Thomas%20C.%20Edwards,%20Jr.

1
MODEL-BASED STRATIFICATIONS FOR ENHANCING SURVEY
DETECTION RATES OF RARE SPECIES

• Thomas C. Edwards, Jr.
• USGS Utah Cooperative Research Unit

Richard Cutler, Mathematics Statistics, Utah
State University
Niklaus Zimmermann Swiss Federal Research
Institute WSL
2
RARE ECOLOGICAL EVENTSIN TIME AND SPACE

• Overview
• A (Biased) Historical Perspective of the PNW
  Forest Plan
• The Case of Survey and Manage Species as Rare
  Events
• Design and Sampling Issues
• Detection of rare events
• Example Analyses
• Sampling issues related to rare ecological
  events lichens as an example
• Some Final Thoughts

3
RARE ECOLOGICAL EVENTSIN TIME AND SPACE
Historical Overview

• The Context
• Northern spotted owls like old forest
• Timber companies like old forest
• A Socio-Economic, Political, Ecological collision
  led to
• Listing under the ESA
• And the Northwest Forest Plan

4
RARE ECOLOGICAL EVENTSIN TIME AND SPACE
Historical Overview

• More Context
• Northwest Forest Plan Record of Decision
  identified gt350 rare species to be surveyed for
  management, including lichens, bryophytes, fungi,
  and a few token vertebrates
• These species are identified as Survey and Manage
• They represent species for which little to no
  information is known

5
RARE ECOLOGICAL EVENTSIN TIME AND SPACE

• Objectives of survey and manage effort were to
  obtain estimates of, and/or determine, for EACH
  of the gt350 SM species
• Abundance Is the species abundant at local and
  regional scales?
• Spatial distribution Is the species
  well-distributed across the area of the Northwest
  Forest Plan?
• Persistence Do management activities ensure
  long-term persistence?

6
RARE ECOLOGICAL EVENTSIN TIME AND SPACE
Objectives of Survey and Manage

• Information to meet objectives comes from
• Existing data
• New data
• Expert opinion
• All must be merged so that simple policy
  decisions can be made for each species
• Decision framework must be multi-faceted

7
RARE ECOLOGICAL EVENTSIN TIME AND SPACE
Objectives of Survey and Manage

• Meeting these objectives required significant
  exploration into issues of
• Sampling,
• Estimation,
• Non-Spatial Modelling, and
• Spatial Modelling
• Can we detect, model, and eventually estimate,
  attributes of rare species at landscape scales?

8
RARE ECOLOGICAL EVENTSIN TIME AND SPACE

• Some constraints affecting ability to meet
  objectives

• Rare species are, well, rare!
• Limited life history information available
• Some populations exhibit irruptive behaviors,
  necessitating multiple site visits through time
• Efficient sample designs a must

9
RARE ECOLOGICAL EVENTSIN TIME AND SPACE

• Analytical approach
• Develop models for common lichens based on
  topographic and weather (DAYMET) variables
• Translate these models into spatially explicit
  maps
• Use maps as basis of stratification for sampling
  associated rare species
• Evaluate with independent data and determine if
  the models increase detection rates of rare
  species

10
RARE ECOLOGICAL EVENTSIN TIME AND SPACE
Example Analysis Lichens

• Characteristics of data
• Forest Service CVS/FIA plots were basis of sample
  design
• All plots visited number of visits variable
• Only first visit considered in subsequent
  analyses
• All lichen species searched for at each plot

11
Modeling Survey Manage DataCase Studies

• Model Families applied to common species
• Linear logistic regression (GLM)
• Additive logistic regression (GAM)
• Classification trees (CART)

12
Modeling Survey Manage DataCase Studies

• Internal Validation
• 10 fold cross-validation.
• (delete-one jackknife for logistic regression)
• External Validation
• Pilot and other random grid surveys

Training
Validation
13
Modeling Survey Manage DataCase Studies
Rare Common overlap ()
Common
LobaOreg LobaPulm PseuAnom PseuAnth
- 78.7 83.0
- - 96.0 76.0
76.0 - 76.9
100.0 - - 77.4
87.1 - 88.9
88.9 77.8 -
Rare
LobaScro NephLaev NephOccu NephPari PseuRain
14
Modeling Survey Manage DataCase Studies
Summary statistics for Lobaria oregana

• Differences in mean values for presences and
  absences for
• Topographic Elevation, Easting, and Northing
• Weather Minimum temperature, Relative humidity,
  Rainfall

15
Modeling Survey Manage DataCase Studies

• Classification tree for Lobaria oregana

• Measures of model fit
• PCC 94.5.
• PCCAbsent 94.8.
• PCCPresent 82.7.

16
Modeling Survey Manage DataCase Studies

• 10-fold internal cross-validation of Lobaria
  oregana model

• Measures of model fit
• PCC 90.5.
• PCCAbsent 95.9.
• PCCPresent 72.0.

17
Modeling Survey Manage DataCase Studies

• External validation of Lobaria oregana model

• Measures of model fit
• PCC 81.2.
• PCCAbsent 90.0.
• PCCPresent 51.1.

18
Modeling Survey Manage DataCase Studies

• Measures of error () for classification tree
  models for three other common lichen species used
  to model rarer species

Cross-validation
Model
Prediction

• LobaPulm 15.2 18.3 19.3
• PseuAnom 12.6 15.4 15.0
• PseuAnth 10.2 13.2 15.3

19
Modeling Survey Manage DataCase Studies

• Models of common species applied to spatial data
  for PNW and probability of lichen occurrence
  estimated for each location
• Estimated number of detections for each rare
  species using stratifications based on common
  species

20
Modeling Survey Manage DataCase Studies
Detection likelihoods for rare species LobaScro

21
Modeling Survey Manage DataCase Studies
Detection likelihoods for rare species PseuRain

22
Modeling Survey Manage DataCase Studies
Observed / Expected (Efficiency)
Common
LobaOreg LobaPulm PseuAnom PseuAnth
- 13/26 (2.0) 13/36 (2.8)
- - 19/23 (1.2) 19/48 (2.5)
19/60 (3.2) - 1/5 (5.0)
1/5 (5.0) - - 7/14
(2.0) 7/16 (2.3) - 2/1 (0.5)
2/5 (2.5) 2/5 (2.5) -
Rare
LobaScro NephLaev NephOccu NephPari PseuRain
23
Modeling Survey Manage Data Conclusions

• Stratification applied to independent region for
  field validation
• Expected detections for rare species should be
  apportioned across likelihood bins
• Ideal concordance would be 45 line

24
Modeling Survey Manage Data Conclusions

• Common problem when designing surveys for rare
  species is sufficient detections for analysis
• Design-based approaches provide least biased
  estimates, but can lead to low detections
• Model-based stratification using more common
  species can improve probability of detecting more
  rare species
• 2 to 5-fold gains in detection realized when
  process applied to rare epiphytic lichens

25
Modeling Survey Manage Data Conclusions
Questions?

• Edwards et al. Enhancing survey detection rates
  of rare species using model-based
  stratifications. In press, Ecology.
• Download at
• ella/gis.usu.edu/utcoop/tce