2.2 Where, What, and When to Sample
29
2.2.1.1.2 Random versus Nonrandom Site Selection
One of the most important considerations in site selection is the determination of whether the sampling sites will be selected using a randomized sampling design. Streams or lakes should be randomly selected for sampling if
the goal is to characterize populations of surface waters for a defined area
too big or impractical to census. This enables the statistics obtained for the
sampled waters to be applied to the full population of waters in the designated sample frame within the area. Generally, for a statistically based survey of surface waters, some form of stratified random sampling will be used
because this approach allows the sample population to be stratified such that
streams or lakes that are of greatest interest can be included in amounts that
are disproportionate to their frequency of occurrence in nature. Such a stratified random-sampling process preserves the ability to make population-level
extrapolations while maximizing the collection of data for the sites of greatest interest. For example, when it is known that landscape properties such as
bedrock or land cover account for spatial variation in surface water quality
or susceptibility to degradation, randomized selection and sampling of sites
within strata defined by influential landscape properties may allow multiple
subpopulation-level extrapolations that collectively provide more information about surface waters in a region than nonstratified randomized sampling.
A carefully targeted and stratified random sampling does not necessarily have
to entail a large and expensive sampling program. Random surveys of aquatic
resources conducted by the Environmental Protection Agency (EPA) have
often been large efforts that sampled hundreds to more than a thousand water
bodies. These have included the Wadeable Stream Survey (WSS), National
Lake Assessment (NLA), National Surface Water Survey (NSWS), and various
surveys conducted as part of the Environmental Monitoring and Assessment
Program (EMAP). Nevertheless, smaller surveys could also be conducted using
a random-sampling structure, thereby allowing extrapolation to a population
of waters of particular interest.
If all streams or lakes are included for potential sampling, accessibility may
complicate a totally randomized sampling design. This is particularly relevant
in remote areas with poor access. Remote sites may require extended periods of
time to reach, which lengthens the period over which the survey is conducted
and may introduce complications regarding sample holding times and costs.
This can also be problematic because environmental sampling conditions (e.g.,
stream flow) may vary during the survey if some of the sites take several days
to access. This can be important because data collected from surface waters
sampled during low-flow conditions are generally not comparable to those
determined during high-flow conditions.
Note that stratification can be performed on more than one variable or
characteristic. For example, within a randomized sampling design, candidate
29
2.2.1.1.2 Random versus Nonrandom Site Selection
One of the most important considerations in site selection is the determination of whether the sampling sites will be selected using a randomized sampling design. Streams or lakes should be randomly selected for sampling if
the goal is to characterize populations of surface waters for a defined area
too big or impractical to census. This enables the statistics obtained for the
sampled waters to be applied to the full population of waters in the designated sample frame within the area. Generally, for a statistically based survey of surface waters, some form of stratified random sampling will be used
because this approach allows the sample population to be stratified such that
streams or lakes that are of greatest interest can be included in amounts that
are disproportionate to their frequency of occurrence in nature. Such a stratified random-sampling process preserves the ability to make population-level
extrapolations while maximizing the collection of data for the sites of greatest interest. For example, when it is known that landscape properties such as
bedrock or land cover account for spatial variation in surface water quality
or susceptibility to degradation, randomized selection and sampling of sites
within strata defined by influential landscape properties may allow multiple
subpopulation-level extrapolations that collectively provide more information about surface waters in a region than nonstratified randomized sampling.
A carefully targeted and stratified random sampling does not necessarily have
to entail a large and expensive sampling program. Random surveys of aquatic
resources conducted by the Environmental Protection Agency (EPA) have
often been large efforts that sampled hundreds to more than a thousand water
bodies. These have included the Wadeable Stream Survey (WSS), National
Lake Assessment (NLA), National Surface Water Survey (NSWS), and various
surveys conducted as part of the Environmental Monitoring and Assessment
Program (EMAP). Nevertheless, smaller surveys could also be conducted using
a random-sampling structure, thereby allowing extrapolation to a population
of waters of particular interest.
If all streams or lakes are included for potential sampling, accessibility may
complicate a totally randomized sampling design. This is particularly relevant
in remote areas with poor access. Remote sites may require extended periods of
time to reach, which lengthens the period over which the survey is conducted
and may introduce complications regarding sample holding times and costs.
This can also be problematic because environmental sampling conditions (e.g.,
stream flow) may vary during the survey if some of the sites take several days
to access. This can be important because data collected from surface waters
sampled during low-flow conditions are generally not comparable to those
determined during high-flow conditions.
Note that stratification can be performed on more than one variable or
characteristic. For example, within a randomized sampling design, candidate