CALCULATION SYSTEM FOR LARGE-SCALE FOREST INVENTORY
KariT.
Korhonen
Metsäntutkimuslaitoksen
tiedonantoja
505CALCULATION SYSTEM FOR LARGE-SCALE FOREST INVENTORY
Kari T. Korhonen
Metsäntutkimuslaitoksen
tiedonantoja
505tiedonantoja
Calculationsystem for
large-scale
forestinventory
Kari T. Korhonen
Joensuu 1994
Korhonen, system
large-scale inventory.
Metsäntutkimuslaitoksen
tiedonantoja
505. The Finnish Forest Research Institute, Research Papers505.36pp.ISBN951-40-1371-9,ISSN0358-4283.Summary:
Thispaperdescribesa calculationsystem forlarge-scale
forestinventory.
National ForestInventory
datafromtheForestry
BoardDistrictofKainuuare usedfordemonstrating
the calculationsystem. Mixedestimationis usedforestimating
localizedvolumes functionsfrom thesample
tree data measuredintheinventory
anddatameasuredin aprevious inventory.
Ordinary
leastsquares-technique
isusedforestimating growth
modelsandvolumefunctionsby
timber assortments. Estimated models and some basic statistics calculated for the testarea arepresented.
Keywords:
forestinventory,
mixedestimation, models,volume,growth
Author'saddress: Korhonen, KariT. TheFinnishForestResearch Institute, JoensuuResearch Station, P.0.80x 68, FIN-80101 Joensuu, Finland.
Publisher:TheFinnishForestResearchInstitute,
Department
ofForestResources,project
3001.Accepted by:
AameReunala,professor,
24.5.1994Distributor:TheFinnishForestResearch Institute, JoensuuResearch Station, P.0.80x 68, FIN -80101 Joensuu, Finland.
Data used in this
study
were collectedby
the fieldgroups of theproject
National ForestInventory
of Finland. Several personsworking
in the sameproject
in the Finnish Forest ResearchInstitutehavedonevaluableworkinfurtherprocessing
oftheHelddata.Especially
the workofMr.Alpo
Aarnio,Mr.ArtoAhola,Mr.MattiKujala
andMs. HelenaMäkelähavemade itpossible
to carry out thisstudy.
Discussions withpeople
mentionedaboveand with Ms.HelenaHenttonen,Mr. AnttiIhalainen,Mr. Juha
Lappi,
Mr. SakariSalminenand Mr. ErkkiTomppo
havehelped
insolving
severalproblems
meatduring
the research work. Mr. JuhaLappi,
Mr.SakariSalminenandMr.ErkkiTomppo
havereadthemanuscript
andgiven
valuable criticalcomments. Ms.JoannvonWeissenberg
hasrevisedtheEnglish
text.Theauthorwishesto thankallthese
people
for theirefforts.Joensuu 1.6.1994 Kari T. Korhonen
1. BACKGROUND 8
2. MATERIAL AND METHODS 9
2.1
Study
material 92.2Methods 11
3. DESCRIPTIONOF SYSTEM 12
3.1
Processing sample
treedata 123.1.1Estimationofvolumes 12
3.1.2 Estimation of volume increment 13
3.2Estimationofvolumesandincrementfor
tally
trees 153.2.1 Volume functions 15
3.2.2 Growth models 18
3.3
Generating
reports 194. APPLICATION OF THE SYSTEM FOR KAINUU DISTRICT 22
4.1 Estimated models 22
4.2
Examples
ofcalculatedresultsforKainuudistrict 235. DISCUSSION 25
APPENDICES 29
Concepts
andNotationThe
following
concepts relatedtothe measurementsof thefielddataareusedinthispaper.sample point
= apoint
wherearelascope plot
is measuredrelascope plot
= a set of concentric circles with each circlehaving
a fixedradius for eachdiameter
(the
radiusis afunctionofthecross sectional areaofthetree)
restricted
relascope plot
=arelascope plot having
amaximumradius; treeswith distancefromthe
sample point
greaterthanthemaximumradiusarenot talliedplot
section=sectionofaplot,
whentheplot
isnear aland-classboundary
plot
factor=relativesize of aplot;
whenaplot
is locatedneara land-classboundary
and thecenter
point
oftheplot
isonforestry
land,theplot
factorindicatestheproportion
ofthe wholeplot
circle madeby
theplot
sectiononforestry
landplot
stand=standcontaining
trees talliedon arelascope plot, usually
aplot
containstrees fromonly
onestand; delineationofstandsisbasedonthecharacteristics ofthesite andthegrowing
stocktally
tree =a treebelonging
to the(restricted) relascope plot
sample
tree =atally
tree forwhichmore detailedmeasurements aretakentimbertree =
tally
treecontaining high enough
timberquality
foratleastone sawlog
non-timbertree =
tally
tree whose dimensionsand/orquality
are notenough
forany sawlogs
The
following
notationsforthemostcommon treeandstandvariablesare usedinthispaper.1. BACKGROUND
In
large-scale
forestinventoriessampling
methodsare usedtoobtainarepresentee sample
ofthe
population.
