Previous Page Table of Contents Next Page

Chapter V
Relation Between the FRA 1990 RS Survey and
Other Survey Efforts on Forest Resources Monitoring

Several programmes are presently under way, which share the broad common objectives of assessing and monitoring tropical forest resources and the aim to provide the international community with appropriate information to facilitate the development of adequate forest policies and to support remedial action in those areas where forest depletion is evident. However, there are wide variations on how this common goal is perceived, which mensurational approaches are adopted, what information is considered adequate, and even on what can be defined as tropical forest.

In order to define the range of validity of these different survey efforts, to assess the comparability of their results and gauge their possible complementarity, it is essential that one understand their main distinctive features. Table (5) 1 summarizes some basic features of four large-scale surveys.

Table (5) 1: Main features of four large-scale tropical forest resources surveys
 Study areaData sources and temporal referencesCoverageLand cover classificationChange classificationState of progress
FAO-FORIS (Forest Resources Information System)All tropical and non-tropical developing countriesExisting national and sub-national surveys. Single date and multi-dateComplete, at sub-national levelForest Non-forest (closed forest open forest woodlands plantations)Deforestation (afforestation) based on forest-non forest classificationSurvey completed and published (FAO, 1993) for tropical developing countries
FAO-RS SurveyEntire tropical zoneMulti-date Landsat images. Close to 1980 and close to 1990Sample of 10% to be increased in future survey roundsClosed forest Open forest Long fallow Fragmented for. Shrubs Short fallow Other land cover Water Plantation9×9 possible transitions, including: Deforestation Fragmentation Degradation Amelioration Afforestation and othersFirst sampling round completed
TREES Phase IHumid tropicsMulti-date NOAA AVHRR 1 km resolution (LAC or HRPT) dated 1990–93 + ERS SAR data 100 m resol. 1994AVHRR Wall to wall (complete) ERS SAR full mosaic of Central AfricaDense evergreen forest; Fragmented evergreen. forest; Dense seasonal forest; Non-forestNot done as yet but planned in TREES Phase II (starting 1996)Forest map of humid tropics completed (1 km resolution) - Analysis of ERS Africa mosaic in progress
LANDSAT PathfinderHumid areas of: Amazon and Orinoco basins; Central Africa; SE continental Asia and West IndonesiaMulti-date Landsat images. Mid '70s, '80s and '90sAll Landsat images taken on appropriate dates available for the study areaForest Non-forest (in some area also deforestation and regrowth)Deforestation (afforestation and, in places, regrowth)Mid '80s coverage under completion for America and Asia; remaining dates and areas in progress

The first two large-scale surveys represent Phase I and Phase II, respectively, of the FAO Forest Resources Assessment 1990 Project; the third one refers to the Project TREES Phase I (Tropical Ecosystem Environment Observation by Satellite) by the Joint Research Centre of the Commission of the European Community; the fourth one refers to the NASA LANDSAT Pathfinder Tropical Deforestation Project, a collaborative research effort carried out under the auspices of the University of New Hampshire, the University of Maryland and the NASA Goddard Space Flight Center.

In order to avoid duplication of work and achieve the synergetic harmonization of all these efforts, and thus to meet more efficiently the underlying common objectives, it is necessary to review the distinctive features of each survey and analyze promising levels of integration.


As reported in several documents of the Organization (FAO Forestry Papers 112 and 124) and in the Introduction to the present report, two distinct complementary approaches have been followed by the Forest Resources Assessment 1990 Project to determine the extent of tropical forest cover and its rate of change:

As expected, the results of the two approaches are different, which may create some confusion for the users of FAO data. The purpose of this section is to guide the user by highlighting their distinctive characters, clarifying their scope and their complementary role in a new forest resources estimation system.

Data sources

The data sources of the FORIS database range from repetitive and highly accurate inventories to non-documented single date estimates of varying, and often unknown, reliability. The mapping scale of such data is also quite heterogeneous. This variability prevents the assessment of overall FORIS accuracy or spatial resolution but it remains an irreplaceable compendium of the best information available at national and sub-national levels.

