Post on 13-Jul-2020
LiDAR as a tool to monitor biodiversity
Roland Brandl University of Marburg
Jörg Müller Bavarian Forest National Park
Thomas Nauß University of Marburg
Background
• Maps of biodiversity are important in basic
and applied ecology
• Sampling biodiversity is costly
• Mapping biodiversity using remote sensing
opens up perspectives
– large extent
– fine grain
Habtitat suitability
Capercaillie
LiDAR in the scientific literature
Web of Science: LiDAR Web of Science: LiDAR & …
Passive vs. active sensors
LiDAR
1.
2. 3. 1.
1.
● Scan angle
● Runtime
● Intensity
(individual images by openclipart.org)
Horizontal Resolution
cm (airborne) …
m (spaceborne)
LiDAR: Costs
• Compared to field
surveys LiDAR is
cheap if one
considers the extent
• LiDAR surveys are
partly available from
public resources
LiDAR Field campaign < 10 € ha-1
Processing 5 € ha-1
< 15 € ha-1
Ground survey 30 to 100 € ha-1
Vegetation density measured by space filling:
2 % 50 % 65 %
2 % 50 % 65 %
Müller et al. Oecologia 2012
20 m Radius
Courtesy D. Nill
LiDAR: Use as control variable
LiDAR: Use as predictor variable
f(LiDAR variables)
Variables derived
from LiDAR
Eco
log
ical
vari
ab
le
Statistical Model
Müller et al. BAE 2011
Use cases: Birds and arthropods
• Bavarian Forest National Park
• 171 / 223 plots of 1 ha along
4 transects
• Habitat characteristics
– field measurements
– aerial photographs
– airborne LiDAR
0
10
20
30
40
50
0
50
100
0 50
100
Can
op
y h
eig
ht
[m]
X [m
]
Y [m]
0
10
20
30
40
50
0
50
100
0 50
100
X [m
]
Y [m]
0
10
20
30
40
50
0
10
20
30
40
50
0
50
100
0 50
100
X [m
]
Y [m]
A B C
Open stands Mixed forest Mature beech forest
Müller et al. BAE 2011
Plot characteristics
Number of tree species 2.71 1 6 0.855 2.23
Number of cavity trees 1.66 0 12 1.93 3.49
Volume of snags [m³ ha-1] 37.6 0.00 411 2.86 9.03
Maximal diameter in breast height [cm] 57.6 8.00 130 0.57 1.08
Age of the oldest tree [years] 131 0 400 2.16 6.92
Gaps without regeneration [m²] 272 0.00 7376 5.48 36.1
Young broadleaf forest, height 0 to 6 m [m²] 801 0.00 8799 2.59 6.18
Young coniferous forest, height 0 to 6 m [m²] 446 0.00 9108 4.03 16.7
Middle aged broadleaf forest, height 6 to 12 m [m²] 1114 0.00 8641 1.97 3.39
Mature broadleaf forest. height above 12 m [m²] 2568 0.00 9620 0.751 -0.577
Mature coniferous forest. height above 12 m [m²] 3051 0.00 10,000 0.824 -0.428
Edge length of patches [m] 568 0.00 1257 0.048 -0.381
Mean Min. Max. Skewness Kurtosis
Mean canopy height [m] 14.6 1.75 28.6 -0.336 -0.242
Standard deviation of canopy height [m] 8.29 2.87 12.4 -0.286 -0.110
Maximum canopy height [m] 37.7 22.6 51.7 0.261 -0.0276
Penetration rate 5 to 1 m above ground [%] 71.3 35.1 93.5 -0.593 -0.508
Penetration rate 10 to 2 m above ground [%] 63.5 21.1 76.3 -0.367 -0.523
Variables derived from LiDAR
Variables extracted from aerial photographs
Variables from field measurements
Müller et al. BAE 2011
