Estimation of Site Occupancy

1
Estimation of Site Occupancy

• Estimation of site occupancy and related turnover
  dynamics (extinction and colonization) from
  presence-absence data is important for monitoring
  efforts, wildlife-habitat models, and
  metapopulation studies.

• Is a defined patch or site occupied by 1
  individual of a species of interest?

• Field approach involves visiting a number of
  sites and recording whether species is present
  (coded as 1) or absent (coded as 0).

• POTENTIAL PROBLEMDoes an absence represent a
  true absence or a false absence that reflects
  species are detected with probability lt1.0?

2
False Absences from Imperfect Detection

• Underestimate site occupancy and colonization
  rates.

2. Overestimate local extinction rates.
3. Bias predictive wildlife-habitat models if
detectability is related to measured habitat
variables.
If detection probability can be estimated, then
unbiased estimators of occupancy, colonization,
and extinction can be derived.
3
Occupancy Modeling

• Follows from general ideas of closed-population
  mark-recapture models.
• In occupancy models, sites are equivalent to
  individuals for M-R models.

• ASSUMPTIONS
• Closed systemif site is occupied, it is always
  occupied during primary sampling period.
  Necessitates sampling during relatively short
  time period.
• No false presences (but false absences possible)
• Species detection at site is independent of
  detection at other sites.

4
Occupancy Modeling

• Data for single site takes the form of an
  encounter history
• 1001
• where 1 detection and 0 nondetection.

• Because species was detected at least once at
  site, we assume site was occupied for all four
  occasions, but species was not detected during
  second and third visits.

5

• We define psi (?) as the occupancy probability
• We define p as the detection probability

Likelihood statements for four encounter
histories
11 psip(1)p(2) 10 psip(1)1 -
p(2) 01 psi1 - p(1)p(2) 00 psi1 -
p(1)1 - p(2) (1 - psi)
Note that p could vary over time, but psi cannot
(closed system).
6
Covariates

• Covariatean explanatory variable that is
  continuous instead of categorical (like the
  groups that we worked with last week).

• Occupancy models can include covariates using a
  logistic model.

• Occupancy might be related to site-specific
  covariates.
• Likewise, detection might be related to
  site-specific covariates.

7
Extension to estimate extinction and colonization
rates
11001 10001 00000 00010
Primary sampling period when closure is assumed.
Same framework as ROBUST DESIGN for
mark-recapture analysis.
8
Occupancy of pitcher plant bogs by wintering
Henslows sparrows (Ammodramus henslowii)

• Migratory species that breeds in moist grasslands
  of central and eastern US and migrates to
  southeast where it winters mainly on open pine
  savannas and pitcher plant bogs.

• Between 1966-1996, breeding Henslows sparrows
  declined and average of 8.8 per year. Loss of
  breeding habitat important, but loss of wintering
  habitat could be contributing to decline.

• Gulf Coast pitcher plant bogs have been reduced
  to lt3 of former area.

• Natural fire regimes of pitcher plant bogs
  include fairly frequent fires that occur during
  growing season (late spring and summer).

(data from Tucker, JW, and WD Robinson. 2003. Auk
12096-106)
9

• Data set includes 42 bogs that were sampled two
  times for occupancy by sparrows during 2-month
  period in winter of 1999-2000.

• One observer conducted all of the surveys by
  walking adjacent transects across area and
  flushing sparrows (they rarely vocalize in
  winter).

• Measured 10 covariates (see handout).