Designs for environmental monitoring

1
Designs for environmental monitoring

• ?Gerry Quinn Mick Keough, 2002
• Do not copy or distribute without
• permission of authors.

2
Some consistent terminology

• Time
• divided into two major Periods, Before and After.
  
• Within these periods are Times (e.g. years)
• and subdivisions of times
• Spatial areas
• divided into two major groups, Control and Impact
• Within each group are Locations
• larger spatial units
• different areas in which the same kind of
  management
• i.e., true replicates of the management activity
• and subsamples within locations

3
Controls Impacts

• Control areas lack the management action
• Reference areas
• Impacts are effects of management action

4
Restricted Designs
CONTROL IMPACT
Control
Impact
BEFORE AFTER
Before
After
5
Better DesignsMultiple controls impacts
CONTROL SITES
IMPACT SITES
6
Williamstown Rifle Range

• Replicate open (C) and closed (I) locations
• Single point in time
• Four harvested species
• Three unharvested species
• Keough et al. (1993) Cons. Biol.

7
km
0
1
2
Rifle Range
WILLIAMSTOWN
RR2
ALTONA
A1
W1
RR1
W4
W3
W2
A2
8
(No Transcript)
9
(No Transcript)
10
Turban Snails
Marine Research Group, Coastal Invertebrates of
Victoria
From Edgar, Australian Marine Life
11
WRR early study

• Three of 4 harvested species larger in protected
  areas
• No differences in unharvested species
• Power calculations for unharvested species

12
Three common models

• BACI BACIP
• Stewart Oaten / Green
• Beyond-BACI
• Underwood
• MBACI
• Keough Mapstone

13
BACI(P) DESIGN
BEFORE IMPACT
AFTER IMPACT
Time
CONTROL SITE
IMPACT SITE
14
BACIP

• Impact and single Control location
• Sampling through time (B A)
• Times within B A
• Test is divergence of two locations after
  management activity
• uses C-I as test measure
• Only one control location of interest
• does Impact change relative to standard?

15
Calculations in BACI design
d
Time
Time
16
Procedure

• Plot values for each location through time
• Calculate di Ci - Ii for each time
• Compare d values between groups (Before After)
  - ANOVA or t-test.

Time
17
BACIP replication power

• Replicates are times within B A
• Each location represented by single value in
  analysis
• Subsampling of locations useful to characterize
  each location well
• Indirect effect on power
• Power primarily determined by time
• Needs longer sampling period,
• not more frequent samples

18
Assumptions of BACIP

• Independence of C I
• No serial correlation
• Value of d at time i independent of value at time
  i1
• C I track well through time in absence of
  impact
• Control is well-chosen

19
Choosing control locations

• As similar as possible to I, except for presence
  of activity in question
• Physically similar
• Fauna flora similar
• Far enough away to be free of effects of activity

20
Choosing controls a freshwater example. Within
a catchment

• substrate/geology/suspended load
• discharge regime/stream or catchment size
• riparian vegetation/zone
• catchment land use/vegetation/soils
• gradient
• altitude/stream order/dist. from source
• current speed/riffle pool structure
• spatial proximity/external variables
• water quality including detritus levels
• channel form/geometry

21
Choosing controls within a catchment

• algal cover
• where I'm prepared to drink the water
• logistical practicality
• Chance
• presence of fish
• aspect

22
Choosing controls between catchments

• substrate
• discharge regime/stream size
• catchment land use/vegetation
• riparian vegetation
• gradient
• geographic proximity
• geomorphic forms/hydraulic habitat
• altitude and aspect/temperature
• water quality/geology
• can't be guessed/not solvable/no answer

23
Steps to choosing controls

1. Conduct a literature review
2. Draw up a list of criteria ordered from most
   important to least
3. Carry out location visits
4. Are there sufficient control locations?
5. Revisit the criteria
6. Does this improve the number of control locations?

