Characterization and Classification of Vegetation Canopy Structure And

Characterization and Classiﬁcation of Vegetation Canopy Structure and Distribution within the Great Smoky Mountains National Park using LiDAR

Jitendra Kumar1, Jon Weiner2, William W. Hargrove3, Steven Norman3, Forrest M. Hoﬀman1, Doug Newcomb4

1Oak Ridge National Laboratory, 2University of California Berkeley, 3USDA Forest Service, 4U.S. Fish and Wildlife Service

US – International Association of Landscape Ecology, Asheville, NC, USA April 06, 2016 Introduction

◮ Forests form a complex mosaic of diverse tree and coexisting plant and animal species. ◮ The structure of vegetation reveals information about stand age and height; forest composition, health, and disturbance; and suitability as habitat for birds and other animal species. ◮ Airborne Light Detection and Ranging (LiDAR) enables large scale remote sensing of topography, built infrastructure, and vegetation structure. ◮ Multiple laser “returns” produce “point clouds” used to map the ground surface, buildings, roads, and utility infrastructure, and to reconstruct the structure of vegetation canopies. ◮ Large data volumes pose signiﬁcant computational challenges to employing LiDAR to monitor and manage forests and animal habitats. Great Smoky Mountains National Park (GSMNP)

◮ The GSMNP is the most visited national park in the U.S., and it hosts a rich ecosystem of plants and wildlife. ◮ The Park encompasses 816 sq. miles in Tennessee and North Carolina and ranges in elevation from 876 to 6,643 feet above mean sea level.

1 23 4 5 6 7 8 9 10 11 12

T E To Knoxville To Knoxville To Newport To Newport N N E SEVIERVILLE S 321 S E E 40 411

R 32 I

V 441 E R

r T 411 r Stream Crossings e r CHEROKEE NATIONAL FOREST e y r m Exit Litt T a A l n m ) e i a w er 443 Nearly all park trails cross small streams—making very wet crossings ta 1.0 C k t P n r in 129 ig u t a w n P during flooding. The following trails that cross streams with no bridges e o 0.3 n i o M u r s d n ve e o e can be difficult and dangerous at flood stage. (Asterisks ** indicate the Ri ab os G M cl 0.4 r ( McGhee-Tyson L most difficult and potentially dangerous.) This list is not all-inclusive. e s ittl 441 ll Airport e w i n o Cosby h o 0.3 L ot e Beard Cane Trail near campsite #3 Fo Pig R R ive iv r Beech Gap Trail on Straight Fork Road er Cold Spring Gap Trail at Hazel Creek 0.2 W Eagle Creek Trail** 15 crossings e 0.3 0.4 SNOWBIRD s e Tr t Ridg L Fork Ridge Trail crossing of Deep Creek at junction with Deep Creek Trail en 0.4 o P 416 D w IN r e o k Forney Creek Trail** seven crossingsA G TENNESSEE T n Wa E a N g n BB IN Gunter Fork Trail** five crossingsU S MOUNTA p Exit 451 O 0.1 32 Hannah Mountain Trail** justM before Abrams Falls Trail L i NORTH CAROLINA tt Jonas Creek Trail near Forney Creek le Little River Trail near campsite #30 PIGEON FORGE C 7.4 Long Hungry Ridge Trail both sides of campsite #92 Pig o 35 Davenport eo s MOUNTAIN n b mere y r MARYVILLE Lost Cove Trail near Lakeshore Trail junction Cam Trail Gap Waterville nt 9 R Pittman u C 1.