Inmanycases it isnotpossible
orrationaltomeasuredirectly
thosevariablesthat we are interested in. Therefore, mathematicalmethods
(also
methodsotherthansimple
summation)
are neededtoderivethestatisticsoffinalinterestfromthecharacteristicsmeasuredforthe sample.
A
complete
calculationsystem ofinventory
resultsshouldincludefollowing
components:1.
checking
offielddata,2. derivationof volumes, volume
growth
etc. forsample
treesusing existing
modelsandmeasured data,
3.
generalization
ofvolumesandothercharacteristics fortally
trees,4. summationofstatistics for
any chosen calculationstratum,and
5. estimationof
reliability
oftheresults.Theaimofthis
study
wastodevelop
a systemofcalculationfor theNationalForestInventory
ofFinland(NFI).Thesystem shouldbe basedontestedanddocumentedmethodsandshould
cover estimationof areas of different strata
(e.g.
foresttypes),
mean volume,growth
andpercentages oftimber assortments. Estimationsoffuture
growth
andcutting possibilities
are, however,excludedfromthe system. Norareprocedures
fordetecting
errors inthe fielddatanorestimationof
sampling
error withinthescope ofthisstudy.
Thecalculationsystem is tested
using
Kainuudistrictasastudy
area.2. MATERIAL AND METHODS
2.1Studymaterial
Datafromthe7thNationalForest
Inventory
ofFinland(NFI7)
forthe wholecountryanddatafromtheBth8thNationalForest
Inventory
ofFinland(NFI8) forKainuudistrict(seeFig.
1) wereusedin this
study.
The NFI7wascarriedoutduring
1977-1984(Kuusela
& Salminen1991).
The NFIB data for Kainuu district were measured in 1992.
In both inventoriesthe
sampling
methodwassystematic
clustersampling.
In the NFI7 thedistance between clusters was 8 km andeach clusterconsistedof 21
relascope plots.
In theNFIB in the Kainuu district the distance between clusters was 7 km and each cluster consisted
of15 restricted
relascope plots.
Figure
1. LocationofKainuudistrict.Inbothinventories,severalvariables
describing
thesiteandgrowing
stockoftheplot
stand(s)wererecorded. Talliedtreeswereselectedwitha
relascope.
Inthe NFI7arelascope
withabasalareafactorof2was used.IntheNFIBinKainuudistrictarestricted
relascope plot
withabasalarea factor of 1.5 and a maximum radius of 12.45 m was used. In both sets of data the
following
variableswererecorded foreverytalliedtree:tree
species
diameterquality
classdescribing
thequality
ofthestem,latercalled'tally
treequality
class'crown class.
IntheNFI7data,the
living
treesweredividedinto3quality
classes: non-timbertrees(based
onthe dimensionsofthe stem), non-timbertrees(basedondefectson thestem),and timbertrees.
In theNFIBdataamoredetailedclassificationwasused todescribe,e.g. whetheratimbertree
is of
good
orpoorquality.
In theNFI7,
tally
treesmeasuredatfourplots
ineveryclusterwere used assample
trees.Inthe NFIB every7thtalliedtree was measured as asample
tree. In bothsetsofdatathefollowing
variableswere
registered
forsample
trees:height
-
age
length
andlocationofdifferenttimberassortments(=saw log quality
classesA,B, andC;pulp
wood;cull)
diameterincrementfor thepast
5-year period
height
incrementforthepast5-year period (only
forconifers).Inadditiontoabovementionedvariables,diameteratsixmeters
height
andthicknessofbarkatbreast
height
wereregistered
forallsample
trees in the NFI7data andfor asub-sample
ofsample
treesin theNFIB data.In the NFIB dataavariablecalled
'sample
treequality
class' was alsorecorded. This variabledescribes the cruisers
opinion
ofthequality
ofthe stemaftermeasurementofthesample
tree.The
'tally
treequality
class'describes thecruiser'sopinion
aboutthequality
beforethesample
tree was measured. The cruiser may
change
hisopinion
about thequality during
detailedexaminationofthestem when
measuring
thecharacteristicsof thesample
tree.2.2 Methods
The first
phase
in the calculations is to derive volume,growth
and percentages oftimberassortments forevery
sample
tree measured.Thevolumesof thetrees were calculatedusing
volumefunctionsofLaasasenaho(1982).
