Longterm Monitoring Project Design and Statiscal Analyses

1
Long-term Monitoring Project Design and Statiscal
Analyses

• Paul Montagna
• Texas AM University-Corpus Christi, TX, United
  States

Presentation made to the Office of Polar
Programs National Science Foundation October,
13 2006
2

• Design Principles
• Choose stations to falsify null hypotheses
• Replicate at treatment level
• Reuse stations to test several hypotheses
• No confounding factors
• Balanced designs (equal sampling effort)
• Sufficient power to detect change
• Meta-analysis in the end

3
Stick with the fixed sampling design
4
In-the-end

• We have a multivariate problem
• Solution to analyzing large multivariate data
  sets
• Solve using two approaches
• Parametric to compare environmental variables
• Non-parametric to compare community responses
• Sediment quality triad approach
• Chemical contaminant background
• Biological response of toxicity
• Ecological response of community change
• Benthic index of biotic integrity

5
In-the-end (continued)

• Parametric to compare environmental variables
• Creates a new set of uncorrelated variables in
  order of decreasing variance
• Therefore, a variable reduction technique that
  transforms the data set and packs most of the
  information in the first few columns
• Correlation of new columns to test for biological
  responses to environmental factors
• Protects experiment-wise error rate
• Non-parametric to compare communtiy responses
• Multi-dimensional scaling (MDS)
• Solves the problem in community data where there
  are matrices with lots of zeros (i.e., lack of
  co-occurrence is not interesting)

6
In-the-end (continued)

• Integrated SQT-BIBI
• Simply relate the environmental variability to
  variability in the community and contaminant
  background
• Lots of long-term data promotes powerful tests

7
Sediment Quality TriadLong and Chapman
1985Montagna and Green 1996
Meiofauna
Macrofauna
8
BIBIEngle et al. 1994Carr et al. 2000
9
Integrated BIBI / SQT

• Carr et al. 2000, Long et al. 2003
• Multivariate analysis to reduce data
• Principal Components Analysis is best
• PCA on a BIBI
• PCA on Chemistry
• PCA on Toxicology
• Correlate metrics to PCAs

10
Identifying Contaminant Effects

• In field studies, contaminants always covary with
  each other and other natural chemical,
  geological, and physical factors.
• Principal components analysis can be used to
  create new variables that summarize and
  distinguish these natural and anthropogenic
  gradients.
• For example, in Corpus Christi Bay, outfall
  contaminants group into metals (PC1) and organics
  (PC2).

11
Relationship to Toxic Effects

• Normal Sea Urchin development and Mysid growth
  rates were inversely related to metals (PC1), but
  not organics (PC2)
• Similarly, a benthic index of biotic integrity as
  also inversely related to metals, but not
  organics.

12
(No Transcript)