1 S ILVA F ENNICA Silva Fennica vol. 50 no. 1 article id 1518 Category: research note www.silvafennica.fi ISSN-L 0037-5330 | ISSN 2242-4075 (Online) The Finnish Society of Forest Science Natural Resources Institute Finland Francesco Chianucci A note on estimating canopy cover from digital cover and hemispherical photography Chianucci F. (2016). A note on estimating canopy cover from digital cover and hemispherical photography. Silva Fennica vol. 50 no. 1 article id 1518. 10p. Highlights • Comparison of fisheye (DHP) and cover (DCP) photography for estimating canopy cover. • Digital photographic estimates validated against artificial images with known cover. • Accuracy of cover estimates from DHP is influenced by mean gap size and actual cover. • Accuracy of cover estimates from DCP is not influenced by mean gap size and actual cover. Abstract Fast and accurate estimates of canopy cover are central for a wide range of forestry studies. As direct measurements are impractical, indirect optical methods have often been used in forestry to estimate canopy cover. In this paper the accuracy of canopy cover estimates from two widely used canopy photographic methods, hemispherical photography (DHP) and cover photography (DCP) was evaluated. Canopy cover was approximated in DHP as the complement of gap fraction data at narrow viewing zenith angle range (0°–15°), which was comparable with that of DCP. The methodology was tested using artificial images with known canopy cover; this allowed exploring the influence of actual canopy cover and mean gap size on canopy cover estimation from photog- raphy. DCP provided robust estimates of canopy cover, whose accuracy was not influenced by variation in actual canopy cover and mean gap size, based on comparison with artificial images; by contrast, the accuracy of cover estimates from DHP was influenced by both actual canopy cover and mean gap size, because of the lower ability of DHP to detect small gaps within crown. The results were replicated in both DHP and DCP images collected in real forest canopies. Finally, the influence of canopy cover on foliage clumping index and leaf area index was evaluated using a theoretical gap fraction model. The main findings indicate that DCP can overcome the limits of indirect techniques for obtaining unbiased and precise estimates of canopy cover, which are comparable to those obtainable from direct, more labour-intensive techniques, being therefore highly suitable for routine monitoring and inventory purposes. Keywords foliage cover; crown cover; forest canopy; fisheye photography; cover photography Addresses Consiglio per la Ricerca in Agricoltura e l’Analisi dell’Economia Agraria – Forestry Research Centre, viale Santa Margherita 80, 52100 Arezzo, Italy E-mail [email protected]Received 4 November 2015 Revised 2 December 2015 Accepted 7 December 2015 Available at http://dx.doi.org/10.14214/sf.1518
10
Embed
A note on estimating canopy cover from digital cover and ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
SILVA FENNICASilva Fennica vol. 50 no. 1 article id 1518
The Finnish Society of Forest Science Natural Resources Institute Finland
Francesco Chianucci
A note on estimating canopy cover from digital cover and hemispherical photography
Chianucci F. (2016). A note on estimating canopy cover from digital cover and hemispherical photography. Silva Fennica vol. 50 no. 1 article id 1518. 10p.
Silva Fennica vol. 50 no. 1 article id 1518 · Chianucci · A note on estimating canopy cover from digital cover…
1 Introduction
Canopycoverisacommonlyusedvariableinforestry(Jennings1999;Rautiainenetal.2005).ThisvariableisstronglyrequiredforaccuratemodellingofleafareaindexL using radiative transfer theory (Majasalmietal.2014;Nilson1999;NilsonandKuusk2004).Inaddition,canopycoverisamajordeterminantofforestreflectancefromopticalremotesensingdata(Dawsonetal.1999).It is also often included in national forest inventories (Angelini et al. 2015). Accordingly, accurate in situ estimatesofcanopycoverarecentralforawiderangeofforestrystudies.