Volumesoftimberassortments werecalculatedusing
thetaper curve modelsofLaasasenaho
(1982)
asa functionofd,d«,
andh.Volumeincrementofthe
sample
treeswas estimatedaccording
to themethodsdescribedby
Salminen(1978)
andKujala (1980).
Regression analysis (Ordinary
LeastSquares, OLS)
andmixedestimationwereusedtoestimate thevolumeandincrementoftally
trees.Mixedestimationiswidely
usedinproblems requiring
combinationoftwoormoredatasets(Teräsvirta 1981).
Korhonen(1992, 1993)
hasshownthatmixedestimationisefficientfor
combining sample
treedatafromtwo inventories.SAS statistical software was usedfor
studying
therelationships
betweendifferentmeasuredvariables inorder to determinethe correct formof thenecessary models
(SAS
InstituteInc.1989).Theparametersofthemodelswereestimatedwith
Fortran-programs
madeby
theauthor.IMSL-routineswere usedfor matrix
operations (such
asinversion)
in theseprograms(IMSL
library...
1982). The reason forselecting Fortran-programs
instead of available statistical software was thatFortran-programs
makesitpossible
to simulatesampling
andthus test themethodsusedinthecalculationsystem.
Fortran-programs
werealso usedtoderivethevolumes ofdifferenttimberassortments forsample
trees measured.Volumeand
growth
oftally
trees were estimatedwithFortran-programs developed by
theauthor. The treewisecharacteristics were summed upinto statistics forthe wholecalculation
area with SAS statistical software.
3. DESCRIPTION OF THE SYSTEM
3.1
Processing sample
treedata3.1.1 Estimation of volumes
IntheNFIB data,dandhwere measuredforevery
sample
tree.Upper
diameter,dg,
however,was measured
only
forasub-sample
ofthesample
trees. The firstphase
inthecalculationof volumeswas toconstructmodelsforestimating
the upperdiameterofallsample
trees(higher
than8meters).
Function(1)
wasapplied
asthe model(Kolhonen 1992)
butwasusedinitsfullform
only
forpine
and spruce. For otherspecies, only
variables d2,h
2, d/t, and were
significant
regressors; therest ofthe variableswereexcludedfromthemodel.Theparametersofthemodel
(1)
foreach treespecies
wereestimatedusing
mixedestimation.In the first stage of theestimationprocess, first-levelestimates ofparameters were obtained
using
NFI7dataforthewholecountry.Inthesecondstage,second-levelestimatesofparametersa,-aswereobtained
using
NFIBdatafromKainuudistrict(Korhonen1992).
Parametersrelatedtothecoordinatesarenotestimetedinthesecondstagebecausethedatameasuredinthisstage
are
quite
few andgeographically
notrepresentative.
In
(dj
=ao+a,*d
2+ a3*h2+ +a3*ln(d6L)+a«*ln(G)
+a7*YC+
a,*YC
2+a,*XC
+alO*XC2+
+
a„*YC*XC,
(1)Whenall
sample
treeshavemeasuredvalues forvariablesdandh andalltreeshigher
than 8meters haveameasuredorestimatedvalueforvariable thestemvolume
(from
theestimatedstump
height
to the top of thetree)
can be estimated with the functionspresented by
Laasasenaho (1982).Estimationoftimber assortments is based on thedimensionsof thestem
(d, d«, h)
andthemeasured
lengths
ofdifferentquality
classes. Thesemeasurements wereusedtodividethestemintosaw
logs
thatfulfilledthedimension(top
diameterandlength) requirements.
Thestemwasdivided into saw
logs by maximizing
the value of the stem witha'complete
enumeration' method.Inthismethod,allpossible
solutions(combinations
ofsawlogs
ofdifferentlengths)
are testedandthe solutionthat
gives
thebestvalueischosen.Thistimberruling
wasmadewitha
Fortran-program
in whichthedimensionrequirements (minimum
andmaximumlengths
and minimumdiameters)and relativevaluesofdifferentassortments areoptional
parameterssothatthey
caneasily
bechanged.