Meigs Creek Trail 18 crossings k i o h EE v Big Creek W e M 1.0 e B O e Mt LH e Center 73 Mount s I Noland Creek Trail** both sides of campsite #62 r r 321 Hen Wallow Falls t 2.1 H C r Cammerer 6 n C Cammerer C r e u 321 1.2 e Panther Creek Trail at Middle Prong Trail junction 0. t e w Trail Br Tr k o

L Pole Road Creek Trail near Deep Creek Trail M

6.6 2.3 321 a 34

321 il Rabbit Creek Trail at the Abrams Falls Trailhead d G ra

Gatlinburg Welcome Center 5.8 d ab T National Service Park U.S. Department the of Interior 1 e Roundtop Trail** crossing of Little River at Little River Road r s Mountain e Tr Cosby o g id 4.6 2.5 National Park Information Center n Trillium Gap Trail at Grotto Falls (icy in winter) e R v 2.1 il en Low k o e ra D G e

C N T Tra Upper Hazel Creek Trail hiker-only portion I L B e a Low Gap ek il r A it s a p re C T tl er Albright Grove ld k C Wet Bottom Trail follow signs for Elijah Oliver Trail as alternate route N e tl a T 5.1 g U 73 P t 33 T n r i i e a B B O g Albright Grove r 29 S i 2.5 Walland M e S 0.7 l ig a o d Loop Trail 2.4 B m x 129 321 n l o te GATLINBURG R O 2.3 t 6. r i 6.1 t 1 v Inadu Knob 37 o C

e Cosby r Greenbrier IER PINNA B

E r NBR CL e 411 E E Camel Gap t

V RE Knob e Gatlinburg Bypass Road u

O 441 G k

C 36 n l

Roaring 3.7 4.1 T

E r

Wear Valley V a

F Cam W i el G l O one-way ork ap 1.0 To Chattanooga C Sugarlands M Greenbrier Trail Ramsey Tra o 6 s il t Cove e Cascades S ( o 7. d w y R c a 4.0 38 a r c A l R 32 y s r a w E Visitor Center o M mse a T 1.8 k L i a C Mount Guyot l N R

W s i r it Little d dd k lo Mount t Park Headquarters e g le 4.1 r a le C a Pro o w h d e ng F Sterling P Greenbrier t l 4.0 Ster 321 e u d rai r t ling Mount Sterling Gap r ar T te F M T

Trail 8. T o Baskins in r ey n o r e p r L

4 w k a alsam Mt Gu k l Historic r B n o R T Luftee i 1.9 i e Creek G 1.0 rail Tr n C i Nature n e a 9 il w Knob i a v C ra l WATERVILLE o n T O Trail r g s Little 2. ve Mountai C T 1.4 e 0. Trail T l 4 r i r r n LAKE l e c a i r Brier h B L e t Gatlinburg a i Tricorner Knob l u h 1.4 a k e l 4. e i

Gap Tr u s rd y n r a t W Trail 2. 9 a r

r T R ) Mount Chapman K T k o 7.5 e O r d 7 c TUCKA l a Pretty Hollow Gap 3.6 LEECHEE COVE P l Porters B T o R l Little Greenbrier r d Brushy M 40 F o i F o Grotto l P A r ra alls 1.9 i Flat e P a k

u T n o g n T School S Mountain d e dtop 3.1 r N g Falls a L i r i e R l a e u 4.5 r r S 3.9 e r T t 5.3 C i w L g t

l e A i t Metcalf t a illiu 2.9 0.4

1.7 r m n r f T y t r i M l s n o l i Laurel Falls l an s Tra 0.6 G a g 73 Bottoms Tr e d a t o 0 u 0.4 0. n Townsend p n t H Ace l n i n P R u Cr Mount Sequoyah 9 l 1. 3.5 Be r o i a o e i The Sinks d d g i e e Gap a a l o Fighting n alls T T Bru M k p l t Townsend R F s a l r e 6.0 ra hy p r Balsam e b 2.7 S o M Little Riv il c il e 5.6 C Creek o o a h a A w e u w i T t a N Visitors Center e 3.1 l n B l High Top Moun 4.0 r I 3.3 r i r i r h i r Gap a u a T g a c A IN 3. y Tr R i T A s l l T G c i l o e 5 v Laurel Top n 1 N A M Rainbow a e G C e He 3.6 o T 2.0 a Balsam e 4. l U l r p G r p 1 i a t d Laurel Gap N e a a Falls 6.0 h a

