Previous PageTable Of ContentsNext Page

Value of low-intensity field sampling
in national forest inventories

T. Thuresson

Tomas Thuresson is
Forest Analyst at the
National Board of Forestry,
Jönköping, Sweden.

On-the-ground forest inventory with relatively low sampling intensity can provide needed information for country-level decision-making at acceptable cost.

Forests are best assessed from the ground


One of the most important objectives of forest resources assessments is to support decision-making (although data on forest resources are often collected for other reasons). Reliable information on the status and trends of forest resources helps give decision-makers the perspective necessary for orienting forestry policies and programmes.

Yet the available information about the forest resources of the world is much more rudimentary than decision-makers may realize. Existing data sets are often outdated and may include severe systematic errors. In some areas, for example, both growth figures and harvesting rates have been shown to be largely underestimated (c.f. Thuresson, Drakenberg and Ter-Gazaryan, 1999). In addition, cutting activities (legal or illegal) have in many areas reduced the validity of old data sets. Conclusions about forest development are thus often based on incorrect and/or insufficient information, leading to faulty decisions.

To improve knowledge about forests in countries where the existing forest database is of poor quality, a ground-based, low-intensity systematic national forest inventory, supported if possible by aerial or satellite photographs, may be effective in providing an overall qualitative assessment at acceptable cost. A forest inventory with relatively low sampling intensity, high accuracy in measurements and known random errors in the estimates can give estimates for the most important and sought-after variables on the country level, although the level of precision does not permit breakdown into smaller areas or classes. FAO is currently developing a project model for supporting national forest assessments along these lines. This article examines the reliability and benefits of such low-intensity sampling.


In the Global Forest Resources Assessment 2000 (FRA 2000), and also in FRA 1990, the lack of basic on-the-ground forest inventories was to some extent compensated by analyses of satellite remote- sensing data. The remote-sensing survey may have gained valuable information for regional and global analyses of forest area and forest area change. However, for many other similarly important variables, such as above-ground woody biomass, growth and yield, merchantable wood, non-wood forest products and biodiversity indices, satellite data are weak or useless without sample-based forest inventory plots. Furthermore, for management variables and for evaluating the dynamics of natural or human impacts on the forest, remote-sensing data are of little or no value unless combined with a large enough sample of forest inventory plots (see article by Tomppo and Czaplewski in this issue).


To estimate the precision of any kind of inventory beforehand is a difficult task. The information needs of decision-makers, the forest types and other conditions, and the variability within the areas of interest on all levels (subplot, plot, landscape and region) vary widely among and within countries. In this article, two examples have been chosen to illustrate different possible outcomes from different parts of the world and with different intensities of inventory: the Swedish national forest survey (Li and Ranneby, 1992) and a low-intensity sampling carried out in Costa Rica (Kleinn et al., 2001).

Swedish national forest survey

The Swedish national forest survey 1983-1987 was designed as a systematic cluster sampling, where the distance between tracts is regular and the same within counties. The clusters were circular sample plots (5 to 10 m in radius) along the sides of square tracts. The length of the tract sides varied from 400 to 1 800 m, and the distance between tracts ranged from 5 to 22.5 km, depending on the region of the country. The tract system was designed for a team of three to four persons working on one tract per day (including travels).

Within each tract there were on average four to five volume plots, nine to ten stump plots and also regeneration plots. On all plots, land use classes were registered. The survey inventoried in total about 2 250 tracts, 20 000 volume plots (10 000 on forest land) and 20 000 stump plots (on forest land) per year.

A rough method was used to simulate the coefficient of variation (relative standard error) in percentage for a single-year inventory with different numbers of tracts (50, 100, 250 and 500) spread systematically over the country, based on figures presented by Li and Ranneby (1992). The simulation was performed for the following variables: area of forest land (ha); total standing volume (m3/ha over bark) on forest land; harvested area (ha of clear-cuttings) on forest land; and total harvested volume (m3/ha over bark) on forest land.

The results indicate that total area and standing volume can be estimated with fairly high precision even with relatively few tracts (Table 1). However, clear-cut area and harvested volume were estimated with low precision.

