Sei sulla pagina 1di 8

Pores;~;ology

Management
ELSEVIER Forest Ecology and Management 71 ( 1995) 187-l 94

Evaluation of permanent sample surveys for growth


and yield studies: a Swiss example
Michael KiihP*, Charles T. Scottb, Andreas Zingg
Swiss Federal Institute for Forest, Snow and Landscape Research, Ziircherstrasse 111, CH-8903 Birmensdorf; Switzerland
WSDA Forest Service, Northeastern Forest Experiment Station, 359 Main Road, Delaware, OH 43015, USA
Accepted 24 January 1994

Abstract

This paper compares experimental growth and yield study plots with sample plots from forest inventories. The
difference in research objectives leads to a situation where sample plots are available that are representative of the
total population, but give only limited information on site conditions and management history. However, detailed
information on site condition and management history is available for growth and yield plots, but does not repre-
sent the total population. As an example, Swiss growth and yield plots are compared with survey plots in terms of
species distribution, stem form, slope class and elevation.
A critical review is given of the two most common sampling designs for regional forest surveys. Continuous
forest inventory is recommended over the more efficient but more complex sampling with partial replacement
design for use in conjunction with growth and yield studies.
The limitations of both growth and yield plots and sample plots are discussed. Recommendations for linking
both types of plots to close information gaps are given to improve both forest surveys and growth and yield studies.

Keyworuk Forest inventory; Continuous forest inventory; Sampling with partial replacement; Forest monitoring; Permanent
plot

1, Intmduction but to assessvarious treatments to the plots to


obtain information on the effect of distinct man-
In Europe growth and yield studies based on agement methods. Regardless of the initial re-
experimental growth and yield study plots have search objectives, the data from these plots are
a long tradition. The first plots of the Swiss Fed- often used to develop growth and yield models
eral Institute were established as early as 1888. and management guidelines.
Growth and yield models-in Europe almost ex- Permanent sample survey plots were intro-
clusively presented as tables or charts-were de- duced into forestry by Stott in 1938 (Stott,
rived primarily from growth and yield (G&Y) 1947). Rather than based on experimental de-
plots. The plots were established not only to study sign, plots are located based on sample survey
the growth of single tree speciesor provenances, design. The sample represents the treated and
untreated plots in proportion to their occur-
l Corresponding author. rence. In national forest surveys such as in the
0378-l 127/95/$09.50 0 1995 Elsevier Science B.V. All rights reserved
SSDZO378-1127(94)06098-4
188 h4. Kiihl et al. /Forest Ecologyand~Uanagement 71(1995) 187-194

USA, Scandinavia, Germany, Austria or Switz- Concerns regarding the representativeness of


