Upload
others
View
6
Download
0
Embed Size (px)
Citation preview
Avaliando padrões emergentes e processos ecológicos por meio
de simulação de movimentos de aves em paisagens fragmentadas
MILTON CEZAR [email protected]
Referata biodiversaINPE, Julho 2010
Assessing simulated bird species responses to landscape structure:spatial explicit IBM
RIBEIRO * FORTIN * METZGER et al., in reviewUofT-CANADA * USP-BRAZIL
MILTON RIBEIRO – [email protected]
STUDY BACKGROUND:• Habitat Loss and Fragmentation
• Negative effect on Biodiversity
HABITAT MATRIX
Habitat loss WITHOUT Fragmentation
Habitat loss WITH Fragmentation
STUDY BACKGROUND:• Fragmentation Threshold:
10-30% of habitat (Andrén 1994; Fahrig 2003)
Radford et al (2005)
STUDY BACKGROUND:
• Brazilian Atlantic Forest : 12% of remaining habitat
• ~90% of habitat for forest species persistence was lost
• How species respond to habitat configuration?
• How habitat qualityinfluences speciesdistribution?
Ribeiro et al 2009
OBJECTIVES:
• What influences the response patterns of simulated species?
Habitat AMOUNT * Habitat QUALITY * AGGREGATION
at patch level at landscape level
APPROACH:
• Individual Based Models• Using generic species profiles that we setup for
the Atlantic Forest bird species
• What influences individuals to move/disperse?• Habitat quality• Degree of crowdingness
• Why Individual Based Models?• Generate testable hyphothesis• Check if expected patterns occurs
APPROACH:
• Individual Based Model
• When individual are dispersing:
• The mortality rate is high on matrix
• Change the movement pattern• Routine maximum distance < explorative
distance
APPROACH:
• Individual Based Model
• “Cost” to move:
• Depends on position of individual• Interior of patch -> low cost• edge proximity -> medium cost• Matrix -> high cost
• Habitat quality:• High quality -> low cost• Medium quality -> medium cost• Low quality -> high cost
Simulated landscapes (n=10,000)
• Habitat amount: from 5 to 70%
• Aggregation: from 0.1 (random) to 1.0 (clumped)
• Habitat Quality: from low (0.1) to high (1.0)
METHODS
Continuous variables – Markov Chain Monte Carlo - MCMC
High Quality Medium Quality Low Quality/Matrix0.7 – 1.0 0.4 – 0.7 0.1 to 0.4
Simulated Landscape: 512x512 – resolution 30m
Continuous variables – Markov Chain Monte Carlo - MCMC
AMOUNT=60% AGGREG=0.85
AMOUNT=42% AGGREG=0.47
AMOUNT=25% AGGREG=0.32
AMOUNT=12% AGGREG=0.21
Simulated Landscape: 512x512 – resolution 30m
Continuous variables – Markov Chain Monte Carlo - MCMC
AMOUNT=55% AGGREG=0.92
AMOUNT=42% AGGREG=0.47
AMOUNT=25% AGGREG=0.32
AMOUNT=12% AGGREG=0.21
EDGE SENSITIVE HABITAT-DEPEND. – DON’T USE CORRIDOR
EDGE AND CORRIDOR TOLERANT – USE CORRIDOR
GAP CROSSING OF 240M
Species profiles:
X
GAP CROSSING OF 60M
GAP CROSSING OF 120MEDGE TOLERANT-DON’T USE CORRIDOR
120 M
GAPCROSSING
PROBABILITY
Gap Crossing - Awade & Metzger 2008 (understory birds)
Method = Play backThamnophilus caerulescens
HABITAT MATRIX
Response to Edge – Hansbauer et al 2008 (understory birds)
Method = Telemetry
DISTANCE TO EDGE (m)
PreferenceIndex
>200 M
C. caudata
MATRIX HABITAT
Home-range & use of Non-Forest habitatsHansbauer et al 2008 (understory birds) -Method = Telemetry
C. caudata
MATRIX = agriculture; pasture; rural building and silviculture
500 m
Habitat
Matrix
Starting
population size =
Habitat amount/HR
Define position
Start simulation
Map database:
Landscape +
auxiliary maps
Species profile: HR,
routine distance,
dispersing distance
landscape
FOR time step=1 to N
Overcrowding on patch?
