is a dynamic process.
The approach must address the causes and effects of deforestation, as these have become important issues in view of global concern about the possible negative effects of deforestation and forest degradation on the CO2 cycle, loss of biodiversity, land degradation, etc.
The methodology must be easily transferable to developing countries. This would contribute to the development of sound programmes for national forest resources assessment from which the global assessments could benefit.
As clearly stated above, the assessment was to be implemented on a statistically sound basis, not only to provide reliable results but also to provide a time-series of estimates, of known precision, and to contribute to knowledge about the process of deforestation. It may be noted that consistency and continuity of assessment have implications for the continuity of institutions at global and national levels.
To meet the need for more detailed and accurate information mentioned earlier, FAO Forest Resources Assessment 1990 (FRA 1990 Project) carried out an assessment of forest resources of the tropical countries for the reference year 1990 based on existing reliable data (from national or sub-national forest inventories, surveys, etc.). This activity constituted Phase I of the Project.
A detailed description of the methodology used and results is given in FAO Forestry Paper 112: Forest Resources Assessment 1990: Tropical countries (FAO, 1993) and a summary is provided in FAO Forestry Paper 124:Forest Resources Assessment 1990: Global Synthesis (FAO, 1995). This assessment was implemented according to the following steps: establishment of database modelling techniques and production of standard results by country for the reference year 1990. The report covers many important forest resources parameters such as area of natural forests and plantations; forest-harvesting, land and forest area by ecological zone and the rate of deforestation during 1980–1990, by country and ecological zone.
A Forest Resources Information System (FORIS) has been developed to store and retrieve the data collected previously and new data which will become available in the future. The linkage between FORIS statistical data and geo-referenced databases, including vegetation cover, ecological zones and sub-national boundaries, has opened new opportunities for carrying out the analysis of deforestation on an ecological basis.
Forest Resources Assessment 1990 based on FORIS constitutes an important contribution to FAO global forest resources assessment; however, in view of the limitations inherent in the heterogeneity of its data sources, it does not satisfy all the requirements for the study of global change listed earlier in Section 1.1. Lack of global consistency in source data with respect to classification, reference period, scale, precision, etc., strongly reduces their suitability for the study of global changes at the required depth.
These were some of the considerations included in the development and implementation of a statistically sound survey of tropical forests based on the concept of continuous forest inventory (see Forestry Paper 112: page 8) which constitute Phase II of the Project.
This complementary approach meets all five requirements specified earlier by providing a more detailed, informative and objective description of the change processes. Moreover, its results can be easily integrated with those of FORIS within the framework of a two-phase survey approach.
The present report presents the activities and results of Phase II of the Forest Resources Assessment 1990 Project.
As mentioned above, a survey was required primarily to provide a reliable estimate of tropical deforestation to international policy makers, an estimate which was not available nor possible from existing data (FAO, 1993).
An equally important reason was the scientific community's demand for a time-series of estimates for the study of global change. This called for the establishment of an assessment system capable of providing forest cover change information on a consistent and continuing basis.
Information on the processes of change was also an important motive in conducting the survey. In the context of the Tropical Forest Action Programme (formerly TFAP, now called the National Forestry Action Programme, NFAP), reliable statistics concerning the fate of deforested land were needed in order to take appropriate policy measures to control deforestation. The situation was often known in qualitative terms, but not in quantitative terms.
The only satisfactory way to provide reliable information on the processes of change is to establish a forest resources monitoring system, wherein the land area under consideration is observed on a systematic basis over time, using comparable techniques. This provides reliable and location-specific change information.
In consideration of costs, as well as the precision and timeliness of results, a sampling approach was designed and used to cover the entire tropical zone.
The specific immediate objectives were:
to achieve the highest level of consistency and precision in the assessment of forest cover changes at global and regional levels, and to contribute to the improvement of the global forest area estimate;
to develop and disseminate a simple and effective monitoring technique for producing estimates of forest cover state and change at global and regional levels, suitable for application also at national level; and
to provide spatial and statistical data for estimating class-to-class changes of land use and forest cover categories between the two dates of interpretation at the sample locations and for producing change matrices at regional and global levels.
In order to optimize the use of the results deriving from the two phases of the Project, the first providing wall-to-wall sub-national level data and the second providing statistically reliable data on forest cover and change processes at global and regional levels, it is essential to integrate the two data sets into a unique resources assessment system. This integration, discussed in further detail in Section 6.2 could be considered as the development objective of the present remote sensing survey.
Available data
An intensive review of the literature carried out during the first phase of the Project revealed an almost complete lack of reliable data on the subject. Table (1.4) 1 below reports the level of information available from the tropical countries on forest resources state and change, in concise form. Even without analyzing the quality of the individual surveys it is clear that the data available are rather inadequate to provide reliable information on the changes occurring in the forest cover.
| Table (1.4) 1: State of forest inventory in the tropics at the end of 1990 | |||||
| Region | Number of countries under assessment | Forest area information | |||
| One assessment | More than one assessment | ||||
| No assessment | before 1981 | 1981–90 | |||
| Africa | 40 | 3 | 23 | 12 | 2 |
| Asia & Pacific | 17 | 0 | 1 | 6 | 10 |
| L.America & Caribbean | 33 | 0 | 15 | 9 | 9 |
| Total | 90 | 3 | 39 | 27 | 21 |
Source: FORIS database of the FRA 1990 Project
Below are some relevant comments expressed in FAO Forestry Paper 112 on the current state of the country forest inventories:
There is considerable variation among regions with respect to completeness and quality of the information, with Asia faring better than Latin 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. In most cases, the surveys have been carried out independently from the previous ones and therefore present differences that are not only due to forest cover changes.
