APPLICATIONS WITH GIS

1
APPLICATIONS WITH GIS

• MODULE 9
• RSF WORKSHOP
• JANUARY 11-12, 2003

2
RSF and GISAdvantages

• Use of GIS increasing as software is more readily
  available and user friendly (e.g., ARCVIEW)
• Allows greater flexibility in studying/varying
  scale, different definitions of availability
• Allows for mapping results, useful management and
  research tool
• Can investigate many variables
• Do not have to field sample habitat

3
RSF and GISDisadvantages

• Should not be used as a substitute for thinking
• Too many variables, highly correlated
• E.g., 10 variables 210-1 models (no interactions
  or quadratic terms considered)
• Model selection difficulties
• Derived variables may not relate to selection
• Interpretation
• E.g., what does it mean for moose to be selecting
  for an increasing interspersion and juxtaposition
  index

4
CASE STUDY

• Moose winter habitat selection
• Data collected by Robert Skinner
• Study area Innoko NWR, AK
• 4 million plus acres
• Innoko river corridor (1994)
• Western strata (1996)
• See Erickson, McDonald, and Skinner (1998)
  Resource Selection Using GIS Data A Case
  Study, Journal of Agricultural, Biological, and
  Environmental Statistic (JABES), v3, p. 296-310

5
Objectives

• Develop a model for winter habitat selection.
• Apply model to the study area for aid in
  identifying winter range and strata for other
  work (e.g., browse surveys).

6
Field Methods

• Aerial line transect surveys March, 1994 and
  1996.
• Locations of moose groups fixed using GPS
  interfaced with a laptop computer.

7
COVARIATESMoose Habitat Selection, Innoko NWR

• Landcover map (30 m X 30 m resolution) developed
  via Landsat TM and standard classification
  methods.

8
Innoko River Vegetation Map
9
Buffers

• Data generated
• Proportion of each vegetation type within buffer
• Quadratic effects considered

10
(No Transcript)
11
Why Buffers?Moose Habitat Selection, Innoko NWR

• Buffers around used and available locations
  required to calculate pixels in each class
• Exact moose locations were not known.
• Can calculate area based metrics

12
Why Buffers?

• Buffers should be generated for both available
  and used.
• Example
• 25 of each of 4 types at the used location
• Almost always 100 of one type at available
  locations
• Available locations show less diversity

13
Why Sub-sample Available Locations?

• No need to place available buffers around every
  pixel
• Can get arbitrarily close to true parameters with
  sub-sample.
• Sub-sampling reduced data requirements
• A sample of 5170 available buffers was drawn
• 18 million buffers would be required to sample
  every pixel

14
(No Transcript)
15
Innoko River Corridor

• Best subsets regression techniques used to choose
  the best model (AIC criterion).
• BEST MODEL
• w(X)exp((-3.1P3)(4.1P4)(0.5P8)-(8.2P10)-
  (8.5P16)-(1.8P18)(6.1P22)-(5.8P42)-
  (10.2P222))
• P4Birch/White spruce forest P8Poplar forest
  P22Willow, Etc.......

16
(No Transcript)
17
Varying Scale

• Varying buffer sizes changes resolution of
  selection
• Red patches dominated by willow and birch forest

18
Application

• Modeling and mapping relative probability of use
  for blackpoll warbler, savannah sparrow, and
  alder flycatcher on the Innoko National Wildlife
  Refuge

19
Study Area and Transect Locations
20
FIELD METHODS

• Line transect surveys walked during June
• (3 am-11 am)
• GPS location on transect line and approximate
  location of bird

21
GIS DATA LAYERS

• Landsat TM bands 1-5, 7, from August 26, 1991
• elevation (from USGS 163,360 scale quads)
• slope (derived from elevation)
• aspect (derived from elevation)
• distance to river
• distance to lakes

22
(No Transcript)
23
Variables
x

• Experimental unit
• (7 x 7 pixel, or 210 m x 210 m area)
• Means of 49 pixels for each data layer
• Standard deviations of Landsat data (each band)
• Variables calculated for bird locations and
  systematic sample of available pixels (i.e,
  within floodplain)
• Design 1, Sampling Protocol A

24
Statistical Analysis

• Correlation analysis to reduce number of
  variables (12)
• Logistic regression for estimating coefficients
  of RSF model
• w(X)exp(b1X1 b2X2... bnXn)
• where Xi is the proportion or a function of the
  proportion of a vegetation class
• bi estimated from the logistic regression
  analyses

25
Model Selection and Reporting
Model importance determined by BIC BIC
-2log(Likelihood) p?log(n)
Model averaging with top five models based on
differences
Model weights
26
Alder flycatcher Empidonax alnorum
27
Alder flycatcher results




Model





Standard

Wald



Model

BIC

Weight

Variable

Estimate

Error

Chi
-
Square

p
-
value

1

911.36

0.376

river

-
0.00077

0.000285

7.3402

0.0067




band 4

0.0412

0.00701

34.54
63

lt.0001




std band 3

0.2278

0.0347

43.1197

lt.0001




std band 4

0.0393

0.0102

14.835

0.0001







elevation

-
0.1191

0.0226

27.8688

lt.0001


28
(No Transcript)
29
Importance Values
30
(No Transcript)