0173-B1

Forests Without Trees? A Note on Problematic Forest Definitions and Change Assessments

S. Wunder^[1]

Abstract

Three decades after deforestation appeared on the global agenda of key environmental problems, we are still fumbling for adequate forest definitions and reliable data. The introduction of new assessment methodologies often causes much larger changes in forest estimates than genuine changes in vegetation cover. This paper looks at national forest assessments from a user's perspective, giving examples of data problems observed in an applied study of land-use changes in five tropical countries.

Four critical issues are discussed: canopy cover, spatial resolution, sample size and time scale. Each of these factors entails potential pitfalls that may severely limit the comparability over time of forest estimates. It is shown that ideological motives to absolve certain forest-clearing activities from charges of deforestation, e.g. slash and burn by shifting cultivators or clear-cuts by forestry companies, give estimates that are not scientifically defendable under the highly dynamic land-use scenarios in the tropics.

Based on this methodological discussion, it is recommended that the forests without trees be eliminated, and that future forest assessments be based exclusively on objectively measurable tree-cover criteria, rather than arbitrary temporary usage criteria. This also implies a number of recommendations to international donors and forestry agencies. First, we should recognise the weakness of existing forest assessments and form a consortium with the best experts to combine ongoing efforts and improve methodologies. We should make more and better use and of existing satellite imagery. We should generally give higher priority to wall-to-wall measurements, and use differentiated canopy-cover criteria instead of only a single and highly controversial one. We should produce more frequent results that allow us to monitor, analyse and counteract deforestation processes in a much more timely way. Otherwise, we risk global forest assessments lagging hopelessly behind new international demands, e.g. in relation to biodiversity and especially vis-à-vis the global carbon cycle.

1. Deforestation: what does it mean?

Box 1 shows current definitions of ‘forests’ and ‘deforestation’ by FAO (with my added keywords).

2. Canopy cover

The 10% canopy-cover threshold in the FAO’s definition has been much criticised: areas are being counted as forest even though they have lost up to 90% of their originally full canopy-crown cover, as well as many of their environmental forest functions. It should certainly be kept in mind that the FAO uses a ‘broad’ definition of forests, and therefore a ‘narrow’ definition of deforestation. This ‘narrow’ definition thus tends to view deforestation as a conversion process normally involving a drastic intervention causing land-use, distinguishing it from forest degradation (logging, over-hunting, over-grazing, etc.) (Wunder 2000).

Other international organisations have chosen to adopt a more differentiated criterion, and to use several thresholds. For instance, the TREES project has classified forest cover greater than 70% as ‘dense forest’ and the 40-70% range as ‘fragmented forest’. The IGBP has used a 60% threshold for forests and 40-60% for woodlands, while UNEP (2001) uses 40% for closed forests and 10-40% for open or fragmented forests. The scientific argument for FAO's 10% has been that it is ‘the minimum canopy density where naturally occurring formations of trees exist as communities...as opposed to areas where trees exist scattered in the landscape or in rows’ (FAO 2000a: 5). However, arguments for a higher cut-off may seem at least equally valid.^[2]

The choice of a cut-off criterion can make a large difference in landscapes with open forest formations or forest-savannah transition zones. The FAO formerly applied a 20% crown-cover threshold to countries to the group of developed countries. In the FRA2000, this was lowered to 10% in order to obtain a unified global criterion. For Australia, with a lot of open vegetation formations, this reclassification meant that an additional 118 million ha of forest were suddenly ‘discovered’.^[3] In developing countries with a high proportion of fragmented forests, the differences can also be large. For India, national forest cover using the 10% criterion for crown cover was estimated at 19% of land area, while using a 40% criterion reduced this figure to just over half (11% of land area) (UNEP 2001). Obviously, changing crown-cover criteria can severely limit the comparability of forest assessments over time. So the first question one has to pose for each deforestation figure is: Which cut-off criterion was used?

3. Spatial resolution

Secondly, one must ask a related question: At what spatial resolution was forest cover measured? Table 1 provides a numerical illustration, based on numbers that were derived from Gabon. Let us first look at two patches of moist forest, as indicated by an imaginary set of satellite photos or aerial photographs. One image shows a primary, unlogged forest; the other forest has recently been logged. Using the FAO criteria, we are examining vegetation cover at five different pixel sizes that increase stepwise: 1 m², 10 m², 100 m², 1000 m² and 10 km².