The present RS survey, being consistent in terms of data sources used (Landsat data), approach and resolution, makes the assessment of statistical and non-statistical errors, if not easy, possible.


From a global viewpoint there is little consistency among the classifications adopted in national or sub-national surveys which, in view of their original scope, applied purposive definitions of “forest”, often subject to national perspectives. Countries with large areas of rich forest tend to exclude from their statistics sparse and fragmented formations while countries with only sparse formations tend to do the opposite.

The RS survey have overcome this problem by adopting a standard classification system, sufficiently detailed to describe consistently the entire range of forest conditions and change processes encountered in the tropics.

Temporal resolution

The reference years of FORIS data sources range between the early 1970s and the late 1980s, with an approximate mean around 1981; in addition, multitemporal observations exist for only 21 of 90 countries surveyed (see Table (1.4) 1 in the Introduction), and such observations are not distributed homogeneously among the geographic regions. These factors are very important in the estimation of deforestation rates during the 1980–1990 period. As the existing country level data set resulted to be, on average, rather out-of-date, one had to resort to the use of mathematical models or adjustment functions to bring forest areas and deforestation rates to the desired reporting period. The risk exists, however, of introducing a certain bias due to (i) the period covered by the existing multitemporal data set; and (ii) the unbalanced geographic distribution of such multitemporal data sets.

Reference dates of RS observations are much closer to the 1980–1990 reporting period although satellite image availability often imposed a certain deviation (see Figure (2.5) 1, Section 2.5, and the list of images in Annex (3.1) 1). The integration of the RS sampling units with the national multitemporal data sets in developing regional models will improve greatly both their precision and temporal consistency.

Thematic resolution (deforestation versus processes of change)

The heterogeneity of classifications peculiar to FORIS sources and the poor consistency of most multi-date data sets placed a heavy toll on the thematic resolution of the final results. Practically only forest for the state and deforestation for the change were applied; these are terms which, in spite of their neatness, carry a lot of ambiguity.

On the contrary, the RS results achieved an unprecedented level of thematic resolution in the analysis of pan-tropical changes, thanks to the consistency of the classification used and the interdependent interpretation of multi-date satellite data. In addition to quantitative estimates of deforestation, for example, the RS survey provides, with its change matrices, valuable insight into the change processes which helps one to understand their dynamics.

Range of validity

FORIS reports country level data but also includes sub-national level information in its statistical and spatial databases. Its thorough wall-to-wall coverage and sub-national statistical and spatial data allow for detailed analyses carried out in combination with socio-economic, demographic and eco-floristic information layers. In this respect, it is unique and irreplaceable.

RS results represent the first round of a continuous pan-tropical survey based on a ten percent sample. With this sampling intensity the results are reliable at pan-tropical and regional levels, where they show relatively small sampling errors. RS results do not replace the FORIS country estimates but can contribute significantly to the description of both regional and ecological trends. With subsequent sample rounds, in the future, the reporting units can be reduced to sub-regional level and will include analysis of variations in rates of change.

Comparison of FORIS and RS estimates

The results of the RS survey and FORIS cannot be directly compared in view of the differing land area of reference. The RS results refer to the land area from which the SUs were selected, which excluded, from all satellite scenes covering the tropical regions, those scenes with less than one million hectares of land or estimated, according to the project's pan-tropical vegetation map, to have less than ten percent forest cover (see Section 2.1.1). The land area actually surveyed represents 62 percent of the entire tropics.

A comparison can be made only when the results of both approaches refer to the same land area. In order to achieve this consistency, FORIS statistical data, viz., forest cover and deforestation rate by sub-national unit, had to be distributed spatially to match the RS surveyed land area. Such spatial distribution of FORIS data by nominal Landsat scenes was carried out through intermediate spatial variables derived from the project's pan-tropical vegetation map. This process of “spatialization” (M. Lorenzini and A. Marzoli, 1994) is based on a multi-variate analysis of correlation between FORIS sub-national statistics and the (64!) classes of the vegetation map. As a result of this analytical process, the area of forest and the area of deforestation of the FORIS database could be distributed spatially and, subsequently, cut into Landsat frames.