Functional patches
Habitat qualityDistance to edge/patch
change to disperser
movement state
yes
no
ALIVE?
Not move
no
Dispersing?
yes
Follow
disperser rule
Follow routine rule
noyes
Estimate prob.
of death, movement
cost
New position
Summary for landscape
Dispersal rate Mortality rateMovement costMovement variability
RESPONSE VARIABLES – *at landscape level*DISPERSAL RATE:
• % of individual dispersing after a simulation
MOVEMENT VARIABILITY:
MORTALITY RATE:• % of mortality after a simulation
MOVEMENT COST: 1
(position * habitat quality)
TOTAL LENGTH
Euclidean DISTANCE
0.0
0.2
0.4
0.6
0.8
1.0
Distance from edge (m)
Sa
fen
ess
-50
0
-24
0
-18
0
-12
0
-90
-60
-30 0
30
60
90
12
0
18
0
24
0
50
0
Edge sensitive
Edge tolerant
Edge and corridor tolerant
Generalist low gap crossing
Generalist medium gap crossing
0.0
00
00
.00
10
0.0
02
00
.00
30
Distance from edge (m)
Mo
rta
lity p
rob
ab
ility
-50
0
-24
0
-18
0
-12
0
-90
-60
-30 0
30
60
90
12
0
18
0
24
0
50
0
Edge sensitive
Edge tolerant
Edge and corridor tolerant
Generalist low gap crossing
Generalist medium gap crossing
Experimental Design
• Home range size (HR): 10 ha
HR=Surrogate for Carrying capacity, used to define Starting Population Size
• Time step per simulation: 500 movements
• Maximum ROUTINE movement distance: 30 m• maximum EXPLORATIVE movement distance: 150 m
• N replicates: 10,000 runs, for EDGE AND CORRIDOR TOLERANT species profile
Expected Patterns:
HABITAT AMOUNT
HABITAT QUALITY AGGREGATION
Mortality
Dispersal rateMov. Variability
Movement cost
Landscape Ecology And Conservation Lab - USP
IBM simulations – time steps = 500Population size = 100
Edge sensitive Edge and corridor tolerant
Low mobility Medium mobility
Amount of habitat (%)
Dis
pe
rsa
l ra
te (
%)
0
5
10
15
20
10 20 30 40 50 60
Landscape-response:Each points is one simulation (500 time steps)For one landscape, and for one species profile
RESULTS
Fitted by Generalized Additive Model (GAM)
Statistical Approach: Model Selection (AIC)
DISPERSAL RATE
Amount of habitat (%)
Dis
pe
rsa
l ra
te (
%)
0
5
10
15
20
10 20 30 40 50 60
Aggregation
Dis
pe
rsa
l ra
te (
%)
0
5
10
15
20
20 40 60 80
Quality of landscape
Dis
pe
rsa
l ra
te (
%)
0
5
10
15
20
0.3 0.4 0.5 0.6
HABITAT AMOUNT AGGREGATION HABITAT QUALITY
Amount of habitat (%)
Mo
rta
lity
ra
te (
%)
0
2
4
6
8
10 20 30 40 50 60
Agregation
Mo
rta
lity
ra
te (
%)
0
2
4
6
8
20 40 60 80
Quality of landscape
Mo
rta
lity
ra
te (
%)
0
2
4
6
8
0.3 0.4 0.5 0.6
MORTALITY RATE
HABITAT AMOUNT AGGREGATION HABITAT QUALITY
Amount of habitat (%)
Mo
ve
me
nt co
st
50
100
150
200
250
300
10 20 30 40 50 60
Agregation
Mo
ve
me
nt co
st
50
100
150
200
250
300
20 40 60 80
Quality of landscape
Mo
ve
me
nt co
st
0
50
100
150
200
250
300
0.3 0.4 0.5 0.6
MOVEMENT COST
HABITAT AMOUNT AGGREGATION HABITAT QUALITY
Amount of habitat (%)
Mo
ve
me
nt va
ria
bility
5
10
15
20
10 20 30 40 50 60
Agregation
Mo
ve
me
nt va
ria
bility
5
10
15
20
20 40 60 80
