Airborne laser scanning for the site type identification of mature boreal forest stands

Remo e Sens. 2011, 3, 100-116; doi:10.3390/ s3010100 Remo e Sensing ISSN 2072-4292 www.mdpi.com/jou nal/ emo esensing A icle Ai bo ne Lase Scanning o he Si e Type Iden i ica ion o Ma u e Bo eal Fo es S ands Mikko Vehmas 1,*, Kalle Ee ikäinen 2, Jussi Peuhku inen 1, Pe e i Packalén 1 and Ma i Mal amo 1 1 School o Fo es Sciences, Uni e si y o Eas e n Finland, P.O. Box 111, 80101 Joensuu, Finland; E-Mails: [email p o ec ed]i (J.P.); [email p o ec ed] (P.P.); [email p o ec ed] (M.M.) 2 Joensuu Resea ch Uni , Finnish Fo es Resea ch Ins i u e, P.O. Box 68, 80101 Joensuu, Finland; E-Mail: [email p o ec ed] * Au ho o whom co espondence should be add essed; E-Mail: [email p o ec ed]i; Tel.: +358-13-2515-297; Fax: +358-13-2513-634. Recei ed: 8 No embe 2010; in e ised o m: 17 Decembe 2010 / Accep ed: 27 Decembe 2010 / Published: 10 Janua y 2011 Abs ac : In Finland, o es si e ypes a e used o assess he need o sil icul u al ope a ions and he g ow h po en ial o he o es s and, he e o e, p o ide impo an in en o y in o ma ion. This s udy in oduces ai bo ne lase scanne (ALS) da a and he k-NN classi ie da a analysis echnique applicable o he si e quali y assessmen o ma u e o es s. Bo h he echo heigh and he in ensi y alue pe cen iles o di e en echo ypes o ALS da a we e used in he analysis. The da a a e o 274 ma u e o es s ands o di e en sizes, belonging o i e o es si e ypes, a ying om e y e ile o poo o es s, in Koli Na ional Pa k, eas e n Finland. The k-NN classi ie was applied wi h alues o k a ying om 1 o 5. The bes o e all classi ica ion accu acy achie ed o all he o es si e ypes and o a single ype, we e 58% and 73%, espec i ely. The conclusion is ha when conduc ing la ge-scale o es in en o ies ALS-da a based analysis would be a use ul echnology o he iden i ica ion o ma u e bo eal si e ypes. Howe e , he echnique could s ill be imp o ed and u he s udies a e needed o ensu e i s applicabili y unde di e en local condi ions and wi h da a ep esen ing ea lie s ages o s and de elopmen . Keywo ds: k-NN classi ica ion; ege a ion; heigh dis ibu ion OPEN ACCESS Remo e Sensing 2011, 3 101 1. In oduc ion In Finland, o es s ands a e classi ied in o o es si e ypes acco ding o hei unde s o ey ege a ion. This si e classi ica ion echnique is based on Cajande ‘s [1] o es si e ype heo y and is a classic example o he indi ec si e quali y es ima o s in o es y con ex [2]. Vege a ion o coni e domina ed bo eal o es s is a composi ion o a ew ee species and a ious plan s wi h di e en shapes and sizes [3]. The e a e also many biological (succession s age, dominan ee species, e c.) and physical ( opog aphy, soil and geology) ac o s a ec ing he ege a ion cha ac e is ics o o es s. The cu en o es si e ypes o Finnish o es s ands ha e been loca ed and mapped in con en ional s and-based o es in en o ies [4]. The delinea ion o o es s and bo de s is subjec i ely made by o es in en o y pe sonnel om ae ial pho og aphs and ield measu emen s and e o s in he de e mina ion o he bo de lines a e common [5]. Despi e he limi s o he in en o y me hod i is widely app o ed and commonly used in o es y. The e is, howe e , a need o new me hods in o de o inc ease he accu acy and cos -e iciency o la ge-scale o es in en o ies. Ae ial pho og aphs ha e ypically only been used o delinea e di e en o es s ands, since he pho os ob ained a e no de ailed o accu a e enough o use o he pu poses o la ge-scale