Comparison of grid-based and segment-based estimation of forest attributes using airborne laser scanning and digital aerial imagery

Remo e Sens. 2011, 3, 945-961; doi:10.3390/ s3050945 Remo e Sensing ISSN 2072-4292 www.mdpi.com/jou nal/ emo esensing A icle Compa ison o G id-Based and Segmen -Based Es ima ion o Fo es A ibu es Using Ai bo ne Lase Scanning and Digi al Ae ial Image y Saka i Tuominen 1,* and Reija Haapanen 2 1 Finnish Fo es Resea ch Ins i u e, Me sän u kimuslai os, PL 18, 01301 Van aa, Finland 2 Haapanen Fo es Consul ing, Kä jenkosken ie 38, 64810 Vanhakylä, Finland; E-Mail: eija.haapanen@haapanen o es consul ing. i * Au ho o whom co espondence should be add essed; E-Mail: saka[email p o ec ed]; Tel.: +358-10-211-2167; Fax: +358-10-211-2202. Recei ed: 2 Feb ua y 2011; in e ised o m: 4 Ma ch 2011 / Accep ed: 2 May 2011 / Published: 12 May 2011 Abs ac : Fo es managemen planning in Finland is cu en ly adop ing a new-gene a ion o es in en o y me hod, which is based on in e p e a ion o ai bo ne lase scanning da a and digi al ae ial images. The in en o y me hod is based on a sys ema ic g id, whe e he g id elemen s se e as in en o y uni s, o which he lase and ae ial image da a a e ex ac ed and he o es a iables es ima ed. As an al e na i e o a complemen o he g id elemen s, image segmen s can be used as in en o y uni s. The image segmen s a e pa icula ly use ul as he basis o gene a ion o he sil icul u al ea men and cu ing uni s since hei bounda ies should ollow he ac ual s and bo de s, whe eas when using g id elemen s i is ypical ha some o hem co e pa s o se e al o es s ands. The p opo ion o he so-called mixed cells depends on he size o he g id elemen s and he a e age size and shape o he s ands. In his s udy, we ca ied ou au oma ic segmen a ion o wo s udy a eas on he basis o lase and ae ial image da a wi h a iew o delinea ing mic o-s ands ha a e homogeneous in ela ion o hei o es a ibu es. Fu he , we ex ac ed lase and ae ial image ea u es o bo h sys ema ic g id elemen s and segmen s. Fo bo h uni s, he ea u e se used o es ima ing he o es a ibu es was selec ed by means o a gene ic algo i hm. O he ea u es selec ed, he majo i y (61–79%) we e based on he ai bo ne lase scanning da a. Despi e he heo e ical ad an ages o he image segmen s, he lase and ae ial ea u es ex ac ed om g id elemen s seem o wo k be e han ea u es ex ac ed om image segmen s in es ima ion o o es a ibu es. We conclude ha es ima ion should OPEN ACCESS Remo e Sens. 2011, 3 946 be ca ied ou a g id le el wi h an a ea-speci ic combina ion o ea u es and es ima es o image segmen s o be de i ed on he basis o he g id-le el es ima es. Keywo ds: o es in en o y; ai bo ne lase scanning; ae ial pho og aphy; image segmen a ion 1. In oduc ion In Finland, he o es in en o y o o es managemen planning has adi ionally been based on isual in en o y by s ands. In his me hod, he o es s ands ha a e delinea ed on he basis o ae ial pho og aphs and hei g owing s ock and si e- ela ed cha ac e is ics a e measu ed o es ima ed in he ield. The me hod equi es a la ge amoun o ieldwo k. The e o e, he isual in en o y me hod is o be eplaced wi h a new-gene a ion o es in en o y me hod. The new-gene a ion o es in en o y me hod will be based on in e p e a ion o ai bo ne lase scanning (ALS) da a and digi al ae ial image y using ield sample plo s as e e ence da a. Lase scanning has been conside ed he mos p omising emo e sensing echnology in o es in en o y, and i has been widely applied o s and-le el o es in en o ies (e.g., [1–4]). On he o he hand, ALS da a