4 Methodology

In small forest areas of few hectares, measurements of the forest characteristics can be made on all trees composing the survey area. It is a 100% survey where the accuracy of the estimates will be high and cost will not be very great. In large forest areas, sampling techniques are used instead for cost and time considerations. A representative sample of the total population is statistically withdrawn, where various tree parameters and forest characteristics are measured and assessed. For totals, the sample estimates are extrapolated to the total areas and populations.

The size of the forest population in this study extends to the entire land surface of the forest domain in the world. Individual countries will be independent survey and compilation areas for forest resources assessments. For a population of this size and diversity, the sampling design to be selected should be statistically and technically adequate in order to generate reliable and cost-effective results. In view of the size and characteristics of the survey area and population and of the allocated resources, the sampling may be random or systematic, stratified or unrestricted and simple, multi-phase or multi-stage.

This methodology will, thus, be designed for the survey of the world forest resources, where the forest domain expands over a large variety of ecological regions ranging from tropical, to temperate to boreal; from desertic to dry to wet; and from low land to high mountains. It also extends to countries with contrasting social and economic conditions. Any methodology should therefore take into account the social, economical and ecological particularities of each geographic entity whether it is country or region. While, for all countries, the statistical principles should be identical or of comparable outputs, the methodology should be adapted to the specific conditions of each country.

The fundamental precepts in the formulation of this methodology is that information on forest resources from the tropics, as the world reservoir of biological diversity, is sporadic, scattered, often obsolete and of questionable value when reliability is concerned. At the same time, statistics from the developed countries are far more consistent, comprehensive, detailed and reliable. It is therefore essential to try to bring the sets of information from the two blocks to agreeable levels of reliability and to common and harmonised formats.

4.1 Objectives of the survey

The long-term objectives of the global forest survey (GFS) will be to:

• improve management of forests by providing decision makers and stakeholders with the best possible, most relevant and cost-effective information for their purpose at local, national and international levels;

• ameliorate and consolidate the baseline information at country and FAO levels from a statistically verifiable data (not relying on secondary statistics) for policy-makers, scientists and the general public on the state of the forest resources and biodiversity at global, regional and country levels. Thus, economically, socially and environmentally, the survey should aim at collecting information that provides a broad view on the present state of the resources in their current distribution, and a reliable projection in the future;

• promote multilateral co-operation and international partnerships for conservation, development and management of the global forest resources for the sustainable use of their components.

The short-term objectives of the GFS will be to:

• develop a practical approach for the assessment and monitoring of the forest resources at country and global levels that meets the concurrence of the member countries and donor community;

• establish, under the leadership of the FAO, partnerships and co-operation linkages with well-established national and international institutions for the carrying out of world wide forest survey on continuous basis;

• support the member countries in establishing continuous forest resources assessment and monitoring systems and in developing their national capacities for the purpose;

• develop a world database on forest resources in which statistical, textual and photographic information are attached to digital thematic cartographic data;

• provide technical assistance to member countries in building national capacities in the areas of methodology development, use of new technologies and establishment of forest monitoring systems;

• secure a continuous reporting on the progress of the survey, the interim results and the state of the forest resources at any point of time.

4.2 Design considerations

While very few initiatives for forest resources assessments and monitoring have seen light in the developing world, the industrialised countries are far more advanced in this area and have designed and put into practise a number of statistical sampling methods for the periodic surveying of their resources. These are mostly based on systematic sampling using various designs of grid nets with different spacing depending on the countries. For the purpose of the GFS, parts or all of these existing sampling schemes will be utilised, and this depending on their sampling intensities.

The world’s forests were mapped by the USGS, which classified natural vegetation into closed forest, open or fragmented forest, and other wooded land. Other Land Cover (OLC) is recognised in the map as an independent unit. In order to have a holistic view of the resources, the OLC map unit will be part of the survey as it is judged to comport a substantial amount of natural vegetation along hedges, rivers and streams, or of planed cover in line planting as hedges, roadside, homestead and wood-lot planting, etc. The vegetation categories together with the OLC will constitute the survey area.

Considering the results published by FRA, the world’s forest map classifies the total land surface of the planet into forest and non-forest. Out of 129,358,670 km² of total land surface, the world forest domain was estimated at 34,423,690 km² in which there are 20,099,120 km² in the developing countries and 14,324,570 km² in the developed ones.

The world land surface is divided into countries where the forest cover is being distributed. The survey will concentrate on the forestland within each country apart. Each country will constitute an independent stratum in this sampling design. The chief orientation of this methodology is that the forest survey will be carried out by national or regional institutions in partnership with the FAO. National capacities will be built up under the GFS Project.

As to integrate the existing sampling schemes in many countries into the GFS, a number of sample plots will be selected systematically within each country independently. Selection should follow certain basic principles that will be common for all countries, such as use of certain design of dot grid, sampling intensity, sample unit shape and size, etc. Depending on the size of the individual countries, sampling intensity of permanent sample units will be fixed to the level of affordable surveys, but will be open to increase by temporary plots when a level of precision is desired for specific variables. Sample unit shape and size will likely vary from country to another as, in most cases, these characteristics do not influence significantly the sampling error when incorporated into sound statistical design.

Remote sensing data and techniques will be essential tools for vegetation stratification and total area estimation of different strata, and for assessing the state and change of forest cover at the sampling units, country, regional and global levels, within the multiphase sampling design. Well correlated models, from sample area, between remote sensing data and field variables will allow predicting these variables at acceptable precision. The remote sensing data would include Landsat TM or SPOT Imagery, aerial photographs, etc.