As direct measurements are impractical and time-consuming, canopy cover was oftenestimatedfromopticalinstrumentsusingarestrictedzenithanglerange(typicallyabout0°–15°).Althoughtheapproachissomewhatincorrect,inthatmeasurementsarenotstrictlyvertical,useofnarrow-angleofviewisoftenconsideredduetoitssimplicity.Forexample,thewidelyusedhemisphericalsensorssuchasLAI-2000PlantCanopyAnalyzer(Li-COR,Lincoln,NE,USA)ordigitalhemisphericalphotography(alsocalledfisheyephotography;DHP)havebeenfrequentlyemployedtoobtainanestimateofcanopycoverfromgapfractiondataatnarrowviewingzenithanglerange(Korhonenetal.2006;Kuchariketal.1999;Rautiainenetal.2005;SeedandKing2003).However,thegapfractionreadingsobtainedatthisviewweretypicallynoisyinhemispheri-calsensors,becauseoflimitedspatialsamplingatthisview.Whilesomestudieshaveproposedcorrectionforinsufficientsamplingusingtheseinstruments(e.g.,NilsonandKuusk2004),Mac-farlaneetal.(2007)recentlyproposeddigitalcoverphotography(DCP),arestricted-viewanglemethod.Theauthorsuseda70mmequivalent-focal-lengthtoobtainveryfinespatialresolutionatanapproximately30°fieldofview(FOV),whichwascomparablewiththeFOVoftheuppermostringofLAI-2000.TheresultingcombinationofhighresolutionandmainlyverticalsamplinginDCPallowedtoseparatetotalgapfractionintolarge,between-crownsgapsandsmall,within-crowngaps,leadingtodistinctestimatesofcanopycover(crowncoverandfoliagecover;Macfarlaneetal.2007,butseealsosection2.1).However,althoughsomestudiesspeculatedthatDCPcanyieldmoreaccurateestimateofcanopycoverthanDHP(PekinandMacfarlane2009;Ryuetal.2010),owingmainlytoitshigherzenithalresolution,nopreviousstudieshaveevaluatedtheaccuracyofcoverestimatesfromboththesemethodscomparedwithknownreferencecanopycoverdata.
Theobjective of this studywas testing the accuracyof digital canopyphotography forestimatingforestcanopycover.Forthepurpose,estimatesobtainedfromcoverandhemispheri-calphotographywerecomparedwiththoseobtainedfromanartificialtargetwithknowncanopycover;thisallowedexploringtheinfluenceofactualcanopycoverandmeangapsizeoncanopycover estimation from digital photography.Resultsfrombothphotographicmethodswerealsocomparedinrealforestcanopies.Finally,theimpactofcanopycoveronmodellingcanopyattrib-uteslikeclumpingindexandleafareaindexwasillustratedusingatheoreticalgapfractionmodel(Nilson1999).
Silva Fennica vol. 50 no. 1 article id 1518 · Chianucci · A note on estimating canopy cover from digital cover…
thegapareathresholdwassetbasedonapreviousstudy(Chianuccietal.2014).ConsistentlywiththeterminologyofMacfarlaneetal.(2007),crowncoverwasestimatedasthefractionofpixelsthatdonotlieinbetween-crownsgapsandfoliagecoverwasestimatedasthecomplementof total gap fraction.
whereN is tree density (trees m–2), S( θ ) is the area of projection of the average tree crownenvelopeat thezenithangleθ, c( θ ) isanauxiliaryparameterwhichcorrects themeancrowncoverage, P1( θ )thegapfractionwithinasinglecrown,G( θ ) is the foliage projection function, L istheleafareaindex,BAIisthebranchareaindex,GIistheFisher’sgroupingindexoftreedistributionpattern.Themodelestimatesclumpingindex(Ω,Eq.4)andleafarea(L, Eq. 5) as:
Ω(θ ) = c(θ )NS(θ )cosθG(θ )(L+ BAI )
(4)
and
L = −N(κ +α )G(θ )
S(θ )× ln 1−1− exp (1−GI ) lnP(θ )NS(θ )⎡⎣ ⎤⎦
1−GI⎡
⎣⎢⎢
⎤
⎦⎥⎥0
π /2
∫ cosθ sinθdθ (5)
whereκ istheshoot-levelclumpingindexandα isthebranchareatoleafarearatio.Toapplytheformulas, in addition to the estimated value of P( θ ), the model requires input parameters includ-ingstanddensity,treeheight,crowndepth,shoot-levelclumping,foliageprojectionfunctionandcanopycover.Themodelwasruninthebirch(Betula pendula Roth)standinJärvselja,Estonia,belongingtotheRAMI(RadiationtransferModelIntercomparison)testsites,whichhavebeenwidelyusedtobenchmarkingradiativetransfermodelling(e.g.Kuusketal.2009;Piseketal.2011).Alltheinputparameters(includingcanopycover)wereobtainedfromtheworksbyKuusketal.(2009)andNilsonetal.(2011);theinputcanopycoverwasthenvariedtoevaluatetheinfluenceoftheseattributesonLandΩ,theotherinputparametersheldconstant.Ascanopycoverdirectlyinfluencesthegroupingindex(GI),theinputcanopycoverwasvariedwithinarangesuitabletoallowthetheoreticalformulastogivereliableresults,i.e.,whereGI(S(0)) < 1.
