comparison of canop

7/28/2019 Comparison of Canop

http://slidepdf.com/reader/full/comparison-of-canop 1/12

577

Silva Fennica 40(4) research articles

www.metla.f/silvaennica · ISSN 0037-5330 The Finnish Society o Forest Science · The Finnish Forest Research Institute

-6Ê

1 Introduction

Forest canopy cover, also known as canopy cover-age or crown cover, is dened as the proportion o

the orest foor covered by the vertical projectiono the tree crowns (Jennings et al. 1999). Estima-

Estimation of Forest Canopy Cover:a Comparison of Field MeasurementTechniques

Lauri Korhonen, Kari T. Korhonen, Miina Rautiainen and Pauline Stenberg

Korhonen, L., Korhonen, K.T., Rautiainen, M. & Stenberg, P. 2006. Estimation o orest canopy cover:a comparison o eld measurement techniques. Silva Fennica 40(4): 577–588.

Estimation o orest canopy cover has recently been included in many orest inventory pro-grammes. In this study, ater discussing how canopy cover is dened, dierent ground-basedcanopy cover estimation techniques are compared to determine which would be the mosteasible or a large scale orest inventory. Canopy cover was estimated in 19 Scots pine orNorway spruce dominated plots using the Cajanus tube, line intersect sampling, modiedspherical densiometer, digital photographs, and ocular estimation. The comparisons were basedon the dierences in values acquired with selected techniques and control values acquiredwith the Cajanus tube. The statistical signicance o the dierences between the techniqueswas tested with the nonparametric Kruskall-Wallis analysis o variance and multiple com-parisons. The results indicate that dierent techniques yield considerably dierent canopycover estimates. In general, labour intensive techniques (the Cajanus tube, line intersectsampling) provide unbiased and more precise estimates, whereas the estimates provided byast techniques (digital photographs, ocular estimation) have larger variances and may alsobe seriously biased.

Keywords orest canopy, canopy cover, canopy closure, Cajanus tube, line intersect sampling,spherical densiometer, digital photographsAddresses University o Joensuu, P.O. Box 68, FI-68101 Joensuu, FinlandE-mail [email protected] 12 April 2006 Revised 22 September 2006 Accepted 26 September 2006Available at http://www.metla./silvaennica/ull/s40/s404577.pd

tion o orest canopy cover has recently become animportant part o orest inventories. First, canopycover has been shown to be a multipurpose eco-logical indicator, which is useul or distinguish-

ing dierent plant and animal habitats, assessingorest foor microclimate and light conditions, and



578

Silva Fennica 40(4), 2006 research articles

estimating unctional variables like the lea areaindex (LAI) that quanties the photosynthesiz-ing lea area per unit ground area (Jennings et al.1999, Lowman and Rinker 2004). Secondly, manyremote sensing applications involve estimation o either canopy cover (Gemmell 1999) or individualtree canopy area (Kalliovirta and Tokola 2005) asan intermediate stage in distinguishing the signalsrefected rom orest canopy and orest foor, aterwhich, or instance, estimation o timber volumebecomes possible (Bolduc et al. 1999, Maltamoet al. 2004). Canopy cover is also an impor-tant ancillary variable in the estimation o LAIusing empirical or physically based vegetationrefectance models (Jasinski 1990, Spanner et al.1990, Nilson and Peterson 1991, Knyazikhin etal. 1998, Kuusk and Nilson 2000). The validationo canopy cover estimates obtained rom remotelysensed data and development o new remote sens-ing techniques require eld-based canopy covermeasurements. Finally, the international deni-tion o a orest is based on canopy cover: theUnited Nations Food and Agricultural Organiza-tion (FAO) has dened orest as land o at least0.5 ha with potential canopy cover over 10% andpotential tree height o at least ve meters (FAO2000). To ensure compatibility o internationalorestry statistics, orest canopy cover needs tobe included in national orest inventories.

Measuring canopy cover accurately involvespractical and theoretical diculties. For inter-

preting the measurements the dierence betweenthe concepts o canopy cover and canopy closure

must be recognized. Canopy cover , dened hereas the proportion o the orest foor covered by

the vertical projection o the tree crowns, shouldbe distinguished rom canopy closure, whichis dened as the proportion o sky hemisphere

obscured by vegetation when viewed rom a single

point (Fig. 1) (Jennings et al. 1999). The dier-ence between these concepts is clear: i canopy ismeasured with instruments that have an angle o view (i.e. measure a larger area than just a verticalpoint), like cameras (Kuusipalo 1985) or spheri-cal densiometers (Cook et al. 1995), the resultsare estimates o canopy closure. In other words,canopy closure or “site actor” (Anderson 1964)is just a percentage gure describing the ractiono non-visible sky within a certain angle, whereascanopy cover describes the raction o ground areacovered by crowns.

According to the denition by Jennings et al.(1999), i canopy cover is to be measured cor-rectly, the measurements should be made in exactvertical direction. I instruments with an angle o view are used, canopy cover is usually overesti-mated, because the trees seem to “all” towardsthe centre o the observed area (Bunnell and Vales1990, Cook et al. 1995, Jennings et al. 1999). Asthe size o the area sampled increases, the biasalso increases. Another issue worth noting isthat tree height and length o the live crown donot aect the estimates o canopy cover, whereascanopy closure increases as the trees become

taller, and as the height to the live base o thecrown decreases (Jennings et al. 1999).