Depending on the way remote sensing data is exploited, it influences the operation costs with either increase or decrease. Very broad stratification with few created strata results in higher heterogeneity within each stratum, which requires an increase in field plots. But, although detailed stratification allows achieving better precision, field plot intensity can seldom be reduced. The increased use of stratification may actually increase the need for field data. Stratification should therefore be optimised for cost-effectiveness.

Overall cost of a national or global forest survey is governed by the precision requirements of inventory totals and the level of information details to be gathered. The predefined information needs and the allowable standard error for the global survey should be optimised in such a way that they fall within the country, donor and implementing institution capacities.

In view of sampling design options, there will be need to reconcile between different requirements such as large span of variables and then wide range of information, and meeting the usual constraints of limited budgets. An efficient optimisation of the entire survey will be needed in point of view of both cost effectiveness and the resulting consequences for future surveys. The sampling should be flexible to allow any change in the design when it is desired to increase or eventually decrease precision of estimates, to expand variables or to introduce new technologies such as remote sensing technologies, GPSs, etc.

For equal precision, there are a number of alternatives for sampling designs. Considering the objectives of the survey, the size of the inventory area and the need for repeated measurements, the multi-phase and multi-stage sampling has been proved more cost-effective than other techniques. Multi or two-phase sampling with stratification can be used to increase the efficiency of the survey over time by improving remote sensing data sources and other stratification criteria.

Continuous monitoring of the forest resources through permanent field plots will require highly accurate location of the plots on the ground. GPS technology offers an excellent and probably the most cost-effective alternative especially after the USA Administration decided to discontinue intentional degrading of the GPS.

4.3 Design approaches

Two options are presented below for considerations. The first involves remote sensing data sources for forest cover stratification and area estimation from a defined sample to be selected according to the two-stage sampling method using a 30x30km size grid net. Within the selected primary units, secondary units, constituting field plots, will be selected systematically from a 3x3km dot grid. The second is designed as three-phase sampling where the land area of each country will, in the first phase, be divided into blocks of defined sizes. Each block will be subdivided, in a second phase, into sub-blocks that will be interpreted for area calculation and sample size determination (see figures below). The third phase will consist of selecting field sample plots per forest cover class to be located in the field for measurements.

4.3.1 Option 1

In this option, the proposed sampling revolves around a statistical design that will produce at the end the planned outputs at the desired precision. The approach would be multiphase sampling with stratification; remote sensing will play a role in stratifying the forested areas within a selected set of primary sample units, out of which field plots will be selected and visited for measurements. Primary and secondary sample units selection will be done according to the two-stage sampling method. The choice is made on systematic sampling, as is considered more representative than randomised allocation of samples since field plots are uniformly spaced over the entire population distributed over a given area. The advantage of the two-stage sampling approach over others is that, while securing data collection according to sound statistical approach and maintaining sampling error low, the amount of fieldwork is significantly reduced and the logistics are substantially facilitated.

Phase 1. The first phase is relative to classifying the total grid blocks falling in a given country into forest and non-forest land, on the basis of the FAO vegetation classification system incorporating terms and definitions that will recognise trees outside forest. This first level stratification would be carried out using the available and affordable remote sensing data (AVHRR, Landsat TM or SPOT images, Aerial photographs, etc.) in combination with existing maps and other sources of information. Considering the land area of each country separately as an independent stratum, its forest area will be surveyed independently according to a global, coherent and GFS standardised inventory method.

Phase 2. Under this phase and based on the pre-stratification, some grid blocks will be selected systematically among those classified as forestland according to the fixed sampling intensity. The selected blocks, that will serve as sample units for vegetation classification using remote sensing data, will be visually or automatically interpreted on the basis of the above indicated FAO vegetation classification (see annex 2). The interpreted remote sensing units will constitute the Primary Units (PUs) and will serve for allocating filed sample plots and for total stratum area determination.

Phase 3. The third phase will consist of selecting field plots within each primary unit. The plots will constitute the secondary units where field data on forest and tree variables as well as on environment and local population will be collected. In the existing experiences from the developed countries, field plots are frequently distributed in clusters of two or more, although there is a number of countries where concentric plots are used. For economy of time and budget, clustered plots are believed to be better suitable to this project especially for the developing countries where little or nothing has been done. From field experiences also, a cluster of 4 plots of 0.5 hectare each is an average working day for a field crew composed of two technicians and two labourers. Plots should be laid out with a spacing of not more than 500m apart and an average amount of variables to be measured.

a. First phase

Using a created gridlines overlaid onto the map of land area of a given country at 30km interval, the total population of the PUs will be determined for the whole land mass. Excluding the industrialised nations, the total land surface of the developing countries was estimated at 75,936,970 km². This gives a total number of Primary Units as 84,374 PUs.

The procedure will therefore consist of preliminary classifying the 84,374 PUs into forest or non-forest lands. Each primary unit classified as forested one should include at least 1% of its total surface a forest vegetation (trees and/or shrubs), natural or planted, according to the terms and definitions of the FAO vegetation classification. The 1% could be a contiguous block of natural or planted forest, scattered forest vegetation or in line along river banks, hedges or roadsides or in patches around homesteads.

For the sake of homogeneity of the survey results, the land area of the developed world should also be classified into forest and non-forest using the 30x30km developed grid. Countries with classified land following methods compatible with the GFS may be discarded from the exercise but results of their own classification should be used by the project. The given area of the developed countries raises to 53,421,700 km² or 59,357 PUs. This gives a total number of 143,732 primary units for both developed and developing countries.