a e no well sui ed o es ima ion o ee species p opo ions o dominance a he applied poin densi y (e.g., [5]). Acco dingly, op ical image y is needed o complemen he ALS da a. Spec al ea u es o op ical images a e ypically used o sepa a ing di e en ee species, whe eas ex u al ea u es o op ical images a e mainly connec ed wi h he size and spa ial a angemen o he ee c owns. In Finland, ae ial images ha e been widely used in o es in en o y since he 1950s [6], and hei a o dabili y and a ailabili y a e good (e.g., [3,7]). S a is ically, he new-gene a ion o es in en o y me hod is based on wo-phase sampling wi h s a i ica ion, whe e he in en o y da abase is based on a sys ema ic g id o sample uni s (i.e., g id elemen s as sample uni s), and he size o he g id elemen s should co espond o he size o ield plo s. Field measu emen s a e alloca ed in o s a a ha a e commonly de i ed on he basis o ea lie s and in en o y da a. Typical emo e sensing da a sou ces used in he new-gene a ion o es in en o y sys em a e low-densi y ALS da a ( ypically 1–2 pulses/m2) and digi al ae ial image y wi h a spa ial esolu ion o app oxima ely 0.5 m con aining he ollowing spec al channels: blue (B), g een (G), ed (R) and nea -in a ed (NIR). The o es a iables a e es ima ed o each elemen o he in en o y g id, and each g id elemen is commonly de ined as a squa e a ea. As an al e na i e o he g id-based app oach, use o au oma ic s and delinea ion by image segmen a ion has been s udied o de ining in en o y uni s (e.g., [8]). Au oma ic segmen a ion o ALS da a has also been applied o delinea ing e y small segmen s wi h a iew o de ec ing indi idual ee c owns (e.g., [9,10]). In es ima ing s and le el o es a ibu es (a ea based app oach) he segmen size is la ge han single ee c owns. S ands delinea ed au oma ically on he basis o emo e sensing image y (i.e., image segmen s) ha e an ad an age o e g id elemen s (e.g., [11,12]). They can be delinea ed in such a way ha hey exac ly ollow he ac ual s and bo de s, whe eas he g id elemen s a e spa ially ‘spa se’ in ela ion o he ac ual s and bo de s in he o es and Remo e Sens. 2011, 3 947 hey do no ollow he bo de lines accu a ely, ins ead o en in e sec ing ees om mo e han one s and (e.g., [11]). On he o he hand, he g id elemen s a e unambiguously de ined by hei coo dina es, so he same uni s can be used in subsequen in en o ies. In delinea ion o he o es s ands, he p ima y inpu a iables a e he mean heigh o he ees and he ee species composi ion (o dominance). F om he s and delinea ion pe spec i e, s and densi y usually is a seconda y pa ame e . The heigh o he ees can be de i ed on he basis o he ALS da a. As s a ed abo e, ALS da a wi h he applied pulse densi y do no se e well he pu pose o ecogni ion o ee species. The e o e, again, op ical ae ial image y is gene ally used o dis inguishing o es s ands on he basis o hei ee species composi ion. The geome ically h ee-dimensional na u e o ALS da a makes i possible o ex ac a la ge numbe o s a is ical ea u es. When combining he ALS da a wi h ae ial pho og aph da a, one inds ha he numbe o a ailable ea u es inc eases u he . Consequen ly, he dimensionali y o he ea u e space inc eases g ea ly, and he da a become spa se in ela ion o he ea u e space dimensions and he con as be ween objec s in he ea u e space weakens, making, o example, he nea es neighbo sea ch uns able [13,14]. Fo he es ima ion p ocedu e, he dimensionali y o da a mus he e o e be educed, and a subse o ea u es wi h good disc imina ion abili y ound. In an ideal case, he analys would be able o in e he