Fig. 1. Canopy cover (let) is always measured in vertical direction, whereascanopy closure (right) involves an angle o view.



579

Korhonen, Korhonen, Rautiainen and Stenberg Estimation o Forest Canopy Cover: a Comparison o Field Measurement Techniques

In publications concerning canopy cover andcanopy closure, it has oten been unclear whichterm to use when describing the results (Sarvas1954, Kuusipalo 1985, Bunnell and Vales 1990,Ganey and Block 1994, Cook et al. 1995, Nuttle

1997). The denitions used here are similar tothose published by Jennings et al. (1999). Thesedenitions are not yet commonly established,but in the uture it would be preerable i authorsconsistently use dierent terms when reerring todierent measurements.

Another diculty in dening canopy cover hasbeen deciding whether the gaps inside tree crownsshould be counted as canopy. The traditional de-nition o canopy cover includes an “outer edge” or“envelope” o a crown, inside o which the cover

is thought to be continuous. In practice the “outeredge” is sometimes very dicult to observe.Another approach is not to consider the gapsinside the crowns as part o the canopy so thateach crown comprises only the leaves, branchesand stem and not the empty spaces between them.Rautiainen et al. (2005) introduced the concepto “eective canopy cover ” to distinguish themeasurements that do not include gaps in thecover rom measurements made according to thetraditional denition. This study mainly concernsthe measurement o traditional canopy cover, andwhen eective canopy cover is discussed it isalways mentioned separately.

In addition to an accurate denition o canopycover, the technique used to acquire the canopycover inormation is crucial. There are three alter-native approaches or obtaining this inormation:(1) eld measurements at the place o interest,(2) statistical models, i stand parameters suchas basal area, number o stems, and diameterat breast height are known rom the area, or(3) remotely sensed inormation such as aerialphotographs (Pitkänen 2001, Culvenor 2003),satellite images (Iverson et al. 1989, Gemmell1999) or laser scanner data (Næsset et al. 2004).Preliminary results indicate that predictions o canopy cover obtained with regression modelsbased on stand parameters, such as basal areaand breast height diameter, could be used i directmeasurements are not possible (Korhonen 2006).However, more data are needed to conrm these

results. Remote sensing o canopy cover may alsobecome a requently used method along with the

improved availability and reduced cost o accu-rate high resolution remote sensing materials andthe ast development o physically based orestrefectance models to extract the inormationembedded in this data. Nevertheless, eld meas-

urements are needed or testing and validating allremote sensing methods.

In most cases, eld measurements are the onlyway to dene the true vertical projection o acanopy. In structurally complex orests, indi-rect estimates achieved with regression modelsor remote sensing always include at least somerandom variation. On the other hand, obtainingaccurate (i.e. unbiased and precise) eld meas-urements o canopy cover is very time consum-ing, and attempts to save time usually lead to

inaccurate results. In Finland, the best-knowneld-based method is the Cajanus tube (Sarvas1953, Rautiainen et al. 2005), which can be usedto measure both the traditional and the eectivecanopy cover. The Cajanus tube is a simple sight-ing tube equipped with an internal mirror thatallows the observer to look upwards through thetube, and then estimate i the crosshair at the top o the tube points directly at part o the canopy. Thetube is placed on a holder, which allows the tubeto hang reely so that the measurement is madein direct vertical direction. A supportive stick isused to keep the holder steady. Instruments simi-lar to the Cajanus tube have been described byvarious authors (Walters and Soos 1962, Bonnor1967, Jackson and Petty 1973, Johansson 1984,Stump 1993), but they are all used in the sameway. Canopy cover is estimated by measuring agrid o points on the survey plot with the tube. Theresult o each individual measurement is recordedas 1 i the view is obstructed and 0 otherwise.Canopy cover is then estimated as the mean o these binomial (Bernoulli) variables. The varianceo this unbiased estimate o canopy cover (cc) isgiven by cc(1 – cc)/n, where n is the number o measurements. The variance estimator assumesthat measurements are not correlated. Thus, theestimator is biased or systematic grid designs.The number o measurements needed to obtaina specied accuracy thus depends on the canopycover itsel, but is independent o the area o theplot. In practice it has been observed that at least

200–250 points should be measured to arrive atstable estimates (Johansson 1984, Jennings et al.



580


1999, Rautiainen et al. 2005).The standard alternative to point-based sam-

pling is called line intersect sampling or LIS(O’Brien 1989, Jennings et al. 1999, Williams etal. 2003). Instead o a grid, transects (or ‘lines’)

are used. The points on the lines where canopycover begins and ends are recorded using a tapemeasure and the Cajanus tube, ater which per-cent canopy cover can be calculated as a ratioo the length o the transect covered by canopyand the ull length o the transect. The lines tobe measured should cover the entire plot, but theplacement o the lines can be either systematicor random (Jennings et al. 1999, Williams et al.2003).