If, for the developing countries only, the reported forest cover area of 20,099,120 km² is considered and assimilated as contiguous blocks of forest (which is not the case), the total population of PUs falling inside the forest domain will be 22,332 PUs. At 10% sampling intensity, the sample size will be 2,233 primary units or 2,009,912 km² approximately will be analysed for detailed vegetation classification, forest type area estimation and field plot allocation. The 10% sampling intensity at the first stage level will be selected systematically from those classified as forested units.

When the reported forest area in the two groups of countries is taken, the forest area sums up to 34,423,690 km² or roughly 38,249 PUs inside the forest domain. The preliminary classification will allow producing the correct number of the forested primary units at global, regional and country levels that will be sampled. These assumptions are given only for procedure description purposes. The actual forest classified area at country level and at global level will be much higher because it will incorporate vast tracts of land with very low forest cover. Will be excluded areas composed of permanent sand dunes (deserts, coastal dunes, permanent snow cover, water bodies, significantly large built-up areas, etc.).

Based on the same assumptions and on the basis of 10% sampling intensity, the sample size will be around 4,758 PUs. This makes approximately 4,282,200 km² will be interpreted for area determination of the various natural vegetation strata (closed and open forests, fragmented forest, fallow system with regrowth from natural vegetation, shrub, wooded grassland, forest plantation, homestead and roadside tree planting, riverine vegetation, etc). This number of selected PUs is result of interpretation outcome and of adjustments for countries with less than 10PUs. A minimum of 10PUs will be required for countries where the 10% sampling intensity produces less than that. Preliminary calculation of the PUs per country is given in annex 1.

As to simplify the exercise and secure involvement of national experts, this first land stratification should be carried out for each country separately. An amount of efforts will be required to standardise the grid nets among countries or to design comparable gridlines in terms of sampling point of view. Assistance to the countries in this matter should be given within the framework of forest inventory implementation.

Table 4. Example for determination PUs and sample plots

Remote sensing is certainly the appropriate mean for classifying the 1,293,587 primary units into forest and non-forest lands. The level of precision of classification of these PUs will depend on the remote sensing material utilised, such as aerial photographs or satellite recorded data and on the correlation of land use/land cover classes with the remote sensing set of data. Vegetation classifications detailed and/or using criteria uneasy measurable with the required accuracy on satellite imagery are source of higher interpretation errors.

Use of aerial photography for this work allows producing more precise results in comparison with the known efficiency of satellite captured data in estimating areas of various vegetation types. Even taken at high altitude, aerial photographs will be a costly operation if to be considered for a global level such as for the GFS or even when it is just limited to the developing countries. Classification of the first set of PUs should not be considered merely on the basis of air photos although these must not be discarded completely.

Classification of the PUs set can cost-effectively be done with the help of the satellite imagery. AVHRR data can be used for homogeneous and least disturbed ecosystems, but Landsat TM and SPOT images are the sources of data that can confidently be recommended for difficult situations as they have been proven with higher level of reliability. AVHRR, TM and SPOT should however be keyed by: a sample of recent aerial photographs; field data collected during air and/or ground reconnaissance trips in representative sites with various ecological and social conditions; and existing maps or descriptive reports.

Results of the classification should be checked on the ground or from the air for validation. Ground or air trips for reconnaissance or validation should be undertaken with experts from relevant countries.

PUs classification should make use of the FAO prepared categorisation system based on land cover percentages. Woody vegetation, natural or planted, but assimilated as forest component according to FAO terms and definitions, should be grouped into crown cover classes as shown below, with the help of a template to be designed for the purpose. The land will therefore be classified as follows:

Figure 1. First level classification system

Interpretation results of this phase would serve as base for the first GFS exercise as well as for future inventories. No interpretation would therefore be needed for subsequent surveys.

b. Second phase

For developed countries, the second phase work has mostly been done, as plots at grid points have been classified into forest or non-forest lands, areas of forest types estimated and field plots selected and located on the ground. The proposal will therefore be not to duplicate work done by these countries but just to rely on the outputs from a part of the sample plots used for their surveys. The plots to be elected for the GFS will be selected in a way as close as possible to the sampling design adopted by the developing countries.

Under this phase, which corresponds to the first stage of the sampling, and based on the pre-stratification, 10% of grid samples of 30x30 km²⁽¹⁾ will be selected systematically among those classified as forestland within each country. These sample units will be denominated as Primary Units (PUs). The withdrawn sample units will be interpreted for stratifying the forest vegetation on the basis of the FAO vegetation classification. Considering the size of the primary units and the pre-stratification criteria, it is expected that for the majority of the countries, most of the land area will be included, which will lead to high population of primary units. Countries with substantial area covered by non-forest cover such as desert, snow, agriculture, urban zones, etc. will have lower number of PUs compared to their total land surface. But if the estimated forest land area by FRA1990 is taken as the land proportion that will be surveyed, the number of the PUs to be part of the sample will be around 4,758 or less than 4,282,200 km². For this task, Landsat TM and/or SPOT data will be required, together with information collected during field reconnaissance such as aerial photographs, thematic maps, reports, etc.