o es in en o ies [6]. Howe e , Ai bo ne Lase Scanning (ALS), which p o ides spa ially accu a e h ee-dimensional (3D) in o ma ion on o es s and is al eady being applied in p ac ical o es y, could eplace con en ional ield in en o y me hods o de e mining ee s ocking quan i ies [7,8]. The 3D na u e o ALS da a has p o en o p o ide excellen in o ma ion on landscapes (e.g., [9]), and especially in o es y applica ions he heigh abo e g ound is o he g ea es in e es [8,10-14].Va ious echo ypes (e.g., i s , las , in e media e and only echoes) can be iden i ied by p ocessing he backsca e ing ene gy o a single lase pulse. In addi ion, he in ensi y alue which desc ibes he amoun o backsca e ing o he echo can be u ilized. The heigh cha ac e is ics o ALS da a ha e been used in a ious s udies: When analyzing he modeled canopy uel pa ame e s o e ical o es s uc u es o he pu poses o i e beha io assessmen [15,16], dis inguishing dominan and unde s o ey laye s o ege a ion [17-19], analysis o na u al egene a ion [20]. and o p edic he cha ac e is ics o dead wood in a gi en sample a ea [21,22]. In ensi y alues ha e been s udied by B ennan and Webs e [23], o example, who ound hem o be sui able o dis inguishing be ween di e en su aces, bu i is only e y ecen ly ha hei applicabili y o he de e mina ion o o es cha ac e is ics has been in es iga ed (e.g., [24,25]). This is mainly due o di icul ies in scaling and no malizing in ensi y alues o lack o knowledge in hei in e p e a ion [26]. Due o he ac ha di e en heigh a ibu es a e es ima ed accu a ely om lase da a [11,12], he ALS echnique has also been applied o he de e mina ion o s andwise si e quali y indica o s based on he heigh dis ibu ion cha ac e is ics [27,28]. These emo e sensing based app oaches, in ac , co espond o he adi ional g ow h and yield s udies, in which he si e classi ica ion is based on he dominan heigh -age dependency (c . [29]). In Finnish o es in en o ies he si e quali y by o es s ands is, howe e , de e mined using Cajande ‘s [1] o es si e ype classi ica ion sys em, which ope a es on assessable s and cha ac e is ics, i.e., g ound ege a ion cha ac e is ics and indica o species, a he han explici ly measu able ee a iables. In he s udy, Pi känen [30] showed ha he connec ion be ween di e en s and s uc u es and a ia ion in g ound ege a ion exis . This phenomenon can be u he s udied by applying ALS da a echniques. In he ecen s udy by Remo e Sensing 2011, 3 102 Ko pela e al. [31] hey ound ha ALS da a can be used in assessmen o di e en bo eal mi e su ace pa e ns, ege a ion and habi a s. The applicabili y o ALS da a o he iden i ica ion o o es s ands wi h high he baceous plan di e si y was ecen ly s udied by Vehmas e al. [32]. One o he disc imina ion echniques applied by Vehmas e al. [32] was he k-nea es neighbo (k-NN) me hod: In hei s udy he he b- ich o es s ands we e dis inguished om less e ile o es s ands. The k-NN me hod disc imina ion echnique u ilized he di e ences in lase heigh dis ibu ions o hese wo o es e ili y classes [32]. Howe e , high he baceous o es s ands co e less han 1% o he o al a ea o o es s and he e o e he applicabili y o he me hod is a he limi ed. The nea es neighbo me hods ha e been widely used o es ima ing con inuous o es a iables (e.g., by [8,33,34]) and some ex en o de e mining disc e e o es a iables (e.g., [32,35]). Peuhku inen e al. [36] s udied species-speci ic diame e dis ibu ions and saw log eco e ies wi h he k-NN me hod by using i s pulses o he