3.1.2 Estimation of volume increment
The
sample
treevariables thatarerelatedtotheestimationofgrowth
arediameterandheight
increment
during
thepast5-year period
and thickness ofthebark. Thebark is measuredonly
forsome ofthe
sample
trees.Therefore,aregression
modelwas constructedforestimating
thethicknessofthebark.Function
(2)
was foundasa suitablemodelforpine.
where b = thickness of bark.
For other
species
alogarithmic
model(Function 3)
was foundto be necessary to solve theproblem
ofheteroscedasticity.
Fordecidioustreesheightincrementwas notmeasured.Therefore,this variable was estimated
b =&o+
a,*d
+(2)
ln(b) =a„+
a,*ln(d)
+(3)
using
thetablesofIlvessalo(1948,
seeKujala 1980),
in whichheight,
age,andcrown classofthe treeare
independent
variables.Whenthebark modelsare estimated,each
sample
tree havemeasuredorestimatedvaluesfor variablesrelatedtovolumegrowth:
diameterandheight
incrementandthicknessofthebark.No characteristics aremeasuredtodirectly
describe thechanges
in stem formand inthicknessofthebark
during
thepast5-year period.
Whencalculating
thevolumeincrementthesechanges
can betakenintoaccountwiththe method
presented by
Salminen(1978)
andKujala (1980).
In this methodit is assumed thatthechange
inv/gj (ratio
ofvolumeand cross sectionalarea)
during
thepast5-year period
canbeestimatedwithhelp
ofafunction(v/g,=f(h))
estimatedfromthepresentv,g(
andhofthe trees.
Todescribe themethodofSalminen
(1978)
andKujala (1980),
letus note that:r(h)
=afunctionthatestimatesr asa functionofheight,
(Vuistheunitvolumeofa
sample
tree=volumeofthetreedividedby
itscross sectionalarea), and(SVu
is 'seedvolume'ofarelascope
tree =the volumeofthe tree 5 years agodividedby
itspresentcross sectional
area)
r
=v/g„
(4)Vu =
v/g (5)
SVu= Vj/g (6)
Formula (6) for SVu can befurtherwrittenas follows:
Using
notationVuforv/g,
r(h) forestimatedv/g„
andf(h
s)
forestimated Function (7)can be written:Afterthe 'seedvolume' ofatreeisestimated
using
Function(8)
andareasin cross sectionandheights
now and5 years ago are measured, the volume5 years agocan be calculated withFormula (9):
3.2 Estimationofvolumesandincrement for tallytrees
3.2.1 Volume functions
Functionsfor
estimating
thevolumeof the wholestem fromstumptothe topof thetreewereconstructed
using sample
trees measured in the NFI7 and the NFIB. The two data werecombined
using
mixedestimation(Korhonen 1993).
Atthefirststageoftheconstructionofthevolume functions, NFI7 data were used for
determining
the form of the models andforobtaining
first-levelestimatesoftheparameters. Inaprevious study (Korhonen 1993)
Function 10was shown to workwellforpine.
SVu =gis
/g
*(vj/g
i3)<=>
SVu =gi3
/g
*(v/gi
-(y/g-,
-v,/g
i5))
<=>
SVu = *
v/g,
- gis/g
*(V/&
-Vj/g
i5)
<=>
SVu =gis
/g
*(g/g,
*v/g
-(v/g,
-Vj/gi,)) (7)
SVu =
gis
/g
*(g/
gi *Vu -(f(h)
-?(h
5)). (8)
v 5 =
g * SVu(9)
whereRDIST=relativedistancefromthe seacoast
(see 'Concepts
andNotation').
Function
(10)
was alsofoundtobesatisfactory
forspruce andbirches. Forotherspecies,
the variable RDIST was excluded from the model.Atthesecondstage intheconstructionofthevolumefunctions,mixedestimationwas usedfor
obtaining
second-level estimates of the parameters(Korhonen 1993). Only
data from the calculationarea(typically
onedistrictoftheCentralBoardofForestry
inFinland)
wereusedforthere-estimation.Becausethe numberof
sample
treesmeasuredwasquite high
forpine,
spruceand birches, modelswereestimated
separately
foreachsiteclass.Only
constant andparametersofvariablesd, d
2,and
ln(G)
wereestimatedatthe second stage;forothervariables, first-level estimates were used (Korhonen 1993).Regression
models were also constructed forestimating
the volumes of different timberassortments: timberandcull.Thevolumeof
pulp
woodquality
wasestimatedby subtracting
the estimatedvolumesoftimberandcull fromtheestimatedstemvolume.Afunctionwithaformof
Equation (11)
was foundtobesuitable.BecausetheformoftheFunction
(11)
fortimberassortmentwisevolumesdiffersmarkedly
fromtheFunction
(10)
forwhole-stemvolume,itwasalso necessarytoestimateamodelofaformsimilarto
Equation
(11)forstemvolume.Otherwise,allerrors duetothe formofthefunctionswouldhavebeensummedupin thevolumeestimatefor
pulp
wood.In somecases thiscouldevenhaveledto
negative
estimates forpulp
wood.Finalestimates for,e.g. thesawlog
volume ofatally
treewerethenobtainedwithformula(12).