erland, the total area in these countries is repre- growth and yield plots arise. This becomes criti-
sented by permanent sample (PS ) plots. cal when small data sets are used to develop
The objective of this paper is to compare the growth and yield models which are then applied
use of growth and yield plots versus permanent to regional survey updates and to forest planning
sample survey plots for the development of and management.
growth and yield models. We will show that
growth and yield models are not fully represent-
ative for updating regional forest surveys. PS 3. coIliparisoll of growth aad @t&l versus
plots may not provide the range of treatments of pemmt in %vi&akd
interest, but are representative of all forest types
and conditions. Continuous forest sampling de- The study of growth and yield in Swiss forests
signs for use in estimation and for the develop- was one of the most important research objec-
ment of growth and yield models are evaluated. tives right after the foundation of the Federal In-
We give recommendations on combining both stitute of Forestry Research in 1885. Growth and
methods for inventory assessment and yield yield plots were established in 1888 to study spe-
studies. The findings are equally applicable for cific species or provenances. One of these-plots
the design of monitoring programs and the selec- is still monitored. The plot design and data ac-
tion of permanent observation plots in forest de- cesshas changed over time, resulting in incon-
cline studies. sistent data, particularly from old plots. Much
effort was necessary to convert the growth and
yield data into a database system.
2. Growth and yield @As A critical problem was the accessibihty of plots
in these early times when &id crews had to use
Growth and yield plots are typically installed public transportation. Most of the growth and
aspart of an experiment. The objective is to study yield plots established before the 1950sfollow the
the effect of different planting methods, growth main railroad lines. Hence, these plots are not
patterns of distinct provenances, thinning re- representative spatially.
gimes and cutting practices. Plots are typically The comparisons shown below are based on
large and surrounded by a buffer strip to ensure data from growth and yield plots between 1977
uniform treatment and response across the plot. and 1988. The total area covered was 98.2 ha and
Several plots are installed in each treatment both was partitioned into 474 blocks with 184 ongo-
within the site and across sites for replication. ing research projects. The total number of trees
Unfortunately, replication was very often ne- registered on these plots was 90 735.
glected in early growth and yield studies. Mea- The first Swiss National Forest Inventory
surements and assessmentsare very intensive, (NFI ) was conducted in the mid- 1980s (EAFV,
particularly relating to site conditions and treat- 1988). A uniform 1 kmx 1 km grid of perma-
ment history. nent sample plots was assessedacross all of
Growth and yield plots are established on Switzerland. Of these, about 12 000 plots were
treated and untreated stands in accessibleareas. forested and measured by lieldgroups. The NPI
The focus is put on commercially valuable tim- plots are representative of all forest conditions in
ber types; other timber types are rarely studied. Switzerland. However, the plots are small and
No e&ones or edge conditions are permitted. treatment history or site data are more difficult
Becauseit is expensive to establish and maintain to determine.
plots, few plots are taken, resulting in a low sam- Switzerland can be divided into~fave produc;
pling intensity. Samples are often confined to a tive regions, which r&ct jmd site
narrow geographic range, e.g. low elevations and conditions. The distribution of growth and yield
easily accessible sites in mountainous regions. and permanent plots in the productive regions is
IU. Kiihl et al. /Forest Ecology andManagement 71(1995) 187-194 189

shown in Table 1. The Jura, Prealps and south- slope (Table 3) of the plots. Approximately 80%
ern slopes of the Alps are sufficiently repre- of the growth and yield plots are located at ahi-
sented. Most growth and yield plots are located tudes below 1000 m. The results of the NFI in-
in the Plateau region; the Alpine region is under- dicate that more than 50% of the Swiss stands
represented. are above 1000 m. A similar situation can be
Other indicator variables for site and growth found for the slope of the plots. Roughly 20% of
conditions are the elevation (Table 2) and the the NFI plots are located in areas with slopes
larger than 60%, but only 5.3% of the G&Y plots
Table 1 were established on these steep slopes.
Distribution of plots in various productive regions of The comparison of the proportion of growth
Switzerland and yield plots and permanent plots by produc-
Region Growth and NPI plots tive region, elevation and slope indicate that the
yield plots (%I G&Y plots do not represent the growth and site
(%) conditions of Swissforests. The growth and yield
plots tend to be established in lower elevations
Jura 13.3 16.4 and relatively flat terrain. A large proportion of
Plateau region 35.7 19.2
Prealps 21.8 18.3 Swiss forests are located in the unique condi-
Alps 11.9 32.2 tions of higher elevations, but these conditions
Southern slopes of 17.2 13.9 are poorly represented by G&Y plots.
the Alps Almost 29% of the growth and yield plots were
establishedto study selectively cut forests, but the
NFI results show that only about 20% of the Swiss
Table 2 forest have been managed using this regime.
Proportion of plots in each elevation class The growth and yield plots do not reflect the
proportions of tree speciesin Switzerland. Table
Elevation Growth and NFI plots 4 showsthat regarding number of stemsand basal
(ml yield plots (%I area, fir, spruce, oak, and a few other speciesare
(%I
fairly well represented. Beech is one of the most
> 1800 3.0 7.8 frequent tree speciesin Switzerland, but is poorly
1601-1800 2.9 8.6 represented in growth and yield studies.
1401-1600 2.2 10.2 Stem form can be expressed as the ratio of tree
1201-1400 7.2 11.7
1001-1200 4.8 13.5 height (h ) to diameter at breast height (d). Trees
801-1000 23.8 13.3 with a high h/d ratio are slender and are more
601-800 13.1 16.0 desirable for timber production. The h/d ratios
< 600 42.9 19.0 were grouped into three classes and are pre-
sented in Table 5. On growth and yield plots, trees
Table 3
Proportion of plots in each slope class have higher h/d ratios than those on NFI plots,
which means that poor stem forms are under-
Slope Growth and NPI plots represented and good stem forms are over-rep-
(%I yield plot (Oh) resented. This can create problems, such as in the
WI development of taper and volume equations. Any
> 100 1.8 model deficiencies will be accentuated by the
81-100 0.1 4.6 differences in how well the G&Y plots represent
61-80 5.2 12.8 the entire population.
41-60 12.8 23.3 Data on growth and yield plots were used to
21-40 24.6 27.2 develop Swissgrowth and yield tables for spruce,
<21 40.8 30.5
fir, larch and beech (Badoux, 1983a,b,c,d). They
190 M. Kiihl et al. /Forest EcologyandManagement 71(1995) 187-194