According to the denition, i unbiased canopy

cover estimates are desired, instruments with anangle o view should not be used. However, i theangle o view is narrow, the bias is not signicant(Garrison 1949, Bonnor 1967, Bunnell and Vales1990). These results suggest that instrumentswith narrow angles o view can be successullyused in the estimation o canopy cover, and theyalso speed up the measurement considerably,as the number o individual measurements canbe decreased when a larger area is observed ateach measurement point. For methods that use anangle o view wider than 30 degrees, the bias incanopy cover becomes signicant (Bunnell andVales 1990, Ganey and Block 1994, Cook et al.1995), and these methods should only be usedor estimating canopy closure. In the estimationo canopy closure, it is best by denition to usean angle o view o 180 degrees (Jennings et al.1999). When this is not possible, the angle o viewused in the measurement should be reported.

Cameras (Kuusipalo 1985, Strandström 1999)and spherical densiometers (Lemmon 1956) havebeen the most popular angle-o-view instrumentsin orest canopy measurements. When estimatingcanopy closure, the canopy is photographed whilethe camera is kept in a vertical position. The dig-ital or scanned camera images are rst convertedto binary images so that the canopy and the skyare shown in dierent colours, usually so that thecanopy is black and the sky is white. Next, thepercentage o black and white pixels in the binaryimage can be calculated (Teraoka 1996, Strand-

ström 1999, Ishida 2004). Also the gaps inside thecrowns are observed, which means that these gaps

displayed in the image should be painted overwith black or better estimates o canopy coverin its traditional sense. The spherical densiometer(Lemmon 1956) consists o a small wooden boxwith a convex or concave mirror, engraved with

24 squares, placed in it. The densiometer is usedby holding it at breast height so that the observer’shead is refected rom the edge o the mirror justoutside the graticule. The curved mirror refectsthe canopy above, and canopy closure can beestimated by calculating the number o squares(or quarters o squares) that the image o thecanopy covers.

I no other instruments are available, ocularestimation o canopy cover (Sarvas 1953, Bunnell& Vales 1990) is an alternative. However, ocular

estimates are always subjective, and the resultscan vary even with changing weather (Jenningset al. 1999). Objectivity can be increased in theprocess by dividing the plot into smaller sec-tions and counting the average o estimates madeor each section (Sarvas 1953, Bunnell & Vales1990). This will, however, increase the neededtime, and results would probably be more consist-ent i an instrument was available or measuringthe subplots.

The impetus behind this study is to nd a suit-able canopy cover estimation method or theFinnish national orest inventory (NFI). Althoughmany dierent methods exist, none is suitable orlarge-scale inventories, which require an inexpen-sive method that allows quick and accurate esti-mation o canopy cover. An ideal method shouldsatisy all o these conditions. In addition, a bal-ance between costs, speed and accuracy should bepossible. In this study, the objectives are to test theground-based techniques described above, and toevaluate their perormance in estimating canopycover in Scots pine (Pinus sylvestris L.) andNorway spruce (Picea abies (L.) Karst) stands.

2 Materials and Methods

The data used or comparing ground measurementtechniques were collected during summer 2005at Suonenjoki, central Finland. There were two

separate study sites located approximately 20 kmapart: the Hirsikangas site (62˚38´N, 27˚01´E)



581


and Saarinen site (62˚40´N, 27˚29´E). The plotsat Hirsikangas site were mostly rather poor,Scots pine dominated heaths, whereas the plotsat Saarinen site were more ertile and dominatedby Norway spruce. The plots were located so thatthe data would be as diverse as possible, with theollowing conditions: 1) the dominant tree specieshad to be either Scots pine or Norway spruce, 2)the site type had to be a heath, and 3) tree height inthe stand had to be at least two meters. Altogether19 plots were measured, o which ten were pinedominated and nine spruce dominated. In mixedstands, the species with the largest basal area wasconsidered as the main species.

For all the plots, canopy cover was estimatedwith the Cajanus tube, LIS, convex spherical den-siometer, and digital camera images. In addition,ocular estimates were used in the comparison.The Cajanus tube results were used as controlvalues. For the Cajanus tube and the densiometer,dierent sample sizes or the number o measuredpoints were also compared. The canopy cover

estimates were gathered rom circular plots witha radius o 12.52 meters, which corresponds to the

maximum radius o the relascope plots used in theNFI. The circular plot included 195 measurementpoints in a 1 m × 2.5 m grid (Fig. 2).

In each plot, all 195 points were measured withthe Cajanus tube, and the control values or com-parison with other methods were calculated as anaverage o these points. To examine the eect o sample size on Cajanus tube results, samples o 102, 49, and 23 were also picked rom the data;these sample sizes correspond to selecting everysecond, ourth or eighth point on each transect.

During the eld campaign, several decisionsconcerning special situations had to be made. I a measurement point was situated on the edge o a tree crown, the decision as to whether the pointwas covered was made according to whether mosto the tube’s eld o view was below a crownor below the sky. Especially in sapling stands,it happened occasionally that at the eye level(1.7 meters), where the measurement was madethrough the tube, only sky was visible, but thetwigs o a close sapling extended to the measure-

ment point below eye level. In such a situation, thepoint was considered to be covered, i the sapling

Fig. 2. Plot design with 12.52 m radius and 1 m × 2.5 m measurement point grid.