5
Silva Fennica vol. 50 no. 1 article id 1518 · Chianucci · A note on estimating canopy cover from digital cover…
3 Results
Bothphotographicmethodsprovidedaccurateestimatesofcanopycover,whichsignificantlyagreedwiththoseobtainedfromartificialimageswithknowncanopycover(Pearson’sr-test, p < 0.01). Nonetheless,thetwomethodsshoweddifferentperformancedependingonactualcanopycover(Fig.1a)andmeangapsize(Fig.1b). DHPshowedalargertendencytounderestimatecanopycover,inparticularinmedium-densecanopies(Fig.1a);inaddition,DHPshowedanincreasingtendencytounderestimatecanopycoverwithdecreasinggapsize(Fig.1b).Closerinspectionofartificialchessboard-patternedimagesrevealedthatDHPpossessedmoremixedpixels(Fig.2),beinghighlysensitivetogapfragmentation,becausemixedpixelsarelocatedmainlyonthecanopy-skyedges.Thiseffect,inturn,haveastrongimpactonpixelclassificationsincefragmentedcanopyimagesaremorepronetobemisclassifiedintoskyorcanopyduringthresholding(Fig.1).Conversely,DCPpossessedveryfewmixedpixels(Fig.2),andshowedlowervariationsinestimatedcoverassociatedwithactualcanopycover(Fig.1a)andmeangapsize(Fig.1b).
Atheoreticalgapfractionmodel(Nilson,1999)indicatedthatcanopycoverwasinverselycorrelatedwiththegroupingindex(Pearson’sr-test p < 0.05)andthereforewiththefoliageclump-ingindex,thesparsercanopiesexhibitingmoreclumpeddistributionoffoliageandthedensercano-piesexhibitingmorerandomlydistributedfoliage(Fig.4).Inaddition,canopycoverwasstronglycorrelatedwithleafareaindexinvertedfromgapfractiondata–thehigherthecanopycover,thesmallertherespectiveinvertedleafareaindexvalue.Inoursimulationincreasingcanopycoverfrom0.65to0.95decreasedtheestimatedleafareaindexbyvaluesrangingfrom24%to51%.
Fig. 1. a) Differencebetween theactualandmeasuredcanopycovercalculatedfromartificialcanopy imageswithknowncanopycoverusingdigitalhemispherical(DHP)andcover(DCP)photography;b)Variationofcanopycoverwithmeangapsizecalculatedfromartificialcheckerboard-patternedcanopyimagesusingdigitalhemispherical(DHP)andcover(DCP)photography.Meangapsizerangesfrom0.02%(3mm)to0.2%(30mm)oftheartificialcanopyimage area.
6
Silva Fennica vol. 50 no. 1 article id 1518 · Chianucci · A note on estimating canopy cover from digital cover…
ThestudydemonstratedthatDCPprovidesrobustestimatesofcanopycover,whoseaccuracyislargelyunaffectedbyvariationinactualcanopycoverandmeangapsize.Bycontrast,theaccuracyofDHPwasaffectedbybothactualcanopycoverandmeangapsize,mainlybecauseitslowernearlyverticalresolution,whichresultsinalowerabilityindetectingsmallgapsnearthezenithandameasurementofaquantitythatisanalogoustosomethingbetweencrownandfoliagecover(PekinandMacfarlane2009).Ithasgenerallybeennotedthatlowresolutionimageshavemoremixedpixels(Blennow1995;Leblancetal.2005;Macfarlane,2011),whichcouldobscuresmallgapswithincanopies.Thiseffect,inturnmaypreventtheaccuracyofgapsizedistributionesti-mates,particularlyinhighlyfragmentedcanopies(Songetal.2014).ThisimpliesthatuseofDHPisparticularlycriticalindensecanopies,whicharemainlycharacterizedbysmallwithin-crowngaps;DHPmaybeunabletodetectgapsinsuchcanopies,leadingtoanincreasinglyprobabilityto achieve saturated canopy cover values. This effect, in turn, may prevent the inversion of canopy attributesfromtransmittancedata,sincethelogarithmofzerogapfractionisundefined.Therefore,DCPappearsparticularlysuitedindenseforestcanopies,becauseitshigherabilitytodetectsmallgapsallowsmoreaccuratecanopycoverretrievalinthesestands.Inaddition,thehigherverticalresolutionofDCPallowsseparatingtotalgapfractionintolarge,between-crownsgapsandsmall,within-crowngaps,leadingtoeffectiveestimatesofcanopytransmittance(Rautiainenetal.2005).Further,previousstudiesdemonstratedthatDCPwaslesssensitivetocameraexposure,lensvignet-tingandskyluminancethanDHP(Macfarlaneetal.2007;Macfarlane,2011).