Table 4
Representation of tree species in Switzerland in studies concerning number of stems and basal area

Tree species No. of stems Basal area


--_-_
Growth and NFI plots Growth and NFI plots
yield plots yield plots
(%) (96) (WI i%)

Spruce 30.0 39.4 28.0 28.6


Fir 11.0 11.8 22.6 i 3.6
Pine 5.0 4.2 3.4 1.7
Larch 19.4 4.2 17 3 6.2
Mountain pine 0.6 0.8 1.5 1.o
Other coniferous 9.9 0.4 1.1 9.8

Beech 4.6 19.2 8.1 ! 7.7


Maple 6.6 3.6 0.3 6.6
Ash 0.6 3.6 0.2 5.X
Oak 4.0 2.4 7.4 3.8
Chestnut 2.4 1.3
Other deciduous 8.4 8.0 3.4 9.9

Table 5 determined on NFI plots. The NFI sh-owedthat


Proportions of trees surveyed in Switzerland in three stem- even-aged pure stands occupy about 18% of the
form classes
total forested area of Switzerland (Table 6 ) . Only
h/d ratio Growth and NFJ plots in these stands can the growth and yield tables
yield plots (%) be applied.
(%I The comparisons shown above arrive at the
conch&on that growth and yield plots are not
<0.8 45.4 67.7
0.8-l .O 30.9 22.3 representative of Swiss forests. For the commer-
>l.O 23.7 10.0 cially important areas, especially in the Plateau
region and in the Prealps, sufficient infktion
a Ratio of height to diameter at breast height.
is available for some tree specieslike oak, spruce
and fir. Spruce stands in mountainous regions
are applicable for even-aged, pure stands with and chestnut stands on thesouthern slopes of the
heavy thinning. The thinning methods cannot be

Table 6
Forested area ( 1000 ha) where growth and yield tables can be applied

Species Jura Plateau Prealps Alps Southern Alps Switzerland


region

Spruce 12.6 22.0 42.2 67.8 10.3 154.9


Fir 2.7 3.7 3.3 0.6 0.4 10.7
Larch 0.2 0.4 11.3 4.8 ! 6.7
Beech 10.6 6.6 3.3 4.1 7.5 32.1
Total 26.1 32.7 48.8 83.8 23.0 214.4