582


was taller than 1.3 meters. Seedlings shorter than1.3 meters were classied as undergrowth anddid not infuence the canopy cover estimate. Allliving twigs and branches o trees taller than 1.3meters were taken into account when measuring

the cover, no matter how sparse a cover they cre-ated. Dead branches were ignored in the measure-ments, unless they signicantly hindered the lightcoming rom the sky. For example, i the pointwas situated under a dead spruce whose crownstill intercepted most o the light, the point wasclassied as covered.

The same transects located inside the circularplot (Fig. 2) were also used in LIS and sphericaldensiometer measurements. In LIS, the beginningsand ends o connected crowns were recorded with

an accuracy o 0.1 meter. The spherical densiom-eter, on the other hand, was used at the same 49and 23 points grids as the Cajanus tube. For thisstudy, a new modication o the densiometer wasdesigned to decrease the instruments angle o view rom the original 60 degrees. In the modi-cation only the our squares out o 24 that werelocated closest to the observer were used. Theseour were selected so that the light coming romthe zenith would refect in the direction o theobserver’s eyes; at the centre o the convex mirror,the light coming rom above refects back to thezenith. With this modication, the angle o viewcould be reduced to a third o the original (i.e.to about 20 degrees). Along with the systematicdensiometer measurements, a subjective sampleo ten points was tested as a aster alternative.In this measurement scheme ten representativepoints were selected by the measurer.

Digital photographs were taken with a standarddigital camera, with an angle o view o about55 degrees. Five photographs were taken at eachplot – one at the centre and the other our incardinal points at an 8.5 meters distance romthe centre. The original images were convertedto binary images with a standard image analysissotware. Because the camera method actuallymeasures canopy closure, which correlates betterwith eective canopy cover, the crowns in binaryimages were painted black so that the result wouldbe closer to the traditional canopy cover esti-mate. However, the results achieved with both

painted and unpainted images were used in thecomparison.

The ocular estimation o canopy cover was car-ried out by three observers. Observer A estimatedthe percent canopy cover by eyesight beore meas-uring the control value with the Cajanus tube, andwas thus able to learn rom previously measured

plots. Observers B and C were experienced or-estry proessionals, who had been making similarestimates throughout the previous summer. Theyvisited the plots ater the eld campaign hadended and were not given eedback until estimateswere obtained or all plots.

In addition to the canopy cover measurements,routine stand inventory parameters were alsomeasured at each plot to characterize the struc-ture o the stands (Table 1.). The parametersincluded plot coordinates, soil type and site class

classication, stand age, basal area, stand densityby size class, diameter at breast height (DBH),tree height and crown length. The data includedour sapling stands and also some mixed orests(species: spruce, pine, birch and other deciduousspecies).

The comparison o measurement techniquesand sample sizes was based on the arithmeticdierences between the results acquired with agiven method and the control values (i.e. Cajanustube measurements rom 195 points). Becausethe comparison was made using the dierencesobtained rom dierent plots, the results romdierent measurements could be handled as inde-pendent samples. The mean dierence indicateswhether the method is biased: a positive meandierence indicates that the method producesoverestimates, and respectively, a negative meandierence indicates an underestimate. The stand-ard deviation o the dierences describes theprecision o the method, i.e. how much the resultstypically dier rom the mean. An ideal methodwould thus be unbiased and precise (i.e. accurate),and, simultaneously, quick and inexpensive. Theproblem with using arithmetic means and stand-ard deviations was that or most methods, thedistributions o the dierences were not normal;instead, they were usually skewed and includedsome outliers. Thereore medians and quartileswere also used as supplemental measures. Thebox-and-whiskers plot (Moore and McCabe 1993)is one useul median-based method or illustrating

the distributions.The statistical signicance o the observed



583


dierences was tested with the nonparametricKruskall-Wallis analysis o variance (Zar 1984, p.176). Analysis o variance allows comparison o multiple methods in a single test, which decreasesthe probability o nding a statistically signicantdierence by chance when it in act does notexist; this would probably happen i a series o T-tests were used instead o an analysis o vari-ance. A nonparametric test was chosen becausethe dierences were not normally distributed. Thenonparametric Kruskall-Wallis test uses ranksrather than the actual dierences. This decreasesthe infuence o skew distributions and outliervalues on the results. Since the output o the testshows whether a signicant dierence exists,multiple comparisons were used to dene whichmethods actually diered rom the control. Themultiple comparisons used nonparametric com-parisons o a control to other groups (Zar 1984, p.201). The test results or each method were thencompared to a critical value obtained rom Q-table(Zar 1984, p. 569), and i the test coecient waslarger than the critical value, the dierence wasconsidered statistically signicant.

3 Results

The main dierences between the measurementmethods can be easily seen rom the box-and-whiskers plot (Fig. 3). The detailed data o themeans, medians, standard deviations etc. are pre-sented in Table 2.