da a and no ed ha he me hod hey in oduced can also be applied in di e en classi ica ion p ocedu es. Ea lie indings by Vehmas e al. [32] sugges ed ha he ALS echnology could be applied o dis inguish he o es si e ypes because o di e ences in c own s uc u es and e ical p o iles ha di e be ween di e en o es si e ypes. In his s udy, we applied an ALS da a k-NN me hod [36] o dis inguish o es si e ypes o o es s ands in wall- o-wall co e age by employing wo al e na i e uses o da a: (1) Whole da a wi h lea e-one-ou c oss- alida ion; and (2) an example calcula ion wi h sepa a e se s o modeling and es da a o be compa ed in e ms o classi ica ion accu acy. In he analysis we used a ious echo ypes. 2. Ma e ial and Me hods 2.1. S udy A ea and Fo es In en o y Da a The o es a ea conce ned he e is loca ed in he Koli Na ional Pa k (29°50′E, 63°05′N) in eas e n Finland, on he bo de line be ween he sou he n and middle bo eal o es ege a ion zones a e Kalela [37] (Figu e 1). The o al a ea o Koli Na ional Pa k is abou 3,000 ha, o which 930 ha was he s udy a ea. Ex ensi e a eas in he no he n pa o he Na ional Pa k ha e been le unmanaged o decades, whe eas o es managemen ope a ions we e ca ied ou in he sou he n pa un il he ea ly 1990s. The a ea is cha ac e ized by a highly a iable bo eal landscape and ee species s uc u e, wi h al i udes a ying be ween 95 and 347 m a.s.l. [38]. Fo es s in he a ea a e domina ed by No way sp uce [Picea abies (L.) Ka s .] and Sco s pine (Pinus syl es is L.) wi h a highly a iable admix u e o sil e bi ch (Be ula pendula Ro h.), downy bi ch (B. pubescens Eh h.), Eu opean aspen (Populus emula L.) and g ey alde [Alnus incana (L.) Moench] [38,39]. This a he small a ea includes a ious ypes o o es soils om in e ile o e y nu ien - ich. Bo eal e y ich o es s a e cha ac e ized by mix u es o b oad-lea ed and coni e ous ees, and by soils wi h s uc u al complexi y and alues o pH indica ing nea neu ali y [40]. Thei unde s o ey ege a ion is he e o e dense bo h e ically and ho izon ally han in less e ile soils. Following he classi ica ion o Cajande [1], he o es si e ypes iden i ied in Koli Na ional Pa k we e in i e classes: (1) Ve y- ich (e.g., Oxalis-Maian hemum ype, OMaT); (2) ich (Oxalis-My illus, OMT, he b- ich hea h o es ); (3) medium (My illus ype, MT, mesic hea h o es ); (4) a he poo (Vaccinum ype, VT/EMT, subxe ic hea h o es ); and (5) poo (Calluna ype, CT/MCClT, xe ic hea h Remo e Sensing 2011, 3 103 o es ). The in en o y da a we e collec ed by he Finnish Fo es Resea ch Ins i u e in 2004. We used ALS echo dis ibu ion da a o classi y hese i e o es si e ype classes (Figu e 2). Figu e 1. Map o he Koli Na ional Pa k in eas e n Finland, wi h loca ions o he s ands o he di e en e ili y classes. OMaT deno es e y ich, OMT ich, MT medium, VT a he poo and CT poo o es si e ypes [1]. Fo es ege a ion zones a e Kalela [37]: Sou h Finland (1 Hemibo eal and 2 Sou he n bo eal), Pohjanmaa-Kainuu (3 Middle bo eal), and Pe äpohjola and Me sä-Lappi (4 No he n bo eal). Remo e Sensing 2011, 3 104 Figu e 2. Fi s and only pulse ( o) heigh dis ibu ions in di e en e ili y classes by Cajande [1]. The o al numbe o s ands wi hin he s udy a ea was 680. The s ands we e delinea ed acco ding o he ules documen ed by Da is and Johnson [41], whe e he o es s and is conside ed o be homogeneous in espec o, o example, soil and g owing s ock. In e ms o de elopmen classes, he s ands used in he analyses we e ma u e s ands, as hese closed-canopy s ands ep