q

Preliminary Sampling

Preliminary sampling helps to ensure that the population of interest is being

sampled and that its distribution is being evaluated. Preliminary sampling or

previous testing helps avoid the problem of collecting large sets of useless data

because of ineffective gear, improper sample preparation, or preservation. The

target population can be easily missed. especially for biological monitoring.

Properties of Estimators

The goal for sampling is the collection of a time series of data that can be

summarized for a

time period with a single estimator. Most often the time

series of interest consists of either years or seasons. The estimator is the expected

value, or the mean or another estimate of central tendency (e.g., the median). For

regression analysis, other parameters, such as slope, may be of interest. The

properties of the estimator should be considered so that it relates to the needs of

the sampling program.

Statistical analysis involves testing the properties of sample estimators and their

data sets. The monitoring objective, design, and the degree to which these

assumptions are met by a data set determine the appropriate statistical test.

Normal Distribution

Knowledge of the distribution of water quality variables is important for charac-

terization of water quality and also to determine applicable statistical techniques.

In addition, much more information (e.g., spread, skewness) is contained in the

data distribution as compared to only using point estimates of central tenancy such

as the mean or median.

The normal, log normal, and the gamma distributions are common theoretical

distributions that water quality variables exhibit. The log normal distribution may

be the best for many water quality and hydrologic variables and is widely used in

water quality studies. If the logarithms of the random variable are normally

distributed, then the random variable itself has a log normal distribution.

The sample estimator is a true estimate of the population.

normal population,

the sample mean and the sample median (center value of an ordered set) are

unbiased estimators.

Variability in the y variable at any value of x is independent of x-value and is

randomly distributed. The data scatter should be the same for either high or low

values along the x axis.

Time correlation (temporal autocorrelation) is found when the value of one

measurement is dependent on the previous measurement. If a value is dependent

upon the value of a parameter at another location, then there is spatial

lation. Dependencies such as these must be known and accounted for. Sequential

samples taken during a storm are not independent because they are subject to a

common influence, the storm flow. This must be considered when analyzing the

data.

4.14