Table 1. Forest cover and spatial resolution levels.

A hypothetical example with three landscape types

Figure 1. Forest-cover measurement and spatial resolution - Primary and logged forests compared

Source: Table 1

Note: Differences between unlogged primary moist forest and logged forest shows deforestation

The unlogged, primary forest includes natural tree-fall gaps. At an extremely detailed scale of 1m² pixel size, these gaps cover approximately 5%, so that forest cover is 95%. At the next pixel size (10 m²), they are reduced to 1%, not being registered at all at lower resolution levels: the small openings are ‘averaged out’, and the area will appear to be completely covered by tree-canopy cover. For the logged-over forest, there will be more canopy openings due to tree-fall gaps, skidder trails, and roads. At 1 m² pixel size, 20% of the area appears to be without forest cover, but that figure diminishes steadily with resolution levels, because the openings become relatively smaller. At 10 km² pixel size, resolution is too coarse to register any logging activity: the area appears to be fully forested. The numbers from the logged-unlogged forest comparison are also illustrated graphically in Figure 1. Note that the difference between the two forest types, which can be interpreted as deforestation caused by logging, shrinks from 15% points to 12%, 7%, 2% points, and finally to zero. So, as can be seen in Figure 1, estimated deforestation is inversely proportional to pixel size. Different thresholds of detectability produce different estimates of deforestation.

The example of logged forest was just one of several options in what one might more generally call a landscape of ‘cleared islands in a sea of forests’. Alternatively, one could have chosen patchy shifting cultivation, small-scale artisan miners, or oil exploration in an otherwise undisturbed, closed forest. But, a third type of landscape would be ‘forest islands in a sea of open space’, e.g. forest patches in a savannah area or in an agricultural landscape. Note that the relation between pixel size and forest cover is precisely the opposite in this case: the coarser the resolution level, the less our vegetation screening captures forest fragments. So, at 10 km² pixel size, none of the forest fragments is large enough to occupy 10% of the pixel. Hence, there would appear to be no forest cover. Estimated deforestation of forest fragments is positively proportional to pixel size.

What does this small example imply for deforestation time series in the tropics? Many of the nation-wide historical forest maps available from the 1950s, 1960s and 1970s are of a coarse-resolution type, for example, at the 1:500,000 or 1:1,000,000 scale. For the last decade or two, we have more nation-wide high-resolution maps available. But these are not directly comparable to the old, coarse-resolution ones. In countries with a predominance of the ‘forest islands in a sea of non-forests’ situation (e.g. Ecuador or Venezuela), these newer forest maps have revised national forest cover significantly upwards. This is because many previously ignored forest fragments are now being detected. In countries where the ‘cleared islands in a sea of forests’ are more common, e.g. because of a lot of shifting cultivation (e.g. in Central Africa), new forest-cover estimates have been revised downwards.

Disparities between national forest stock estimates can thus be huge, especially for fragmented landscapes - which become more important as humans penetrate more into tropical forests. Figure 2 compares for five study countries the TREES estimates from the early 1990s with those from FAO's FRA1990, FRA2000 and the IUCN et al. Conservation Atlas. While most of the IUCN and TREES figures only vary by 5-10%, FAO's mostly model-projected stocks from FRA1990 were 10-30% lower for Venezuela, Gabon and Ecuador, but higher for Cameroon. For FRA2000, the deviations are even larger (e.g.38% more for Ecuador, but 37% less for Cameroon) - not only vis-à-vis TREES, but also compared to the FRA1990 (e.g. Gabon, PNG). The disturbing implication is that when we compare different sources, or when assessments like FAO's FRA change methodology, altered assumptions come to quantitatively dominate over what we would really want to know: What happens to forest cover over time?