O 3 4.3 n Elkmont G Mount Le Conte a t p e Point c 2. r U R l p T T

M T ai g k 2.0 M o a

Tr Rid e Li r Campﬁres Prohibited r

I p T ttle Huskey Gap 31 o PISGAH

a C

O T C To r T a l

r R r y 0.5 k r T i

r e i i

a a l O a

e T H t v 2.2 l

i r e 5.9

M a b t Great Smoky Mountains National Park North Carolina/Tennessee l e u i

e r R T a k n l 20 r i t 2.4 l U e l s m o i h C NATIONAL C e s n Ch u a ra e Pecks Corner N 2 e R a 4.2 L 19 4.1 u d T 39 le Palmer C 4.1 1. Mountain L B T t Gold Mine n E R H Chimney e M i Schoolhouse 9 igs it o Charlies A it House a H c e Tra C 2. t 2.1 v 44 l 2.3 FOREST d h il u 4 l a u i L Trail 0. C C r T Tr M c e Tops Trail l I Look Rock 8 M 1. Gap u Alum Cave e a N M Bunion a r 0 m C va

d N T b r r T e t g Great Smoky Mountains er 5. d Trail H B Cataloochee n G S 0. T a M 3.3 R Trail A B 2. n i a U Bluffs 4 lsam t l p u H i n d T G m Grassy n 3 T r il r 4. A 1. 9 T Pin Oak a 4 t o 7 r 0. H 11 T Institute at Tremont g C r r d a 1 S R Chimney u U p r T in 1.1 2. r R u LA l 1.7 3 Branch 0.8 o P l g h a Gap k a M e T 21 a N e o t e iv 1.0 rlan D A G re 0.8 n g k Tops D e G 2.5

1. p u e e d 0 Icewater Trail C Cove Creek Gap e o 5 3.6 P H t s H 1.7 8 r M G t d Blanket e r 5. r 2.8 r

) e ap t i s r r 9 y h Palme l r v T y Ro te r o e o C 2. 0.9 E c i

a n a R M Spring e a d i r c 1.2 Mountain oun 2.5 a w G S d n r tain S e r C h in Ind 1.8 R W s T B T ian g S t D T o D T c n e 0.8 k t

r i 3.4 18 r k 1.0 l Cabin v r a 0.8 L k e R u i 1.1 e e re 2. R e

T l R 6 i a C 1.1 i t o 1 5. p e c Flats i 1.7 R C 40 6 r tl J h 5 e 2. d i l r n 5 y a 2.3 Crooked e 27 g T 9 d i e E 2. t ed C g g e r r Trail 49 E e l a R u d s e T i Tr Sw e e r k o Arm r v o 2.8 a e 3.7 B e

l k k

W 0.9 Little M c u e e ree Jakes 0.8 1.7 a G Round Bottom ig T R ( r r C i a d r Rich Mtn a n R t p E 3.6 1.8 i o

O B 17 Ca r l M H T e Ridge Tr u L n H o 2.3 r F 1.9 v a

A M e Gap e l i t h b T y d L c r Loop Tr 2.9 h P i 0.6 o 47 Spruce Mountain k I Abrams Creek to t am 2.2 le id ant 2.3 O ron 1.7 fe Kephart 3 T e H m a n dle P Miry 26 g Trai r Cr K r C D i l R 4.6 C s H s F N U Tr 3. a one-way r P a e k Wet T 2.5 ro 2.5 R 3.3 4.8 i i e l 7 C r n 24 Newfound Gap l I T l ek R r i 2. T F ade Cove Road T g N p C r i o 3.9

r a s L p i 2.4 ) ai 1.2 d a d l y oo ( F

Bottom p 2.8 M r h H Trail Spruce Mountain g r

4.2 ll a T k d c e T g

a Abrams L r s T e a a l T

R G 1.8 A L 1.6 t o Trail r E

Tr e 3.2 n e r e r o o B

a ib E n A oogerma Falls a e w k r t F R s B 40

b T ree C a 1.8 G I i D C T N N e h r GREAT i SMOKY MOUNTAINS NATIONAL PARK a k d 2. I b l