v/d2=ao+ a l
*d+
a^d
2+ a3*RDIST+04*111(0
+a,*YC
+a^YC
2+a,*XC
+a^XC
2+a,*YC*XC, (10)
vt/d
2
=a„+
a,*ln(d)
+ (11)Because
logging
ruleshavechanged
since theNFI7,itwas notpossible
touse theNFI7dataasprior
information.Timberassortmentwise modelswere estimatedusing
OLS andsample
tree data measured from the calculation area.Naturally,
theproportions
of timberassortments varymarkedly by
tree class. Therefore, thesample
treedataweregrouped by
treeclasses asfollows.1 = non-timber tree
2 =a
good
sawlog
tree3 =apoor saw
log
tree.As mentionedinSection 2.1, inthefieldmeasurements two differentcodesfor treeclass were
recorded forthe
sample
trees. The firstof these describes thequality
'at firstglance',
whichrefers tothe way itiscodedfor
tally
trees(later called'tally
treequality
class'). The secondone is coded
according
to moredetailedsample
tree measurements(later
called'sample
treequality class'). Usually,
thecodes match,butinsomecases thecruisermay
change
thequality
code whentaking sample
tree measurements. Thus,asample
tree withtally
tree class code'non-timbertree' can includesaw
log quality,
and viceversa.Only
trees with thesample
treequality
class code'good
sawlog
tree'or'poor
sawlog
tree'were usedin
estimating
theabove mentionedregression
modelsfortimberassortment volume.Atthe
application
stagetally
treesweregrouped
intotherespective quality
classesaccording
tothe
tally
treequality
class. To avoidpossible
bias causedby
differencesin thetwoquality
classifications anadjustment
was made as follows. For allsample
trees the mean of themeasuredtimbervolumeand the meanoftheestimatedtimbervolumewere calculated
by
treespecies.
Thetimbervolumeestimateof asingle
tree was thenmultiplied by
theratio ofthese V,=v/v
2*
v, ,
(12)
where v,=finalestimateforv„
vt=estimateforv,obtainedwiththe modelfortheformof
Equation (11)
v, =estimatefor vobtainedwiththe modelforthe formof
Equation (10),
andv2
=estimatefor vobtainedwiththe modelfortheformof
Equation
(11).two means. Asimilar
adjustment
wasmadefortheestimatesofcull volume.3.2.2 Growth models
Model
(13)
was selectedas a basic modelforestimating
thevolumeincrementofthepast5-year
period.
For somespecies
notallthe variablesintheequation
weresignificant
rcgressors.Inthesecases,
only significant
variableswere used(see
Section4.1).
whereSI, S2,
S 3 are dummy
variablesfordifferentsiteclasses, andSOILis adummy
variableto separatemineralsoils form
peatlands.
When the
growth
models were estimated, treesgrowing
onpoorly productive
land wereseparated
fromtreesgrowing
on forestlandfortwo reasons:1. the
growth
of treesgrowing
onpoorly productive
landismarkedly
different from thegrowth
oftrees onforestland,and2. mostofthevariables
describing
thegrowing
stockare not measuredonpoorly productive
land.
Equation (14)
was usedasagrowth
modelfortreesonpoorly productive
land.The natural
logarithm
ofi^d
2is usedas andependent
variable inmodels(13)
and(14).
