http://www.ices.dk/reports/FTC/2005/WKSAD05.pdf

Random vs Systematic Sampling:

  • Systematic Sampling: they used a grid of 64 equally spaced points, the location of initial point was randomly selected. basically, they divided the survey area into 64 equal boxes. They then randomly chose a point in one box, and then collected the sample at the same relative location in each of the 64 boxes. I think that this was continued (i.e. they generated another regular grid of 64 points) until they had 1000 points in total.
  • Random: start with 64 stations, then add random stations and use an algorithm to see how long it takes to travel from each station to the next one. They continued adding random stations until they reached a maximum time limit (making the number of sample locations variable).

They ran these two survey types on two different simulated oceans of fish. One ocean had high variability and low spatial autocorrelation (don't know what that is), and the other had low variability and high correlation. (each ocean had 1x10^7 fish).

The random sample did a better job estimating the population in the high variability ocean, but the systematic sample did a better job with low variability. Implies that sample method selection should correlate to spatial distribution of fish. ("Further investigation of a wider range of surfaces with different properties should help to refine the parameters that influence the point at which different survey strategies are more efficient estimators...")

Stratification:

The international bottom trawl survey used stratification based on depth. There is evidence that in some areas, bottom sediment type makes a significant difference in fish populations, so characteristics of the seabed will be incorporated into future stratification designs. Stratification can become a confounding factor in samples of several different species where the species have different spatial distribution characteristics. Preliminary analysis is need to know how to stratify when surveying several species.

(this is where I got bored of reading most of the document, and it is 174 pages, so the rest is just a summary of a skim reading).

There is a detailed graphic on printed page 24 that describes how to chose what type of survey to use to get good population estimates.

Starting on printed page 39 the report talks about integrating trawl survey data and SONAR data.

Printed page 49 starts talking about estimating population parameters from survey data.

Quotation:

Recommendation

  1. The spatial distribution of the fish should be considered when designing and
    analysing surveys. A decision tree has been provided to assist in the choice of
    methods available. Survey planners should be fully aware of the assumptions allied
    to any model-based estimation technique.
  2. The survey specific effect of tow duration, should be investigated in individual
    surveys. Shorter tows should be implemented if found to provide an improvement
    in the precision of the survey.
  3. Covariates should be used, if available, where they provide an improvement in
    the precision of the survey. Be aware that the covariates must have a good relationship
    with the response and be available over the entire sample space (not just
    the sampled area).
  4. Inverse variance weighting should be considered to combine survey data.
    When combining indices of the same resource, the inverse variance of the individual
    indices is a useful weighting scheme.
  5. The effective sample size to determine biological parameters should be investigated.
    The effective sample size of fish selected for ageing, measuring, etc. can
    be much smaller than the actual number of animals sampled, it is, therefore, important
    to account for this when reporting information on biological parameters.
    In cases where this can demonstrated to be smaller than current sample sizes
    more effort can be incorporated into sampling other species (including non-fish
    species) for consideration of an ecosystems approach (e.g. to compile community-
    based indicators).
  6. Quantiles of individual distributions can be used to map biological data rather
    than interpolating a summary statistic (e.g. mean length).
  • No labels