Stratification may be carried out visually or automatically. Land classification according to the vegetation cover, will allow recognising and delineating different types of forest formations whether they are forests (refer to definition of forest) or trees outside forest in lines (hedges, riverine vegetation, roadside trees, etc) or in wood-lots (homestead planting, relics of forests, etc.). Visual interpretation of the satellite data and delineation of the forest and other land cover types is preferably to be carried out on a computer screen so as to avoid additional work for digitising the interpretation results. The digitised polygons of forest and other land cover types will therefore constitute the reference-state of the natural vegetation to be extracted and updated during subsequent monitoring activities and for the purpose of producing new vegetation state and vegetation cover change overlays. The previous interpretation layers will be maintained as independent layer(s) in the GIS.

Interpretation checking in the field should be carried out where doubtful delineation or difficult situations have emerged during the process of stratification. More checking will be done during the fieldwork for data collection in field sample plots.

In case of use of aerial photography for forest classification into homogenous forest types, the suitable scale will be 1:40,000. At 60% forward overlap and 80% overlap between two successive bands, the size of the effective area in each photo will be around 27km². This gives a number of approximately 34 photos (33 stereo-models) to cover each primary unit or 125,950 to cover all the PUs.

The photo coverage may be produced on digital format using digital cameras from which orthophotos can be processed using ground control points. The digital orthophoto data is helpful for visual interpretation on computer screen but with the help of stereoscopically checked units under a mirror stereoscope.

For mountainous terrain and due to high relief and/or tilt displacements, interpretation should be carried out first under mirror stereoscope, then digitised using the orthophoto data.

Use of aerial photographs is more precise work compared to the other existing remote sensing data recorded on board of earth orbiting satellites. However, the cost of aerial photography and exploitation is higher and may be a limitation.

Recourse may be made to Landsat TM or eventually SPOT images, which when used with enough support data from the field may give reliable results. Depending on the location of each primary sample unit in comparison with the area covered by a satellite scene, whether is inside it or extending over two and up to 4 image frames, the number of scenes to be used will be at least 4,758. Overlaying the gridlines onto the image frames will check this number.

Figure 2. Flowchart of the vegetation classification system used by FAO

Details of the classification and definitions are given in annex 2, relative to vegetation classification designed and used by FAO for global forest resources assessments.

c. Third phase

The third step, which coincides with the second stage of sampling, will consist in sampling the forest vegetation within each stratified PU, in each country. A 1x1km dot grid will be overlaid over the PU and 10% of the dots falling within the forest types will be selected systematically. The location of the dots will coincide with that of the field plots. The forest, as defined in FAO classification, includes artificially or naturally established vegetation, open or closed, disturbed or undisturbed cover. It also includes isolated wood-lots such as riverine forest, homestead plantations, line plantations, e.g. roadsides, hedges, etc. Up to 90 field plots can be selected in a PU. The number will depend on how much forested these units are. Concentration of the plots within areas of 900km² will render the logistic easier and the project cost lower.

Figure 3. Schematic illustration of grid net and sample plot selection

Table 4. Summary table resulting from sample unit selection

4.3.2 Size and shape of field plots

Sample plots can be laid out using a relascope (point sampling) or with a fixed area (circular, square or rectangular). The size and shape of the field plots are governed by a number of factors namely stand density and topography of the survey. In open forest, as it occurs frequently in southern Africa where woodland formations are widespread by their limited stories and number of trees per hectare, 0.5 ha plot is the optimum size when only trees above 20cm DBH are recorded. Trees with smaller DBH than 20cm are recorded in sub-plots of 0.05ha. In the same region a smaller plot size produces reliable information in dense forest or closed canopy woodlands. For this type of vegetation density a 0.2 ha will be suitable for large tree diameters and 0.02ha for smaller DBHs. An average number of 30 trees per plot will be required if sufficiently low sampling error should be reached.

In closed rain and moist forests, strips or concentric plots could successfully be used. Large plot area will be for large DBH tree population and smaller plot sizes for smaller diameter trees from recruitment and regeneration. 0.1ha plot will allow having the number of observations that can yield reliable estimates for most trees and stand variables. Depending on the regrowth intensity, small plots can be 100 to 200m².

Depending on local experiences and forest and terrain conditions, sample plots may be circular or rectangular. However, for as large as 5000m² sample plot, the suitable sample form is the strip with reduced width but extended length. Strips allow to go better through changing vegetation types, render the survey easier when thick undergrowth and tall grass are present and the terrain is rugged. Lay out of equal areas within circles of 80m diameters or squares of 71m side is more time consuming. It also gives more work to the crews, especially to clear sights to “IN” trees in forward and backward movements, instead of progressing in a line without return from start to end of plots. As a proposal, the sample plot may have the following sizes and forms:

a. Dry Tropical Forest

b. Wet and Moist Tropical Forest

c. Temperate and Boreal Forests

a. Dry Tropical Forest

• Open canopy forest: 20mx250m for tree DBHs larger than 20cm with a sub-sample of 20mx25m for smaller tree DBHs.

Figure 4. Shape and dimensions of a sample plot for open canopy forest inventory

• Closed canopy forest: Plot size: 20mx100m for trees with DBHs larger than 20cm and a sub-sample of 20mx25m for smaller tree DBHs.

Figure 5. Shape and dimensions of a sample plot for closed canopy forest inventory

b. Wet and Moist Tropical Forest

• Open canopy forest: 20mx250m for trees with DBHs larger than 20cm and a sub-sample of 20mx25m for smaller tree DBHs. (See figure 4).

• Closed canopy forest: Plot size: 20mx100m for trees with DBHs larger than 20cm and a sub-sample of 20mx10m for smaller tree DBHs.