Inapplications,
unbiasedestimatesfori^d
2 inthearithmetic scale are needed.Themostcommonway tocorrectthebiasduetothenon-lineartransformationistoaddtheterm 1/2*MSEtothe
logarithmic
estimate. This correction is based on theassumption
that theresiduals of thelogarithmic
modelarenormally
distributed.Ifthisdoesnot holdtrue,following
estimatorcanbe
justified according
toSnowdon(1991):
ln(iyd
2) =a<>+a,*d
+aj*ln(d)
+aj*G
+a^*DD
+ + +
a,*S2
+ag*S3
+a,»SOIL
(13)ln(i/d
2)
=a,,+a,*d
+a2*ln(d) (14)
y, =measuredvalueforobservationi,and
ft
=estimatedlogarithmic
valueforobservationi.Estimator
(15)
wasusedinthisstudy
to correctthebias duetothenon-lineartransformationingrowth
models.3.3
Generating
reportsUsing
the modelsdescribedinSection3.3, theHelddataare transformedintoa file, which is suitableforfurtherprocessing
with SASstatisticalpackage.
Inthisfile,eachplot
standhasonerecordforcharacteristics
describing
the locationoftheplot
andsiteandthegrowing
stock ofthe stand; and one record for each measured
tally
tree. The recordof atally
tree containsfollowing
data:- tree
species
-
tally
treequality
class -d- v/d2
-tyd
2-OJd
2
-
U
2.
(Volumes
andandgrowth
are dividedby d 2 before they
are stored in the file in order to decreasethe errorsduetorounding off).
Whenareaestimatesare calculatedfordifferentstrata,e.g. site classes, the
sample point
isusedasone observation.AreaestimatesareobtainedwithFormula(17)(Salminen1993).
ft
=c*exp(p
i), (15)
2y.
where c= ,
S exp(pi)
Volume sum and mean statistics for a calculation area or different strata of the area are
compiled by summing
thevolumesandplot
factorsoftheplots
overthestratainquestion.
Themeanvolumeis calculated
by dividing
thevolumesumby
thesumoftheplot
factors.Nomean volumesforsingle
standsof arelascope plot
are needed.Infact, itisimpossible
tocalculatesuchmeanvolumesin theNFIdatafor standsthatdonot coverthewhole
relascope plot.
In the case of a restricted
relascope plot
with maximumradius of 12.45 m (as inNFIB inKainuudistrict) thevolumesumofa
plot
(section)iscalculatedasfollows.It should be noted, that Formula
(17)
does not differentiatebetween wholeplots
andplots
sections. Thesizeof aplot
is takenintoaccount laterwhenthemeans or sums are calculatedfor the calculation strata.
Afterthe volumesums arecalculatedforeachpart
plot,
mean volumeestimates fordifferentstrata ofa
inventory
areaare obtainedwithFormula(18).
Äj
=m/M
* AREA,(16)
V=
X
n,• V,,(17)
IV,
Vi
= ,(18)
iFi
4.APPLICATION OF THE SYSTEMFORKAINUU DISTRICT
4.1 Estimated models
Theupperdiameterfunctionsestimated
using sample
treesfromKainuudistrictareinAppendix
2.Separate
modelswereestimatedforfollowing species: pine
(Pinussylvestris),
spruce(Picea
abies),
whitebirch(Betula pendula),
silverbirch(B. pubescens),
aspen(Populus tremuloides),
alder (Alnus incanaand A.
glutinosa).
As described earlier, NFI7 data were used asprior
informationforsomeoftheparameters and
only
asinformationforotherparameters(Korhonen
1992).
Thebark modelsfor
sample
trees of Kainuudistrictare inAppendix
3.Separate
modelsforpine,
birches(no
differencebetweenwhiteand silverbirch),
aspenandalderwereused. The modelswere estimatedusing
NFI7 andNFIBsample
tree datafromKainuudistrict.Volumefunctionsestimatedfordifferent
species
forKainuudistrictareinAppendix
4.Forpine,
spruce and birches the final
parameterestimates were calculated
separately
for 4 site classgroups.For aspen,alderandthe groupofother
species
thesample
tree datawere too few todistinguish
betweensiteclasses.Themodelspresented
inAppendix
4are notlogical
forsmall sized trees.Therefore,generalvolumefunctionswere usedfor treeswithd <3 cm.The
regression
modelsfor timberassortmentwise volumesarepresented
inAppendix
5. Thedatafor
pine
andspruceweredividedintotwosite classgroups.Separate
modelsfortimberandcull volume modelswereestimatedforthe two groups.Thedataforbirches were too few to
makeadistinctionbetweensiteclasses. For aspen
only
fewsample
trees were codedas timberquality
trees,andthereforenomodelfortimbervolumewereestimated.Attheapplication
stage, timbervolumemodelofbirches were usedfor thoseaspens thatwere codedas timberquality
trees.For other
species,
no timbertreeswere measured;modelswere estimatedonly
for cull volume.The correctionfactorsfortimberassortmentwisevolumes(see
Section3.2.1)
are alsogiven
inAppendix
5.Theestimatedmodelsfor thepast
5-year
volumegrowth
ofdifferentspecies
are inAppendix
6.
