Table of Contents Next Page


For nearly 50 years (1946 – 1990), the FAO has routinely assessed the state of the world's forest cover. These assessments have provided some of the most comprehensive and useful information available regarding the world's forests. This information is in great demand by scientists, practitioners and decision-makers.

In order to increase reliability and better respond to information needs, the 1990 Forest Resources Assessment utilized a two-phased approach. The first phase of the assessment was made using existing national forest inventories and maps in conjunction with an adjustment model used to standardize the results to a common reference date. Information on the methodologies and results of the first phase are available in Forestry Paper No112 (Forest Resources Assessment 1990: Tropical Countries). The second phase of the 1990 assessment was conducted in the tropical zone only and employed a statistical survey using remote sensing to sample the forest cover.

The approach for the second phase was based on the comparison of satellite imagery from two dates by the same analyst at the same time, using a uniform classification throughout the tropics. By using this approach, class to class changes in land cover could be detected and depicted in change matrices according to regions and climatic zones. Information on class to class changes is new and adds substantially to the understanding of the processes of vegetation degradation and deforestation. Statistical estimates of the standard error were also generated and used to compare the results derived from existing inventory data (first phase) to those based on remote sensing (second phase). This report describes the methodology used in the second phase and the related findings.

The results of this work have special importance as they demonstrate that the state of forest cover for 1990 and the rate of change from 1980 to 1990 estimated by the two phases lie within the confidence interval established from the sample survey data (second phase). Consequently, concerns regarding the consistency and reliability of global forest resources assessments were addressed for the first time. The results of the remote sensing sample survey do not replace the forest cover state and change data published in Forestry Paper No112, but confirm them and complement them by providing a new wealth of information on the processes of forest change.

David A. Harcharik
Assistant Director-General
Forestry Department


The survey of tropical forest resources was supported by the Governments of Finland, the Netherlands, Sweden and the United States (through U.S. Forest Service) as well as the European Union through a multi-donor trust fund. Personnel assistance was given through the Associate Professional Officers scheme of the Governments of Germany, the Netherlands, Sweden and the United States. Regional workshops on forest monitoring methodology with the participation of representatives from some forty tropical countries, were supported by the Unit of Technical Cooperation among Developing Countries of the United Nations Development Programme (Bangkok, 1991), by the U.S. Forest Service (Nairobi, 1991, and Mexico City, 1993), by the European Community and FAO Technical Cooperation Programme (Yaounde, 1994).

Major in-kind contributions were made by cooperating institutions including the U.S. Forest Service, the Swedish University of Agricultural Sciences, the University of Freiburg, Germany, and the Overseas Agronomic Institute, Florence, Italy. The Environmental Protection Agency of USA (EPA) provided a grant for acquiring historic satellite images for all the tropical zone and the National Research Council of Thailand provided satellite data of south-east Asia for shifting cultivation studies.

The Forest Resources Assessment 1990 Project, located at FAO, Rome, was conducted by a project team led by Dr. K.D. Singh, who conceived the sample survey design on the lines of continuous forest inventory technique and secured necessary resources and collaboration for its implementation.

Mr. R. Drigo made a major contribution to the remote sensing component of the survey including development and dissemination of the interdependent image interpretation technique, review of interpretation for all sampling units, study of change processes and writing the present report.

Dr. R. Czaplewski and Professor B. Ranneby made important contributions to the statistical survey design and analytical models, and Professor D.R. Pelz led a IUFRO review of the design.

Mr. M. Lorenzini developed the Project's Geographic Information System and, in cooperation with Mr. A. Marzoli and Mr. G. Mu'Ammar contributed to the definition and stratification of the sampling universe through the combined analysis of geographic and database information.

Mr. A. Hildeman and Mr. H. Simons significant contribution to the dissemination of the monitoring methodology through the organization of regional workshops and training sessions.

Mr. R. Baltaxe contributed to the development of the monitoring methodology and provided useful comments to the present report.

Ms. E. Milani contributed to the production of the various maps that are shown in the present report on the basis of the Project's Geographic Information System.

Ms. C. Hardy and Ms. J. Rechter contributed to the preparation of the present report through proof reading and language editing.

In addition, the survey benefited from the collaboration with an extended network of institutions and individuals, remote sensing centers, and participants to the regional workshops, who all contributed in the interpretation of remote sensing images, by providing source information, consultancies and expert help. The Forest Resources Assessment 1990 Project is greatly indebted toward this large number of contributors that made this challenging idea possible and successful.