esen ed ad anced successional s ages o bo eal o es s wi h an ad anced g ound ege a ion and we e he e o e ideal o si e ype classi ica ion by he me hod o Cajande [1]. The whole da a used in his s udy included 274 selec ed o es s ands co e ing an a ea o 337 ha, which included s ands om he no he n and sou he n pa s o he Na ional Pa k. As an example, he da a we e also andomly assigned in o modeling da a and es da a. The modeling da a consis ed o 184 o es s ands and comp ised an a ea o 241 ha, while he es da a consis ed o 90 o es s ands and co e ed a o al a ea o 96 ha. The desc ip i e a ea s a is ics o whole da a (Table 1) we e simila o he model and es da a. Classi ica ion accu acies we e compa ed in di e en e ili y classes and di e en s and size classes. The size classes in ha we e 0.05–0.25, 0.26–0.50, 0.51–0.75, 0.76–1.0, 1.1–2.0 and 2.1–12.1 (maximum s and size). The numbe o s ands (n) in each size class was 45, 48, 40, 39, 54, and 48, espec i ely. Table 1. Desc ip i e s a is ics o s and a eas in ha by o es si e ype in he whole da a. The numbe o s ands is n; Min is he minimum, Max is he maximum, Mean is he a e age a ea o he o es si e ypes in ha and S.D. he s anda d de ia ion. Ve y- ich deno es OMaT (1), ich OMT (2), medium MT (3), a he poo VT (4) and poo CT (5) o es si e ypes by Cajande [1]. Fo es si e ype Whole da a 336.7 ha (n = 274) n Min Max Mean S.D. Ve y- ich (1) 61 0.03 5.80 0.73 0.91 Rich (2) 60 0.13 12.13 1.48 2.12 Medium (3) 60 0.16 7.77 1.38 1.37 Ra he poo (4) 60 0.12 6.46 1.29 1.14 Poo (5) 33 0.04 7.02 1.30 1.58 Remo e Sensing 2011, 3 105 2.2. Ai bo ne Lase Scanne Da a The ALS su ey was pe o med on 13 July 2005, using an Op ech ALTM 3100 lase scanning sys em. A o al o nine ALS lines (Figu e 2) was lown a an al i ude o 900 m and a ligh speed o 75 m/s. The a ea co e ed was app oxima ely 2,200 ha. The lase pulse epe i ion a e was 100 KHz and he scanning equency o a swa h was 70 Hz, a an angle o ±11 deg ees. The pulse densi y o he da a was 3.9/m2, bu because o nominal side o e lap (35%), and a ia ion in he e ain, he ac ual g ound hi s a ied om app oxima ely 3.2/m2 o 7.8/m2. The da a echoes collec ed included coo dina es (x,y) and heigh alue (z), ligh line numbe s, in ensi y alues ( ange om 1 o 180) and echo ypes in ou classes: 1 = only echo, 2 = i s echo, 3 = in e media e echo and 4 = las echo. A digi al e ain model (DTM) was p oduced om he las and only echo da a using a pixel size o 1 m × 1 m, employing Te aScan so wa e, which uses he me hod p oposed by Axelsson [42]. In o de o analyze he ALS da a, he i s s ep was o con e he o home ic heigh s o an abo e-g ound scale by sub ac ing he DTM om he co esponding ALS heigh s [43]. The lase echo cha ac e is ics a e p esen ed by o es si e ypes in Table 2. Table 2. Numbe s o lase echoes/m2 by o es si e ypes and p opo ions (%) o he di e en ypes o echoes including he z alue and in ensi y alue in he whole da a. The le e s , o, l and i indica e he i s , only, las , and in e media e echoes, espec i ely, whe eas o is he sum o i s and only echoes. Fo es si e ype all o / o, % o/ o, % l/ o, % i/ o, % all/ o, % Ve y- ich (1) 7.3 5.0 37.3 62.7 38.0 6.2 144.2 Rich (2) 7.0 5.0 35.7 64.3 36.3 5.6 141.9 Medium (3) 6.8 4.8 34.9 65.1 35.7 5.1 140.8 Ra he poo (4) 6.4 4.7 32.1 67.9 32.8 4.0 136.8 Poo (5) 6.2 5.0 22.5 77.5 22.9 1.7 124.6 2.3. k-NN Classi ica ion In o de o dis inguish o es s ands by o es si e ypes using he k-NN classi ie , we applied wo di e en ways o analyze