Figure 6. Shape and dimensions of a sample plot for closed canopy forest inventory in moist tropical region.

c. Temperate and Boreal Forests

- Large tree DBHs: plot size between 500 and 700m².

- Small tree DBHs: plot size ˜ 100m².

Figure 7. Schematic illustration of concentric circular plots

The plots may be:

• Concentric circles:



R: Radius of large plot for large tree DBHs
r : Small radius for small tree DBHs
x: Plot centre

Figure 8. Schematic illustration of concentric circular plots

Or

• Concentric squares

• as they may be point sampling using Relascope:

Figure 9. Sample plot layout and tree selection for measurement in point sampling

The plot size and shape used by the industrialised countries in the temperate and boreal regions should be used as they are. Countries from the same regions without permanent sample can inspire from the existing experiences and decide on the suitable plot sizes for them.

In developing countries, the proposed sampling intensities can be increased when it is required for improving precision of estimates by including additional sample plots in the cluster scheme. To each permanent plot, one or more plots may be added with not less than 250m interval between neighbouring plots.

Figure 10. Layout of cluster composed of 4 sample plots

Layout of clusters

Layout of the sample plots within a cluster when more than one plot is needed. P1 is permanent. P2, P3 and P4 are optional. Distance between plot borders should not be less than 250m.

4.3.3 Plot location and layout

Continuous monitoring of the forest resources and of the dynamics within ecosystems requires periodic observations of the population components through measurement and assessment of a selected set of variables. Second and subsequent observations on the same population necessitate permanent observation sites or sample plots. To secure revisiting the same areas within their boundaries as defined initially and re-measuring the entire population within these boundaries, the plots should be accurately located in the field and which geographic co-ordinates should accurately be measured and catalogued.

Properly registered maps in a geographic information system allow an accurate reading of the X and Y co-ordinates of plot centres. With the help of a differential GPS, the GIS generated co-ordinates may be correctly located to the nearest meter in the field and revisited whenever needed for re-measurement.

4.3.4 Variables

It is important to reconcile the need for a comprehensive data set at country level for management actions with the requirements for global assessment that looks at the causal factors (natural or human) causing positive or negative changes.

This work should be essentially a national undertaking that should primarily meet the national needs of information for policy formulation, general forestry activity planning and resources management. The gathered information when aggregated should also meet the needs of international organisations such as FAO for the global monitoring of the resources, the ecosystems and the depending local populations.

One of the major outputs from permanent plots is information on growth of forest stands and species, which is widely lacking in the tropical world and hence limiting seriously adequate management planning and forestry development in general. The design of a sampling based partially on permanent field plots should secure yielding precise growth models of the forest stands and of some important species selected socially, environmentally and economically.

The survey should extend to tree species as well as to stand characteristics, which can provide a comprehensive understanding of the economical, social and ecological values of the tree species and forests. It should extend to site parameters and social environment within the forest. As far as forest is concerned, the various functions should be assessed for protection, production and aesthetic.

Protection includes soil and water conservation, biodiversity preservation (faunal and floral components), carbon sequestration, etc. Production refers to timber, other wood and non-wood products.

Having these considerations in mind, the following set of variables may be included in country surveys for their measurement in the field. A team of national experts should refine the final set, in collaboration with FAO as leading agency for this work and main user of the results for global resources assessment. The survey should cover the attributes that describe aspects of:

a. Economy (forest production)

b. Environment (protection, recreation, etc.)

c. Social environment in and around the forest

a. Forest production

- Forest area

- Forest structure (species composition, layers and density, etc.)

- Species names (local, commercial if any and scientific)

- Woody biomass

- Timber Volume

- Other non-wood products (type and size)

- Stand height

- Diameter

- Increment (MAI)

- Growth (drain, removals, mortality, increment, etc)

- Timber and other non-timber product quality

- Forest management system indicating cutting and other exploitation systems (concession systems: timber, wildlife, other wood and non-wood products)

- Origin of the forest (planted, naturally regenerated, mixed, coppice forest)

- Shape and size of occurrence (riparian, roadside, hedges, homestead plantation, wood-lots, fragmented, contiguous blocks of forests)

- Natural regeneration

- Mixture type of the forest (tree by tree or in groups)

- Development stage

- Stand age when relevant

- Crown closure (percentage of plot area covered by the vertical projection of the tree crowns estimated at 10% increment)

- Forest edges surrounding (built up area, agricultural land, pasture, water, rocks, others)

b. Environmental aspects

- Ecological zone (e.g. wet, moist, dry, very dry, arid, desertic, etc)

- Forest ecosystems (e.g. mountains, lowland, gallery, swamp, mangroves, etc.)

- Vegetation community (species composition, frequency of occurrence, etc)

- Health of forest

- Forest degradation (removal of selected species, elimination by fire, change forest from closed canopy to open or fragmented forest cover, etc.)

- Deforestation (clearing of forest cover for the aim of transforming the forest land into temporary cultivation or permanent crop land)

- Type of spatial pattern of the forest area (scattered gaps of non-forest land, oblong gaps of non-forest land, island of forest, belt of forest, none)

- Forest fires and other damages

- Endemic species (from existing lists)

- Endangered and threatened species (from existing lists)

- Soil

- Water resources (erosion)

- Wildlife and wildlife habitat

- Aspect of relief terrain (N, NW, NE, W, SW, SE, S and E), slope, elevation.