Uses: Birds – predictor variables
Canonical root 1Aerial photographs
-0.1 0.0 0.1 0.2
Ca
no
nic
al
roo
t 1
LiD
AR
-0.2
-0.1
0.0
0.1
0.2 a
Canonical root 1Field measurements
-0.1 0.0 0.1 0.2
1 2 3 4 5
0
0.5
1.0
1 2 3 4 5
0
0.5
1.0
b
Pairwise correlations between sets: < 0.5
But canonical correlations: > 0.8
Müller et al. BAE 2011
Uses: Birds – predictor variables
„predictive power“:
LiDAR > field measurements
Müller et al. BAE 2011
predictive power R²
0 0.1 0.2 0.3 0.4 0.5
Ph. trochilus
Ph. collybita
P. modularis
F. coelebs
T. troglodytes
S. atricapilla
S. europaea
L. curvirostra
C. familiaris
T. viscivorus
P. ater
R. ignicapillus
T. merula
E. rubecula
T. philomelus
C. spinus
Py. pyrrhula
P. caeruleus
P. palustris
P. cristatus
R. regulus
G. glandarius
P. major A B
unique contribution
0 0.1 0.2
Performance – R² (test data set)
Aerial photographs
LiDAR
Uses: Birds – abundance
223 plots
Predictive R²
0 0.2 0.4 0.6
Tenuiphantes alacris
Macrargus rufus
Drapetisca socialis
Walckenaeria cucullata
Amaurobius fenestralis
Cybaeus angustiarum
Pardosa ferruginea
Centromerus sellarius
Callobius claustrarius
Micrargus georgescuae
Diplocephalus picinus
Cryphoeca silvicola
Tenuiphantes tenebricola
Histopona torpida
Saloca diceros
Coelotes terrestris
Robertus scoticus
Pardosa lugubris
Harpactea lepida
Gnaphosa badia
Pardosa sordidata
Diplocephalus latifrons
Alopecosa taeniata
Centromerus pabulator
Eurocoelotes inermis
Ground
Lidar
Frequency
0 50 100
A B
„predictive power“:
LiDAR > field measurements
Vierling et al. Ecol Appl 2011 Performance – R² (test data set)
171 plots
pitfall traps
Eurocoelotes inermis
Centromerus pabulator
Uses: Spiders – abundance
Space
[3]
Assemblage of forest passerines
31 species – 20.6%
15.6 % 5.1 % 13.5 %
14.7% 5.2%
2.0%
21.0 %
0 20 40 0 20 40 0 20 40 0 20 400 20 40 0 20 400 20 40 0 20 40
0 20
Joint effects
Independent effects
Müller et al. BAE 2011
Uses: Birds – community composition
Müller & Brandl J Appl Ecol 2009
Uses: Arthropods – community
Pitfall traps
Flight-interception traps
Total
R2
Total
95% CI
LiDAR
%
Biotic
%
Abiotic
%
Total
R2
Total
95% CI
LiDAR
%
Biotic
%
Abiotic
%
Individuals 8.9 –0.3–19.9 17.7 –8.0 54.2 43.8 30.5–55.7 99.8 75.8 54.8
Richness 3.0 –8.7–10.7 99.5 99.0 5.9 26.4 13.6–39.4 89.1 30.1 48.4
Diversity 3.7 -6.5–19.8 99.0 54.3 -25.7 23.8 14.3–34.0 94.8 60.5 66.5
Body size 27.1 14.3–39.6 87.6 60.6 19.0 14.7 3.9–27.2 66.8 31.0 14.0
171 plots
50,910 individuals
782 species
Müller & Brandl J Appl Ecol 2009
Uses: Communities – species richness
Bässler et al. Biodivers Conserv 2010
Uses: Forest types
• LiDAR
• Climate and soil
• Vegetation (Cover-abundance of species)
a) b c)
LiDAR
Accuracy: 69%
Climate and soil
Accuracy: 68%
Vegetation
Accuracy: 67%
Bässler et al. Biodivers Conserv 2010
Asperulo-Fagetum
Picea-forests
Bog-forests
Luzulo-Fagetum
Uses: Forest types
40 50 60 70 80
0
10
20
30
40
50
Fre
qu
en
cy