TABLE 1. Simulated relative standard errors of a single inventory, given different numbers of tracts, based on the design of the Swedish national forest survey 1983-1987

Number of tracts

Coefficient of variation in percent of estimate

Total forest area

Clear-cut area per year on forest land

Total standing volume

 Total volume harvested on forest land





















The Swedish national forest survey provides an example of the different possible results with different intensities of inventory - shown here, a spruce stand in Sweden


Costa Rica pilot inventory

The recent pilot forest inventory in Costa Rica (Kleinn et al., 2001) was designed as a systematic grid sampling using both aerial photographs for interpretation of land use classes and clusters of on-the-ground measurements. A total of 159 photographs, with on-the-ground sides of about 2.7 to 4.5 km, representing two-thirds of Costa Rica's area, were available. Forty locations close to the photographs were sampled, and 34 square forest inventory tracts were measured. These consist of a 1 x 1 km2 reference unit with an inner tract of 500 x 500 m2. On the sides of this tract 150 x 20 m2 subplots were established. Within these subplots all trees greater than 30 cm diameter at breast height (DBH) were measured. In addition, in three smaller nested plot levels (20 x 10 m2 plots and circular plots of 3.99 and 1.26 m radius) smaller trees and saplings were measured and counted. The time spent in the field varied from 7 to 12 full working days per tract (excluding preparation and office work). The fieldwork included not only tree measurements, but also interviews with landowners on topics such as land uses and needs from the forest.

Four variables from the Costa Rican inventory were studied: total forest area; tree resources outside the forest, which is all land outside forest with the capacity to hold trees; commercial volume (>30 cm DBH) in forest; and commercial volume (>30 cm DBH) in tree resources outside the forest. The first two variables were estimated from the 159 photographs, and the last two were estimated based on the 34 sample inventory tracts. The coefficient of variation in percentage was simulated for 50, 100, 250 and 500 photographs (for the first two variables) or tracts (for the last two variables), systematically spread over the country.

The results show that the coefficients of variation for aerial photo interpretations of forest area and tree resources outside the forest were fairly low with a few hundred photos interpreted (Table 2). The commercial volume estimates show higher coefficients of variation, but considering that only trees above 30 cm diameter were included in the estimate, the figures seem fairly good, depending of course on the precision required.

TABLE 2. Simulated relative standard errors, given different numbers of photographs (first two columns) or tracts (last two columns), based on a single inventory in Costa Rica

Number of photos or tracts

 Coefficient of variation in percent of estimate

Total forest area

Tree resources outside the forest

Commercial volume (>30 cm DBH),  forest

Commercial volume (>30 cm DBH),  tree resources outside the forest






















A variable presented by FRA 2000 which has historically generated great interest is the forest cover change estimate. The Costa Rican inventory and the Swedish national forest inventory did not estimate forest cover change. However, the Swedish inventory estimated the area of clear-cuttings. The simulated coefficients of variation from direct plot measurements on clear-cut area are high, 48 percent with 50 tracts and 15.2 percent with 500 tracts. Thus in Sweden, where the clear-cut area is usually about 0.8 percent of the forest area (a deforestation rate that could be typical of many areas), the absolute confidence limits are 0 to 1.6 percent with 50 tracts and 0.6 to 1 percent with 500 tracts. With 50 tracts the results would offer little value for country decision-makers. However, with 500 tracts the results could be informative for many decision-makers and much better than the information available today in many countries.


In the Costa Rican inventory the costs were estimated to be about US$2 000 per tract. This is a fairly high cost, but it entailed not only measurement of forest variables, but also measurement of different land use areas and interviews with the local population. Moreover, this estimate included indirect costs such as planning and travel, which were proportionately high since little more than 30 tracts were inventoried. An inventory with more tracts would also probably reduce the labour time per tract, with the training factor and the fixed cost being lower per tract. Thus the cost per tract could probably be reduced for similar inventories in the future.

If it is assumed that the cost per tract could be halved with efficiency gains in larger and better-planned inventories, the cost of a 500-tract inventory would be about US$500 000. Is that too high a cost? Perhaps the answer is to turn the question around: what are the costs of decisions based on incomplete and incorrect information?