ASCIIAmerican Standard Code for Interchange of Information
AVHRRAdvanced Very High Resolution Radiometer (instrument of NOAA satellites)
COFOCommittee on Forestry
ECEuropean Community
EDCEros Data Center of the US Geological Survey
EOSATEarth Observation Satellite Company
FAOFood and Agriculture Organization of the United Nations
FORIS (1990)Forest Resources Information System (baseline 1990)
FRA 1990Forest Resources Assessment 1990 Project
FSIForest Survey of India
GACGlobal Area Coverage (AVHRR data at 4 km resolution)
GISGeographic Information System(s)
HRSDHigh Resolution Satellite Data
IAOIstituto Agronomico per l'Oltremare (Overseas Agronomic Institute), Florence, Italy
IBAMAInstituto Brasileiro de Meio Ambiente e Recursos Naturais Renováveis, Brazil
IBMInternational Business Machines
INPEInstituto Nacional de Pesquisas Espaciais, Brazil
IUCNInternational Union for the Conservation of Nature (The World Conservation Union)
LACLocal Area Coverage (AVHRR data at 1 km resolution)
MSSMulti-Spectral Scanner (sensor of Landsat satellites)
NASANational Aeronautics and Space Administration (United States of America)
NOAANational Oceanic and Atmospheric Administration
RSRemote Sensing; used also to indicate the remote sensing component of the FRA 1990 Project
SPIAFService Permanent d'Inventaire et d'Aménagement Forestiers, Zaire
SUSampling Unit
SUASSwedish University of Agricultural Sciences
TCDCTechnical Cooperation among Developing Countries
TMThematic Mapper (sensor of Landsat satellites)
TREESTRopical Ecosystem Environment observations by Satellites; joint project of the Commission of the European Community
UNUnited Nations
UNCEDUnited Nations Conference on Environment and Development (Rio de Janeiro, Brazil, June 1992)
UNEP GRIDUnited Nations Environment Programme - Global Resources Information Database
USFSUnited States Forest Service
WRS (1 or 2)World Reference Systems (1 or 2) of the Landsat Programme

Executive Summary


The survey of tropical forests reported here was developed and implemented with the aim of providing detailed and reliable information on the process of on-going changes in the tropical forest cover, in order to meet the needs of researchers, policy makers and the public at large. It addresses the following types of questions:

The only satisfactory way to provide reliable information on the process of change is to establish a forest resources monitoring system, using a globally compatible and consistent methodology. This provides reliable and location specific change information. In consideration of cost, precision and timeliness of results, a sampling approach based on remote sensing was designed and used to cover the entire tropical belt.

The specific objectives of the survey were:

  1. to achieve the highest possible level of consistency and precision in the assessment of forest cover state and change at global and regional levels;
  2. to develop and disseminate a simple and robust monitoring technique for producing the forest cover state and change estimates at global and regional levels with application also at national level; and,
  3. to provide spatial and statistical data for estimating class to class changes of land cover and forest cover categories between the two dates of interpretation at the sample locations and for producing change matrices at regional and global levels.

Though applied only in the tropical zone, the survey methodology is general and can easily be applied worldwide for all the forest formations to provide reliable information on the extent and the process of change on a consistent and continuing basis.


The survey covers all the tropical regions. The World Reference System 2 (WRS2) of the LANDSAT satellites is used to construct the sampling frame. LANDSAT scenes covering approximately 3.4 million hectares serve as sampling units.

In view of cost-benefit considerations, the sampling units have been selected from all LANDSAT scenes with a minimum land area of 1 million hectares and a forest cover of 10 percent or more, estimated on the basis of existing vegetation maps. This has restricted the area surveyed to some 62 percent of the total tropical land area but containing some 87% of the total tropical forest.

The main characteristics of the survey design are as follows:

This first survey round is based on a 10 per cent sample consisting of 117 sampling units randomly selected. The distribution of the selected sampling units by region is 47 in Africa, 30 in Asia and 40 in Latin America. This sample size was chosen to estimate forest cover at global level with a standard error less than ±5 percent. At each sample location, satellite images of the best quality and appropriate season, separated by an approximate ten-year interval, have been selected for observation. The image close to 1990 provides assessment of the current state; whereas the area in common between the “1990” and “1980” images provides the assessment to be made of the changes over time.

A distinctive feature of the methodology lies in the fact that it provides not only forest cover change data, but also maps and change matrices for each sample location. This enables estimations of class to class changes of land cover and forest categories between the two dates of interpretation at sample, regional and global levels: this is essential information for understanding the complex processes taking place, such as deforestation, fragmentation, degradation and afforestation.

The salient features from a remote sensing point of view are the following:

  1. Standard classification of various forest classes (closed, open, with shifting cultivation, fragmented) on a pan-tropical basis;
  2. Interdependent interpretation procedure: this interpretation approach secures the highest level of thematic and spatial consistency between historical and recent image classification. This procedure is the most important element of the methodology since it reduces the error associated to the estimate of changes and makes production of change matrices possible;
  3. Image archive: all images used represent permanent reference as part of a continuous time series; in future these images will be used to estimate the speed of change (3 or more time series);
  4. Low sophistication: in spite of its sound conceptual basis, the methods and procedures developed for this survey are simple and robust and require very low technological inputs; the methodology has been designed for implementation in average developing country conditions;
  5. Flexibility: although applied here in a global survey context, the monitoring methodology can be applied at national or sub-national level without substantial modifications.

The interpretation was implemented at selected regional and national forestry and/or remote sensing institutions, which have a good knowledge of the sample locations and are traditionally involved in forest resources assessment activities. With the two-fold objective of strengthening national capacities for forest monitoring and improving the quality of image interpretation, the Project organized three regional workshops and eight training sessions with national institutions, benefiting 27 countries and 81 participants.