LiDAR
40 50 60 70 80
0
10
20
30
40
50
Overall accuracy [%]
Fre
qu
en
cy
Climate and soil
40 50 60 70 80
0
10
20
30
40
50
Overall accuracy [%]
Vegetation
Bässler et al. Biodivers Conserv 2010
Uses: Forest types
Reduced training data (50%)
Predictive R²
0 0.2 0.4 0.6
Tenuiphantes alacris
Macrargus rufus
Drapetisca socialis
Walckenaeria cucullata
Amaurobius fenestralis
Cybaeus angustiarum
Pardosa ferruginea
Centromerus sellarius
Callobius claustrarius
Micrargus georgescuae
Diplocephalus picinus
Cryphoeca silvicola
Tenuiphantes tenebricola
Histopona torpida
Saloca diceros
Coelotes terrestris
Robertus scoticus
Pardosa lugubris
Harpactea lepida
Gnaphosa badia
Pardosa sordidata
Diplocephalus latifrons
Alopecosa taeniata
Centromerus pabulator
Eurocoelotes inermis
Ground
Lidar
Frequency
0 50 100
A B
predictive power R²
0 0.1 0.2 0.3 0.4 0.5
Ph. trochilus
Ph. collybita
P. modularis
F. coelebs
T. troglodytes
S. atricapilla
S. europaea
L. curvirostra
C. familiaris
T. viscivorus
P. ater
R. ignicapillus
T. merula
E. rubecula
T. philomelus
C. spinus
Py. pyrrhula
P. caeruleus
P. palustris
P. cristatus
R. regulus
G. glandarius
P. major A B
unique contribution
0 0.1 0.2
Challenges
• Biodiversity modelling with LiDAR is
phenomenological
Why differ species in predictability?
• Hypothesis: Predictability as a species trait?
None Strong
Pagel´s
Body size
all values P < 0.003
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Challenges: Phylogenetic signal
100 possible trees from https://www.birds.org/
None Strong
0.0 0.2 0.4 0.6 0.8 1.0 1.2
100 possible trees from https://www.birds.org/
Pagel´s
Predictability
all values P < 0.003
predictive power R²
0 0.1 0.2 0.3 0.4 0.5
Ph. trochilus
Ph. collybita
P. modularis
F. coelebs
T. troglodytes
S. atricapilla
S. europaea
L. curvirostra
C. familiaris
T. viscivorus
P. ater
R. ignicapillus
T. merula
E. rubecula
T. philomelus
C. spinus
Py. pyrrhula
P. caeruleus
P. palustris
P. cristatus
R. regulus
G. glandarius
P. major A B
unique contribution
0 0.1 0.2
Challenges: Phylogenetic signal
Body size female [mm]
0 2 4 6 8 10 12
Pre
dic
tive R
²
0
0.2
0.4
0.6
Frequency
20 40 60 80 100
Niche position Shading
0.4 0.8 1.2 1.6
Pre
dic
tive R
²
0
0.2
0.4
0.6
Niche position Moisture
-0.4 0.0 0.4 0.8
R² = 0.08p = 0.17
R² = 0.12p = 0.08
R² = 0.27p = 0.01
R² = 0.01p = 0.56
Vierling et al. Ecol Appl 2011
open closed moist dry
Eurocoelotes inermis
Centromerus pabulator
Challenges
Summary
• LiDAR provides fine grained information of the vegetation structure
• LiDAR information captures similar characteristics than ground
measurements
• LiDAR information can be used to model biodiversity
• The performance of models differs considerably between…
– Species
– characteristics of assemblages
– composition of assemblages
• These variations in performance occurs also with other data sets
• Hypotheses and tests to predict predictability are badly needed