- Protection status (national park, game and/or gene reserve, official hunting area, open hunting and other resources collection area, etc)

- Land and resources tenure (state, private, community, etc)

c. Social environment

- Land-use type (assigned function to the land: wood production, grazing, agriculture, environmental protection, other infrastructure scheme)

- Type of agriculture (permanent farming system, shifting cultivation, agro-forestry)

- Use of bush fire (land clearing, grazing, hunting, household and people’s protection)

- Settlements (permanent, temporary, construction in concrete or with local material)

- Number of house holds

- Main activities of local population (agricultural activities, trading, services, etc)

- List of non-wood products extracted from the surrounding forest

- Forest-man relation by eco-region through a rapid rural appraisal (questionnaire for RRA to be designed with the help of a sociologist).

General variables: date of assessment, province, district and locality, and names of field crew members.

Plot variables: Forest name, compartment, sub-compartment number, plot number, X and Y co-ordinates.

(Refer to variable template in annex 3)

4.3.5 Estimates and Sampling Error

The PU will have variable sizes according to where they are located. Along the international borders or the cost-lines, PU’s area will be smaller than that for those located fully inside the country. Taking a single forest type or stratum, the size will vary greatly from PU to another. In some PUs, there will be forest types which will be extensive and others very small or non-existent. Since field plot selection will be done systematically and shapes and occurrences of forest type polygons are randomly distributed, plot allocation to different strata within each PU and in the country as whole will be proportional to size.

Analysing each primary unit separately will not produce reliable results for every forest type. It will do it if the sampling will be performed with unequal probability and strata are allocated not less than a certain minimum number of plots fixed in the design. Considering this limitation, processing will be done per strata, and the population totals will be weighted. If stand volume is considered, the processing will be performed as follows:

• Tree Volume:

V_t = (p x DBH2) x Ht x Fc

4

where DBH = Tree diameter measured at 1.30m from ground level

H_t = Total height of the tree

F = Form factor for total volume correction

• Volume in plot « p » in forest type« i »:

V_pi = SVt_pi

where V_pi = timber volume per plot in forest type « i »

Vt_pi = timber volume of a tree in plot « p » in forest type « i »

• Mean volume per plot in forest type « i »: V_m = SV_pi/N_i

where Ni = number of plots in forest type « i »

• Mean volume per hectare: V_h = ( V_m)_*10,000

A_p

where A_p= Area of a plot

• Variance in forest type « i »: S² = (?(V_pi)² - (?V_pi)²/N_i)

(N_i-1)

Standard error of the mean: S_v = S/vNi

Coefficient of Variation: CV = (S)*100

Vm

Sampling error SE = Sv * t

then SE² = T² * Sv² = t² * S²/N_i and

The number of sample plots for a fixed sampling error is given by:

Ni = (t² * S²)/SE²

In percent, these quantities can be calculated using the CV as fallows:

Sampling error : SE% = t * CV/vN_i

N_i = t² * (CV)²/ (SE%)²

Combined sampling error for all forest types in a given state or province:

__________________________________

E = 100 * v(A₁²*SE₁² + A₂²*SE₂² +........+ A_k²* SE_k²)

(V₁ + V₂ + .... + V_k)

where A₁....A_k = Area in hectare in forest types 1 to k

SE₁...SE_k = Sampling error computed for forest types 1 to k

V₁....V_k = Total volume estimate for forest type 1 to k

Overall sampling error for the country:

______________________________________

E = 100 * v(A_d1²*SE_d1² + A_d2²*SE_d2² +........+ A_dk²* SE_dk²)

(V_d1 + V_d2 + .... + V_dk)

where: A₁....A_k = Area in hectare in district or province d1 to dk

SE₁...SE_k = Sampling error computed in district or province d1 to dk

V₁....V_k = Total volume estimate in district or province d1 to dk

4.3.6 Monitoring

At the following appointments for forest resources assessment, the forest survey will rely essentially on field data from the permanent field plots without discarding remote sensing that will continue to be used for change assessment. Field plots will be revisited at fixed intervals (5 or 10 years) for measuring and recording the current state of the forests, sites and social environments. The new set of data form the sample plots when compared with the historical information will allow deducing and assessing various changes such as deforestation, degradation of forest cover due to uncontrolled burning, disorganised resources harvesting, or to other natural factors, recovery of vegetation, forest growth, etc.

Information on quantified changes from the fixed field plots and remote sensing would be used for modelling and prediction of changes in other sites. Depending on the degree of correlation between the variables, a number of models may be established for some estimates.

Correlation Additional photo measurements

(n₁ plots) (n₂ photo plots)

Field plots

Photo (image plots)

4.3.6.1 Modelling

a. Using field and radiometric data

Various models may be established for predicting forest area changes and many other forest variables. Taking forest cover change as one of the most important parameters that gives indications on trends and tendencies of the forest, models may be established using a set of interdependent variables measured in the field and on satellite images or aerial photographs.

Derived forest cover changes from field plot data may be correlated to a set of spectral reflectance in the Infrared domain of the same areas measured on satellite data within the field plots. In this correlation, plot areas with varying levels of deforestation may be correlated to differences of reflectance between two successive images taken at intervals as close as possible to the field measurement dates. Provided the two variables are well correlated, this approach would lead to reduction of field plots in the future and consequently of fieldwork and survey cost while the precision may well be improved. Infrared may be used separately for this relationship or in composition with reflectance in the red and green bands. As an example, the following data is given below:

Table 5. Example of variables that may be used for forest change determination

Figure 11. Example of a model established between Infra Red reflectance and actual forest cover change

b. Using field, climatic and social data

Changes in forest cover caused by man are often closely correlated with some climatic and social variables. Under certain social environment, the lack or abundance of rainfall influences the behaviour of people in respect to deforestation for both traditional and large-scale farming. This tends to decrease with the decrease of the annual precipitation amount as crops become more hazardous under drier conditions. High and well distributed precipitation over the year facilitates the recovery of vegetation under shifting cultivation system as growth rates of vegetation are higher for longer growing periods. Vegetation growth rates are also influenced by air temperature. Speed of transformation of the forests into other land cover/Land uses under given social and economic conditions depends on the growth of the depending local populations.