Although at local level (single Landsat frames or small aggregations) the error of “spatialization” could be significant, due to the limitations of scale, 1:5 million, and classification of the vegetation map, it was considered that for larger areas the results of this process were acceptable and, particularly, that their accuracy improved with the size of the aggregation.

Figure (5.1) 1 a and b show the stratum level estimates of forest cover and deforestation rates from RS and FORIS for the sampling frame from which sampling units were selected (see Section 2.1 “Statistical Design”). The RS definition of forest adopted here corresponds to the intermediate F2 definition (described in Section 2.1.1), that better matches the FORIS standard definition. According to this definition, the land cover classes that constitute the forest are:

closed forest + open forest + 2/3 of fragmented forest

The correlation coefficient of FORIS and RS forest cover estimates at stratum level is 0.969, while that for deforestation estimates is 0.793. In spite of the elements that distinguish the two estimation approaches, the two sets of data are highly correlated, particularly for forest cover estimates. In view of the strong common ground represented by the highly-correlated forest cover, and in consideration of the limitations of the multi-date FORIS database, the lower correlation of deforestation estimates indicates that there is both scope for, and need of, further analysis and, probably, that the RS deforestation estimates could contribute significantly to the refinement of FORIS change estimates.

A comparison of pan-tropical FORIS and RS forest cover estimates at the years 1980 and 1990 is given in Figure (5.1) 2. The correspondence between mean RS and FORIS estimates is so high to be almost identical; this fact, however, should not be over-emphasized since the forest cover, at 95 percent probability, is not exactly the mean, rather it lies anywhere within the error margin, defined here as plus or minus two times the standard error.

In Figure (5.1) 3 the comparison is between FORIS and RS regional forest cover estimates for the year 1990. At regional level the correspondence between FORIS and RS estimates is lower, but still significant. Two elements probably concur to make the discrepancies in regional forest cover estimates: (i) sampling intensity at regional level, which is less than optimal; and (ii) discrepancies in forest classifications between the national statistics, base of FORIS data, and the RS classification.

a)Figure (5.1) 1 a
b)Figure (5.1) 1 b

Figure (5.1) 2: Comparison of FORIS and RS forest area estimates at years 1980 and 1990 Pan-tropical level (RS sampling frame)
Figure (5.1) 2

Figure (5.1) 3: Comparison of FORIS and RS forest area estimates at year 1990 by geographic region
Figure (5.1) 3

1 Estimates based on Forest Resources Information System (FORIS) database.
2 Mean forest cover (definition F2) plus two standard errors based on remote sensing results (RS).
3 Mean forest cover (definition F2) based on pan-tropical remote sensing survey (RS).
4 Mean forest cover (definition F2) minus two standard errors based on remote sensing results (RS).

Complementarity and integration

From the aspects described above, it is evident that there is a strong complementarity between the two approaches and, at the same time, a strong correlation, as shown in the two graphs in Figure (5.1) 1. This correlation would be best exploited by integrating the two data sets into a unique two-phase resource estimation system.

The integration of the two data sets, which has yet to be carried out, will result in: (i) more reliable prediction models and, consequently, more detailed, homogeneous and up-to-date country level data; and (ii) more efficient stratification systems to optimize future sample rounds and the expansion of RS results.


Project TREES Phase I (Tropical Ecosystem Environment Observation by Satellite)

From a technical point of view, the approach followed by the TREES Project and that followed by the FRA 1990 Project's remote sensing survey are complementary:

With the wider objective of contributing to the coordination of international survey efforts (namely those of the United Nations and European Commission) related to the monitoring of tropical deforestation and degradation, the two projects have always cooperated closely and have identified areas of integration, of data and methodologies, that will be of mutual benefits. The strengthening of technical links between the two projects is now being formalized into a cooperative operational programme, with the following objectives:

  1. Integration of TREES' results and TFIS data (Tropical Forest Information System) into the FORIS database of the FRA 1990 Project and integration of FORIS databases and RS results with the TFIS of the TREES Project.
  2. Use of the NOAA-based forest map and other relevant TREES data to optimize the FRA 1990 Project's high resolution survey approach which would, in addition to improving its focus on forest area changes, satisfy the TREES' requirements for map validation and calibration.
  3. Improve TREES' calibration phase and contribute, with the high resolution change maps, to the development of spatial and statistical modelling of forest cover change and to the identification of areas exposed to a high risk of deforestation.

