You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Proceedings of the Symposium on Global and Environmental Monitoring

Simo Poso
University of Helsinki
Department of forest Mensuration and Management
Unioninkatu 40 B, 00170 Helsinki, Finland
ISPRS Commission VII
The principles of two-phase sampling procedures for forest resource assessments are described. In the
first phase, data from auxiliary sources (e.g. maps, satellite imagery, aerial photographs, old forest
inventory data, forest growth models) are used singly or in combination to form strata. In the second
phase, actual ground measurement data are then transferred to the first phase units; each plot belong
ing to the same stratum obtains egual ground truth estimates.
The problem of using many data sources is that the number of strata tends to grow too high causing
difficulties in getting a sufficient number of ground truth measurements. The solution introduced in
the paper is to apply many separate stratifications based on different data sources which produces many
estimates for first-phase sample units. The final estimates may be based on weighting using of inverse
of error variance or on a specific expert system.
KEY WORDS: Forest inventory, Remote sensing, Two-phase sampling
There is a large number of alternative approach
es that can be adopted for forest resource as
sessments. These approaches can be roughly di
vided into two categories: (1) to mapping for
area classification and (2) to sampling proce
dures. The differentiation of the above catego
ries is not always unambiguous as many combina
tions may exist.
In mapping approaches the whole area is divided
into a certain number of different categories
such as forests, woodlands, other land uses.
Different categories, i.e. classes, are diffe
rentiated by drawing borderlines on maps or oth
er base material, such as satellite imageries
and aerial photographs. If remote sensing is
used distortions should be known and elimina
ted. The areas of different classes can then be
measured on the base material.
Sampling approaches involve locating a number of
sampling units, usually circular plots or relas-
cope points, in a systematic pattern in the in
ventory area. Data are measured or estimated for
each unit and the inventory results, including
area distributions and mean values as well as
reliability estimates, are calculated on the ba
sis of the samples.
The principal advantages and disadvantages of
mapping approach may be listed as follows:
+ traditional, easy to understand
+ suitable for overall viewing
- borderlines between classes are often ambig
uous and subjective
- variation within a class is often large
- classes are often difficult to measure accu
rately in the field (estimates often based
largely on ocular estimations)
- monitoring of changes is difficult if based
on more or less subjective area delineation
Generally, it is recommended to use a sampling
technigue for national forest inventories in
order to get detailed and non-biased informa
tion for planning purposes.
The purpose of this paper is to illustrate an
application of sampling technigue together with
the use of remote sensing in order to avoid the
disadvantages associated with mapping. The tech
nigue demonstrated is based on two-phase samp
ling. In the first phase, auxiliary data is ob
tained from many kinds of sources, such as maps,
satellite imagery, aerial photographs. In the
second phase, data are usually measured in the
field. The sample for the first-phase data is an
intensive one and the sample for second-phase
data is a sub-sample of the first-phase sample.
The final results are the better the higher is
the correlation between the first- and second-
phase data.
A combination of remote sensing and ground truth