There are other factors, which influence the rates of deforestation. The legal status of the forest (protected areas), land and resources tenure, physical characteristics of the forest areas (mountainous areas are often less frequented by people than lowlands), and country development policies. Except for policies, these factors have localised effects compared to the above mentioned ones.

Change in forest cover is likely to be closely correlated to population growth, annual precipitation, length of wet seasons and to air temperature. Population growth may be replaced by population density in the locality, district and province, as shifting cultivation does not depend on population growth but also on soil fertility.

Putting forest cover change as function of these variables will require specific study to find how the change in annual precipitation, length of wet seasons, temperature and population growth affect the change in forest cover in a given region or a country. It is also to find the model that best defines the relationship and to define the process of building the model. As there is more than one independent variable, the most suitable regression equation that defines the relationship should be determined.

Putting: Y: Forest cover change in hectares

X1: Annual precipitation

X2: Length of wet season

X3: Mean air temperature

X4: Population growth

The equation could be a polynomial of first order, second or even third order. The following is a polynomial model of first order that can be contemplated:

Y = a₁(X₁*X₂^-1) + a₂X₃ + a₃X₄

4.3.7 Option 2

With the aim of conducting the survey at lower cost and with less logistical implications, while keeping the sampling error of estimates at acceptable levels, it is judged more practical to follow the two stage sampling approach where field samples are concentrated in selected areas called primary units. Under this option, the approach will consist of subdividing the country into blocks of areas of certain sizes (depending on the country). Out of those blocks a sample of 10% will be withdrawn systematically. A 1x1km grid will overlay each block. A fixed sample size for each country, composed of 1x1km sample units, will be selected from each. The sample areas will be interpreted using aerial photography for the classification of the vegetation into major cover classes. A frequency histogram of the forest cover categories will be prepared for the country and intensity of field sampling will be defined for each vegetation cover.

4.3.7.1 Constraints for a Common Sampling Design

The major constraints in designing a detailed methodology common for all countries, is that the area of the blocks will depend on a number of aspects such as:

• The size of the country itself. There are countries with very large areas such as Brazil, China, India, Argentina, Democratic Republic of the Congo, Algeria, Sudan, Indonesia, etc and many others with tiny areas.

• The topographic maps exist at various scales (1/50,000, 1/100,000, 1/250,000, etc) and projections (UTM, Mercator, Lambert, etc).

• The satellite Landsat TM and SPOT grids, although common for all countries, are not convenient to small size ones, which are covered with a few number of scenes in statistical point of view.

• A common sampling intensity for all countries will not be practical, as homogeneity of population (forest variables) to be inventoried varies from a country to another. A small country with large variety of ecosystems and social environments would require higher sampling intensity than an immense country with less human interference on forest cover state and lesser but large ecosystems.

• Accessibility to remote areas could not be possible in many countries for lack of infrastructure such as roads, rivers, etc. Reaching such areas from the air will render the forest inventory a very costly operation.

Thus, although the statistical principles will be the common foundation of the design for all countries, the methods will vary somewhat from country to country in function of the extents and heterogeneity of the forest resources, the access to all parts of occurrence of these resources, the available cartographic framework, etc.

4.3.7.2 Sampling design

The design will be based on the two-phase sampling for stratification where remote sensing will be used for stratification of the country land surface into various forest cover classes.

a. Phase 1: Creation and Selection of Primary Units

The first phase will consist of dividing each country into blocks of areas of certain sizes (blocks of unequal sizes depending on their location, either inside or at the edge of the land of the country). The subdivision of the country into blocks will depend on the block size that will be retained for the country. It could be the size of a topographic map, or sequential areas of UTM grid, or of satellite scenes or a quarter of them. It could also correspond to blocks of a grid that will be designed for the purpose. The area of a block could vary from few hundred km² to more than thousand km².

Out of the total number of blocks or primary units a number of PUs will be selected. The number will depend on the allowable sampling error of estimates, but also on the available timeframe and funds. Country-wise, the sampling will consist of units of unequal sizes and unequal sampling intensities. This will not hamper computation of the overall sampling error of estimates from weighted results, but variances will be highly variable from country to country, depending on the sample size and the population homogeneity.

The sampling intensity should not however descend below the threshold of 10% if estimates will be measured at acceptable levels of reliability. When small countries are concerned, the subdivision into N primary units should allow selection of not less than 10% PUs (n). In specific situation sampling intensity n may be increased for these countries to allow the statistically acceptable minimum number of sample units.

The leading objective for selecting primary units where secondary units will be selected for further analysis is to lower the cost of the survey and to minimise the logistical implications. Costs and logistics can be controlled through a well-organised survey programme for each country independently according to its specificity in terms of human resources, financial capacities, size of the survey area, and accessibility.