International processes are becoming more and more demanding when it comes to data variables and the quality of the data (see articles by Braatz and Prins in this issue). Even if there are many practical difficulties with forest inventories, it is obvious that most of the important variables must be measured, counted or observed in the forest.

Satellite-based "inventories" often attract interest because of their low cost and "high tech". However, the high technology is superfluous from the information point of view if the infrastructure is missing. Moreover, the images cannot provide all the information wanted.

Although the examples in this article are few and simplified, some conclusions can be made. Different variables call for different inventory methods. However, most often the chosen inventory design is a compromise between different solutions. Similarly, the inventories illustrated in this article may not produce the best possible precision for the variables presented, but they provide fairly low standard errors for some of the more important variables. The simulations of standard errors from the Swedish and Costa Rican examples show that total forest area and standing volume can be estimated with fairly good precision even with relatively few inventory tracts. Thus low-intensity forest inventories can gain information of use for both local and global decision-makers, with or without additional information from remote sensing data.

The costs of such low-intensity forest inventories might seem high for many of the developing countries, but in an international perspective and from the decision-makers' point of view the benefits might be worth the cost. However, decision-makers will have to be convinced that data from such inventories are better than other available information if money for national inventories is to be raised. Here, the international community probably holds the key to future information about forest resources.

Even a low-precision inventory might, as in the Armenian case (see Box opposite), give information that changes the whole perspective on a country's forest resources. If decision-makers do not have hard data, decisions will be based on other available information, "guesstimates" or just pure assumptions. But when decision-makers have more data available and see the strength of basing the decisions on hard data, they will ask for more - other variables, higher resolution and time series of data. In this way, the low-intensity forest inventory might lead to further inventory work with capacity building and better information as result.

It is important to note that ideally both levels of inventory are needed: systematic, unbiased sample-based inventory with known precision for strategic use (decision-making), and full-cover inventory for operational use (management planning).

To conclude, low-intensity national forest inventories may not fulfil all information demands on the national or global level from a forest inventory point of view. However, they can provide fairly advanced data for use in decision-making, and the information from such inventories may promote national forest policy work and new inventories that can further improve the national professional capacity.

A crew member of Costa Rica's national forest inventory measures the length of a permanent sample plot in a young broadleaved forest



Sampling results promote change in forest policy: the case of Armenia

Armenian forest inventory team navigating to find the sample plot


Sometimes local administrations have data compiled from stand-based subjective forest inventories, and the data are often old. In Armenia, for example, the database available in 1998 still contained old former Soviet Union inventory results. A strategic objective inventory of the Armenian forests, financed by Sida (the Swedish international development agency), was carried out to obtain better decision support for the ongoing forest policy development in the country (Thuresson, Drakenberg and Ter-Gazaryan, 1999). In this low-intensity forest inventory, 270 subcompartments, stratified by volume and age from the old inventory data, were inventoried (with an average of ten circular plots of 10 m radius in each compartment). On average three persons inventoried one subcompartment per day, excluding preparation work.

The results were surprising for decision-makers. For example, growth was measured to be 2.86 ± 0.17 m3 per hectare per year, or twice the previous official figure of 1.4 m3 per hectare per year. Stump measurements indicated that cuttings totalled about 600 000 m3 per hectare per year (on 215 000 ha), about six times the official maximum allowable cut of 100 000 m3 per hectare per year.

The inventory results became one incentive for change in Armenian forest policy. A seminar based on the new information was held for Armenia's policy-makers. Many found it difficult to accept the large differences between the old State figures and the results of the stratified random sampling. However, all participants concluded that illegal cuttings were a big problem and that cutting was not generally carried out in a sustainable manner. The forest policy had to be updated, and for this process further local inventory processes were needed.

In this case the objective inventory served as a wake-up call, and it also induced further inventory activities by the local authorities - with the eventual goal of an operational stand-based full-cover inventory.

Calipering the diameter of a tree on a sample plot in the Armenian inventory; measurements indicated that tree growth was twice the previous official figure


Previous PageTop Of PageNext Page