Separate
modelsfortreesgrowing
atforestlandandpoorly productive
landwereestimated.For forest land separate models for
pine,
spruce,birches and aspenwere used. Forpoorly
productive
landbirchesandaspenwere combinedtothegroupofotherspecies.
Thecorrectionterms
(see Equation (16))
are alsogiven
inAppendix
6.4.2
Examples
ofcalculatedstatistics forKainuudistrictAccording
to thestatisticsoftheNationalBoardofSurvey,
thetotallandareaofKainuudistrict is2 156690ha. Estimatedareaof forestlandis 1 664015ha.Figures
2and 3present twoexamples
ofestimatedpercentages ofdifferent strata.InFigure
2 is thedistributionofforestland
by
dominantspecies. Figure
3 presents the age class distributionon the forestland inKainuu district.
Theestimatedmeanandtotalvolumesandtimberassortmentwise meanand totalvolumes
by
species
onforestlandandpoorly productive
land aregiven
in Table1. Thegrowth
statistics obtainedusing
thecalculationsystem arepresented
inTable2.Figure
2.Dominanceoftreespecies
onforestland.Figure
3.Age
structure on forestlandTable 1.Meanandtotalvaluesofstemvolumeand timberassortmentwisevolumes
by species
for forestandpoorly productive
land.Table2.Meanandtotalvalueestimates for
growth by species
forforestandpoorly productive
land.Meanvolumes, m 3
/ha Totalvolumes, 1000m
3
saw
log pulp
total sawlog pulp
totalPine 13.3 21.9 37.7 25089 41288 71062
Spruce
6.1 9.8 16.9 11427 18427 31947Birches 0.2 6.2 8.7 414 11763 16382
Others 0.0 0.7 1.1 29 1273 2009
All 19.8 44.8 64.4 36959 72750 121401
Pine
Spruce
Birches Others Allm
3/ha 1.35 0.41 0.38 0.06 2.20
1000m
3 2552 782 715 104 4153
5. DISCUSSION
Thesystemofcalculation
presented
inthispaperdiffersfromthepresentcalculationsystemforNFIin the
following
ways:1. Stemvolumeandthe timberassortmentwise volumesare
generalized
fortally
treesusing
regression
modelsinsteadofclasswise meanvalues.2. Growthestimatesarecalculatedfor
tally
trees withamethodsimilarto thatforestimating
the timber assortmentwise volumes. In the
present calculation
system
growth
is notgeneralized
fortally
trees(growth
isestimatedusing growth
percentagesestimatedfromthesample
treedataanddiameterdistributionsestimatedfromthetally
treedata).
Theabove-mentionedsolutionsmakethe system flexible.Area,volumeand
growth
estimatescan
easily
becalculatedforanysub-classofthedata.Plotwisevolumeandgrowth
estimatescaneasily
beusedasa'ground
truth' forprocessing
satelliteimages (Tomppo 1992).
Charactericsof'final
interest',
suchastimberassortmentwisevolumesandvolumegrowth,
wereusedas
dependent
variableswhenthemodelswereestimatedfromthesample
treedata.Anotherpossibility
wouldhavebeento use variablesdescribing
thedimensions(d,
h,b, id,ij
andquality (lengths
ofdifferentquality
classes) asindependent
variables.The modelswouldthengive
estimatesofallsample
treevariablesforeverytally
tree.Theseestimatescouldbe usedasindependent
variables for further estimations, e.g. of the volume orgrowth
oftally
trees.Variousmodificationsofthismethoddescribed
by
Kilkki (1979)arewidely
usedininventory
systems. The method hasone serious drawback, however: residual varianceof thedifferent
models mustbe taken into account when, e.g.,
using
estimatedh,i,,,
and ih asindependent
variablesin
estimating
volumeincrement(Kilkki 1979).
Thejoint
distributionsoftheerrorsofthe different models are difficulttoestimate. Therefore, aftersome trials this
approach
wasrejected.
As stated, in this system ofcalculation
sample
treevariables such asheight
and age are notgeneralized
fortally
trees.Therefore,
thedatagenerated
fortally
treescannotbeused topredict
the future
development
of forests with simulation systems like the FinnishMela-system
(Siitonen 1983).