The linkage with the TREES' programme is well defined, particularly in view of the clear complementarity of the two approaches, and could be integrated into a cooperative multi-phase global survey.

NASA LANDSAT Pathfinder Tropical Deforestation Project

The Pathfinder Project is presently progressing with the classification of the first layer of Landsat data, with the mid 1980s as the time reference (see Table (5) 1). Multi-date analysis has been carried out, up to the present, for the Brazilian Amazon Basin, based on Landsat data acquired around years 1978 and 1988. The mid '80s forest cover classification will be followed by a historical layer, based on Landsat data of the mid 1970s, and by a recent one based on Landsat data of the early 1990s.

In addition to the forest cover maps derived from the classification of Landsat data, an important contribution made by this project to the community of investigators, involved both in local and global level studies, consists of the large archive of multi-date Landsat data being built up as a result of the in-depth review of all currently available archives and receiving stations' catalogues.

The main features that distinguish the Landsat Pathfinder approach from the FRA 1990 Project's approach can be summarized as follows:

Study area:Major forest areas of humid tropics versus entire tropical coverage.
The Pathfinder programme is concentrated on the large remaining blocks of wet and very moist tropical forests.
The FRA 1990 Project's study area includes all ecological zones down to very dry conditions, as discussed in the present report.
Coverage:Wall-to-wall versus statistical sampling.
Within the study areas the Pathfinder programme is carrying out a complete mapping exercise, limited only by missing data (the areas which resulted to be permanently cloudy or non-recorded). The output of this large scale mapping exercise will find wide use beyond the estimation of tropical deforestation; digital maps of tropical forest cover, with high spatial resolution, will be of great support to national and regional planning.
FRA 1990 coverage entailed a statistical sampling approach of subsequent sample rounds (each covering ten percent of the study area), following the principles of the continuous forest inventory design. Some relevant features of the sampling approach are: (i) the possibility of producing results at various reporting levels, with known confidence limits; (ii) that such results can be released at comparatively short time intervals, thus responding to changing information requirements; and (iii) that, with increasing sampling intensity, such results become progressively more detailed and accurate.
Classification:Coarse versus detailed classification.
The Pathfinder classification includes one forest class: forest cover, and the following non-forest classes: deforestation, non-forest vegetation, water, clouds and other.
The FRA 1990 classification includes ten Main Classes, comprising four forest classes, and 13 additional ones (optional) of greater detail, as presented in Section 2.2. The influence of the land cover classes on the analysis of changes is obvious, as discussed in Section 2.2.1.
Data analysis:Independent digital classification versus interdependent visual interpretation of multi-date satellite data.
The pathfinder approach consists basically of producing and comparing forest maps based on satellite data sets acquired in three different periods, while the FRA 1990 method is based on the interdependent interpretation of multi-date images with the main focus of achieving the highest possible level of consistency and reliability in the detection of land cover changes.

In considering of the above elements, the Pathfinder approach could be considered as being more extensive while the FRA 1990 approach could be considered intensive.

The first approach aims at the “core” of the tropical deforestation problem, by mapping and quantifying deforestation where the forest is still abundant, but with comparatively coarse thematic resolution. Its major contribution is likely to consist of geographic evidence on: (i) the extent and distribution of humid tropical forests; and (ii) the location of current active deforestation fronts.

The second approach aims at providing a consistent, detailed description and qualification of land cover change processes (not only change in terms of deforestation) for the entire tropical areas; with its detailed description of change processes, FRA 1990 results help one to understand the mechanisms of change, an understanding which is essential to the development of appropriate policies and to the efficacy of remedial action plans.

To date, the main cooperation between the FRA 1990 Project and the Pathfinder Project has consisted in the sharing of Pathfinder's Landsat data and FRA 1990's spatial monitoring results of the Central Africa sub-region.

Previous Page Top of Page Next Page