s 23 o w P c d l

r 9 m A D B 48 a O s i r 0.5 o d T 3 V

t 1. b C s 2.0 r n a 42 Exit 20

a A n 3.7 n r I 28 r o

a B l h o 5 i e m

CADES COVE o i a a l

Cre t k

P e r 2. n o e R v

k T i N r D h v a d l 0 d r 2.9 a g l e M 4 T l r amp Pron 8.1 i R e n g k T e w a o C e I e t r 1. s l e o n M l a 1.6 T T n f 4. r D

r o o O e C i n R y i n l a l y c u F a w M L B n s e y Mount Collins ( n T y o d l n Pr d U e G w r l 1.4 C l r c l 16 o r T d u l C t Greenbrier n a Ga N e a K 2.3 Appalachian Highlands 5.1 a i a r g p t E i C n 41 15 e o R T s n e V k e M T o d k A r R a F t r e h e k Ridge Trail t b e H 2.5 4.4 a d A a e Science Learning Center

y r r O o l a o

t ams C T il e e h r B d I F

p a e c ) r 1.2 30 1.8 r i r

o one-way 1.9 o N y C t m

r k n p a

n t T 10 Cold Spring o F n h r Purchase Gap

o a r T 1.7

o o a h g n o F H T D r l S g u

B Knob k i 3.0 a R k u R o

4. 5.1 2. m 3.5 Cades Cove 3.5 s 9 r f a R h Buckeye Gap d

d t F r l 2 n o r k o o 50 t 276

3.9 F e e e 9 5 a R r T M e id e k S r i 2.4 g ti k o C F 3. e e 1.8 Visitor Center 1.7 Thunderhead m a u Derrick 3.3 n g a m Balsam Mountain n e ll TENNESSEE K T Mountain Double Spring n T o t g 3.9 F e Knob i o 2.2 r a Double Gap

T w r s l l r n a a i s Clingmans Rocky C a d o Gap i n u i R t t l Chilhowee Spence Field 1.9 l L l F R i

14 Top 6.2 Dome v o S a Russell Field e o t Polls Gap B

NORTH CAROLINA C r p r

r 2.8 i

r n e 0.5 2.6 l a T l k

g i ) Silers Bald 1.9 E i r r e 53 k h

l te 2.9 N p

l G oad

n i l T R m E

o i r i e E

7.6 d T r H

w a D l r E a l Clingmans Dome C

I a a r l

6.4 a i

r l H

CHILHOWEE in i B T R n M T T Bypass Trail C

d a 2.2 ove k r

l r O e d o r n C d 5.5

s Mollies Ridge i u o O

LAKE E o o l 1.7 T Newton

l i n t a ) L c k a t Smokemont A ( 3.7 ea F CAT e ain -S w r n

a r s o e o G -T r e Bald N e a

r T 6 R one-way t T h C D D 3. n I 4.9 4.7 t e Tow String e i 12 7 Andrews i a g n R 68 v T e g l g w n 0. i k r i d id a 8.1 Bald d B n i i g J e a o o 52 e R J R a n e e n l v i h l a i 3.3 r 2. i 129 y E i s a 54 l a r c l a le R C l a Black Camp Gap i il C 7.4 r o 8 R L i

d r r o e d o ra e T a g a I a a T T P d a

g e e

o l 4.1 d T 97 k 4.5 61 r o r

R 4. Gre e 0 C i s A

g r 55 T o T r r o ) y a e T To Asheville r M

r v 5. o

1 a

h g t i e r G k S l

4 50036 80251 0 c t e R d T 3.1

p a i c

n in i e r i n

e k l D

T i

Doe Knob e R A n rkw ( i

a r e a

r w p e g p P a

C r d 56 g e y e Dellwood

B n C h l i e e 9 H

i 8.7 c 69 i R M u g

A 3. e d l a E e

r g 13 ld l i

$1.00 a i T s u n B e d a e D O l d Tr 2.0 l y G Martins 57 d i R B o g e i e D H R Maggie Valley s 1.2 I s Parson S W n s a l k r Gap Trail R r e E R s a L E z e o T o Gregory 96 Bone Valley Tr N