op imal ea u e combina ions om he cha ac e is ics o he independen and he dependen da a, bu in he case o a physically complex and a ying objec , such as a o es , his is no possible, and au oma ed ea u e selec ion me hods mus be used. The objec i es o his s udy we e: (a) o ind a sui able combina ion o lase and ae ial da a ea u es o au oma ic s and delinea ion; (b) o ind a sui able combina ion o lase and ae ial da a ea u es o he es ima ion o o es a ibu es; and (c) o compa e g id elemen s and au oma ically delinea ed s and polygons in he es ima ion o o es a ibu es. 2. Ma e ials and Me hods 2.1. S udy A eas The lase -scanning and ae ial-image-based es ima ion was es ed in wo s udy a eas. S udy a ea 1 was loca ed in he municipali y o Lammi, in Sou he n Finland (app oxima ely 61°19'N and 25°11'E). The a ea co e ed app oxima ely 1,800 ha o s a e-owned o es . The ield da a in S udy a ea 1 consis ed o 281 ixed- adius (9.77 m) ci cula ield sample plo s ha we e measu ed in 2007. The plo s we e loca ed wi h T imble's GEOXM 2005 Global Posi ioning Sys em (GPS) de ice, and he loca ions we e p ocessed wi h local base s a ion da a, esul ing in an a e age e o o app oxima ely 0.6 m. S udy a ea 2 was in Eas e n Finland, in he municipali ies o Kuopio and Ka ula (app oxima ely 62°55'N and 27°12'E), co e ing app oxima ely 36,700 ha o mainly p i a ely owned o es . The ield da a consis ed o 546 ixed- adius (9 m) sample plo s measu ed in 2009. In o de o co e all ypes o o es , bo h s udy a eas we e s a i ied on he basis o ea lie s and in en o y da a and he ield sample plo s we e assigned o hese s a a. The loca ion o he s udy a eas and he sample plo layou s a e p esen ed in Figu e 1. Remo e Sens. 2011, 3 948 Figu e 1. Loca ion o he s udy a eas in Finland and o he sample plo s wi hin he s udy a eas. Some di e ences we e e iden be ween he o es cha ac e is ics o he wo s udy a eas. In S udy a ea 1, he o al g owing s ock was mo e e enly dis ibu ed among he ollowing ee species g oups: Sco s pine, No way sp uce, and deciduous ees, whe eas S udy a ea 2 was clea ly domina ed by No way sp uce. Fu he mo e, S udy a ea 2 had a somewha highe a e age s and olume, as well as g ea e a ia ion in sample plo olumes. The s a is ics o he wo s udy a eas based on he sample plo measu emen s a e p esen ed in Table 1. Table 1. Fo es s a is ics o he s udy a eas: a e age, maximum (Max.) and s anda d de ia ion (S d.) o he sample plo alues. S udy a ea 1 S udy a ea 2 A e age Max. S d. A e age Max. S d. To al olume, m3/ha 178.7 575.4 115.4 191.3 798.5 131.5 Volume o Sco s pine, m3/ha 69.8 560.6 86.9 47.7 561.8 78.9 Volume o No way sp uce, m3/ha 63.7 575.4 94.9 102.9 739.2 128.0 Volume o deciduous species, m3/ha 45.2 312.0 56.2 40.7 400.4 63.9 Basal a ea, m2/ha 19.8 45.5 10.3 22.3 62.0 11.2 Mean heigh , m 17.0 30.5 6.7 16.9 35.6 6.7 Mean diame e , cm 21.1 50.2 9.4 20.7 60.3 10.0 Remo e Sens. 2011, 3 949 2.2. Remo e Sensing Da a In S udy a ea 1, he emo e sensing da a consis ed o o ho ec i ied colo -in a ed digi al ae ial image y (con aining nea -in a ed, ed, and g een bands) wi h a g ound esolu ion o 0.5 m and ALS da a acqui ed om a lying al i ude o 1,900 m wi h a densi y o 1.8 e u ned pulses pe squa e me e . In S udy a ea 2, he emo e sensing da a consis ed o o ho ec i ied digi al ae ial image y con aining nea -in a ed, ed, g een, and blue bands wi h a g ound esolu ion o 0.5 m and ALS da a acqui ed om a lying al i ude o 2,000 m wi h a densi y o 0.6 e u ned pulses pe squa e me e . In addi ion o use o he ALS poin da a, he ALS da a we e in e pola ed o a as e image o ma , o wo ou pu images: heigh and in ensi y. The as e -image pixel alues o he heigh and in ensi y images we e calcula ed wi h A cGis Spa ial Analys ools, using in e se dis ance weigh ed (wi h a powe o 2) in e pola ion based on he wo nea es ALS poin s. The ou pu lase images we e esampled o a spa ial esolu ion simila o ha o he ae ial images. 