Transcript Rick Southgate 2

The track-based monitoring
technique and the estimation of
occupancy and detection rates
Rick Southgate1 and Rachel Paltridge2
1
2
Envisage Environmental Services
Desert Wildlife Services
Outline
• Track-based monitoring
• Types of data
• Occupancy and detection modelling
PRESENCE
• Asserting absence
Bayesian approach
• The way forward
Track-based monitoring: motivation ~2000
Track plots
+
experienced trackers = meaningful data
Track-based monitoring: motivation ~2000
Track plots + indigenous communities = meaningful work
Track-based monitoring: motivation ~2000
2.1 M km2 of sand dunes
Potential application
enormous
Track-based monitoring: motivation ~2006
structured
program
+
national
coordination
Methodology
Federal agencies
Verification
• DEEWR
Training
• DAFF
Accreditation
• DEWHA
Data collation
- NRM
Analysis
- IPA
Feedback
=
positive broad-scale
monitoring &
community benefits
State agencies
Indigenous comm.
NGOs
Consultants
Camel occurrence
Track-based monitoring: 2013
Over 1500 plot locations
Proponents:
• KJ
• CDNTS
• CLC
• NRAW SA
• NRAL SA
• Consultants
•Envisage Env. Ser.
•Desert Wildlife Ser.
•Ecological Horizons
Bilby occurrence
IBRA7 regions
Track-based monitoring: 2 ha plots
Similar to BirdsAustralia 2 ha sample method
Provide a snap-shot of spp. present/absent at a site
(spp. >~100 g)
Standarise effort & approach, repeatable
•
200 x 100 m plot searched
•
25-30 minute
•
Experienced observers
Track-based monitoring - 2 ha plots
Three components to site selection:
•
Spacing between sites to achieve independence
(generally > 5 km)
•
Repeat visits to sites to address imperfect detection
•
Stratify sites on substrate & sub-bioregion
Response variable - 2 ha plots
•
Id species based on track characteristics
•
Age of sign (1-2 day, 3-7, >7 days)
- comparison of small: large animal sign
•
On-plot: on-road
- comparison of transit v non-transit spp
− Juvenile sign
− Abundance of sign
− Diggings, burrows, scats
Site (occupancy) covariates - 2 ha plots
• Potential management factors
Fire age pattern, dist. to community & water
• Threats
Invasive predators, herbivores etc
• Habitat
Substrate, rainfall, veg composition, cover etc
Detection covariates - 2 ha plots
• Time of day (tracks crisp, sun angle, observer fatigue)
• Light intensity (shadow strength: track visibility)
• Track surface continuity (gait visibility)
• Track surface quality (small v. large animals)
Additive:
=> Ordinal detection score
Species detection in relation to tracking conditions
hopm
0.8
greyk
0.7
redk
dectectability
0.6
cat
0.5
fox
0.4
dingo
0.3
camel
0.2
rabbit
0.1
0
4
5
6
7
8
ODS
9
10
11
12
2 ha tbm data by latitude
latitude
bilby
dingo
fox
cat
camel
rabbit
n
16
0.14
0.26
0.00
0.37
0.00
0.00
37
18
0.14
0.21
0.03
0.59
0.26
0.00
220
20
0.20
0.25
0.20
0.65
0.32
0.02
189
22
0.07
0.29
0.13
0.34
0.44
0.02
603
24
0.06
0.16
0.04
0.24
0.18
0.02
71
26
0.08
0.34
0.11
0.34
0.32
0.38
80
28
0.00
0.34
0.62
0.26
0.21
0.62
385
30
0.00
0.19
0.41
0.32
0.03
0.61
133
2 ha tbm data by bioregion
bioregion
bilby
dingo
fox
cat
camel
rabbit n
gsd
0.12
0.27
0.17
0.43
0.51
0.03
306
pilbara
0.00
0.50
0.07
0.25
0.30
0.00
44
lsd
0.03
0.24
0.12
0.26
0.37
0.01
271
gas
0.02
0.28
0.06
0.34
0.28
0.02
47
gid
0.10
0.23
0.00
0.35
0.35
0.00
69
Types of data
Abundance of species at a site
-> ordinal or continuous data
Presence/absence of species at a site
-> binary data: 0 or 1
Binary data from multiple sites
-> propn of area occupied (f)
provides a surrogate for sp. abundance
- true for broad-scale surveys
- true for cryptic, low density species.
- occurrence less expensive than abund.
Problems arise if a species is not detected perfectly
• Non-detection may mean the sp. is not genuinely absent
• Propn area occupied underestimated etc.
Monitoring
Observed state
Detected
Not detected
Actual state
Genuine presence True presence
Genuine absence
False absence
False presence True absence
Monitoring
Observed state
Detected
Not detected
Actual state
Genuine presence True presence
Genuine absence
False absence
False presence True absence
Monitoring
Observed state
Detected
Repeat
surveys
Not detected
Actual state
Genuine presence True presence
Genuine absence
False absence
False presence True absence
Incorrect ids not tolerated:
Validate! If in doubt, leave out
Data types and probability estimates
Revisits to multiple sites -> detection history for each site eg.00101
-> naïve est. (which is of more value than f )
-> prob. of detection (p)
-> prob. of occupancy (psi)
an unbiased estimate of propn area occupied.
Occupancy and detection modelling
PRESENCE
Developed by Darryl MacKenzie and colleagues
use standard maximum likelihood based methods to obtain estimates
logistic models to incorporate covariates
strength covariables associating with detection eg. observer
strength covariables associating with occupancy eg. bioregion
Important parameters:
Prob of occupancy (psi):prob. that a species is present at a site
(constant across all sites)
Prob of detection (p): prob. a species will be detected in a single survey
at a particular site given a site is occupied
-> used to determine sampling effort, assert absence, species status etc
Detection: survey effort
Survey effort (n*) to determine the status of a species
at a site depend on:
• the suitability of a habitat (psi’)
•
•
n*=
the reliability of a survey to detect a species (p)
the probability of the occupancy required when the
survey fails to detect the species (psi). Do we need
95% or 99% confidence?
(log(psi/(1-psi))-log(psi'/(1-psi')))/log(1-p')
where
psi =posterior prob of presence at a site
(confidence you need)
p'= prior belief about detectability of species
psi' = prior prob that the species is present
Summary
• Need to apply standarised techniques
• Revisiting, resampling sites – funding
agencies need to recognise importance
• Data sharing – sort out data ownership,
management and access agreements
Thank you
Acknowledgements
•
•
•
•
•
•
KJ
CDNTS
Maralinga Tjarutja Council
DENR
AWNRMB
ALNRMB