a k

l Bald a F ee o n G e m r I C 19

i r 1.5 c

P H 62 Bald n o

( Trail D C 1.5 k K I E h g 70 y T 3.1 n 1.8 R e 2.8 58 a w 2.9 N D k t H e n Trail I 46 lo J r e E o p u o 5.1 4.6 J V e e l I 4.3 k i n 8. 82 F r e G h n D a a g B c D C r 9 1.2 u 1.5 p Oconaluftee River Trail 1.5

0.7 T ry 113 ea n Y il 59 1.8 2.4 r a S Oconaluftee C r E 63 a R B r 3.8 5.2 id High Rocks r 71 T p C e use N g 1.3 ho e one-way W e g 51 e o in R k k Spr 64 e Visitor Center 0.8 l 0.9 e 7.8 To Asheville o T d O e D 0.5

89 l 4 Soco Gap r 92 3.5 r i 4.2 D

l 5.9 T F a 2. f C CHEROKEE INDIAN RESERVATION

e 83 r d a 8

i T n

95 Cov Sp a Nola N r l r rin r g G t ap 1. Cooper

Los a Trail Cold Spring i T l A 441 T 84

1.9 il 1.3 k (Qualla Boundary) e L 2.2 Creek k Gap e 60 k

R 2.7 e r e CALDERWOOD i Twe r l 91 k O Loop Trail F d n op T ai e e C e Indian

g t Lo Tr r Whiteoak 4.1 r e e N ymil e C 1.1 Shuckstack zel Trail C LAKE ile a r 75 Branch Tr Creek

T CHEROKEE m H C n y

r 2.9 y 90 Lakesho 85 Motor Plott Balsam a re l a

t rai e Indian i T 1.8

i d 2.1 l n 4.4 4.5 n Tunnel n Trail 2.6 e l r Creek 1.0 I CHEROKEE 3.7 e k 1.1 z o e 93 w Ha H Bypass e Falls Deals Gap C F T 88 EL r 0.9 5.2 W Trail 65 C k 0.5 0.6 2.0 Cree NATIONAL 1.0 1.3 d oco WAYNESVILLE 86 n S 0.7 0.4 a T l 1.1 Stone Pile r 4 o Waterrock Knob Twentymile e a N Gap Trail FOREST il 74 0. La Deep Creek L y 2 ke A EOA a 1. vi Horse Trail K CH H 1.0 e E K w Juney Whank 19 c 72 67 Falls Trail 87 Lake a sh M Ela o nton 98 Goldmine re e 76 2.0 Drive Fontana 81 Trai 77 B Loop 66 Deep Creek S l Trail AM Dam 8.7 441 LS BA EE FONTANA 12.3 ue INA Bl 73 BRYSON CITY idge C 78 R OL h Fontana Whittier R e Y o EL Marina LA a LOW KE Tuckaseg TENNESS CA h ee FONTANA 28 R C i P 129 REE ve 74 a K VILLAGE r r TH k R M w G OUN T

TAIN T a S y NO O T L R u P i v c e k a JOYCE KILMER - SLICKROCK r s eg A ee R WILDERNESS AREA p iver p 74 E a 23 G ID la R c h 0.5 i B 441 a ra n d D l L e SYLVA BA il 28 y ra

T F o

Sm r o k OLD k e m o n 0.9 t North SANTEETLAH L o Stecoah Gap NANTAHALA NATIONAL FOREST 3.9 o MAP LEGEND p 23

T LAKE r H 0.3 441 143

Wesser L 0 1 10 Kilometers kywa itt ic S y le en Sc 0 1 10 Miles rohala he T C ROBBINSVILLE o w

e Cheoah Bald r T 1.0 h e e s T 28 N Tr t v n ew ld i ton Ba r i n Unpaved road Appalachian Trail Ranger station R n

e g Hiking only (top)