24
BACIP example San Onofre Nuclear Generating
Station (SONGS)

• Seawater used in cooling system
• Discharge 6000 m3/min
• 1 km2/d, 3-4 m deep
• Temperature raised by 10ºC
• Water released through diffusers
• Plume entrains 60,000 m3

25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
SONGS study
29
SONGS studyGiant Kelp, Macrocystis pyrifera
30
SONGS studyGaribaldi
31
SONGS studyRockfish
32
SONGS studyMobile invertebrates - Fromia
33
BACI as a partly-nested design
34
Before
After
Control
Impact
1
2
3
4
5
6
Sampling Times
35
Before
After
Control
Impact
1
2
3
4
5
6
Plots
36
Factor A (Periods)
Factor C (C-I)
Plots
37
The linear model

• Times (T) nested within Before After (BA)
• Crossed with Control-Impact (CI)
• Time random, but BA and CI are fixed
• DV ? BA T(BA) CI BACI
• No replication (no Times x CI interaction)
• Power depends on of times

38
MBACI

• Multiple Control and Impact locations
• Times within Before After
• Do Impact locations diverge relative to suite of
  Control areas?
• Controls represent whole range of unmanaged areas
• All available times (e.g. years) presumed to be
  sampled

39
1
2
3
4
5
40
MBACI linear model

• Now four factors
• BA - fixed
• Times within Before After - fixed??
• CI - fixed
• Sites within Control Impact - random
• DV ? BA T(BA) CI S(CI) BACI
  CIT(BA) BAS(CI)
• Test BACI using BAS(CI) as denominator

41
MBACI

• Power depends on of sites
• Because test is made with BA x sites, critical
  assumptions are about how the BA changes occur at
  the replicate sites
• Are these changes normal, homogeneous variances,
  etc?

42
MBACI subsampling

• Replicates are locations
• direct effect on power
• Subsamples to characterize location-time
  combinations
• subsamples affect power by reducing L x B-A
  variance
• Replicate times used only to estimate B A
  conditions
• affect power by reducing L x B-A variance

43
MBACI example - Keough Quinn (2000)

• Effect of marine protected areas in the absence
  of strict enforcement
• Areas open to recreational harvesting for gt75 y,
  and areas closed for 75 y, opened in 1992
• DV is mean size of a common limpet
• Factor A Harvesting
• units of replication are individual reefs
• Factor B (reefs) nested within factor A

44
(No Transcript)
45

• Factor C Before and After opening of shoreline
• units of replication years
• Factor D (years) nested within Factor C
• Factors A C fixed
• nested factor B (sites) is random
• nested factor D (years) is fixed

46
ANOVA table
Source of Variation
Num
Den
MS
F
P
Df
Df
Harvesting
1
6
0.010
Before-After
1
6
0.007
H x B-A
1
6
0.015
Sites(Harvesting)
6
Res
0.119
Years(B-A)
6
Res
0.008
Sites x B-A
6
Res
0.951
Year x Harvesting
6
Res
0.714
Residual MS
31
12.593
Key test is H x B-A, which is tested using Sites
x B-A. This effect is significant
47
Cellana tramoserica
40
I
C
30
Shell Length (mm)
20
10
89
90
91
93
94
95
96
97
48
WRR broken protection

• Impact was opening of Rifle Range
• 3 years Before
• 7 years After
• Additional Controls using unharvested species
• Keough Quinn (2000), Ecol. Appl

49
WRR broken protection

• Declines in size, abundance of Cellana, Turbo, no
  change in less harvested species
• No change in unharvested species
• Power

50
Example 2 Liming of streams

• Aim Evaluate restoration methods for Welsh
  streams affected by acid rain
• described by Downes et al., (2002), 8.4.1
• Impact would be enhancement of salmonid fish
  invertebrates following raising of pH
• Impacts can be good!
• Control is streams with no restoration

51
Liming

• Locations were tributaries of major river
• Prior sampling showed similar pH, CaCO3
• Variation in vegetation, but C I matched
• pine forests
• 3 C 3 I tributaries
• Sampling
• Invertebrates sampled annually (spring)
• Water quality at 1-2 wk intervals
• Sampling 1985-1992
• Liming in late 1987

52
Beyond-BACI

• Multiple Control Impact
• Before After
• Range of scales of sub-sampling
• Do Impact areas change relative to Controls?
• at various spatial and temporal scales
• comparisons of variance
• Controls as sample of population

53
Sampling design
Select variables
Effect Size
Risks of errors
Estimate variance
Optimization
N
N
Y
Y
54
General comments

• Precise management questions lead to precise
  tests
• Clearly specified statistical models have
  unambiguous power rules
• but
• Power calculations are approximate
• Final analysis must be flexible

55
General comments

• For monitoring, variance structures likely to
  vary between places
• Effect Size locally determined
• No attempts to produce single best design
• Standard approach to optimising, with different
  local designs