Sampling Design

1
Sampling Design

• Lloyd Morrison
• and
• Michael DeBacker

Heartland Network
2
Challenges of Sampling Design

• Making inferences
• Issues with stratification
• Integration (co-location/co-visitation)
• Intensity vs. spatial coverage

3
Inference

• We can only sample a very small proportion of
  most parks.
• We need to make scientifically defensible
  inferences to areas beyond the limited sites we
  sample.
• Statistical inferences can only be made to areas
  that have a chance of being included in the
  sample (i.e., need a probabilistic design).

4
Stratification

• Strata must be carefully chosen in long-term
  studies, so criteria for strata do not change.
• Multiple protocols may need to use different
  criteria for stratification.
• Later addition of protocols may not fit within
  defined strata.

5
Co-location (Co-visitation)

• Co-location will provide information on multiple
  vital signs for the same sites, and can increase
  sampling efficiency.
• Co-visitation will aid overall sampling
  efficiency, as multiple vital signs can be
  sampled at the same time.

6
Intensity vs. spatial coverage

• Should we sample relatively few sites more
  intensively (i.e., measure more variables), or a
  greater number of sites more superficially?
• Should we sample a few sites every year, or a
  greater number of sites less frequently?
• Sampling intensity must be considered along both
  spatial and temporal dimensions.

7
Sampling Design Examples from Phase III
Monitoring Plans
Caveats

• Only have time for a few examples
• May oversimplify
• Focus on networks currently in Phase III
  development
• Look at diversity of approaches

8
Designs highlighted last year

• Integrated terrestrial design using a systematic
  grid (birds, plants) HTLN
• Systematic, unequal probability sampling (plants,
  soil, etc.) - NCPN
• Two stage systematic sampling (plants, birds,
  etc.) - CAKN
• GRTS design for large river systems (fish,
  aquatic invertebrates, etc.) HTLN

Available online http//science.nature.nps.gov/im
/monitor/meetings/Austin_05/austin.htm
9
GRTS Generalized Random Tessellation Stratified

• generates a spatially well-balanced sample with
  a random component.
• allows for easy addition/removal of sites while
  maintaining spatial balance.
• use of strata or unequal selection probabilities
  may decrease overall degree of (apparent) spatial
  balance.

Simple random
Systematic
GRTS
10
Southern Colorado Plateau Network (SCPN)
Upland vital signs
Co-locate and co-visit vegetation, hydrologic
function, and soil/site stability.
Stratify based on ecological sites (i.e.,
landscape divisions with specific physical
characteristics).
11
Initial sampling frame is a systematic grid of
evenly-spaced points.
12
Inaccessible areas are eliminated from frame
travel time will affect selection probabilities.
13
Ecological sites also influence selection
probabilities.
14
GRTS (with unequal selection probabilities) is
used to select the sites to be sampled, and these
are divided into panels.
15
Revisit schedule depends upon park size
For small parks
16
For medium-size parks, a split panel 2-3,1-4
design will be used
17
For large parks, a split panel 2-6, 1-9 will be
used
18
Southern Colorado Plateau Network (SCPN)

• Integrated riparian design The sample frame is
  composed of points that are equal distances
  apart.
• Co-locate and co-visit riparian vegetation,
  stream flow, and channel morphology.
• Stratify based on stream order (i.e., main stem
  vs. tributaries).

19
Sampling frame is composed of points spaced equal
distances apart.
20
A GRTS sample is used to select the sites to be
sampled, with unequal selection probabilities
based on geomorphic settings and travel costs.
21
Sonoran Desert Network (SODN)
Terrestrial Co-location of vegetation and soil
vital signs
22
(No Transcript)
23
Problem For some vital signs, sampling is
relatively rapid, but travel times are high.
Solution use a two-stage sampling scheme

• Begin with a systematic grid of points employ
  GRTS with selection probabilities based on
  accessibility.
• Establish sample locations in regular clusters
  around selected locations.

24
Sonoran Desert Network (SODN)

• Aquatic co-location of 7 different vital signs
  along each reach (co-visited when possible).
• GRTS will be used to select sample sites.
• Each reach is divided into 10 segments.

25
Southwest Alaska Network (SWAN)
Three-tiered sample approach for lakes, rivers,
and fish
Three-tiered sample approach for monitoring
lakes, rivers, and fish
26
(No Transcript)
27

• Sampling is based on a combination of targeted
  and random selection procedures.
• All water bodies categorized into three tiers,
  prioritized according to access, level of
  use/management issues, and ecological and spatial
  coverage.
• All Tier 1 water bodies will be sampled annually
  a subset of Tier 2 3 water bodies will be
  sampled every 2-5 and 10 years, respectively.
• Tier 2 3 lakes and rivers will be stratified by
  size, water type, and accessibility prior to
  drawing a GRTS sample.

28
Northeast Temperate Network (NETN)
Hierarchical, multi-tiered approach to forest
monitoring
Systematic sample design (tessellated hexagonal
grid) with random point placement (within each
grid cell).
Revisit plan 1-3 (sample every 4 years).
29
25 of plots will be intensively monitored.
Balances competing needs for broad spatial
coverage and intensive quantitative sampling.
30
Potential problems with current sampling designs

• Use of judgment to select sites.
• Stratification is very common.
• Many vital signs have their own unique sampling
  design integration is lacking.
• Some sampling designs are not developed well in
  relation to the issues discussed here.