Furtherstudiesareneededif,
forexample,
thegrid
method(Holm
etal.1979)
is
applicaple
forgeneralizing
thesample
treecharacteristicsfortally
treesinsuch awaythattheresults can be used as a basis both for
calculating
unbiasedinventory
statistics and forsimulating
thefuturedevelopment
ofthetrees.LITERATURE
Holm,S.,
Hägglund,
B.&Märtensson,A.Amethodforgeneralization
ofsample
treedatafrom the Swedish National ForestSurvey.
SwedishUniversity
ofAgricultural
Sciences.Department
of ForestSurvey. Report
25. 94pp.IMSL
library
reference manual.Edition9. 1982.IMSLInc. Houston,Texas.Kilkki, P. 1979. Outline for a data
processing
system in forestmensuration. SilvaFennica13(4):368-384.
Korhonen,K.T. 1993.Mixedestimationincalibrationofvolumefunctionsof Scots
pine.
Silva Fennica 27(4):269-276.Korhonen,K.T. 1992.Calibrationofupperdiametermodelsin
large-scale
forestinventory.
Silva Fennica 26(4) :23 1-239.Kujala,
M. 1980.Runkopuun
kuorellisen tilavuuskasvun laskentamenetelmä.Summary:
A calculation method formeasuring
the volumegrowth
over bark of stemwood. Folia Forestalia441. 18pp.Kuusela,K.& Salminen,S. 1991.Suomenmetsävarat 1977-1984
ja
niidenkehittyminen
1952- 1980.Forestresources ofFinlandin 1977-1984and theirdevelopment
in 1952-1980.Acta ForestaliaFennica220.84pp.Laasasenaho, J. 1982.
Taper
curve and volume functions forpine,
spruce and birch.CommunicationesInstitutiForestalisFenniae 108.74pp.
Salminen,S. 1993.EteläisimmänSuomenmetsävarat1986-1988.
Summary:
Forestresourcesof SouthernmostFinland.FoliaForestalia825. 11l pp.Salminen,S. 1978. Incrementcalculationsonthe basis of
relascope sampling
in theFinnish NationalForestInventory. Paper presented
in lUFROMeeting
of June 18-25, 1978 in Bucharest, Romania.Subject Group
54.02. 7 p. The Finnish Forest Research Institute.Helsinki, Finland.
SASInstituteInc.,SAS/STATUser's
quide,
Version6,FourthEdition, Volumes1and2,Cary,
NC: SAS Institute Inc., 1989.Siitonen,
M. 1983.Along
termforestry planning
systembasedondatafromtheNationalForestInventory
ofFinland. In:Forestinventory
forimproved
management.Proceedings
ofthe lUFROSubject Group
4.02Meeting
in Finland,September
5-9, 1983.University
of Helsinki.Department
ofForestMensurationandManagement.
ResearchNotes 17:195-207.Snowdon, P. 1991. Aratio estimatorforbias correction in
logarithmic regression.
Canadian JournalofForestResearch 21(5):720-724.Teräsvirta,T. 1981.Someresultson
improving
the leastsquaresestimationoflinearmodelsby
mixed estimation. Scandinavian Journal of Statistics 8:33-38.Tomppo,
E. 1992. Multi-source NationalForestInventory
of Finland. In:Proceedings
of IlvessaloSymposium
on National Forest Inventories. Finland 17-21August
1992.Nyyssönen,
A.,Poso,S.&Rautala, J.(ed.).TheFinnishForestResearchInstitute.ResearchPapers
444:52-59.Valtakunnanmetsien 8. inventointi.
Kenttätyöohjeet,
Kainuunja Pohjois-Pohjanmaan
veriso.1992.
[Field
instructionsforthe fieldworkof the8"NationalForestInventory
ofFinlandatKainuuand
Pohjois-Pohjanmaa
districts. InFinnish.]Manuscript
67 pp +appendices.
TheFinnishForestResearchInstitute. Helsinki, Finland.
Valtakunnanmetsieninventoinnin
kenttätyöohjeet
VMI7. 1978.[Field
instructionsforthefield work of the 7th NationalForestInventory
of Finland. InFinnish.] Manuscript.
59 pp +appendices.
The FinnishForestResearchInstitute.Helsinki, Finland.Appendix
1. for ratio of volume over bark and cross sectional area inside bark (r = v/gj (Kujala 1980).
r(h) = 0.39 +
0.39* h +
2/(h.l-3) +o.77*V(h-1.3)
r(h) = 0.44 +
0.355* h +
2/(h.l-3) +o.6s*V(h-1.3)
r(h) = 0.48 + 0.48* h + 3.5/(h.l-3)