DETECTING LOCAL ECOLOGICAL IMPACTS FROM LOCAL HUMAN PERTURBATIONS

1
DETECTING LOCAL ECOLOGICAL IMPACTS FROM LOCAL
HUMAN PERTURBATIONS

• LOCAL IMPACT IN IMMEDIATE VICINITY OF ACTIVITY
  CAUSED BY THE ACTIVITY

• POINT-SOURCES OF PERTURBATION (e.g., SEWAGE
  OUTFALL, COASTAL POWER PLANT)

• ARISE FROM PLANNED DEVELOPMENT PROJECTS, MOST OF
  WHICH SUBJECT TO EXTENSIVE ENVIRONMENTAL REVIEW
  (ADMINISTRATIVE EIR/EIS PROCESS)

2
Ecological Field Assessments

• Requirement of NEPA (CEQA)

• EIS (EIR) Process

Proposed project
Sewage/industrial outfalls
Negative declaration
EIS/R
Mitigated Negative declaration
NPDES permit
EPA/State
Impact predicted
Report discharge
Mitigation activity
Field-based monitoring of environmental impact
Success criteria
remediation
3
HOW TO QUANTIFY ECOLOGICAL IMPACTSFROM
POINT-SOURCES

• CONDUCT FIELD ASSESSMENTS

• GENERALLY NOT WELL DONE

• IMPORTANT TO UNDERSTAND STRENGTHS LIMITATIONS
  OF FIELD ASSESSMENT DESIGNS

4
CONSIDER HOW TO QUANTIFY EFFECTS OF A PARTICULAR
POINT-SOURCE DISTURBANCE (e.g., MUNCIPAL OUTFALL
, OIL PLATFORM, POWER PLANT) ON ORGANISMS LIVING
NEARBY

• CONDUCT A FIELD ASSESSMENT STUDY

• CANNOT ANALYZE USING A TRUE FIELD EXPERIMENT

5
WHY FIELD ASSESSMENT DESIGNS ARE NOT EXPERIMENTS
1. TYPICALLY NO CHANCE TO REPLICATE POINT-SOURCE
DISTURBANCE
2. NO CHANCE TO RANDOMLY ASSIGN SITES TO
TREATMENT TYPES
6
PROPER EXPERIMENTAL DESIGN REQUIRES

• SERIES OF SITES

• RANDOM ASSIGNMENT OF SITES TO TREATMENT TYPE
  (e.g., DISTURBANCED, UNDISTURBED CONTROL)

• REPLICATES OF EACH TREATMENT TYPE

7
EXPERIMENTS ADDRESS A QUESTION THAT OFTEN DIFFERS
FROM THAT ASKED BY REGULATORS

• EXPERIMENTS HELP ESTIMATE AVERAGE EFFECT OF A
  PERTURBATION

• AVERAGE EFFECTS OFTEN THE WRONG GOAL

• RATHER, WANT TO KNOW PARTICULAR EFFECTS OF
  PARTICULAR PERTURBATION IN PARTICULAR LOCATION

8
SPECIFIC GOAL OF ECOLOGICAL IMPACT ASSESSMENT
STUDY
THE QUESTION
HOW DOES AN ECOSYSTEM AT A SITE OF PERTURBATION
DIFFER FROM THE ECOSYSTEM THAT WOULD HAVE EXISTED
THERE HAD THE PERTURBATION NEVER OCCURRED?
CANNOT DIRECTLY MEASURE THIS
9
BASIC TOOL TO ANSWER QUESTION FIELD MONITORING
OF ENVIRONMENTAL CONDITIONS (FIELD ASSESSMENT
STUDY)
OPERATIONAL GOALS

• ESTIMATE STATE OF LOCAL ECOSYSTEM HAD THE
  PERTURBATION NOT OCCURRED

EFFECT SIZE
10
Effect size
Capitella captitata
11
BASIC TOOL TO ANSWER QUESTION FIELD MONITORING
OF ENVIRONMENTAL CONDITIONS (FIELD ASSESSMENT
STUDY)
OPERATIONAL GOALS