Percent of total 13.4 14.3 22.5 22.0 14.0 18.1


forested area
M. Kiihl et al. /Forest Ecology andManagement 71(1995) 187-194 191

Alps are not represented at all but are essential dependency on the permanent plots. Because
forest types in those areas. plots in distinct strata or regions may be lost,
samples may not be representative over time, or
the number of sample plots will become too low
4. Sampling on successive occasions for some units of reference. Visibly marked plots
may be treated differently from the rest of the
In forest surveys, sampling on successive oc- population and become unrepresentative.
casions is commonly based on two sampling
methods: (i) continuous forest inventory (CFI); 4.2. Sampling with partial replacement
(ii) sampling with partial replacement (SPR).
Other sampling procedures for monitoring pur- SPR was introduced into forestry by Ware and
poseshave been described, but are in most cases Cunia ( 1962) based on former work by Patter-
derivatives of either CFI or SPR (Kohl, 1990). son ( 1950). At the first occasion,assessmentand
estimation procedures are the same as with CFI.
4.1. Continuous forest inventory At the second occasion only a portion of the plots
established at the first occasion are remeasured
CFI uses a set of permanent sample plots, and a set of new plots is added. The remeasured
which are established at the first occasion and re- plots are used to develop regression relation-
measured at each successive occasion. CFI was ships to update all unremeasured plots. A second
introduced into forestry by Stott ( 1947) . Krutsch estimator is based on the new plots. Both esti-
established the first permanent sample plots in mates are combined and result in more precise
Germany in 1938, but did not present statistical estimatesof current values than does CFI (Scott,
analysis procedures. Schmid-Haas ( 1983) de- 1984). Cunia and Chevrou ( 1969) extended
veloped a special application of CFI for forest SPR to multiple occasions and Newton et al.
management planning, called control sampling. ( 1974) extended it to the multivariate case,but
This method combines survey sampling and the problems emerged owing to the awkward esti-
control method (Biolley, 1921; Gumaud, 1878), mation procedures.
a traditional forest planning method widely used One advantage of SPR is that it reducesthe risk
in Switzerland and southern Germany. of obtaining a poorly distributed sample. New
CFI plots are selected either randomly or sys- sample plots are added and help make a sample
tematically to represent the entire forest popula- more representative if the old plots were poorly
tion. Plots are monumented, so that they can be distributed initially or if they become poorly dis-
measured on future occasions. The spatial posi- tributed. The dependence on maintaining the
tion of each tree in a plot should be recorded, so permanent plots is lessthan with CFI. SPR also
that analysis based on paired observations of offers the opportunity to place sample plots in
trees are possible. The plots should be marked new areas where no permanent plots were estab-
invisibly to ensure that their locations are not lished initially. A second advantage is that SPR
known by local foresters and that no manage- is simultaneously cost effective for estimates of
ment activities different from those of the entire current values and net change in those values.
population are applied. Finally, SPR is very flexible. The emphasis can
Estimates are easily obtained with CFI and are be altered between the estimation of current val-
particularly good for change estimation. The es- ues and change detection by changing the pro-
timation procedures are simple and results are portion of new and remeasured plots. The esti-
intuitively understood. Additive tables are ob- mation of current values can be improved by
tained, where independent estimates of cell val- adding new plots; remeasuring more of the per-
ues sum up to the row and column totals. Even manent plots can improve the estimation of
for more than two occasions the estimation pro- change.
cedures are clear. A disadvantage of CFI is the Scott (1986a) described many of the prob-
192 M. Kiihl et al. /Forest Ecology andManagement 71(1995) 187-194