2.3. Au oma ic Image Segmen a ion S and delinea ion was ca ied ou in he s udy a eas ia au oma ic segmen a ion o ae ial images and ALS da a in e pola ed o as e o ma . The segmen a ion was ca ied ou in wo phases. In he i s phase, ini ial segmen a ion was pe o med ia a modi ied implemen a ion o he ‘segmen a ion wi h di ec ed ees’ algo i hm, which employs he local edge g adien [15,16]. The objec i e is o ind all po en ial segmen bo de s in his phase. The e o e, his me hod ypically p oduces a e y la ge numbe o small polygons when one is using high- esolu ion emo e sensing da a. In his s udy, he ini ial segmen a ion was based en i ely on ALS heigh , co esponding mainly o s and heigh [17] (see Figu e 2(a,b)). P io o he ini ial segmen a ion, he ALS heigh da a we e p e-p ocessed by Gaussian smoo hing. The size o he smoo hing window was 3 × 3 pixels, and i e sequen ial smoo hing ope a ions we e used, aimed a diminishing he wi hin-s and a ia ion and emphasizing be ween-s ands a ia ion. In he second phase, he ini ial segmen s we e p ocessed ia a egion-me ging algo i hm ha was guided by pa ame e s such as he desi ed minimum size o he inal segmen s and he simila i y o dissimila i y o he segmen s o be me ged [16]. The me ging o egions in o he inal segmen s was ca ied ou on he basis o lase heigh , lase in ensi y, and he NIR/R a io o he ae ial images, wi h he aim o aking in o accoun also he ee species composi ion o he ini ial segmen s (Figu e 2(c)). Two au oma ic segmen a ions, wi h minimum segmen sizes o 350 m2 and 0.1 ha, we e ca ied ou in bo h s udy a eas. Remo e Sens. 2011, 3 950 Figu e 2. Lase heigh da a, ini ial segmen s based on he lase heigh da a, and inal segmen s (min. size 0.1 ha) based on a combina ion o lase and ae ial image da a. Examples om S udy a ea 2. (a) (b) (c) 2.4. Ex ac ion o Lase and Ae ial Image Fea u es Th ee emo e sensing ea u e da a se s we e ex ac ed o each o he s udy a eas. In hese se s, he emo e sensing ea u es we e alloca ed o each sample plo om a squa e window o a segmen in which he sample plo was loca ed. The ea u e se G id was ex ac ed om a 20 × 20 m squa e window cen e ed on each sample plo . The ea u e se Seg350 was ex ac ed om image segmen s Remo e Sens. 2011, 3 951 whose minimum size was se as 350 m2. Fea u e se Seg1000 was ex ac ed om image segmen s wi h a minimum size se as 0.1 ha. The ollowing s a is ical and ex u al ea u es (max. 174) we e ex ac ed om he ae ial images and ALS heigh and in ensi y o i s pulse da a o each ea u e da a se : 1. A e ages o pixel alues o g id elemen s (20 × 20 m) and image segmen s su ounding each plo . 2. S anda d de ia ions o pixel alues o blocks, in o which a 32 × 32 pixel window was di ided. The block sizes co esponded o 1 × 1, 2 × 2, 4 × 4, and 8 × 8 pixels. In addi ion o hese ou s anda d de ia ion alues, he s anda d de ia ion o hese ou alues was compu ed. Fo he segmen s, hese we e calcula ed as a e ages o he a ea co e ed [18]. 