• ESTIMATE STATE OF LOCAL ECOSYSTEM HAD THE
  PERTURBATION NOT OCCURRED

EFFECT SIZE

• PROVIDE MEASURE OF CONFIDENCE IN ESTIMATE

POWER OF TEST
12
KINDS OF ERRORS IN FIELD ASSESSMENT STUDIES
ERROR
REALITY
CONCLUSION
NO IMPACT
IMPACT
FALSE IMPLICATION(TYPE I)
IMPACT
NO IMPACT
FALSE EXONERATION(TYPE II)

• ß risk of making type ll error
• Power (1- ß) prob. we havent made an error.
  Measure of our confidence that we would have
  detected an important effect if one existed

13
FALSE EXONERATION (TYPE II) LIKELY TO BE MORE
WIDE-SPREAD THAN FALSE IMPLICATION (TYPE I)

• INADEQUATE DESIGNS (WITH LOW POWER) UNLIKELY TO
  DETECT REAL DIFFERENCE UNLESS EFFECT SIZE IS HUGE

How can we increase power of the test?
increase effect size greater sample size
increase a
(Quinn and Keough, Ch-7 http//www.statsoftinc.co
m/textbook/stathome.html)

• IDEALLY, WANT DESIGN THAT MINIMIZES LIKELIHOOD
  OF MAKING EITHER ERROR

14
COMMON ASSESSMENT DESIGNS
1. CONTROL - IMPACT
IMPACT SITE - WHERE PERTURBATION OCCURS
CONTROL SITE - SIMILAR ENVIRONMENT TO IMPACT SITE
BUT FAR ENOUGH AWAY NOT TO AFFECTED

• TAKE MULTIPLE SAMPLES AT EACH SITE
• COMPARE STATISTICALLY PARAMETER VALUES BETWEEN
  IMPACT CONTROL SITES

15
CONTROL - IMPACT DESIGN
KELLETIA (SNAIL) DENSITY NEAR FARFROM DIFFUSER
PORT
)
2
10
8
6
DENSITY
(no. per 40 m
4
2
0
NEAR
FAR
CONTROL
IMPACT
IMPACT
1.6 km
250 m
50 m
SITE
16
COMMON ASSESSMENT DESIGNS

• CONTROL - IMPACT CONFOUNDS IMPACT WITH NATURAL
  SPATIAL VARIABILITY
• (C-I impact design requires that sites be
  identical in absence of impact)

HOW CAN WE ACCOUNT FOR NATURAL SPATIAL
VARIABILITY?
17
COMMON ASSESSMENT DESIGNS
1. CONTROL - IMPACT
2. BEFORE - AFTER

• COMPARE PARAMETER VALUE AT IMPACT SITE BEFORE
  AFTER PERTURBATION

• ELIMINATES NATURAL SPATIAL VARIABILITY

18
BEFORE - AFTER DESIGN
PINK SURFPERCH DENSITIES BEFORE AFTER START-UP
OF SAN ONOFRE NUCLEAR GENERATING STATION
19
COMMON ASSESSMENT DESIGNS
1. CONTROL - IMPACT
2. BEFORE - AFTER CONFOUNDS IMPACT WITH NATURAL
TEMPORAL VARIABILITY

• HOW CAN WE ACCOUNT FOR NATURAL SPATIAL
  TEMPORAL VARIABILITY?

20
COMMON ASSESSMENT DESIGNS
1. CONTROL - IMPACT
2. BEFORE - AFTER
3. BEFORE - AFTER - CONTROL - IMPACT
(OPTIMAL IMPACT ASSESSMENT DESIGN)