143 s T

19 e r Horse and hiking Developed campground e

74 2.2 R Hiking trail Mountains-To-Sea iv Trail Auto-access horse camp e r la a h Observation tower ta an Horse and hiking Benton MacKaye N trail Trail Backcountry shelter (by reservation) I 4 Backcountry campsite (by reservation) NAN M TAHAL Roads in park are closed 1.3 Trail distances are OUN MOUN A to commercial vehicles. shown in miles be- 2 Backcountry campsite (non-reservation) IRD TA B INS 129 TAIN tween intersections OW S SN 6 Boat-in only campsite (non-reservation)

Great Smoky Mountains Trail Map To Atlanta 2011 1 2 3 4 5 6 7 8 9 10 11 12 Objectives

◮ Develop computational tools and worﬂow for processing and knowledge extraction from massive LiDAR data sets ◮ Map and characterize the 3-D structure and distribution of the vegetation canopy in GSMNP LiDAR Tiles for GSMNP

Tennessee ◮ LiDAR data for 540 sq. miles of the Tennessee portion of the GSMNP and the Foothills Parkway from 1,658 flight miles were collected during February–April 2011 by the U. of Georgia and Photo Science, Inc. ◮ Four multiple discrete returns per pulse were collected at a rate of 20.2 Hz from a nominal flying height of 1,981 m above ground level. ◮ Overlapping data were split into 724 non-overlapping 1,500 × 1,500 m tiles, which we obtained from the National Park Service. ◮ 724 LiDAR tiles (approx. size 98 GB) projected onto a 3.0 m resolution digital elevation model (DEM) derived from the LiDAR point cloud. ◮ Projection: UTM Units: meters North Carolina ◮ LiDAR data for North Carolina was collected by NC Floodplain Mapping Program in 2005. ◮ Overlapping data were split into non-overlapping 10,000 × 10,000 ft tiles, which we obtained from the NC Floodplain Mapping Program. ◮ 184 LiDAR tiles (approx. size 8.9 GB) projected onto a 3.0 m resolution digital elevation model (DEM) derived from the LiDAR point cloud. ◮ Projection: NC State Plane Units: ft LiDAR Data for GSMNP Computational Workflow and Data Processing

◮ We employed a process-parallel Start

approach to extract and analyze P0 P1 PN LiDAR point cloud data using get subset “las” file list get subset “las” file list get subset “las” file list

check geographical area python. covered by the “las” file

◮ open DEM file and read To estimate vegetation heights above elevation value for the region ground level, elevations from the

check ground elevation at S

LiDAR point and detrend a

m 3.0 m DEM were subtracted from m e

apply noise and k

point cloud data. k

ground filtering o

◮ s

P The resulting points were grouped assign point to appropriate P

0 into 1 m vertical bins, up to 75 m, at 3-D bin × a horizontal resolution of 30 30 m. are there more points in the file? ◮ Anomalous high points (aerosols,

birds) and low points (steep slopes, are there more files to process? surface litter) were ﬁltered out. ◮ Corrections were made for low height write data to file write data to file write data to file

vegetation (shrubs and grasses) and perform k-means clustering

for many returns at the same generate maps and statistics elevation. End LiDAR 101 LiDAR 101 LiDAR 101 LiDAR Point Cloud Example: 30m pixel

a) 3-D LiDAR point cloud extent at b) Raw LiDAR point cloud (3,985 points), 30 × 30 m (black square) shown in a showing imprints of underlying topography. typical GSMNP cove forest.

c) LiDAR point cloud after topographic d) Vertical distribution of LiDAR point detrending and ﬁltering (3,936 points). density in a cove forest dominated by tall trees and a dense understory. LiDAR Point Cloud Example: 30m pixel LiDAR Point Cloud Example: 100m pixel Computational Workﬂow and Data Processing