Figure 2. Differences in the tropical forest-cover estimates by TREES¹, IUCN² and FAO^{3, 4}

Notes:

1. TREES figures from Mayaux et al. (1998)
2. IUCN, WCMC and CIFOR’s Conservation Atlas
3. FAO FRA1990 figures from FORIS (wet, moist and montane forest)
4. FAO FRA2000 figures from FAO (2001a)

4. Sample size

A third critical point concerns the size of selected study areas, in particular the question of spatial sampling versus wall-to-wall measurement. For cost reasons, it has not yet been possible to measure the entire world’s forest cover at an equivalent level of detail. Some observers have argued that this can be effectively overcome by careful spatial sampling. For instance, in the FAO’s global forest assessment, 117 sites representing a 10% sample of tropical forests are analysed using remote-sensing images. The results are then used to extrapolate forest-cover changes for different forest types. The TREES project has now produced new deforestation estimates for the humid tropics, revising FAO's estimates downward by 23%, based on 100 sites with high deforestation that represent 6.5% of tropical land area (Achard et al. 2002). A recent study has even argued that the sub-national level is more appropriate than national data for the study of deforestation processes and causes (Geist and Lambin 2001), but some of their sub-national regions, such as the Brazilian Amazon, are actually larger and more heterogeneous than most tropical nation states.

Measurement options used in extrapolating sub-national results on to the national level depend very much on the type of landscape and land use under analysis. CIFOR research in Cameroon showed that an economic crisis induced an increase in food-crop production and subsequent acceleration of forest-cover loss from the mid-1980s onwards. Different regional studies from the forest zone all confirmed the same trend. Here, a stratified sample would thus give a good reflection of the countrywide picture. On the contrary, in Ecuador spurts in deforestation were much more regionally heterogeneous, with areas being opened up sequentially for settlement by new roads, oil production, specific crop booms, etc. (Wunder 2000). Predicting national deforestation in Ecuador based exclusively on pre-identified regional samples would thus have given severely biased results, because deforestation fronts shifted dramatically over time. In line with an observation by Matthews (2001: 6), in general this should caution us against sample-based deforestation assessments, including the extended sample-based World Forest Survey proposed by the FAO for future assessments (Päivinen et al. 2000: 13). For many countries with heterogeneous land use and multiple fronts of deforestation, only ‘wall-to-wall’ methods can provide us with correct answers.

5. Time scale

Finally, the time perspective relating to deforestation matters a great deal. The FAO explicitly refers to long-run (or permanent) loss of forest cover. The term ‘excludes areas where the trees have been removed, due, for example, to harvesting or logging, and where the forest is expected to regenerate naturally or with the aid of silvicultural measures within the long-term’ (FAO 2001a: 25). An area only truly leaves the forester’s domain once a planned, long-term decision is taken to put it to an alternative use, for example, to enlarge a residential area or to crop it. In that case, the land will obviously not be reforested, and may thus safely be counted as deforested.

However, applying this temporal distinction to the reality of the tropics runs into severe problems. First, tropical land uses tend to be highly unpredictable, due to both volatile economies and insecurity of land tenure. For instance, squatters may occupy areas intended as plantations to grow crops or, vice versa, plantation companies may crowd out local farmers. To predict on the ground whether an area that is cleared now will be tree-covered again in ten years is a formidable task.

All forest-cover removal should count as ‘deforestation’, including areas that have been clear-felled for allegedly temporary purposes. The proposal is thus not to try to distinguish between ‘temporary’ and ‘permanent’ forest clearing, nor between ‘reforestation’ and ‘afforestation’. I use the wording ‘try to’, because much of what is reported as deforested from national governments through to the FAO today already fails to make this distinction in practice. In other words, my proposal is ultimately to delete the ‘usage criterion’ in Box 1 and use exclusively a ‘tree-cover criterion’ to define deforestation. The justification for this proposal lies in two interrelated claims:

I The difference between ‘temporary’ and ‘permanent’ forest-cover removal cannot be measured objectively, other than, at best, historically and retrospectively. For the purpose of monitoring forest loss, the difference between ‘temporary’ and ‘permanent’ forest removal is thus not scientifically verifiable in any consistent manner.

II Even if, hypothetically, this distinction could be objectively measured, any such temporal cut-off criteria would tend to be arbitrary, and would include ‘desirable’ and ‘undesirable’ land-use processes of change on both sides of the selected borderline.