There are statistically signicant dierencesbetween the results produced by dierent meas-urement methods (Fig. 3, Table 2). Some meth-ods, like the Cajanus tube with a sample size o 102 and LIS, produce unbiased results, whereassome o the others, like the ocular method andunpainted digital photographs, provide very largeunderestimates. Also precision varies consider-ably: the Cajanus tube with a sample size o 102 and the LIS-method have very small stand-ard deviations, whereas the ocular estimates anddigital photographs yield signicantly dierentresults at dierent plots.

The analysis o variance conrms the conclu-sions made rom looking at Fig. 3: the initialhypothesis o the equality o medians is rejectedas the test p-value is remarkably small (testing:

χ2

-test coecient = 59.5, d.. = 16, P < 0.01). Themultiple comparisons (Table 3) reveal that the

Table 1. Key inormation o the data.

Plot Dominant Canopy Basal area Stand Height Diameterspecies cover (m2 /ha) density (m) (cm)

(stems/ha)

1 Pine 0.42 19.2 690 26.4 34.6

5 Pine 0.48 14.0 390 16.1 20.57 Pine 0.32 13.6 260 18.7 25.78 Pine 0.83 29.2 2500 15.0 13.39 Pine 0.68 27.2 480 21.0 25.210 Pine 0.72 18.4 3200 13.4 13.411 Pine 0.66 11.4 1600 10.3 11.013 Spruce 0.59 25.6 580 24.2 26.114 Pine 0.47 18.8 490 17.7 22.118 Spruce 0.63 26.8 610 25.6 27.322 Pine 0.51 9.6 2700 5.8 6.324 Pine 0.71 16.4 4700 6.1 6.639 Spruce 0.74 24.8 1800 14.9 17.540 Spruce 0.77 9.0 16000 6.2 6.544 Spruce 0.56 19.6 1100 19.8 21.945 Spruce 0.89 26.0 1700 12.9 15.247 Spruce 0.81 23.4 2300 12.9 13.354 Spruce 0.51 21.2 1200 21.0 23.756 Spruce 0.34 1.0 5100 3.4 3.3



584


Ocular C

Ocular B

Ocular A

Bl. photos

Photos

Dens. 10 subj.

Dens. 9

Dens. 23

Dens. 49

LIS

Caj. 23

Caj. 49

Caj. 102

Method

0.2

0.1

0.0

–0.1

–0.2

–0.3

–0.4

–0.5

D i f f e r e n c e

22

13

40

565656

40

56

56

56

Fig. 3. A box-and-whiskers plot or comparison o the measurement methods andused sample sizes. The numbers close to outlier values reer to individual plots(Table 1).Abbreviations: Caj. = Cajanus tube (ollowed by sample size), LIS = line inter-sect sampling, Dens. = densiometer (ollowed by sample size; subj. = subjectivesample), Photos = unpainted digital photographs, Bl. photos = black-painted digitalphotographs, ocular A–C = ocular estimates by the three dierent observers.

Table 2. Inormation on the dierences between the results o the control method and each evaluated measure-ment technique.

Method N Mean Median Std. dev. Quartile range Min Max

Cajanus tube 102 points 19 0.004 0.007 0.015 0.015 –0.034 0.030Cajanus tube 49 points 19 –0.029 0.011 0.094 0.106 –0.277 0.079Cajanus tube 23 points 19 –0.055 –0.022 0.117 0.099 –0.339 0.060LIS 19 –0.003 –0.006 0.026 0.038 –0.053 0.040Densiometer 49 points 19 –0.001 0.017 0.097 0.094 –0.300 0.150Densiometer 23 points 19 –0.013 0.023 0.111 0.126 –0.338 0.155

Densiometer 9 points 19 –0.019 –0.025 0.119 0.146 –0.338 0.173Densiometer 10 pointssubjective sample 19 –0.056 –0.035 0.081 0.113 –0.255 0.055

Digital photographs 18 –0.143 –0.101 0.140 0.212 –0.407 0.099Black-painted digital

photographs 18 –0.004 0.020 0.123 0.144 –0.277 0.161A’s ocular estimate 14 0.008 0.001 0.080 0.136 –0.132 0.137B’s ocular estimate 19 –0.064 –0.042 0.089 0.152 –0.248 0.085C’s ocular estimate 19 –0.162 –0.178 0.104 0.117 –0.362 0.065

Note: negative mean and median values indicate underestimates



585


methods yielding signicantly dierent resultsrom the control values are (in decreasing order):ocular estimates by observer C, unpainted dig-ital photographs, ocular estimates by observerB, and the densiometer measurements with tensubjectively chosen points. All the signicantdierences were underestimates.

The eect o sample size on the results acquiredwith the Cajanus tube was as expected; the limitunder which the estimates become imprecise isbetween 102 and 49 points. This result is dier-ent rom the previous results that 200–250 pointsshould be measured (Johansson 1984, Jenningset al. 1999, Rautiainen et al. 2005), as the 102points (i.e. every second point rom the circularplot) seem to be enough or this measurementscheme. The maximum dierence compared tousing every point was approximately three per-cent, and the arithmetic mean was practicallyunbiased. When the sample size was decreasedto 49 or 23 points, the standard deviations roseto approximately ten percent, and, in addition,some outlier values appeared. The outliers wereplots 40 and 56 (Fig. 3), i.e. two spruce saplingstands with short average tree heights (Table 1).This indicates that a sparse grid could not nd allthe trees on these plots. It should also be notedthat the dierence in results or grids o 49 and23 points was negligible.