◮ We employed a process-parallel Start

approach to extract and analyze P0 P1 PN LiDAR point cloud data using get subset “las” file list get subset “las” file list get subset “las” file list

check geographical area python. covered by the “las” file

◮ open DEM file and read To estimate vegetation heights above elevation value for the region ground level, elevations from the

check ground elevation at S

LiDAR point and detrend a

m 3.0 m DEM were subtracted from m e

apply noise and k

point cloud data. k

ground filtering o

◮ s

P The resulting points were grouped assign point to appropriate P

0 into 1 m vertical bins, up to 75 m, at 3-D bin × a horizontal resolution of 30 30 m. are there more points in the file? ◮ Anomalous high points (aerosols,

vegetation (shrubs and grasses) and perform k-means clustering

for many returns at the same generate maps and statistics elevation. End Map of 30 Vegetation Canopy Structure Classes Vertical proﬁle distributions were input to the cluster analysis, considering all tiles simultaneously.

This map shows the 30 most-different classes of vegetation canopy structure, randomly colored, as identified by k-means clustering for the Great Smoky Mountains National Park. Map of 30 Vegetation Canopy Structure Classes Vertical profile distributions were input to the cluster analysis, considering all tiles simultaneously.

This map shows the 30 most-diﬀerent classes of vegetation canopy structure, randomly colored, as identiﬁed by k-means clustering for the Great Smoky Mountains National Park.

Overstory Vegetation Cover Map for GSMNP Overstory Vegetation Cover Map for GSMNP GSMNP: Chestnut Oak Forest Chestnut Oak Forests

GSMNP: Northern Hardwood Forest Northern Hardwood Forests

GSMNP: Hemlock Forest Hemlock Forests

Exploring in Google Earth

Great Smoky Mountain Institute at Tremont ◮ The Great Smoky Mountain Institute at Tremont (GSMIT) is surrounded by “Mountain Cove” and “Hemlock” forests with tall, dense canopies and low understory vegetation. ◮ We found strong spatial correspondence between the “Mountain Cove” forest and the vegetation structure classes representing the taller vegetation in the Park. ◮ Individual tree records from “Citizen Science” phenology plots, located at blue circles, were studied for ground truthing.

“Montain Cove” forest from (Madden, Tall canopy vegetation classes 10 and 13 2014). derived from LiDAR. Summary and Conclusions

◮ We developed an approach, parallel software tools, and workﬂow for analyzing large volumes of LiDAR point cloud data in a scalable fashion. ◮ Multivariate Spatiotemporal Clustering (MSTC) provides a valuable quantitative framework for stratifying vegetation canopy structure data derived from LiDAR point clouds. ◮ We applied these tools to LiDAR data from the GSMNP to identify vegetation classes based on overstory/understory distributions. ◮ We used a spatial overlay method to compare the unsupervised clustering results to existing vegetation maps and to validate results ◮ These tools and the resulting maps will inform resource management and conservation planning by forest and wildlife managers, who were not previously able to use large, complex LiDAR data sets. 2015 IEEE 15th International Conference on Data Mining Workshops

Characterization and classiﬁcation of vegetation canopy structure and distribution within the Great Smoky Mountains National Park using LiDAR

Jitendra Kumar∗, Jon Weiner†, William W. Hargrove‡, Steven P. Norman‡, Forrest M. Hoffman∗ and Doug Newcomb§ ∗Oak Ridge National Laboratory, Oak Ridge, TN, USA Email: [email protected] †University of California Berkeley, Berkeley, CA, USA ‡USDA Forest Service, Southern Research Station, Asheville, NC, USA §U.S. Fish and Wildlife Service, Raleigh, NC, USA