• SAMPLE IMPACT CONTROL SITES INTENSIVELY ONCE
  BEFORE AND ONCE AFTER PERTURBATION

• TEST WHETHER IMPACT SITE VALUES CHANGE RELATIVE
  TO CONTROL VALUES FROM BEFORE TO AFTER
  PERTURBATION

21
BEFORE - AFTER - CONTROL - IMPACT
SEA PEN DENSITY AT NEAR (IMPACT) FAR (CONTROL)
FROM OUTFALL BEFORE AFTER PROJECTED
COMMENCEMENT OF DISCHARGE
)
2
12
CONTROL
9
6
DENSITY (no. per 40m
3
IMPACT
0
BEFORE
AFTER
22
COMMON ASSESSMENT DESIGNS
1. CONTROL - IMPACT
2. BEFORE - AFTER
3. BEFORE - AFTER - CONTROL - IMPACT POORLY
ESTIMATES STATE CONDITIONS (e.g. does not detect
time x location interactions)
23
COMMON ASSESSMENT DESIGNS
1. CONTROL - IMPACT
2. BEFORE - AFTER
3. BEFORE - AFTER - CONTROL - IMPACT
4. BEFORE - AFTER - CONTROL - IMPACT - PAIRED -
SERIES (BACIPS)

• CURRENTLY BEST AVAILABLE DESIGN

24
FOUR IMPACT ASSESSMENT DESIGNS
1. CONTROL - IMPACT (CONFOUNDS IMPACT WITH
NATURAL SPATIAL VARIABILITY)
2. BEFORE - AFTER (CONFOUNDS IMPACT WITH NATURAL
TEMPORAL VARIABILITY)
3. BEFORE - AFTER - CONTROL - IMPACT (POORLY
ESTIMATES STATE OF CONDITIONS)
4. BEFORE - AFTER - CONTROL - IMPACT - PAIRED -
SERIES (BACIPS) (BEST AVAILABLE DESIGN)
25
BEFORE - AFTER - CONTROL - IMPACT - PAIRED -
SERIES DESIGN (BACIPS)
MEANING OF SERIES IN BACIPS
REQUIRES SAMPLING MANY TIMES IN BEFORE AND IN
AFTER PERIOD
ESTIMATE OF STATE OF CONDITIONS COMES FROM
SAMPLES THAT ARE REPLICATED IN TIME (i.e.,
TIME-SERIES)
EACH REPLICATE OBSERVATION MUST BE AN INDEPENDENT
ESTIMATE OF CONDITION AT A SITE
26
BEFORE - AFTER - IMPACT - CONTROL - PAIRED -
SERIES DESIGN (BACIPS)
MEANING OF PAIRED IN BACIPS
FOR EACH OBSERVATION, NEED TO SAMPLE CONTROL
IMPACT SITES VERY CLOSE TOGETHER IN TIME (i.e.,
PAIRED SAMPLING)
PAIRED SAMPLING REMOVES EFFECTS OF NATURAL
TEMPORAL VARIATION THAT SIMULTANEOUSLY INFLUENCES
BOTH SITES
27
WHITE SEA URCHIN DENSITIES AT IMPACT CONTROL
SITES BEFORE START-UP OF SONGS
28
SUBTRACT CONTROL VALUE FROM IMPACT VALUE FOR EACH
DATE (? I - C)
MEAN ? IN BEFORE PERIOD
?
29
WHAT DOES MEAN ? (BEFORE) TELL US?
THE AVERAGE NATURAL SPATIAL VARIATION
IF PERTURBATION HAS NO EFFECT, AVERAGE NATURAL
SPATIAL VARIATION SHOULD NOT CHANGE
IF NO IMPACT, MEAN ? (BEFORE) MEAN ?
(AFTER)
30
PREDICTED URCHIN MEAN ?(AFTER) IF NO SONGS IMPACT
BEFORE
AFTER
START-UP
95 CI
31
WHITE SEA URCHINCHANGE IN ? (IMPACT - CONTROL)
FROM BEFORE TO AFTER START-UP OF SONGS
BEFORE
AFTER
START-UP
32
URCHIN DENSITIES BEFORE AFTER AT
IMPACT
CONTROL
33
URCHIN DENSITIES ESTIMATED AT 2 CONTROL SITES
AT IMPACT SITE
IF SONGS IMPACTS URCHINS, EXPECT NO CHANGE IN
MEAN ? BETWEEN 2 CONTROL SITES
PROVIDES A CHECK STRONGER CASE
34
DIFFERENCE IN URCHIN ?s BETWEEN CONTROL SITES
BEFORE AFTER START-UP
35
DID WE FALSELY IMPLICATE SONGS?
CANT TELL BECAUSE WE MADE A MISTAKE
DATA VIOLATED THE CRITICAL ASSUMPTION OF
ADDIVITITY
36
WHAT IS ADDIVITITY?
BACIPS IS ADDITIVE MODEL
? IMPACT - CONTROL
BUT THIS IS NOT HOW POPULATIONS GROW THEY DO NOT
ADD OR LOSE A CONSTANT NUMBER OF INDIVIDUALS
37
POPULATIONS THAT RESPOND IDENTICALLY BUT HAVE
DIFFERENT STARTING DENSITIES
POPULATION AT IMPACT CONTROL
DIFFERENCE
DENSITY
(I - C)
BEFORE
10 1
9
20 2
18
AFTER
ABSOLUTECHANGE 10
1
RELATIVECHANGE 100
100
38
GRAPICAL REPRESENTATION OF POPULATION GROWTH
1
2
4
8
16
39
SIMPLE POPULATION GROWTH MODEL
dN / dt r . N
GROWTH IS MULTIPLICATIVE (NOT ADDITIVE)
BUT CAN CONVERT TO ADDITIVE MODEL
LOG (r . N)
LOG (r) LOG (N)
40
POPULATIONS THAT RESPOND IDENTICALLY BUT HAVE
DIFFERENT STARTING DENSITIES
POPULATION AT IMPACT CONTROL
DIFFERENCE
DENSITY
(I - C)
BEFORE
Log(10) 1 Log(1) 0
1
Log (20) 1.3 Log(2) 0.3
1
AFTER
41
BACIPS VARIATE FOR POPULATION DATA(i.e., DENSITY)
? Log (IMPACT) - Log (CONTROL)
DID SONGS AFFECT DENSITY OF WHITE SEA URCHINS?
42
WHITE SEA URCHIN AT SONGS CONTROL
STATISTICALLY DIFFERENT
EFFECT SIZE
BEFORE
AFTER
43
WHITE SEA URCHIN AT CONTROL 1 CONTROL 2
NOT STATISTICALLY DIFFERENT
BEFORE
AFTER
44
BACIPS CAN BE USED ON ANY KIND OF BIOLOGICAL,
CHEMICAL OR PHYSICAL PARAMETER
DATA MUST MEET ASSUMPTION OF ADDITIVITY IF NOT,
FREQUENTLY CAN TRANSFORM DATA TO MEET ASSUMPTION
45
THE CONTROL IN BACIPS