ou da a. In he i s app oach, we used he whole da a wi h lea e-one-ou c oss- alida ion. In he second app oach he da a we e di ided in o e e ence (modeling) and a ge ( es ) da a. The classi ica ion o o es si e ypes was ob ained by using he non-pa ame ic k-NN classi ie me hod in oduced by Peuhku inen e al. [36]. A leas h ee issues need o be conside ed when using he k-NN me hod: (1) A sui able dis ance me ic; (2) he numbe o neighbo s o be used; and (3) he weigh ing o he neighbo s [44]. This s udy used a Minkowski dis ance me ic o o de one be ween he dis ibu ions. The Minkowski dis ance is applicable o measu ing simila i y be ween objec s and akes in o conside a ion he whole a iabili y and he he e ogeneous s uc u e o lase dis ibu ions. In case o disc e e dis ibu ions, i can be de ined using Equa ion 1: Remo e Sensing 2011, 3 106 i n iipq qpD   1 (1) whe e Dpq is he dis ance be ween ei he lase heigh dis ibu ion o lase in ensi y dis ibu ion o be compa ed, pi is he p opo ion o obse a ions in class i om he all obse a ions o he a ge dis ibu ion (sum (pi) = 1), qi is he p opo ion o obse a ions in class i om all obse a ions o he e e ence dis ibu ion (sum (qi) = 1), and n is he numbe o classes in he dis ibu ions. The alue o Dpq (Equa ion 1) anges be ween 0 ( he dis ibu ions compa ed a e he same) and 2 ( he dis ibu ions compa ed ha e no obse a ions in he same classes). The chosen dis ance me ic is based on he absolu e di e ences be ween he lase echo dis ibu ions o he a ge and e e ence s ands and is sui able in si ua ions in which he p edic o a iables a e dis ibu ions wi h unknown cha ac e is ics (in his case lase echo heigh and in ensi y dis ibu ions) and i is assumed ha he o m o he dis ibu ion con ains mos o he in o ma ion on he a iables o in e es (in his case o es si e ype classes). The dis ance alue can be used in weigh ing he neighbo s by sub ac ing i om he maximum alue, which is 2. When using mo e han one p edic o (i.e., dis ibu ions o lase echoes o di e en ypes), he dis ances a e calcula ed sepa a ely om each dis ibu ion. The inal dis ances a e hen he sum o dis ances. The dis ances a e hen weigh ed using subsequen ly de e mined op imal weigh s o he p edic o s: ji n iji m j jpq qpwD    11 (2) whe e m is he p edic o (lase heigh o in ensi y dis ibu ion o ce ain lase echo ype) and wj is he weigh o he p edic o j. The classi ica ion ule needed o applying he k-NN based classi ie was adjus ed o he case o se e al neighbo s (k) as ollows: 1. k = 1: he alue o he p edic ed a iable is he alue o he nea es neighbo 2. k > 1: he weigh s o he neighbo s a e summed by o es si e ype class and he es ima ed o es si e ype class o he a ge uni is he one wi h he highes sum o weigh s. Fo he k-NN classi ica ion p ocedu e he lase echo heigh s we e classi ied in o 10 cm classes, wi h he nega i e echo heigh s assigned o a class 0 (no e ha some ALS hi s always occu below he DTM le el). This classi ica ion p o ided enough obse a ions o all he app oxima ely 300 classes (0 o 30 m wi h 10 cm in e al). Fu he mo e, he lase echo in ensi ies we e classi ied o 10 e en classes acco ding o he ange o in ensi y alues. The op imal weigh s o he di e en ypes o echoes we e sea ched o by op imizing he o e all classi ica ion accu acy. The op imiza ion algo i hm weigh ed he combina ions sys ema ically so ha e e y echo ype was gi en a weigh om 0 o 1 a in e als o 0.1. In addi ion, all he combina ions o weigh s, which summed o 1, we e examined. The op imiza ion was pe