The LIS-method proved to be another accuratemethod or measuring canopy cover. The aver-

age dierence between LIS and the control was2.5%, and never more than 5%. Both the Cajanustube, with sample sizes o 195 and 102, and LISwere precise and unbiased methods or estimat-ing canopy cover, and they also provided verysimilar results or each plot. The well-knowndisadvantage o these methods is that they aretime consuming: measuring a plot with 195 pointsusually takes, depending heavily on the terrain,more than an hour.

The modied spherical densiometer acted asa compromise between measurement speed andaccuracy. In practice, the results acquired withsystematic sampling seem to be unbiased, eventhough in theory the results should be overes-timates because the densiometer measurementinvolves an angle view. A closer inspection o the data revealed that at a considerable number o plots, the densiometer results were 0–10% higherthan the control values, but in the whole data thedierences were not statistically signicant. Thestandard deviation was approximately ten percentand interquartile range also comparable to theother methods. Decreasing the sample size to 23points or only nine points weakened the results,but there were no signicant dierences betweenthe estimates obtained with smaller sample sizes:the results were slight overestimates with stand-ard deviations o 11–12% and somewhat larger

interquartile ranges. The most distinctive eatureo the densiometer method is that it is not suit-

Table 3. Multiple comparisons.

Method N Mean Dierence Standard Testrank rom control error coecient

Cajanus 195 points (control) 19 182.0 0.0 14.82 0.00Cajanus tube 102 points 19 193.5 11.5 14.82 0.77

Cajanus tube 49 points 19 165.9 –16.1 14.82 –1.08Cajanus tube 23 points 19 144.3 –37.7 14.82 –2.54LIS 19 176.7 –5.3 14.82 –0.35Densiometer 49 points 19 196.5 14.5 14.82 0.98Densiometer 23 points 19 188.2 6.2 14.82 0.42Densiometer 9 points 19 167.3 –14.7 14.82 –0.99Densiometer 10 points subjective sample 19 125.3 –56.7 14.82 –3.83*Digital photographs 18 78.1 –103.9 15.02 –6.92*Black-painted digital photographs 18 189.7 7.7 15.02 0.51A’s ocular estimate 14 187.3 5.3 15.91 0.34B’s ocular estimate 19 120.7 –61.3 14.82 –4.14*C’s ocular estimate 19 58.9 –123.1 14.82 –8.31*

* Statistically signicant dierence at α=0.05 (critical value 2.955).Note: test coecient with negative sign denotes an underestimate.



586


able or stands with tree heights less than vemeters, because the shorter trees are more di-cult to observe. Because the measurement ismade at breast height, the angle o view givesonly little advantage in nding the shorter trees,

i.e. the sample point should be almost inside thesapling canopy to be able to create any cover. Thisexplains the outlier values related to densiometerresults in Fig. 3. I all trees need to be sighted, it isbetter to use the Cajanus tube and a dense grid.

An alternative, quicker approach in using thedensiometer was subjective sampling with onlyten measurement points. This implementationmethod resulted in statistically signicant under-estimates, as the mean and median results wereunderestimates o 5% and 3.5%, respectively.

Selection o measurement points was not an easytask. In heterogeneous mature stands it was otendicult to assess how many points should belocated in large gaps between trees and how manyin dense spots. As the results show, the pointswere typically more oten located in open placesor gaps between the trees, which led to under-estimates. However, the standard deviation andinterquartile range were better than with system-atic measurements, and in more homogenous pinestands the results were practically unbiased. Thisindicates that in some stands subjective densiom-eter sampling could be used as a quick alternativeto the more time consuming methods: the subjec-tive densiometer sampling typically takes aboutve minutes, whereas measuring a systematicgrid o 49 points takes approximately 30 minutes.Because the densiometer results could still easilydier ten percent rom the actual cover even with49 points sample size, using instead the Cajanustube or the LIS method is more advisable.

From the estimates based on digital photogra-phy a clear conclusion can be made: i canopycover estimates are desired, the gaps inside treecrowns must be painted non-transparent to avoidunderestimates. The mean and median values o canopy cover obtained rom unpainted imageswere 15% and 10% smaller than the controlvalues, i.e. statistically signicant underestimates.As the images were painted, the dierences werereduced to 0 and +2%, i.e. practically unbiased.Those values are surprisingly small, considering

the airly large angle o view (55 degrees) whichtheoretically should provide clear overestimates.