lems encountered. The first disadvantage is the and Hansen, 1983). If SPR is used at the third
complexity of the design. The number of plot occasion, change estimation could take only the
types increases geometrically as the number of last two occasions into account. Scott f 1986b)
occasions increases. At two occasions, sample recommended simplification by returning to CFI
sizesfor three different types of plots (unremea- particularly now that change is at least as impor-
sured, remeasured, and new) have to be deter- tant as current values.
mined. At three occasions, the number of plot
types increases to seven.
A second disadvantage is the complex estima-
tion procedures. The first step in the analysis is 5. Link between growth 888 yi&l ad pxmameat
to develop regression estimates. The regression sample plots
equations must be carefully monitored. When
stratification is used, the sample sizesare often Permanent sample plots cover the whole area
too small to develop reliable regression coeffi- of interest and give representative results. How-
cients for individual strata. Often the strata must ever, the sampling intensity in regional invento-
be collapsed, as the number of plots in the strata ries is typically too low and the plots are too small
is too small to estimate regression coefficients to adequately character&e individual stands.
reliably. The estimation procedures become very Treatment history is normally not known and
unwieldy for more than two occasions.Scott and cannot be developed from the data assessedon
Kohl ( 1994) give estimators for SPR and strati- the plots.
fication and for up to three occasions and show On growth and yield plots, site conditions and
their complexity. It is difficult to apply those treatment history are carefuliy monitored. How-
complex procedures in practical applications. ever, they are not representative of the total for-
These estimators can be applied to indepen- est population. Low value species, non-fully
dently estimate individual row and column cells, stocked conditions, and uneven-aged or mixed
as well as row and column totals. However, esti- stands are under-represented. In Swit-zerland,
mates developed in this way are not additive; growth and yield plots established before the
cells will not sum to estimates of row and column 1950s were located in highIy accessibleareas with
totals. For statisticians, non-additive tables are only a few plots in higher elevations or in mixed
acceptable, but forest resource analysts prefer to stands. It is likely that this situation is not unique
use tables that are additive. to Switzerland.
Different methods have been described to To overcome these information gaps, the two
handle the problem of table cells that do not add types of plots can be linked. One use of growth
to row and column totals. For example Li and and yield models is to update forest inventories.
Schreuder (1985) suggested a cell adjustment This approach is used in the north-central US
procedure, which adjusts cells inversely propor- (Hahn and Hansen, 1983). The frequency of
tional to their variances in order to make them forest surveys can be reduced and therefore the
add to the total. However, in forest survey re- cost of field assessments.Becausethe growth and
ports more than one table are presented. Not only yield plots are not representative of the entire
is additivity within tables required but also be- forest area, models may not be available for all
tween tables. conditions encountered.
Experience has been gained with SPR from the Permanent sample plots can be used to com-
1960s (Bit&ford et al., 1963) through the pres- pare the forest characteristicson growth and yield
ent. However, there are only a few practical ap- plots to those in the region of interest. Similarly,
plications of SPR for more than two occasions. they can be used to define the range wheregrowth
The tendency is to abandon SPR and move back and yield models can be appropriately applied.
to CFI or to other techniques like updating tech- Growth and yield models can be cabbrat& using
niques hased on growth projection models (Hahn permanent sample plots, thus increasing the ap
M. Kijhl et al. /Forest Ecology and Management 71(1995) 187-194 193

plicability of the resulting models to new areas have to be adjusted for the presentation of the
and different stand conditions. Finally, the range results. The Continuous Forest Inventory design
of forest characteristics can be used to identify is lessefficient, but due to its simplicity, is more
potential areas for the establishment of new G&Y appropriate for most practical applications and
plots. continuous surveys with more than two
An alternative is to develop models from both occasions.
survey data and from growth and yield studies It is well-known that caution should be used in
for the primary purpose of inventory updating. applying existing growth and yield models to
The resulting models would apply to all forest conditions other than those represented by the
conditions. An example of this type of model is plots upon which they are based. Unfortunately
STEMS (Belcher et al., 1982) which is used only in very rare casesare the users of growth
widely in the US. However, the ability of this type and yield models aware of the underlying condi-
of model to respond to forest management prac- tions and assumptions.
tices is limited unless treatment history is avail- When permanent sample plots are appropri-
able for the sample plots and is accounted for in ately modified, they can be used to augment
the model development. growth and yield plots to develop models which
are more generally applicable.