o med on he whole da a wi h a lea e-one-ou c oss- alida ion echnique in which he a ge uni was le ou o he e e ence da a. In he second app oach he op imiza ion was pe o med on he modeling da a, a e which he es s ands we e classi ied using he p e-de e mined op imal weigh s and he nea es neighbo s we e sea ched o only om he modeling ( e e ence) da a. In he case o se e al neighbo s (k > 1), he p ocedu e Remo e Sensing 2011, 3 107 p o ides no only class es ima es bu also an idea o he closeness o he a ge uni o he o he classes. Howe e , i should be emembe ed ha he o es si e ype classes may exp ess he e ili y le els on an o dinal scale, bu he ee co e o hose ypes a e on a nominal scale. The pe o mance o he k-NN es ima ion me hod was e i ied by calcula ing o e all classi ica ions o i e di e en o es si e ypes and by de i ing classi ica ion ma ices, i.e., wo-dimensional con ingency ables. In addi ion, a Cohen‘s kappa s a is ic [45] was used o measu e ag eemen be ween he classi ica ions wi h di e en numbe s o neighbo s and espec i e o di e en da ase s. 3. Resul s The classi ica ion esul s we e calcula ed o he whole da a as well as he es da a, wi h one, h ee and i e nea es neighbo s, and op imal weigh s we e de e mined o he accu acy o classi ica ion o he s ands in o i e di e en o es si e ypes. Bo h he heigh and he in ensi y alues we e used wi h di e en weigh s (Table 3) o op imize he classi ica ion esul s. The heigh s ha e he highes weigh s wi h one neighbo , whe eas he weigh o he in ensi y inc eases wi h h ee and i e neighbo s. Table 3. Weigh s o a iable dis ibu ions used in he k-NN me hod wi h 1, 3 and 5 neighbo s in wo app oaches: (1) Whole da a (n = 274); (2) es da a (n = 90). is he i s pulse, l he las pulse, i he in e media e pulse, o he i s and only pulses oge he and lo he las and only pulses oge he . Classi ica ion Heigh In ensi y alues (i) me hod sum l i o lo sum i l i o lo 1-NN–(1) 0.8 0.2 0.1 0.5 0.2 0.1 0.1 1-NN–(2) 0.9 0.1 0.2 0.3 0.3 0.1 0.1 3-NN–(1) 0.4 0.1 0.1 0.2 0.6 0.1 0.1 0.4 3-NN–(2) 0.2 0.2 0.8 0.1 0.1 0.4 0.2 5-NN–(1) 0.5 0.2 0.1 0.2 0.5 0.3 0.1 0.1 5-NN–(2) 0.4 0.1 0.3 0.6 0.2 0.2 0.1 0.1 A o es con usion ma ix o all o he i e o es ypes is p esen ed in Table 4, whe e he diagonal shows he co ec classi ica ions. The bes o e all classi ica ion esul (58.0%) in he en i e da a was achie ed wi h he 5-NN and he bes single class classi ica ion (poo , CT 72.7%) wi h he 1-NN. In he case o he whole da a, he dec eased numbe s o neighbo s (1 o 3) al e ed he classi ica ion o some s ands and dec eased he o e all classi ica ion accu acy (Table 4). The accu acy a es ob ained o he classi ica ion o he b- ich o es s, o example, we e 52.5% and 62.3% wi h one neighbo and i e neighbo s, espec i ely. The classi ica ion done using he es da a ga e only sligh ly wo se o e all classi ica ion pe cen ages han wi h he whole da a. In addi ion, some single class classi ica ions we e e en be e when using he es da ase . Mo eo e , he o e all classi ica ion accu acy inc eased no ably (83.3–92.2%) when he nex nea es class was aken as co ec classi ica ion esul (Table 4). Remo e Sensing 2011, 3 108 Wi h he 1-NN he classi ica ion o class 3 (medium, MT) was highes (34.4%) and dec eased o classes 1 (17.8%) ( e y ich, OMaT) and 5 (11.1%) (poo , CT). Wi h 5-NN he classi ica ion o he di e en classes a ied mo e e enly o e he i e classes (Table 4). Table 4. Classi ica ion success a es (%) ma ix ob ained by he k-NN me hod wi h 1, 3 and 5 nea es neighbo s o all o es si e ypes in wo app oaches: (1) Whole da a (n = 274); (2) es da a (n = 90). *1 deno es e y ich o es s (e.g., OMaT), 2 ich (OMT), 3 medium (MT), 4 a he poo (VT) and 5 poo (e.g., CT), including all poo o es si es [1]. 