It is probable that the painting o small gapswithin the crowns did not quite correspond to theclassication o crown/sky in the eld measure-ments. Another explanation or the negative biasis that the method did not work in sapling stands

where underestimation occurred. In addition, theestimated mean and median values are subject torandom variation due to limited numbers o plotsand airly large standard deviations. The digitalphotography was tested mainly as another quickalternative or more accurate methods, similar tothe subjective densiometer sampling. The conclu-sion was that painted photographs can be used i inormation or eective canopy cover or canopyclosure is needed in addition to canopy cover.Even then it would be more sensible to take more

photographs and use a narrower angle o view, i the main interest is in canopy cover. Otherwise,some other method would probably be a betterchoice.

The ocular estimates o the observers B and Cwere also statistically signicant underestimates:C’s mean and median results diered rom thecontrol values by –16% and –18%, and B’s by–6% and –4%, respectively. However, the esti-mates made by A prior to starting the actual meas-urement were practically unbiased. The standarddeviation and interquartile range were approxi-mately the same or each observer. This indicates,as expected, that the ocular estimates achievedwithout proper training might be seriously biasedin either direction, but underestimation seems tobe the most common mistake. However, as A’sresults show, ater some experience and knowl-edge o true canopy cover in dierent stands, thebias o the results should decrease considerably.However, increasing the precision o estimates isprobably more dicult.

4 Discussion

The results o the comparison o measurementtechniques conrm that the conventional methodscannot quickly provide accurate canopy coverestimates; the conclusion is essentially the sameas Sarvas (1953) made more than ty years ago.

I accurate values are desired, unbiased meth-ods such as the Cajanus tube or LIS should be



587


used, and an adequate amount o time shouldbe reserved or eld measurements. I time is alimiting actor, subjective densiometer samplingcould be a possible alternative in mature stands.I even ve minutes per plot is regarded as too

long time or the measurements, the only pos-sibility is to rely on ocular estimates. However,ocular estimation requires training with careullymeasured training sites and careul monitoring o the results. When canopy cover data are gatheredor modelling purposes, using accurate but time-consuming methods is the only alternative.

In the near uture, new methods o canopycover estimation are needed to better satisy thegrowing need or canopy-related inormation.Geographically representative regression models

might be the most cost-eective and easible solu-tion or obtaining canopy cover estimates or largeareas. However, substantial amounts o resourcesand research eorts are required i nation-widemodels are to be developed. In the long run,remote sensing may become a more popular andappropriate means or canopy cover estimation.Nevertheless, or the development and validationo remote sensing techniques, reliable groundtruth measurements and models o canopy coverneed to be obtained and tested rst.

Acknowledgements

This study has been supported by the Universityo Helsinki research unds and the Finnish ForestResearch Institute. The authors wish to thank Mr.Pekka Voipio or assisting in the eld work, Mr.Esko Valtonen or his help in the statistical analy-sis and the reviewers or their valuable commentson the manuscript.

References

Anderson, M.C. 1964. Studies o the woodland lightclimate: I. The photographic computation o lightconditions. Journal o Ecology 52(1): 27–41.

Bolduc, P., Lowell, K. & Edwards, G. 1999. Automatedestimation o localized orest volume rom large-scale aerial photographs and ancillary cartographic

inormation in a boreal orest. International Journalo Remote Sensing 20(18): 3611–3624.

Bonnor, G.M. 1967. Estimation o ground canopydensity rom ground measurements. Journal o Forestry 65(8): 544–547.

Bunnell, F.L. & Vales, D.J. 1990. Comparison o methods or estimating orest overstory cover:

dierences among techniques. Canadian Journalo Forest Research 20: 101–107.

Cook, J.G., Stutzman, T.W., Bowers, C.W., Brenner,K.A. & Irwin, L.L. 1995. Spherical densiometersproduce biased estimates o orest canopy cover.Wildlie Society Bulletin 23(4): 711–717.

Culvenor, D.S. 2003. Extracting individual tree inor-mation. A survey o techniques or high spatialresolution imagery. In: Wulder, M.A. & Franklin,S.E. (eds.). Remote sensing o orest environments:concepts and case studies. Kluwer Academic Pub-

lishers, Boston. 519 p.FAO 2000. On denitions o orest and orest change.

Forest Resources Assessment Programme, Work-ing Paper 33. FAO, Rome, Italy. 15 p.

Ganey, J.L. & Block, W.M. 1994. A comparison o twotechniques or measuring canopy closure. WesternJournal o Applied Forestry 9(1): 21–23.

Garrison, G.A. 1949. Uses and modications or the“moosehorn” crown closure estimator. Journal o Forestry 47: 733–735.

Gemmell, F. 1999. Estimating conier orest coverwith thematic mapper data using refectance modelinversion and two spectral vegetation indices ina site with variable background characteristics.Remote Sensing o Environment 69: 105–121.

Ishida, M. 2004. Automatic thresholding or digitalhemispherical photography. Canadian Journal o Forest Research 34: 2208–2216.

Iverson, L.R., Cook, E.A. & Graham, R.L. 1989. Atechnique or extrapolating and validating orestcover across large regions – calibrating AVHRRdata with TM data. International Journal o RemoteSensing 10(11): 1805–1812.

Jackson, M. & Petty, R. 1973. A simple optical deviceor measuring vertical projection o tree crowns.Forest Science 19: 60–62.