6. Conclusions

Some limitations of both permanent sample References


plots and growth and yield plots have been dem-
onstrated. Growth and yield plots are not appli-
cable to all conditions encountered in forest sur- Badoux, E., 1983a. Ertragstafeln Buche. Eidg. Amt. Forstl.
Versuchswes., 3rd edn.
veys and to all management situations. Although Badoux, E., 1983b. Ertragstafeln Fichte. Eidg. Anst. Forstl.
representative of the forest populations, perma- Versuchswes., 3rd edn.
nent sample plots are not large enough to char- Badoux, E., 1983~. Ertragstafeln Tanne. Eidg. Amt. Forstl.
acterise individual stands and do not have the Versuchswes., 3rd edn.
treatment histories needed to build growth and Badoux, E., 1983d. Ertragstafeln L.%rche. Eidg. Amt. Forstl.
yield models. Versuchswes., 3rd edn.
Belcher, D.W., Holdaway, M. and Brand, G., 1982. A de-
Permanent sample plots can be used to iden- scription of STEMS: The stand and tree evaluation and
tify the range of applicability of growth and yield modeling system. Gen. Tech. Rep. NC-79, USDA Forest
models. They can also be used to identify condi- Service, St. Paul, 18 pp.
tions and areas to establish new G&Y plots, thus Bickford, C.A., Mayer, C.E. and Ware, K.D., 1963. An efft-
expanding the applicability of the models. This cient sampling design for forest inventory: The northeast-
em forest survey. J. For., 61: 826-833.
also applies to establishing forest decline studies. Biolley, H.E., 192 1. Lamenagement des for&s par la meth-
Alternatively, PS plots can be used to augment ode experimentale et spkcialment la methode du contfile.
growth and yield plots in model development. In Neuchatel, Paris, 90 pp.
this case, PS plots should be larger, and better Cunia, T. and Chevrou, R.B., 1969. Sampling with partial
replacement on three or more occasions. For. Sci., 15: 204-
treatment histories should be obtained. 224.
Sampling with partial replacement should be EAFV, 1988. Schweixerisches Landesforstinventar: Ergeb-
applied with great care. This holds especially nisse der Erstaufnahme 1982-1986. Eidg. Amt. Forstl.
when surveys for more than two occasions are Versuchswes., Ber. 305,375 pp.
planned. The number of different plots types in- Gumaud, A., 1878. Cahier lamenagement pour lapplication
de1 al methode par contenance exposke sur la for& des
creasesgeometrically with the number of occa- Eperous, Paris, 160 S.
sions and thus the estimation procedures be- Hahn, J.T. and Hansen, M.H., 1983. Estimation of sampling
come very complex. Non-additive tables are error associated with timber change projection simula-
obtained with SPR-type estimators and usually tors. In: J.F. Bell and T. Atterbury (Editors), Renewable
194 M. Kijhl et al. /Forest Ecology and Management 71 (I 995) I8 7-194

Resource Inventories for Monitoring Changes and Trend, Scott, C.T., 1984. A new look at sampling with partial re-
Proc., 15-19 August 1983, Corvalhs, OR, SAF 83-14, pp. placement. For. Sci., 30( 1): 157-166.
546-549. Scott, C.T., 1986a. An evaluation of sampling with partial
Kohl, M., 1990. Stichprobenkonzepte fur das zweite replacement. In: Use of Auxiliary Information in Natural
Schweizer Landesforstinventar. Schweiz. Z. Forstwes., Resource Inventories, Blacksburg, VA, October i985, SAF
141: 745-75s. 86-01, Society of American Forestry, pp. 74-79.
Li, H.G. and Schreuder, H.T., 1985. Adjusting estimates in Scott, C.T., 1986b. Proposed northeastern forest survey em-
large two-way tables in surveys. For. Sci., 3 1: 366-372. phasizing disturbance detection. In: Use of Auxiliary In-
Newton, C., Cunia, T. and Bickford, C.A., 1974. Multivar- formation in Natural Resource Inventories, Rlacksburg,
iate estimators for sampling with partial replacement on VA, October 1985, SAF 86-O 1, Society of American Por-
two occasions. For. Sci., 20: 106-I 16. estry, pp. 146-I 5 1.
Patterson, H.D., 1950. Sampling on successive occasions with Scott, C.T. and Kdhl, M., 1994. Sampling with partial re-
partial replacement of units. J. R. Stat. Sot. B, 12: 241- placement and stratification. For. Sci., 40: 30-46.
255. Stott, C.B., 1947. Permanent growth and mortality plots in
Schmid-Haas, P., 1983. Swiss Continuous Forest Inventory: halfthe time. J. For., 45: 669-673.
Twenty years experience. In: J.F. Bell and T. Atterbury Ware, K.D. and Cunia, T., 1962. Continuous forest inven-
(Editors), Renewable Resource Inventories for Mouitor- tory with partial replacement of samples. FOT. Sci. Mon-
ing Changes and Trend, Proc., 15- 19 August 1983, Cor- OgT., 3: 40.
vallis, OR, SAF 83-14, pp. 133-140.

Potrebbero piacerti anche