Abstract—Vegetation canopy structure is a critically important of vegetation. Advances in Light Detection and Ranging (Li- habitat characteristic for many threatened and endangered birds DAR) technologies have enabled remote sensing-based stud- and other animal species, and it is key information needed ies of vegetation canopies by providing a three-dimensional by forest and wildlife managers for monitoring and managing forest resources, conservation planning and fostering biodiversity. representation of vegetation structure throughout the canopy. Advances in Light Detection and Ranging (LiDAR) technologies While the application of LiDAR for study of forest ecosystems have enabled remote sensing-based studies of vegetation canopies is becoming more common, the richness of these data sets by capturing three-dimensional structures, yielding information are generally under-utilized due to the large volumes of the not available in two-dimensional images of the landscape pro- data produced by these instruments and lack of computational vided by traditional multi-spectral remote sensing platforms. However, the large volume data sets produced by airborne LiDAR resources and analysis algorithms. Most of the LiDAR studies instruments pose a significant computational challenge, requiring focus on the development of high resolution Digital Elevation algorithms to identify and analyze patterns of interest buried Models, canopy heights and occasionally understory density within LiDAR point clouds in a computationally efficient manner, [1], [2]. While LiDAR dervied metrics have proven to be utilizing state-of-art computing infrastructure. We developed useful for an array of applications [1]–[5], three-dimensional and applied a computationally efficient approach to analyze a large volume of LiDAR data and characterized the vegetation information provided by the LiDAR are left unutilized. canopy structures for 139,859 hectares (540 sq. miles) in the The objective of this study is to develop methods to realize Great Smoky Mountains National Park. This study helps improve the potentials of rich LiDAR data set to map and characterize our understanding of the distribution of vegetation and animal the three-dimensional structure and distribution of vegetation habitats in this extremely diverse ecosystem. canopies. We develop and apply data analytic techniques to identify the ecologically important and understandable struc- I. INTRODUCTION tural types by mining the large and complex volumes of Forest ecosystems are a complex mosaic of diverse plant LiDAR data. and tree species, the location and distribution of which are driven by a number of gradients like climate (ex. temperature, II. MATERIALS Kumar, J., J. Weiner, W.W. Hargrove, S.P. precipitation regimes), topography (ex. elevation, slope, as- pect), geology (ex. soil types, textures, depth), hydrology (ex. A. Study area drainage, moisture availability) etc. Diverse combinations of Norman, F.M. Hoffman, and D. Newcomb. these gradients support diverse composition and distribution of The geographic area for this study was the Great Smoky vegetation which in turn supports an array of wildlife. Under- Mountains National Park (GSMNP), which in part covers standing the vegetation canopy structure is critical to under- the Great Smoky Mountains and the Blue Ridge Mountains, 2015. LiDAR-derived Vegetation Canopy stand, monitor and manage the complex forest ecosystems like encompassing 816 sq. miles across Tennessee and North those in the Great Smoky Mountain National Park (GSMNP). Carolina in the United States. Results presented here focus Vegetation canopies not only help understand the vegetation, primarily on the Tennessee side of the GSMNP (approximately Structure, Great Smoky Mountains National but are also a critically important habitat characteristics of 540 sq. miles). The GSMNP covers complex topography many threatened and endangered animal and bird species for with elevations ranging from 876–6,643 feet above mean sea which the GSMNP is home. level. The GSMNP is ecologically rich and diverse, consisting Park, 2011. ORNL DAAC, Oak Ridge, Remote sensing has been widely used to monitor regional to of about 1,600 species of flowering plants, including 100 global forest ecosystems and for mapping of vegetation types. native tree species and over 100 native shrub species [6]. The However, traditional remote sensing methods for vegetation distribution of vegetation in the park is strongly influenced by Tennessee, USA. classification often use light reflectance from the top layer topography, moisture and other environmental gradients [7].

This research was sponsored by the U. S. Department of Agriculture Forest Service Eastern Forest Environmental Threat Assessment Center (EFETAC) and by the Climate and Environmental Sciences Division (CESD) of the Biological and Environmental Research (BER) Program in the U. S. Department of Energy Oﬃce of Science. This research used resources of the Oak Ridge Leadership Computing Facility (OLCF) at Oak Ridge National Laboratory (ORNL), which is managed by UT-Battelle, LLC, for the U. S. Department of Energy under Contract No. DE-AC05-00OR22725. References

M. Madden. Overstory vegetation at Great Smoky Mountains National Park, Tennessee and North Carolina (reference code: 1047498), 2014. URL https://irma.nps.gov/App/Reference/Profile/1047498.