Quality of Landscape
Mo
ve
me
nt va
ria
bility
5
10
15
20
0.3 0.4 0.5 0.6
MOVEMENT VARIABILITY
HABITAT AMOUNT AGGREGATION HABITAT QUALITY
Response variables Selected Models DAICc wAICc (wAIC)
Movement variability Amount of habitat+Aggregation 0.00 0.89 0.89
Dispersal rate Quality of habitat patches 0.00 0.52 0.80
Quality of landscape 2.48 0.15
Amount of habitat 2.83 0.13
Mortality rate Quality of habitat patches 0.00 0.24 0.69
Quality of landscape 0.15 0.23
Amount of habitat 0.18 0.22
Movement cost Amount of habitat+Aggregation 0.00 0.93 0.93
Expected Patterns:
HABITAT AMOUNT
HABITAT QUALITY AGGREGATION
Mortality
Dispersal rateMov. Variability
Movement cost
Observed Patterns:
HABITAT AMOUNT
HABITAT QUALITY AGGREGATION
Mov. VariabilityMovement cost
MortalityDispersal rate Mortality
Dispersal rateMov. Variability
Movement cost
Future Work
• Other species profiles including those with gap-crossing capability of 60 and 120 m?(functional connectivity)
• Daily-maximum distances >30 m?
• When dispersing-maximum distances >150 m
• Larger HR sizes (>50 ha)? (Top-predators)
Frame-work advantages• A completely continuous modelling
• Habitat amount, quality and aggregation• Individual position completely free (not pixel-
based)• Movement distances (daily or when dispersing):
continuous following a uniform distribution• Home-range size: fixed our variables• Perceptual range position, habitat quality and
mortality • Simulate different species profile, varying all
parameters that allow to adjust to species that can respond to different scales
• Fully spatial explicit response pattern, including density-dependence behaviour
Untangling the effects of landscape spatial
structure, habitat quality, and species
traits on ecological processes
RIBEIRO * MARTENSEN* FORTIN * METZGER - in prepUofT-CANADA * USP-BRAZIL
MILTON RIBEIRO – [email protected]
OUTLINE:• Study background and research motivation• Confounding Effects• BioDIM (Biologically scaled Dispersal Model)• Response patterns• Sensitivity analysis• Results across and within species• Summary
• Future developments for BioDIM
STUDY BACKGROUND:• Biodiversity and ecossystem variability
• How to explain?• How much we can explain?
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
STUDY BACKGROUND:• Biodiversity and ecossystem variability
• How to explain?• How much we can explain?
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
System variability
STUDY BACKGROUND:• Biodiversity and ecossystem variability
• How to explain?• How much we can explain?
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
System variability
Processes(responses)
Patterns(the effects)
DispersalMatingReproductionGene flowMortalityForest sucessionBiomass production
Habitat lossFragmentationPatch size distrib.IsolationHabitat qualityConnectivityLand-cover dynamic
STUDY BACKGROUND:• Biodiversity and ecossystem variability
• How to explain?• How much we can explain?