Regarding the first claim, the example in Box 2 show that one can only safely distinguish between temporary and permanent deforestation ex post facto. In principle, one can only assess whether the initial clearing has proved to be temporary or not ten years after it occurred. The empirical measurement of temporary forest clearing makes little sense in societies that are subject to a high degree of structural change. This applies whether one uses the FAO’s criterion that exonerates forestry operations (the first example), or alternatives that seeks to exonerate shifting cultivators. Some intended temporary clearing inevitably becomes permanent. The two clearing types are better seen as part of a temporal continuum. So, the exclusive ideological focus on permanent changes in land use is simply misdirected.

6. Recommendations for future assessments

This paper has pointed to a number of problems that currently make it difficult for us to monitor national deforestation over time. The following recommendations^[4] may provide a way forward:

• intergovernmental fora must fully recognise the weaknesses of our current forest-cover knowledge

• intergovernmental fora and donors should provide the FAO with a mandate and the resources to form a consortium of the best experts and organisations to produce a more consolidated FRA,

• ‘wall-to-wall’ measurement should be one of the guiding principles for the forest-cover component of a future FRA, combining different resolution levels and different source types

• a single ‘state-of-the art’ methodology should be agreed centrally among the participants of this consortium and applied globally to forest-cover assessment

• baseline establishment should have high priority, enabling historical forest-change analyses

• forests (and deforestation) should be defined by canopy-cover criteria, eliminating arbitrary or ideologically biased usage criteria. Crown-cover criteria should be more differentiated, including at least one cut-off point above the current 10% (e.g. 50% or 70% canopy cover).

• the monitoring system should produce forest-cover indicators more frequently than once a decade.

• the centrally agreed-upon methodology should be promoted by building capacity in tropical countries,

• The growing scientific recognition of the role of forests in the global carbon cycle and global warming should underscore the urgent need to raise the sophistication and frequency of forest cover assessments.

References

Achard, F., H.D. Eva, H.-J. Stibig, P. Mayaux, J. Gallego, T. Richards and J.-P. Malingreau. 2002. Determination of deforestation rates of the world's humid tropical forests. Science vol. 297, 9 August.

FAO. 2000a. FRA2000. On definitions of forest and forest change: FRA Working Paper 33. Rome: FAO.

FAO. 2001a. Results of the Global Forest Resources Assessment 2000: Committee on Forestry, Item 8(b) of the Provisional Agenda, Fifteenth Session, Secretariat Note, Rome, Italy, 12-16 March 2001.

Geist, H. J., and Lambin, E. F. 2001. What drives tropical deforestation? A meta-analysis of proximate and underlying causes of deforestation based on subnational case study evidence: LUCC Report Series, no.4, Louvain-la-Neuve: Ciaco.

Grainger, A. 1996. An evaluation of the FAO tropical forest resource assessment, 1990. The Geographical Journal vol. 162, No. 1.

Malingreau, J.-P. 1993. Satellite monitoring of the world's forests: A review. Unasylva vol. 44, no. 3.

Matthews, E. 2001. Understanding the FRA2000: Forest Briefing No. 1, Washington D.C.: World Resources Institute.

Mayaux, P., Achard, F., and Malingreau, J.-P. 1998. Global tropical forest area measurements derived from coarse resolution satellite imagery: a comparison with other approaches. Environmental Conservation 25(1).

Päivinen, R., Gillespie, A. J. R., Davis, R., and Holmgren, P. 2000. FRA2000. Assessing state and change in global forest cover: 2000 and beyond: Rome: Forestry Department, Food and Agriculture Organization of the United Nations.

UNEP. 2001. An assessment of the status of the world's remaining closed forests: UNEP/DEWA/TR-01-2, Nairobi: UNEP/NASA/USGS/USFS.

Wunder, S. 2000. The economics of deforestation: the example of Ecuador: Houndmills & New York: Macmillan Press & St. Martin's Press, June.

Wunder, Sven. 2003. Oil wealth and the fate of the forest. A comparative study of eight tropical developing countries (forthcoming). London: Routledge.