The following is a model of creating PUs from the total land area of a country using a grid designed for the purpose that can be topographic maps, UTM gridlines, satellite scene frames, etc. The size of a single PU will vary from country to country depending on the considerations quoted above. From that total population of M PUs a sample of m PUs is selected at 10% intensity. As it can be seen there are PUs with full size as they are fully inside the country boundaries, but there are others with reduced area due to their location along the borders. In this case the sampling within the PUs will be proportional to their sizes.

Figure 12. Schematic representation of subdivision of a country into blocks and selection of a sample of PUs

b. Phase 2: Selection of Secondary Units and Interpretation

Under this phase, each PU will be overlaid by gridlines with squares of 1x1km side. From the created Ni 1km² area units a number of (ni) sample units or secondary units (SUs) will be selected systematically. The ni SUs will depend on the sampling intensity to be applied for the country at that level. 10% sampling intensity is recommendable. Smaller sampling intensity can be accepted for large countries or countries judged with homogenous vegetation cover.

Aerial photography at an average scale of 1:40,000 will be required for classifying the total land area of the SUs according to the actual forest cover. Satellite imagery remains an alternative but less accuracy in identifying and delineating forest cover classes. Extensive fieldwork and use of support documents such as existing thematic maps and aerial photographs will be necessary. The FAO forest cover classes of <0% (other land), 1-10% (other wooded land), >10-40% (open forest) and >40-100% (closed forest) and shrub (height: 0.5 to 5m) will be the basis of the vegetation classification. Forest cover can be contiguous block or patchy vegetation in fragmented wood-lots, regrowth of natural forest in shifting cultivation system, line vegetation (riverine and hedge planting). Vegetation classification developed by FAO for forest resources assessment should be used. It may need additional refinement to suit the objectives and methodology of the GFS.

Figure 13. Representation of created SU population within a PU and selection of a sample

Survey for aerial photography, has to be planned for the selected areas. Each area should be completely covered by overlaps of successive photos. No gaps should appear between successive photo strips.

Aerial photographs or satellite images will be interpreted visually on computer screen. In addition to hard copies, air photos should therefore be supplied in digital format in CD-ROM and preferably geometrically corrected to the orthophoto level using ground control points. Rugged terrain provokes extensive relief displacement and tilts during photo taking which are causes of considerable distortions. Direct interpretation of aerial photographs on the screen should be assisted by continuous consultation of stereo-models under mirror stereoscope.

Inventory planning and correct location of areas that will be surveyed for aerial photography require a scrupulous planning using GIS and the national topographic maps. Combination of remote sensing (air photos and/or satellite imagery) implies existence at national level of policy framework, infrastructure and facilities, and a trained national team. It is common that international technical assistance is needed in most phases of forest resources inventories involving a number of areas such as forest inventory, statistics, remote sensing, GIS, etc.

c. Phase 3: Determination and Selection of Field Plots

Areas of each forest cover class will be measured with dense dot grid to minimise sampling errors or, best, using GIS facilities as interpretation results are directly digitised. The delineation should therefore be geo-referenced using ground control points extracted from the existing topographic maps. Resulting area statistics will be processed for the defined levels of the survey (national, state, provincial or district). Results will be entered in a table of the following model:

For each of the established levels, a frequency histogram for the forest cover categories will be prepared on the basis of estimated areas as shown in the example of histogram below:

Frequency histogram

The cover class size, the economical, social and environmental forest cover functions according to national policies, and the vegetation homogeneity will be the basis to decide a field sampling intensity for each vegetation cover class.

The forest survey results will be essentially for policy making, forestry and biodiversity management planning at a large scale in a country. Precision will therefore be low when the results are applied to small forest areas. It increases with the increase of the area to which the results are applied. The sampling intensity should not be thought to produce accurate results at forest blocks level.

The sub-total of each primary unit will allow determination of the number of sample plots to be allocated to each of the forest cover classes to be sampled and distributed within the secondary units. By experience, for a countrywide survey, a sampling intensity of 1 field plot per 100 hectares for the forest cover classes 3 and 4 (cc = >10-40%, cc = > 40%) produces estimates with acceptable level of sampling error. A field plot per 1000 ha for the forest cover class 2 may be considered.

A grid of dots with 500m interval or 1 dot per 25ha should be used (the size of a SU is equal to 100ha). The number of dots falling inside each forest cover class should be counted separately on all SUs of a PU starting from SU 1 to n. This number will not deviate greatly from the one calculated from the measured areas. As for the selection of the field plots locations, 1 dot every 4 ones will be selected from dots falling within Cl3 and Cl4 starting from SU 1 to n. For the Cl2, selection of field plot location will be done at 1 dot every 40. Depending on the size of the Cl2, Cl3 and Cl4 in each SU, the number of field plots will oscillate between 1 and 3, which makes the number of plots per PU variable according to their size and size of the inventory forest cover classes.

Figure 14. Schema representing overlaying a 1km² dot grid on top of interpretation overlay.

SU = 1km² (100 ha)

1 SU covered by 4 dots

1 dot represents 25ha

4.3.7.3 Field Plot Size and Shape

Field plot sizes and shapes described above for dense and open vegetation and for large DBHs and small DBHs could be applied. However, these technical issues may be discussed with the national teams and the suitable plot extent and form will be used.

4.3.7.4 Variables

Tree, forest, soil and other site variables as well as local social environment to be measured within the field plots should be clearly defined. The above indicated variables will be completed and later refined together with the national implementing teams as to focus on issues of interest to the country and to FAO and which reflect best the forestry and social environmental particularities of the country.

¹ The 30x30km correspond roughly to a quarter of a SPOT scene.