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
Processes(responses)
Patterns(the effects)
DispersalMatingReproductionGene flowMortalityForest sucessionBiomass production
Habitat lossFragmentationPatch size distrib.IsolationHabitat qualityConnectivityLand-cover dynamic
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
Full system variability
Pattern 1
Pattern 2
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
Pattern 1
Pattern 2
Processes(responses)
Patterns(the effects)
STUDY BACKGROUND:• Habitat Loss and Fragmentation
• Negative effect on Biodiversity• Habitat quality are important for:
• Species persistence• Population density• Carrying capacity• Reproductive success
• Landscape structure• Influences species movement• Patch accessibility• Dispersal success• Meta-community maintenance
• Species traits• How individuals interact with landscape features• Edge sensitivity / Gap crossing capatity
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
OBJECTIVES:• Estimate the strength of influence for the three
groups of factors on species response patterns:
Landscape structureHabitat QualitySpecies traits
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
Studies bottle neck:• Assess the influencing factors independently
APPROACH:
• BioDIM (Individual Based Models)Biologically scaled Dispersal Model
• Using generic species profiles that we setup for the Atlantic Forest bird species
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
Species profiles:
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
A
B
C
D
E
F
Binary landscape
EdgeTolerant
Gap crossing60 m
Edge sensitive
Edge and corridor tolerant
Gap crossing 120 m
RESPONSE VARIABLES – *at landscape level*
DISPERSAL RATE: % of individual dispersing after a simulation
MOVEMENT VARIABILITY:
MORTALITY RATE: % of mortality after a simulation
ENCOUNTER RATE: number of enconters Male/Female
MOVEMENT COST: 1
(position * habitat quality)
TOTAL LENGTH
Euclidean DISTANCE
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
EXPERIMENTAL DESIGN:
Landscape structure:PLAND: Amount of habitat (5% to 70%)AGGREG: Aggregation of habitat (0.1 to 1.0)
Habitat quality:HQBIN: include quality on model: Yes/No?HQLAND: Habitat quality at landscape levelHQFRAG: Habitat quality at patch level
Species traits:SPECSENSIT: Species sensitivity to spatial heterogeneity
(i.e. Species profiles)HORANG: home-range size, varying continuously from 5 to 30 ha;ROUTDIST: maximum routine distance (30 to 90 m; uniform distribution)EXPLDIST: maximum explorative distance (1.0 and 3.0 times the
ROUTDIST, i.e. 90 and 270 m; uniform distribution)
10 000 replicates / 500 time steps
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
SENSITIVITY ANALYSIS (Saltelli et al. 2008):
Standardized Regression Coefficient:SRC = -1 to +1 & +Std. Error
sensitivity package; Pujol 2008
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
SENSITIVITY ANALYSIS (Saltelli et al. 2008):
Standardized Regression Coefficient:SRC = -1 to +1 & +Std. Error
sensitivity package; Pujol 2008
Background * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
Positive effect Negative effect
+++
Or
- - -
++
Or
- -
+
Or
-
RESULTSBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
Positive effect
Negative effect
Including all species
RESULTSBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
Response variables
Explanatory
variables All species
Edge-
sensitive
Edge-
tolerant
Edge- and
corridor-
tolerant
Low gap-
crossing
Medium
gap-crossing Observation
Dispersal rate SPECSENSIT + + +
PLAND - - - - - - - - - - - always present
AGGREG - - - - - - - - - - - - + strong/trend/always present
HQBIN 0 0 0 0 0 +
HQLAND 0 0 + - 0 - - partial
HQFRAG 0 0 - 0 0 +
HORANG 0 - - 0 + + trend/partial
ROUTDIST + + + + + + + + 0 trend/partial
EXPLDIST 0 0 0 - - - partial
Movement variability SPECSENSIT 0
PLAND + - + + 0 0 + + partial
AGGREG + 0 0 + + + partial
HQBIN - - - - - 0 partial
HQLAND + + + 0 0 0 0
HQFRAG - - - - - 0 partial
HORANG 0 - - 0 0 0
ROUTDIST 0 - 0 - 0 - partial
EXPLDIST + + + + + + + 0 trend/partial
SPECSENSIT 0 0 0 0 0 0
Movement Cost SPECSENSIT ++
PLAND - - - - - - - - - - - - - - - - - - strong/always present
AGGREG 0 - - - + + trend/always present
HQBIN + + 0 0 + + partial
HQLAND - - 0 - - - - partial
HQFRAG + + + 0 + + + partial
HORANG + + + + + + + + + + + + always present
ROUTDIST 0 0 0 0 0 0
EXPLDIST - 0 - - - - Partial