Canopycoverhascertainlyaneffectonaveragegapfractionanditsangulardistribution–thehigherthecanopycover,thesmallerthetotalgapfraction(Nilson2011).Inaddition,thetheoreticalgapfractionmodelindicatedthatcanopycoverinfluencesgapsizeanditsvariance;sparser canopies exhibitedmore clumpeddistributionof foliage,whichcanbe attributed to alargerfrequencyinlargegaps(ChenandCihlar1995)andalargervariationingapsizeoccur-ringatincreasingcanopyspaceavailability.Conversely,densercanopiesshowedmorerandomlydistributed foliage, probably because the lower number of small gaps occurring at saturatingcanopy density (Macfarlane 2011). As the variation in foliage clumping corresponds to different
8
Silva Fennica vol. 50 no. 1 article id 1518 · Chianucci · A note on estimating canopy cover from digital cover…
proportion in sunlight and shaded leaves, it implies that accurate canopy cover estimates are criticalforreliablemodellingfluxesofcarbon,waterandenergy,andtheirdistributionwithinthe canopy (Chen et al. 2012).
AngeliniA.,CoronaP.,ChianucciF.,PortoghesiL.(2015).Structuralattributesofstandoverstoryand light under the canopy.Annals of Silvicultural Research 39(1): 23–31. http://dx.doi.org/10.12899/asr-993.
ChianucciF.,MacfarlaneC.,PisekJ.,CutiniA.,CasaR.(2015).Estimation of foliage clumping fromtheLAI-2000PlantCanopyAnalyzer:effectofviewcaps.Trees29(2):355–366.http://dx.doi.org/10.1007/s00468-014-1115-x.
NilsonT.(1999). Inversion of gap frequency data in forest stands. Agricultural and Forest Meteo-rology98–99:437–448.http://dx.doi.org/10.1016/S0168-1923(99)00114-8.
NilsonT.,KuuskA.(2004). Improved algorithm for estimating canopy indices from gap fraction datainforestcanopies.AgriculturalandForestMeteorology124(3–4):157–169.http://dx.doi.org/10.1016/j.agrformet.2004.01.008.
NilsonT.,KuuskA.,LangM.,PisekJ.,KodarA.(2011).Simulationofstatisticalcharacteristicsofgapdistributioninforeststands.Agricultural and Forest Meteorology151(7):895–905.http://dx.doi.org/10.1016/j.agrformet.2011.02.009.
PisekJ.,LangM.,NilsonT.,KorhonenL.,KaruH.(2011).Comparisonofmethodsformeasur-inggapsizedistributionandcanopynonrandomnessatJärvseljaRAMI(RAdiationtransferModel Intercomparison) test sites.Agricultural andForestMeteorology151(3):365–377.http://dx.doi.org/10.1016/j.agrformet.2010.11.009.
RidlerW., Calvard S. (1978). Picture thresholding using an iterative selectionmethod. IEEETransactions on SystemsMan and Cybernetics 8(8): 630–632. http://dx.doi.org/10.1109/TSMC.1978.4310039.
SongG.Z.M.,DoleyD.,YatesD.,ChaoK.J.HsiehC.F.(2014).Improvingaccuracyofcanopyhemispherical photography by a constant threshold valued derived from an unobscuredovercastsky.CanadianJournalofForestResearch44(1):17–27.http://dx.doi.org/10.1139/cjfr-2013-0082.