The GFS is carried out on a systematic sample of sites over the land area of a country (Figure 2). Systematic sampling is the normal approach to continuous national forest inventories, being (a) a statistically robust method, and (b) a good practical solution for locating field samples based e.g. on national survey maps. Other approaches, such as full-cover stratification and ground sampling within strata, may in some cases be statistically more efficent, but is more difficult to maintain over time as land use changes may alter the sampling frames.
Figure 2. Illustration of a country-wide systematic sample.
The size of the country-wide sample may vary depending on the size and variation of forest cover over the country. If no further (second phase) stratification is wanted (i.e. all first phase samples are visited in the field), the systematic sample should be kept small at, say, 50-500 samples. If stratification is wanted to make the field work statistically more efficient, a larger sample, say 500-5000, can be made and analysed, e.g. using remote sensing. Based on the results, the samples can be stratified, for example based on the proportion of forest, and a smaller second phase sample for field work can be established.
The field site or tract is the fundamental unit of the Global Forest Survey. Almost all information and data compiled refer to a tract, or to sub-units of the tract. The default tract is a 1x1 km square.
The tract is first divided into land use units or stands, following the standard FAO Forestry classification for this purpose, simplified into five classes (Forest, Other wooded land, Other land, Inland water and Sea). The number of stands within the site can thus vary from one to many, depending on the landscape configuration.
Note that the forestry term stand is used to refer to areas with forest, other wooded land and areas with trees outside forests. The other types of land use units are identified, but no survey work is done within these. Each stand, however, is described through a number of variables as further described below. See also Figure 3.
Stands are distinguished not only by the land use classification. If, for example, a forest area on the site is clearly composed of different units (due to varying stand structure, ownership, soils, or variatons in any of the variables listed below), it may be split into several stands which are then described separately.
Similarly, if several stands of the same land use type are identified at different locations within the site, the can be lumped together and described as one stand, to make the field work more efficient.
Usually, the tract is first interpreted using remote sensing, to arrive at a preliminary delineation of stands. The stand delineation is then reviewed and updated in the field and the stands are identified. Each stand is then described in situ using the variable structure outlined below. (Figure 4)
Figure 3. Illustration of a tract with 8 delineated and identified stands. F=Forest, OWL=Other wooded land, OL=Other land. Note that two forest stands adjoin (F1 and F2) - because they apparently are different in some aspects and therefore are described separately. The same applies to the two other land areas (OL6 and OL8) which are split, perhaps because one of the units contains trees outside forests and the other one not.
Figure 4. Process for delineating and describing stands within a tract.
The tracts are too large to make it feasible (i.e. statistically efficient) to measure all trees and to register details about stands and stand management. However, such information is essential for any analysis of the forestry sector. Therefore, the tract is subsampled with a number of plots on which all trees are measured. The standard global design includes 4 strip plots (Figure 5), but this design may vary with specific national conditions or design of existing national forest inventories. The plots are crossed (overlaid) with the stand delineation to create a number of sub-plots (Figure 5). All tree data and stand/stand management data are registered to these sub-plots.
The most efficient way to identify sub-plots is likely to do it as the plots are surveyed. This means that stand delineation control, sub-plot identification and tree measurements are done at the same time.
Figure 5. Illustration of 4 plots laid out within a tract, in this case shaping 8 sub-plots for which detailed stand characteristics and tree-wise data are registered. Note that sub-plots 6 and 8 will be surveyed only if there are trees outside forests on the sub-plots.