3. Tex u al ea u es based on co-occu ence ma ices o pixel alues [19,20] ex ac ed o g id elemen s and de i ed o segmen s as a e age alues o he g id elemen s wi hin segmen s:  Angula second momen   q qp ),( 2  Con as    q qp q ),(*)( 2  Co ela ion )*(/)*),(**( yx q yx qp q      En opy   q qp qp )),(log(*),(  Local homogenei y    q q qp ))(1/(),( 2 whe e: N qM qp /),(),(  M(q, ) = he co-occu ence ma ix o he equan i ied pixel alues q and N = he o al numbe o possible pai s in he image window  x ,  x = he mean and s anda d de ia ion o he ow sums o he co-occu ence ma ix  y ,  y = he mean and s anda d de ia ion o he column sums o he co-occu ence ma ix The ex u al ea u es based on co-occu ence ma ices o pixel alues we e ex ac ed in 4 di ec ions in he ex ac ion window: ho izon ally (0° angle), e ically (90°) and diagonally (45° and 135°). Pixel lag o 3 me e s was applied in ex ac ing hese ea u es on he basis o ea lie s udy [7]. In addi ion, he ollowing ea u es we e ex ac ed om he ALS heigh da a only: 4. Heigh s a is ics o he i s and las pulses o all ALS poin s inside he ield plo a ea o he segmen a ea. These included mean, s anda d de ia ion, maximum, coe icien o a ia ion, heigh s whe e ce ain pe cen ages o poin s (5, 10, 20, ..., 95) had accumula ed, and pe cen ages o poin s accumula ed a ce ain ela i e heigh s (5, 10, 20, ..., 95). Only poin s o e 2 m in heigh we e conside ed in compu a ion o hese a iables. Finally, he pe cen age o poin s o e 2 m in heigh was included as a a iable. Remo e Sens. 2011, 3 952 Fo he es ima ion o o es a ibu es, all ae ial image and ALS ea u es we e s anda dized o a mean o 0 and a s anda d de ia ion o 1. This was done because he o iginal ea u es had e y di e se scales o a ia ion. Wi hou s anda diza ion, a iables wi h wide a ia ion would ha e had g ea e weigh in he es ima ion, ega dless o hei co ela ion wi h he es ima ed o es a ibu es. 2.5. Selec ion o Fea u es and Es ima ion o Fo es A ibu es The k-nea es neighbo (k-nn) me hod was used o es ima ing he o es a iables (e.g., [21–23]). The es ima ed a iables we e o al olume o g owing s ock; he olume o Sco s pine, o No way sp uce, and o deciduous species; basal a ea; mean diame e ; and mean heigh . The alue o k was se o 5 in bo h s udy a eas, which was a comp omise be ween he es ima ion accu acy and he a e aging allowed in he es ima ion esul s. The k-nn es ima ion me hod ypically has an inc easing end in accu acy when one aises he alue o k om 1 o 10 (e.g., [23,24]), bu la ge alues o k ypically esul in e aining less o he o iginal a ia ion in he es ima ion esul s, as well as disappea ance o he a e s a a in he s udy ma e ial. Euclidean dis ances we e used o measu e he closeness in he ea u e space, and he nea es neighbo s we e weigh ed wi h he in e se squa ed dis ances. The accu acy o he es ima es was calcula ed ia lea e-one-ou c oss- alida ion by compa ing he es ima ed o es a iable alues wi h he measu ed alues (g ound u h) o he ield plo s. The accu acy o he es ima es was measu ed in e ms o he ela i e oo mean squa e e o (RMSE) (see Equa ion (1)). y RMSE RMSE *100% (1) whe e: 1 ) ˆ ( 1 2      n yy RMSE n iii yi = measu ed alue o a iable y on plo i ŷi = es ima ed alue o a iable y on plo i y = mean o he obse ed alues n = numbe o plo s. Au oma ic ea u e selec ion was ca ied ou by means o a simple gene ic algo i hm p esen ed by Goldbe g [25] and implemen ed in he GAlib C++ lib a y [26]. The eason o selec ing his me hod was i s success in an ea lie s udy by Haapanen and Tuominen [27]. The GA p ocess s a s by gene a ing an ini ial popula ion o s ings (ch omosomes o genomes), which consis o sepa a e ea u es (genes). The s ings e ol e du ing a use -de ined numbe o i e a ions (gene a ions). This e olu ion includes he ollowing ope a ions: selec ing s ings o ma ing by