1-NN–(1) 56.6; 89.4 3-NN–(1) 56.9; 88.7 5-NN–(1) 58.0; 89.8 *1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 *1 52 23 18 2 5 1 59 21 15 2 3 1 62 20 13 0 5 2 15 45 32 8 0 2 18 47 27 8 0 2 20 43 28 8 0 3 7 15 63 13 2 3 10 18 60 10 2 3 10 20 58 10 2 4 2 3 23 57 15 4 0 8 17 65 10 4 0 8 12 68 12 5 0 0 3 24 73 5 0 0 6 42 52 5 0 0 0 42 58 17 19 30 20 14 19 21 27 24 9 20 20 24 24 11 1-NN–(2) 55.6; 92.2 3-NN–(2) 55.6; 83.3 5-NN–(2) 54.4; 90.0 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 60 20 20 0 0 1 60 10 30 0 0 1 65 25 10 0 0 2 10 50 40 0 0 2 25 45 20 5 5 2 20 40 35 5 0 3 10 15 60 15 0 3 15 15 65 5 0 3 5 30 55 10 0 4 0 5 35 45 15 4 10 10 20 50 10 4 10 15 15 50 10 5 0 0 0 30 70 5 0 0 0 40 60 5 0 0 0 30 70 18 20 34 17 11 24 18 30 18 10 22 24 26 18 10 In bo h da a (whole and es da a) he e was a medium ag eemen in classi ica ion o di e en k-NN classi ie s in e ms o hei kappa alues (0.423–0.469). The co espondence among he k-NN-based classi ica ions was good when he kappa be ween he k-NNs wi h espec i e alues o 3 and 5 was compa ed wi h he kappa alues o he o he wo compa isons. Wi h he k-NN alues o 1 and 3 co espondence was weak (Table 5). S and size had no e ec on he classi ica ion esul s. The e is no end in classi ica ion and accu acies o e di e en e ili y classes (Figu e 3). Classi ica ion accu acy a ies om 14% (OMT wi h s and size unde 0.25 ha) o 100% (MT 0.26–0.50 ha, CT 0.76–1.0 ha and 1.01–2.0 ha). In gene al, he a e age classi ica ion accu acy was 59% a ying om 50% (unde 0.25 ha) o 69% (o e 2 ha). Only in he case o he CT (poo ) class, is he iden i ica ion consis en ly mo e accu a e han in all s ands oge he (All). Remo e Sensing 2011, 3 115 26. Ahokas, E.; Kaasalainen, S.; Hyyppä, J.; Suomalainen, J. Calib a ion o he Op ech ALTM 3100 Lase Scanne In ensi y Da a Using B igh ness Ta ge s. In P oceedings o ISPRS Commission I Symposium, Pa is, F ance, May 2–6, 2006; In The In e na ional A chi es o he Pho og amme y and Remo e Sensing and Spa ial In o ma ion Sciences; ISPRS: Vienna, Aus ia, 2006; Vol. XXXVI, Pa A1, T03-11. 27. Rombou s, J.H. Explo ing he Po en ial o Ai bo ne LiDAR o Si e Quali y Assessmen o Radia e Pine Plan a ions in Sou h Aus alia: Ini ial Resul s. In P oceedings o Fo es Measu emen and In o ma ion Sys ems, Biennial Mee ing 2006, Resea ch Wo king G oup 2, Woodend, Aus alia, No embe 21–24, 2006. 28. Ga ziolis, D. Lida -De i ed Si e Index in he US Paci ic No hwes —Challenges and Oppo uni ies. In P oceedings o ISPRS Wo kshop on Lase Scanning 2007 and Sil iLase 2007, Espoo, Finland, Sep embe 12–14, 2007; In The In e na ional A chi es o he Pho og amme y and Remo e Sensing and Spa ial In o ma ion Sciences; ISPRS: Vienna, Aus ia, 2007; Vol. XXXVI, Pa 3/W52, pp. 136-143. 29. Ee ikäinen, K. A si e dependen simul aneous g ow h p ojec ion model o Pinus kesiya plan a ions in Zambia and Zimbabwe. Fo es Sci. 2002, 48, 518–529. 30. Pi känen, S. Co ela ion be ween s and s ucu e and g ound ege a ion: An analy ical app oach. Plan Ecol. 1997, 131, 109-126. 31. Ko pela, I.; Koskinen, M.; Holopainen, M.; Vasande , H.; Minkkinen K. Ai bo ne small- oo p in disc e e- e u n LiDAR da a in he assessmen o bo eal mi e su ace pa e ns, ege a ion and habi a s. Fo es Ecol. Manage. 