The results and quality of the interpretation undertaken by local institutions were centrally reviewed and evaluated. A database was established and analyses carried out as is presented in the following sections. The project's Geographic Information System (GIS), with its multiple layers of geo-referenced information, has been an integral element of the survey both at the stage of survey design and at the stage of analysis of results.

Considering the extremely poor information available on the processes of change of the tropical forest resources, this project component could be considered an important achievement.

Summary of main results1

Estimates of forest cover and deforestation rates, with associated standard errors, have been produced, at pan-tropical and regional levels, for three different definitions of forest, in order to highlight the important effect of such definitions on the estimation of deforestation rates.

Mean pan-tropical forest cover estimates, at year 1990, range between 40.6 and 54.5 percent of land, depending on the adopted definition of forest, with associated standard errors ranging between 3.5 and 4.5 percent of the mean. Pan-tropical forest cover change estimates, in particular mean annual deforestation rates, range between 0.6 and 0.7 percent of the 1980 forest cover, with standard errors on the range of 12 to 14 percent of the means. These results met the objectives in statistical precision expected during the project's design phase. The higher error in change estimates is explained by the “event” character of change and its consequent high variance with respect to the state.

A most interesting result of the survey is represented by its set of transition matrices, one for each Sampling Unit, that describe in detail the class-to-class changes for the land cover classes. (See box)

The matrices associated with the sampling units can be aggregated at various levels following the standard statistical procedure. Mean transition matrices have been produced and subsequent change analysis has been carried out for the three regions and by three ecological zones (Wet and Very Moist, Moist, Dry).

1 All estimates discussed in the present report refer to the fraction of the sampling frame from which the sample was selected, which represent 62 percent of the total geographical area of the tropical regions. As a consequence, these estimates cannot be directly compared to other estimates referring to the total land area.

Pan-tropical level

Closed forest is the land cover class undergoing the largest number of changes, comprising 51.6 percent of all changes observed, at pan-tropical level, over the period 1980–1990; this change in closed forest can be subdivided by destination categories, as follows:

Land cover classes
Closed Forest
Open Forest
Long Fallow
(forest affected by shifting cultivation)
Fragmented Forest
(mosaic of forest/non-forest)
Short Fallow
(agricultural areas with short fallow period)
Other Land Cover
(agricultural and forestry plantations)

The remaining changes occurred in open forest (12.1 percent of total change), shrubs (11.5 percent) and fragmented forest (8.8 percent), the rest (16 percent) being distributed among the remaining five classes. Over 90 percent of all changes are negative, implying a loss of biomass, while only 10 percent or less are positive, implying an increase of biomass.

Various changes have been observed, with varying environmental implications, which, however, cannot be described through the simplistic dichotomy forest-deforestation; in order to understand these complex land cover dynamics and develop effective preventive or corrective measures, detailed classifications and new analytical tools, such as change matrices and woody biomass flux diagrams, are essential.

Regional level

The detailed analysis of changes occurring shows that there is not only a quantitative difference among the three tropical regions vis-à-vis rates of change but that there are also strong regional characters in the processes of change that indicate distinctive cause-effect mechanisms.

Ecological level

The analysis by ecological zone reveals important elements on (i) the intensity and distribution of forest area changes and (ii) their environmental impact. The analysis of land cover changes by ecological zone shows clearly that, in all geographic regions, the forests of the moist zone have changed, i.e., have been deforested, fragmented, degraded, etc., with much higher intensity than the forests in the wet and dry zones. The percentage of pan-tropical forest area that has changed in the period 1980–1990 is estimated at 10.1 in the moist zone versus 4.8 in the wet zone and 4.6 in the dry zone. However, an indicative analysis of the biomass involved in such changes shows that, owing to the important differences in forest biomass density among the three zones and to the different change processes, the biomass loss in the wet zone is only slightly less than that lost in the moist zone and probably ten times more than that lost in the dry zone.

From the results summarized above it would appear that socio-economic and cultural factors, which are more homogeneous within geographic regions, determine the nature of forest change processes while ecological characteristics determine the intensity of change.

A comparative analysis has shown that estimates of forest cover at tropical and regional level from the FAO's Forest Resources Information System (FORIS)1 are within 95 percent confidence limits of those derived from the present remote sensing (RS) survey, when the data sets are made consistent in respect of study area and definition of forest.

Moreover, FORIS and RS estimates of forest cover and deforestation rate at stratum level are highly correlated, which is an important requisite toward achieving the efficient integration of the two databases aimed at improving the reliability of future country-level estimates for standard reporting years, and at optimizing future RS sample survey rounds. For sake of clarity it should be stated that the RS results do not replace the forest statistics of FORIS but rather complement them, by providing information on the processes of change and insight on the cause-effect mechanisms.

1 Results of the Phase I of the FRA 1990 Project, published in the FAO Forestry Papers 112 “Tropical Countries” [FAO, 1993], and 124 “Global Synthesis” [FAO, 1995].

Top of Page Next Page