multiple covariate distance sampling (mcds) · buckland et al. (eds). advanced distance sampling....
TRANSCRIPT
Multiple covariate distancesampling (MCDS)
• Aim: Model the effect of additional covariates on detection probability, inaddition to distance, while assuming probability of detection at zero distance is 1
• References:
• Marques (F) and Buckland (2004) Covariate models for the detection function. Chapter 3 inBuckland et al. (eds). Advanced Distance Sampling.
• Marques (T) et al. (2007) Improving estimates of bird density using multiple covariate distancesampling. The Auk 127: 1229-1243.
• Section 5.3 of Buckland et al. (2015) Distance Sampling: Methods and Applications
Contents
•Why additional covariates?
•Multiple covariate models
•Estimating abundance
•MCDS in Distance
2nd Lecture:
•Complications• Clusteredpopulations
• Adjustm entterm s
• S tratification
•MCDS analysis guidelines
x
g(x)
x
g(x)
In conventional distance sampling(CDS) analysis all factors affectingdetectability, except distance, areignored
In reality, many factors mayaffect detectability
Sources of heterogeneity:
Object : species, sex, cluster size
Effort: observer, habitat, weather
Why additional covariates?
Examples of heterogeneity 1Effect of time of day on Rufous Fantail birds in Micronesia (point transects). Ramsey et. al. 1987.Biometrics 43:1-11
x
x
g(x)
g(x)
Examples of heterogeneity 2
Effect of sea state (and other covariates) on sea turtles in the Eastern Tropical Pacific(shipboard line transects). Beavers and Ramsey, 1998, J. Wildl. Manage. 63: 948-957
Examples of heterogeneity 3
Effect of cluster size on beer cans. Otto and Pollock, 1990, Biometrics 46: 239-245
Gp Size=1 Gp Size=4
Gp Size=2 Gp Size=5
SIG
HTI
NG
PR
OB
AB
ILIT
Y
SIG
HTI
NG
PR
OB
AB
ILIT
Y
SIG
HTI
NG
PR
OB
AB
ILIT
Y
SIG
HTI
NG
PR
OB
AB
ILIT
Y
DISTANCE
DISTANCEDISTANCE
DISTANCE
Why worry about heterogeneity?
• Pooling robustness works for all but extreme levels of heterogeneity
• Potential bias if density is estimated at a ‘lower level’ than detection function (e.g.density by geographic region, detection function global)
• Could potentially increase precision of detection function estimate
• Interest in sources of heterogeneity in their own right (e.g. group size)
In CDS, we use models that are pooling robust, so why worry about heterogeneity?
Dealing with heterogeneityStratification
Requires estimating separate detection function parameters for each stratum, so often not possible due to lackof data
Model as covariates in detection function
Allows a more parsimonious approach:
- can model effect of numerical covariates
- can ‘share information’ about detectionfunction shapebetween covariate levels
~ 680
~ 320
~ 140
cxpaxkm
jsjj /)()(
1
1
g(x) = Pr[animal at distance x is detected]
Key function
jth series adjustment term
Scaling constant to ensureg(0) = 1
Multiple covariate models Recap of CDS models
CDS models continued
1)(xk
2
2
2
xxk exp)(
Key functions
Hazard rate
Half-normal
Neg. exp.
Uniform
Series adjustments
Cosine cos(jπxs)
Polynomial xsj
Hermite poly. Hj(xs)
xsare scaled distances (see later)
xxk exp)(
bx
xk
exp)( 1
Scale parameter
Shape parameter
Modelling with covariates –ignoring adjustments terms (for now)
J
jjjzz
10 exp)(
2
2
2 )(exp),(
z
xzxk
b
z
xzxk
)(exp),(
1
g(x,z) = Pr[animal at distance xand covariates z is detected]
Assume the covariates affect the scaleof the key function, not its shape.So choose keyfunctions with a scale parameter
Let
e.g. Hazard rate
Half normal
kis used here to denote the “key” function
Modelling with covariates
)exp(.
)exp().exp(
exp)(
)(exp),(
sAA
s
ss
s
xsxg
21
2
10
10
2
2
Example: Dolphin tuna vessel data
Model: half-normal, with no adjustments
Covariate: cluster size, s
~ 680
~ 320
~ 140
From distance outputÂ1 = 2.331Â2 = 0.00895
Estimating abundance without covariatesusing Horvitz-Thompson estimator
ˆˆP rˆ
L
nA
A
Lincludedanim alN
n
i
n
i 22
11
11
)(
)(
)( xg
dxxg
xg
Recall that f(x)= pdf of observed x’s
Because g(0)=1 by assumption, then f(0) = 1/μ
So )(ˆˆ 02
fL
nAN
Estimating abundance with covariates
n
i i
n
i i
n
i zL
A
A
zLincludedanim alN
111
1
22
11
)(ˆ)(ˆP rˆ
Nowz)(
)z,(
)z,(
)z,()|(
xg
dxxg
xgxf
z
Because g(0,z)=1 by assumption, then )()|( zz 10 f
So
n
iif
L
AN
1
02
)|(ˆˆ z
Note similarity to CDS estimator
MCDS in Distance
In Model Definition, choose MCDS analysis engine
See Chapter 9 of online Users Guide
Covariate type:– Factor covariates classify the data into distinct classes
or levels. Can be numerical or text. One parameterper factor level.
– Non-factor (i.e., continuous) covariates must benumerical (integer or decimal). One parameter percovariate + 1 for the intercept.