Jasinski, M.F. 1990. Sensitivity o the normalized di-erence vegetation index to subpixel canopy cover,soil albedo, and pixel scale. Remote Sensing o Environment 32: 169–187.

Jennings, S.B., Brown, N.D., & Sheil, D. 1999. Assess-ing orest canopies and understorey illumination:

canopy closure, canopy cover and other measures.Forestry 72(1): 59–74.



588


Johansson, T. 1985. Estimating canopy density by thevertical tube method. Forest Ecology and Manage-ment 11: 139–144.

Kalliovirta, J. & Tokola, T. 2005. Functions or estimat-ing stem diameter and tree age using tree height,

crown width and existing stand database inorma-tion. Silva Fennica 39(2): 227–248.

Knyazikhin, Y., Martonchik, J.V., Myneni, R.B., Diner,D.J. & Running, S.W. 1998. Synergistic algorithmor estimating vegetation canopy lea area indexand raction o absorbed photosynthetically activeradiation rom MODIS and MISR data. Journal o Geophysical Research 103: 32257–32276.

Korhonen, L. 2006. Havumetsän latvuspeiton mittaa-minen ja ennustaminen puustotunnuksista. M. Sc.thesis. University o Joensuu, Faculty o Forestry.

71 p. Available at: http://cc.joensuu./~lakorhon/ latvuspeitto.pd. (In Finnish).

Kuusipalo, J. 1985. On the use o tree stand parametersin estimating light conditions below the canopy.Silva Fennica 19(2): 185–196.

Kuusk, A. & Nilson, T. 2000. A directional multispec-tral orest refectance model. Remote Sensing o Environment 72: 244–252.

Lemmon, P.E. 1956. A spherical densiometer or esti-mating orest overstory density. Forest Science 2:314–320.

Lowman, M.D. & Rinker, H.B. (eds.) 2004. Forest can-opies. 2nd ed. Elsevier Academic Press. 517 p.

Maltamo, M., Eerikäinen, K., Pitkänen, J., Hyyppä, J.& Vehmas, M. 2004. Estimation o timber volumeand stem density based on scanning laser altim-etry and expected tree size distribution unctions.Remote Sensing o Environment 90: 319–330.

Moore, D.S. & McCabe, G.P. 1993. Introduction to thepractice o statistics. 2nd ed. W.H. Freeman andcompany, New York. 854 p.

Næsset, E., Gobakken, T., Holmgren, J., Hyyppä, H.,Hyyppä, J., Maltamo, M., Nilsson, M., Olsson, H.,Persson, Å. & Söderman, U. 2004. Laser scanningo orest resources: the Nordic experience. Scandi-navian Journal o Forest Research 19: 482–499.

Nilson, T. & Peterson, U. 1991. A orest canopy refect-ance model and a test case. Remote Sensing o Environment 37:131–142.

Nuttle, T. 1997. Densiometer bias? Are we measuringthe orest or the trees? Wildlie Society Bulletin25(3): 610–611.

O’Brien, R.A. 1989. Comparison o overstory canopyestimates on orest survey plots. USDA ForestryService General Technical Report INT-GTR-417.5 p.

Pitkänen, J. 2001. Individual tree detection in dig-

ital aerial images by combining locally adaptivebinarization and local maxima methods. CanadianJournal o Forest Research 31: 832–844.

Rautiainen, M., Stenberg, P. & Nilson, T. 2005. Esti-mating canopy cover in Scots pine stands. SilvaFennica 39(1): 137–142.

Sarvas, R. 1953. Measurement o the crown closureo the stand. Communicationes Instituti ForestalisFenniae 41(6). 13 p.

Spanner, M.A., Pierce, L.L., Peterson, D.L. & Run-ning, S.W. 1990. Remote sensing o temperate

conierous lea area index: The infuence o canopyclosure, understory vegetation, and backgroundrefectance. International.Journal o Remote Sen-sing 11(1):95–111.

Strandström, M. 1999. Landsat TM -kuvalta estimoi-tavien latvuspeiton ja lehtipinta-alaindeksin käyt-tökelpoisuus metsän kaukokartoituksessa. M. Sc.thesis. University o Helsinki, Faculty o Agricul-tural Sciences and Forestry. 55 p. (In Finnish).

Stump, K.A. 1993. The estimation o orest vegetationcover descriptions using a vertical densiometer.[Internet site]. Available at: http://www.grsgis.com/ publications/sa_93.html. [Cited 22 Mar 2006].

Teraoka, Y. 1996. An eective mensuration techniqueor multipurpose management using sh-eye pho-tographic images. In: Saramäki, J., Koch, B. &Lund, H.G. (eds.). Remote sensing and computertechnology or natural resource assessment. Uni-versity o Joensuu, Faculty o Forestry. ResearchNote 48.

Walters, J & Soos, J. 1962. The gimbal sight or theprojection o the crown radius. Faculty o For-estry, University o British Columbia. ResearchNote 39.

Williams, M.S., Patterson, P.L. & Mowrer, H.T. 2003.Comparison o ground sampling methods orestimating canopy cover. Forest Science 49(2):235–246.

Total o 38 reerences

comparison of canop

Documents