• USED TO PREDICT CONDITION AT IMPACT SITE IN
  ABSENCE OF EFFECT OF PERTURBATION

• ESSENTIAL THAT CONTROL BEHAVES SIMILARLY TO
  IMPACT(EXCEPT FOR EFFECT OF PERTURBATION)

• MEAN PARAMETER VALUES TRACK ONE ANOTHER WELL AT
  CONTROL IMPACT

46
POOR TRACKING (LOW COHERENCE)
RAW DENSITIES
? (Log I - Log C)
47
GOOD TRACKING (HIGH COHERENCE)
RAW DENSITIES
? (Log I - Log C)
48
FACTORS INFLUENCING POWER OF BACIPS TEST
1. AMOUNT OF VARIATION IN ?s (NATURAL
VARIABILITY) (HOW WELL CONTROL TRACKS IMPACT)
2. NUMBER INDEPENDENT SAMPLING DATES IN BEFORE
AFTER PERIOD (SAMPLING ERROR)
3. DIFFERENCE IN MEAN ?s BETWEEN BEFORE
AFTER PERIODS (EFFECT SIZE)
49
FINAL IMPORTANT POINT

• TALKED A LOT ABOUT STATISTICAL RIGOR - ITS
  IMPORTANT BUT...

MANAGERS SHOULD NOT BE INTERESTED IN HYPOTHESIS
TESTING(WAS THERE AN IMPACT? YES or NO?)

• REAL AIMS OF FIELD ASSESSMENT STUDY

1. ESTIMATE OF MAGNITUDE OF ECOLOGICAL
CHANGE(EFFECT SIZE)
2. OUR CONFIDENCE IN THAT ESTIMATE
GOAL ESTIMATION, NOT HYPOTHESIS TESTING