applying a use -de ined objec i e c i e ion ( he be e , he mo e copies in he ma ing pool), allowing he s ings in he ma ing pool swap pa s (c oss o e ), causing andom noise (mu a ions) in he o sp ing, and passing he esul ing s ings o he nex gene a ion. In he p esen s udy, he s a ing popula ion consis ed o 300 andom ea u e combina ions (genomes). The leng h o he genomes co esponded o he o al numbe o ea u es in each s ep, and he genomes con ained a 0 o 1 a posi ion i, deno ing he absence o p esence o image ea u e i. The Remo e Sens. 2011, 3 953 numbe o gene a ions was 30. The objec i e a iable o be minimized du ing he p ocess was a weigh ed combina ion o ela i e RMSEs o k-nn es ima es o mean olume, olume o Sco s pine, No way sp uce olume, olume o deciduous species, mean diame e , and mean heigh , wi h mean olume ha ing a weigh o 50% and he emaining a iables weigh ed a 10% each. Genomes ha we e selec ed o ma ing swapped pa s wi h each o he wi h a p obabili y o 80%, p oducing o sp ing. Occasional mu a ions ( lipping 0 o 1 o ice e sa) we e added o he o sp ing (wi h a p obabili y o 1%). The s ings we e hen passed o he nex gene a ion. The o e all bes genome o he cu en i e a ion was always passed o he nex gene a ion, also. Fou successi e s eps (all including 30 gene a ions) we e aken, o educe he numbe o ea u es o a easonable minimum. Only ea u es belonging o he bes genome in each s ep we e included in he nex s ep. Fea u e selec ion was un sepa a ely o bo h a eas and each ea u e ex ac ion uni (G id, Seg350, and Seg1000). The lis o selec ed ea u es o all ex ac ion uni s is p esen ed in he appendix. 3. Resul s In bo h s udy a eas, he ea u es ex ac ed om squa e g id elemen s wo ked be e in es ima ion o he o es a ibu es han he ea u es ex ac ed om image segmen s did (see Tables 2 and 3). Fu he mo e, ea u es om image segmen s de i ed wi h a minimum size o 350 m2 pe o med be e in he es ima ion han did ea u es ex ac ed om la ge segmen s (minimum size: 0.1 ha). S udy a ea 2 had gene ally be e es ima ion accu acy in compa ison o he da a se s o S udy a ea 1. The main eason o his is p obably he highe numbe o sample plo s in S udy a ea 2, which yields a highe numbe o po en ial nea es neighbo s o each sample plo in he k-nn es ima ion. The e we e la ge di e ences be ween he s udy a eas in he es ima ion accu acy o olumes o he ee species g oups. Typically, he highes es ima ion accu acy was seen wi h he dominan ee species. Howe e , he olume o deciduous ees showed be e es ima ion accu acy in compa ison wi h he mino i y coni e ous ee species g oup, since he p esence o he deciduous ees is mo e easily ecognizable in he ae ial images. Table 2. Es ima ion esul s o he ea u e se s: RMSE (%) ela i e o a e age (RMSE a g.) and s anda d de ia ion (RMSE s d.) o S udy a ea 1. G id Seg350 Seg1000 RMSE A g. RMSE S d. RMSE A g. RMSE S d. RMSE A g. RMSE S d. To al olume 27.8 43.0 34.0 52.6 36.6 56.7 Volume o Sco s pine 74.2 59.6 77.1 61.9 99.9 80.4 Volume o No way sp uce 83.9 56.3 87.5 58.7 103.3 69.2 Volume o deciduous species 85.3 68.8 88.7 71.6 93.9 76.3 Basal a ea 25.8 49.8 30.1 58.1 29.8 57.7 Heigh 18.5 46.9 22.4 56.7 25.5 64.7 Diame e 25.5 57.2 27.7 62.1 32.0 71.9 Remo e Sens. 2011, 3 960  Combined s anda d de ia ion o pixel blocks (1 × 1, 2 × 2, 4 × 4, 8 × 8) o ae ial image NIR band  Angula second momen (45° angle) o ae ial image ed band  Angula second momen (45° angle) o ae ial image g een band  Homogenei y (135° angle) o ae ial image g een band  Maximum o i s pulse hi s  S anda d de ia ion o i s pulse hi s (below 