2009, 258, 1549-1566. 32. Vehmas, M.; Ee ikäinen, K.; Peuhku inen, J.; Packalén, P.; Mal amo, M. Iden i ica ion o bo eal o es s ands wi h high he baceous plan di e si y using ai bo ne lase scanning. Fo es Ecol. Manage. 2009, 257, 46-53. 33. Moeu , M.; S age, A.R. Mos simila neighbo : An imp o ed sampling in e ence p ocedu e o na u al esou ce planning. Fo es Sci. 1995, 41, 337-359. 34. Holms öm, H.; Nilsson, M.; S åhl, G. Simul aneous es ima ions o o es pa ame e s using ae ial pho og aph in e p e ed da a and he k nea es neighbou me hod. Scand. J. Fo es Res. 2001, 16, 67-78. 35. Thessle , S.; Sesnie, S.; Ramos Bendaña, Z.; Ruokolainen, K.; Tomppo, E.; Finegan, B. Using k-NN and disc iminan analyses o classi y ain o es ypes in a Landsa TM image o e no he n Cos a Rica. Remo e Sens. En i on. 2008, 112, 2485-2494. 36. Peuhku inen, J.; Mal amo, M.; Malinen, J. Es ima ing species-speci ic dis ibu ions and saw log eco e ies o bo eal o es s om ai bo ne lase scanning da a and ae ial pho og aphs: A dis ibu ion-based app oach. Sil a Fennica 2008, 42, 625-641. 37. Kalliola, R. Suomen kas imaan iede. We ne Söde s öm Osakeyh iö, Po oo, Finland, 1973. (In Finnish). 38. Lyy ikäinen, A. Kolin luon o, maisema ja kul uu ihis o ia. Kolin luonnonsuojelu u kimukse . Vesi- ja ympä is öhalli uksen monis esa ja. 1991; p. 308. (In Finnish). Remo e Sensing 2011, 3 116 39. G önlund, A.; Hakalis o, S. Managemen o T adi ional Ru al Landscapes in Koli Na ional Pa k. Sepa a e Plan o Koli Na ional Pa k; Regional En i onmen al Publica ions 104; No h Ka elia Regional En i onmen Cen e: Joensuu, Finland, 1998; pp. 1-81. (In Finnish, wi h English summa y). 40. Hokkanen, P. Vege a ion Pa e ns o Bo eal He b-Rich Fo es s in he Koli Region, Eas e n Finland: Classi ica ion, En i onmen al Fac o s and Conse a ion Aspec s. Ph.D. Disse a ion, Facul y o Fo es y, Uni e si y o Joensuu, Joensuu, Finland, 2006; Abs ac 27. 41. Da is, L.S.; Johnson, K.N. Fo es Managemen , 3 d ed.; McG aw-Hill: New Yo k, NY, USA, 1987. 42. Axelsson, P. DEM Gene a ion om Lase Scanne Da a Using TIN Models. In P oceedings o XIX h ISPRS Cong ess: Technical Commission IV, Ams e dam, The Ne he lands, July 16–23, 2000; In The In e na ional A chi es o he Pho og amme y and Remo e Sensing and Spa ial In o ma ion Sciences; ISPRS: Vienna, Aus ia, 2000; Vol. XXXIII, Pa B4/1, pp. 110-117. 43. Hyyppä, H.; Yu, X.; Hyyppä, J.; Kaa inen, H.; Kaasalainen, S.; Honka aa a, E.; Rönnholm., P. Fac o s A ec ing he Quali y o DTM Gene a ion in Fo es ed A eas. In P oceedings o he ISPRS Wo kshop Lase scanning 2005, Enschede, The Ne he lands, Sep embe 12–14, 2005; In In e na ional A chi es o Pho og amme y, Remo e Sensing and Spa ial In o ma ion Sciences; ISPRS: Vienna, Aus ia, 2005; Vol. XXXVI, Pa 3/W19, pp. 85-90. 44. LeMay, V.; Temesgen, H. Compa ison o nea es neighbo me hods o es ima ing basal a ea and s ems pe hec a e using ae ial auxilia y a iables. Fo es Sci. 2005, 51, 109-119. 45. Cohen, J. A coe icien o ag eemen o nominal scales. Educ. Psychol. Meas. 1960, 20, 37-46. 46. Hynynen, J.; Ojansuu, R.; Hökkä, H.; Siipileh o, J.; Salminen, H.; Haapala, P. Models o P edic ing S and De elopmen in MELA Sys em; Resea ch Pape s 835; Finnish Fo es Resea ch Ins i u e: Van aa, Finland, 2002; p. 116. © 2011 by he au ho s; licensee MDPI, Basel, Swi ze land. This a icle is an open access a icle dis ibu ed unde he e ms and condi ions o he C ea i e Commons A ibu ion license (h p://c ea i ecommons.o g/licenses/by/3.0/).