The stratification of the first phase units can 
be repeated using different kinds of auxiliary 
data for as many times as required. Each stra 
tification produces a new set of ground vari 
able estimates for each first phase sample plot 
The result is illustrated in Figure 2. 
1,2 
1,2 
1,2 
1,- 
1,- 
3,4 
3,4 
3,4 
3,4 
3,- 
1,2 
1,2 
1,2 
1,- 
1,- 
3,4 
3,4 
3,4 
3,- 
3,- 
1,2 
1,2 
1,- 
1,- 
1,- 
3,4 
3,4 
3,- 
3,- 
3,- 
1,2 
1,2 
1,- 
1,- 
1,- 
3,- 
3,- 
3,- 
3,- 
1,- 
1,- 
1,- 
1,- 
1,- 
3,- 
3,- 
3,- 
“ У ~ 
Figure 2. Estimate vectors for the first 
phase sample units 
The numbers in the Fig. 2 refer to the strati 
fications. Each of the 25 first phase plots may 
obtain four different sets of estimates when all 
stratifications are represented. Some plots are 
represented only in one stratification and hence 
receive only one set of estimates. If no auxil 
iary data exist for a first phase plot it can 
not be included in any stratum and it does not 
receive any estimate directly. In this case 
nearest neighbour techniques, i.e., taking the 
estimates from the geographicly nearest neigh 
bor plots, may be applied. 
In order to estimate the final ground variable 
values for a specific first phase plot, the con 
cept of the reliability of the estimates has to 
be introduced. The final estimates for the 
ground variables, e.g. site, mean height, basal 
area, volume, tree species distribution, volume 
increment, and technical quality of the growing 
stock, is calculated by weighting the separate 
estimates by the reliability of the estimate. 
3. DISCUSSION 
The method, based on one stratification using a 
combination of satellite imagery and map data, 
has been tested in Finnish conditions on several 
occasions with positive results. These experi 
ences include an inventory of two communes with 
forest areas of 18 000 and 120 000 hectares. The 
field sample plots were usually relascope points 
with a basal area factor of 2 nr/ha. 
A study of the relevant possibilities and prob 
lems in estimating final ground truth variable 
values on the basis of more than one stratifi 
cation has been started at the University of 
Helsinki as a part of a thesis for a FI.Sc. deg 
ree. The study deals with updating of old inven 
tory data using remote sensing, forest growth 
models, and maps. 
The applicability of the approach to the tropics 
has not been attempted. The methodology describ 
ed may, however, be simple enough for forest re 
source assessment and monitoring. In the most 
extreme cases only N0AA and some basic map in 
formation may be used for stratification. Land- 
sat TM or corresponding data can be used when 
ever accessible. The major problem would be the 
availability of good quality ground truth data 
as field inventories in tropical conditions are 
very difficult to perform. The need for ground 
measurements may be especially demanding in the 
tropics if much weight is allocated for tree 
species estimation. A single hectare of closed 
tropical forest, for example, may include more 
than 100 tree species each with its own indepen 
dent colony of plants and animals. According to 
Birdsey et al. (1986) tropical species tend to 
have a clustered distribution. The use of clus 
ters as sampling units is recommended (e.g. 
Vazquez Soto 1987). Another general recommenda 
tion for the tropics is the use of permanent 
sampling units (e.g. Schreuder and Singh 1987). 
It is important that all data: plots, pixels, 
air photos, maps, etc., can be located at the 
same coordinate and time system. The correspon 
dence of different data should thus be tested. 
Studies with satellite imagery have shown that 
dislocation of some 20 m may cause a severe de 
crease in the usability of remote sensing in 
conditions where the average stand size is 
small. Problems may also originate from unreli 
able maps. Sometimes, it may be advisable to 
substitute maps for satellite imagery in the 
tropics. The date of the first and second phase 
data may differ and procedures for updating se 
cond phase data are thus required. Forest data 
updating often requires the application of 
growth models. The lack of reliable volume and 
biomass growth models and change models in land 
use emphasizes the need for permanent sample 
plots, especially in the tropics. 
REFERENCES 
Bickford, C.A. 1952. The sampling design used in 
the forest survey of the Northeast. Journal 
of Forestry 50(4), 290-293. 
Birdsey, R.A., Weaver, P.L. and Nicholls, C.F. 
1986. The forest resources of St. Vincent, 
West Indies. SO. 229. LA: New Orleans. 25 p. 
Frayer, W.E. 1979. Multi-level sampling designs 
for resource inventories. Department of Fo 
rest and Wood Sciences, Colorado State Univ., 
Rocky Mountain Forest and Range Exp. Sta., 
USDA, Forest Service. 
Harma, P. 1988. Interpretation of satellite ima 
geries at the department of forest mensura 
tion and management. Proc. of IUFR0 S4.02.05 
Meeting. Research Notes 21. Department of 
Forest Mensuration and Management, University 
of Helsinki. 
LaBau, V. and Winterberger, K. 1988. Use of 
four-phase sampling design in Alaska multi- 
resource vegetation inventories. IUFR0 S4.02. 
05 Meeting, Aug. 29 - Sept. 2, 1988. Research 
Notes 21. Department of Forest Mensuration 
and Management, University of Helsinki. 
Mattila, E. 1985. The combined use of systematic 
field and photo samples in a large-scale fo 
rest inventory in North Finland. Communica 
tiones Instituti Forestalis Fenniae 131. 
Neyman, J. 1938. Contribution to the theory of 
sampling human populations. Journal Amer. 
Stat. Assoc. 33: 101-116. 
Peng, S. 1987. On the combination of Multitem 
poral Image and Field Data for Forest Inven 
tories. Acta Forestalia Fennica 200. 
435