2 m hi s excluded)  Heigh , whe e 20% o i s pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Heigh , whe e 80% o i s pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Pe cen age o i s pulse hi s below 70% o maximum heigh (below 2 m hi s excluded)  Pe cen age o las pulse hi s abo e 2 m heigh  Heigh , whe e 20% o las pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Pe cen age o las pulse hi s below 95% o maximum heigh (below 2 m hi s excluded) Seg1000  A e age o ALS heigh  S anda d de ia ion o ALS in ensi y  S anda d de ia ion o 2 × 2 pixel blocks o ALS in ensi y  Con as (135° angle) o ALS heigh  S anda d de ia ion o ae ial image NIR band  Con as (135° angle) o ae ial image NIR band  Con as (90° angle) o ae ial image ed band  Con as (135° angle) o ae ial image g een band  Heigh , whe e 90% o i s pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Heigh , whe e 10% o las pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Pe cen age o las pulse hi s below 30% o maximum heigh (below 2 m hi s excluded) S udy a ea 2. G id  A e age o ALS heigh  A e age o ALS in ensi y  Con as (135° angle) o ALS heigh  En opy (0° angle) o ALS heigh  A e age o ae ial image NIR band  En opy (135° angle) o ae ial image g een band  En opy (90° angle) o ae ial image NIR band  Heigh , whe e 10% o i s pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Heigh , whe e 40% o i s pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Heigh , whe e 90% o i s pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Pe cen age o i s pulse hi s below 80% o maximum heigh (below 2 m hi s excluded)  Pe cen age o las pulse hi s abo e 2 m heigh Remo e Sens. 2011, 3 961 Seg350  A e age o ALS heigh  Angula second momen (135° angle) o ALS in ensi y  Homogenei y (90° angle) o ALS heigh  S anda d de ia ion o 4 × 4 pixel blocks o ae ial image blue band  A e age o ae ial image NIR band  Angula second momen (90° angle) o ae ial image ed band  angula second momen (0° angle) o ae ial image NIR band  En opy (0° angle) o ae ial image g een band  En opy (0° angle) o ae ial image NIR band  Pe cen age o i s pulse hi s abo e 2 m heigh  S d o i s pulse hi s (below 2 m hi s excluded)  Heigh , whe e 10% o i s pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Heigh , whe e 40% o i s pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Heigh , whe e 80% o i s pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Heigh , whe e 30% o las pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Pe cen age o las pulse hi s below 10% o maximum heigh (below 2 m hi s excluded)  Pe cen age o las pulse hi s below 90% o maximum heigh (below 2 m hi s excluded) Seg1000  A e age o ALS heigh  A e age o ALS in ensi y  Angula second momen (0° angle) o ALS in ensi y  En opy (135° angle) o ALS heigh  Homogenei y (0° angle) o ALS heigh  Homogenei y (90° angle) o ALS heigh  A e age o ae ial image ed band  A e age o ae ial image NIR band  S anda d de ia ion o ae ial image NIR band  A e age o ae ial image blue band  S anda d de ia ion o 8 × 8 pixel blocks o ae ial image blue band  Homogenei y (0° angle) o ae ial image g een band  Pe cen age o i s pulse hi s abo e 2 m heigh  Heigh , whe e 50% o i s pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Heigh , whe e 80% o las pulse hi s ha e been accumula ed (below 2 m hi s excluded)  Pe cen age o las pulse hi s below 10% o maximum heigh (below 2 m hi s excluded)  Pe cen age o las pulse hi s below 20% o maximum heigh (below 2 m hi s excluded) © 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/).