Within species
RESULTSBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
Response variables
Explanatory
variables All species
Edge
sensitive
Edge
tolerant
Edge and
corridor
tolerant
Low gap
crossing
Medium gap
crossing observation
Encounters rate SPECSENSIT + +
PLAND + + + + + + + + + + always present
AGGREG + + 0 + + + + + + + + partial
HQBIN 0 0 - 0 0 0
HQLAND 0 0 0 + 0 0
HQFRAG 0 0 - - 0 0
HORANG - - - - - - - - - - - - - - - - - - strong/always present
ROUTDIST 0 + 0 0 0 0
EXPLDIST 0 0 0 0 0 0
Mortality SPECSENSIT +++
PLAND - - - - - - - - - - - - - strong/always present
AGGREG - - - - - - - - - - - - - - - - strong/always present
HQBIN 0 0 + + 0 0
HQLAND - - 0 - - - 0 partial
HQFRAG + + 0 + + 0 partial
HORANG 0 0 0 0 0 0
ROUTDIST + + + + 0 + partial
EXPLDIST 0 0 0 0 0 0
Within species
SUMMARYBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
• Species sensitivity was the primary factor to explain DISPERSAL rate and MORTALITY
• Amount of habitat: primary factor for MOVEMENT COST
• Home-range size: primary factor for ENCOUNTER Rate
• Movement variability: no strong factor was detected
Including all species
SUMMARYBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
• Amount of habitat and aggregation were the secondary factor for DISPERSAL rate and MORTALITY
• Home-range size and Species sensitivity: secondary for MOVEMENT COST
• Species sensitivity : secondary factor for ENCOUNTER Rate ; amount and aggregation of habitat are of tertiary influence on ENCOUNTER RATE
• Habitat quality not influenced any response variables
Including all species
ImplicationsBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
• Important to define “conservation target”• Species? Taxon?
• Landscape spatial heterogeneity was always important, but its strength of influence varied greatly depending on “process of interest” (i.e. response pattern) and species sensitivity
• Although habitat quality are pointed out as important to population’ s maintenance, this type of information are hard to obtain, and are generally measured at patch scale, not at landscape scale
Within Species
Future WorkFuture studiesBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
• Ecological Thresholds • BioDIM & Landscape genetic• BioDIM & 3D• BioDIM & Continuous surface• Apply BioDIM for sampling issues:
landscape genetic
Future WorkEcological thresholdsBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
HABITAT AMOUNT
MO
VEM
ENT
CO
ST
Future WorkEcological thresholdsBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
HABITAT AGGREGATION
DIS
PER
SAL
RAT
E
HABITAT AGGREGATION
DIS
PER
SAL
RAT
E
HABITAT AMOUNT
HA
BIT
AT A
GG
REG
ATI
ON
DISPERSAL RATE
population
GENETICSUB-MODEL
Individuals
LOCI structure
Proximity perception (PROX m)
Sex (F/M) rate
Gene exchange rate
FOR individual =1 to
popsize
Is Female
Exit genetic module
No
yes
Dist < Prox mno
FOR other individual
yes
Other Is Male
yes
no
Add 1 to # meetings
Change LOCI struct for
potential offspring
Example of LOCI STRUCStart: [0,1,1,0,1,0,1] [1,1] [0,1,0,1,1] [0,0]
End: [0,1,0,0,1,1,1] [1,0] [1,1,0,1,1] [1,0]
Population Size is the sameMovement pattern is the same
Routine daily distance = lower Routine daily distance = higher
Sampling issues on landscape genetic studies
Future studiesBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
Continuous landscapes
Future studiesBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
3D scaled landscapes: a challengeBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
3D scaled landscapes: a challengeBackground * Effects * BioDIM * Responses * Sensitivity * Results * Summary * Future
680
700
720
740
760
780
800
-200 0 200 400 600 800 1000 1200 1400
Perfil1 Perfil2 Perfil3
Perfil4 Perfil5
Distance between paired fragments - m
Altitude (meters)
700
710
720
730
740
750
760
770
780
790
-200 0 200 400 600 800 1000 1200 1400
730
740
750
760
770
780
790
-200 0 200 400 600 800 1000 1200 1400
Perfil3
735
740
745
750
755
760
765
770
775
780
-200 0 200 400 600 800 1000 1200 1400
Acknowledgment:
Brownyn Rayfield, Josie Hughes, Ilona Naujokaitis-Lewis and Danilo Boscolo for fruitful discussion about ecological modelling and birds behaviour at Atlantic Forest
• National Council for Scientific and Technological Development (CNPq)• The State of São Paulo Research Foundation (FAPESP)• Petrobras Company
Landscape Ecology And Conservation Lab - USP