There is little consistency among the classifications adopted in the country surveys and there seems to be little agreement even on the very concept of “forest”. The effect of this poor consistency is that the global synthesis could only report on “forest area” and “deforestation” with no further breakdown.
It was on the basis of this in-depth review of the existing information that the decision was taken to carry out a statistically designed pan-tropical forest survey based on new observations.
Study of survey options
It may be appropriate to mention here that recommendations for a survey of tropical deforestation, on a statistical/remote sensing basis, date back to the United Nations Conference on Human Environment in Stockholm (1972). A FAO/UNEP (United Nations Environment Programme) Expert Consultation, in 1974, conducted an in-depth study of the subject and drafted specific proposals for tropical forest cover monitoring. During 1978–1982, FAO/UNEP jointly carried out pilot studies on forest cover monitoring in the West Africa Region, covering Benin, Togo and Cameroon (Baltaxe, 1980).
In forestry practice, techniques known as Continuous Forest Inventory (CFI) and permanent sample plot (PSP) have been used for several decades. The challenge was to synthesize these ideas and develop a system of forest cover change monitoring based on statistically sound principles in an economically feasible and globally practical way. The possibility of procuring repetitive coverage of the earth's surface from remote sensing satellites provided a unique opportunity for this purpose.
A Project paper, “Analysis of Alternative Sample Survey Designs”, by R. Czaplewski describes the alternative designs considered, and the process used to select the present design. The four design alternatives considered were:
The choice between census (i.e. wall-to-wall coverage) and sampling was an important issue. A sample survey, as is well known, is associated with a sampling error, but has advantages if the area to be covered is large or observations are to be made on a repetitive basis. A wall-to-wall coverage, on the other hand, has no sampling error though it may have larger observation or non-sampling errors, in view of the extensive rather than intensive approach, and it costs more.
At national level it is feasible and even desirable to organize wall-to-wall forest mapping at suitable intervals of time; but, at global level the logistics involved are complex and the costs very high. For the present project, with limited funding and time, it was obvious from the very beginning that sampling would be the only way to proceed. The most important consideration was to minimize the sampling error by utilizing all the existing information and available knowledge on sampling techniques.
Review of past experience
A literature review was conducted to identify techniques of forest cover monitoring using remote sensing. Most techniques were based on a comparison of area information from independently compiled maps of two or more dates prepared in the following manner: a full set of cloud free images covering the whole study area was acquired and interpreted. The interpreted details were transferred to base-map(s). This thematic map was then used to compile area information manually or digitally. The difference between forest cover area from the two or more sets of compiled maps, divided by the time interval between them, provided an estimate of the rate of change for the area studied.
An analysis of the results from the above surveys showed that the derived assessment of changes was usually affected by poor consistency since the surveys carried out at different times over the same area were often inconsistent in classification, quality of remote sensing data, interpretation techniques, etc., and involved significant subjectivity in class delineation. These factors led to an unknown number of pseudo-changes. Though the inconsistencies between results on the two dates under consideration tended to auto-compensate over large survey areas, such as a whole country, they failed to provide a reliable and location-specific estimate of changes. The gathering of consistent information on the process of change from such studies was not possible at all.
In view of the foregoing situation the Project decided to develop a more reliable and consistent system of global tropical forest cover monitoring based on the concept of continuous forest inventory, well documented in the forestry literature as an efficient technique for providing reliable forest state and change information. Moreover, it was clear that consistency, or coherence, between historical and recent observations was of paramount importance in any survey of change.
Remote sensing features
Related to the interpretation of satellite data was the question concerning the classification system for observing and describing changes over a given time period. A simple forest-non forest dichotomy, for instance, could not provide all the information needed to describe forest degradation, e.g., change of density, forest fragmentation, changes into long and short fallow, etc. which are important processes affecting forest biomass and biodiversity. While setting higher goals regarding the types of change to be detected, the limitations associated with high-resolution satellite data were kept in view. The classification system adopted was a compromise between two conflicting considerations (R. Baltaxe, R. Drigo, 1991).
A point interpretation of change over time was a necessary condition for producing consistent “change matrices” which had been accepted as one of the essential objectives of the global survey. How could one ensure the spatial and, most important, thematic consistency required to produce reliable change matrices? How could it be done in a simple, unsophisticated way that would facilitate transfer to and implementation in developing country conditions without compromising the quality of results?
In answer to these questions a new procedure was developed which was termed interdependent interpretation procedure (R. Drigo 1991). This procedure consists of visual interpretation of both images (historical and recent) by the same interpreter (from the region and familiar with the location) and at the same time based on a single interpretation process. This procedure allows for spatial and thematic co-registration of the two interpretations, regardless of any geometric and radiometric distortions of the satellite images used.
The consistency of the classification of satellite data acquired on two separate dates, resulting from this procedure, makes it probably the single most important element of the whole survey.
In order to allow for the immediate processing and verification of results at sampling unit level, a user-friendly data recording/processing system was developed based on customized LOTUS 1-2-3® spreadsheet software and dot grid of appropriate size. The simplicity and low sophistication of the system ensured the smooth implementation of the procedure in all conditions, the only requirements being one personal computer of minimum capacity and a power supply.
The interdependent interpretation procedure, classification system and compilation procedure were discussed and disseminated through (i) four regional workshops (held, chronologically, in Thailand, Mexico, Kenya and Cameroon) with the participation of 31 tropical countries; and (ii) ten training sessions at the level of national institutions.
The interpretation technique was implemented at selected regional and national forestry and/or remote sensing institutions which have a good knowledge of the sample locations and have been traditionally involved in forest resources assessment activities. This decentralized approach was chosen with the twofold objective of strengthening national capacities for forest monitoring and improving the quality of image interpretation.
