Personal tools
You are here: Home → 2. Design → 2.1 Status and Trend Monitoring Design → 2.1.2 Temporal Design → 2.2.2 Choosing an appropriate temporal design
 

2.2.2 Choosing an appropriate temporal design

Overview    Choosing    Results and next steps

Temporal Juggling
Enlarge Image

As discussed in the introduction to step 2 (Monitoring Program Design), it is important to consider all design components collectively because the total cost must be balanced among the cost of the individual components and among potentially competing objectives. 

More information
 

In designing the monitoring program, you will be faced with a problem.  You will have a fixed budget allocated to collecting data at each spatial unit (e.g, a site) during each temporal unit (e.g., year), and an expected budget for the study period.   You will know how much it costs to collect the data at a site (including travel costs).   Your challenge will be to optimize the allocation of sampling across the potential set of spatial and temporal units during the study period.  It might also be important to allocate effort to sampling within the temporal window to obtain an adequate estimate of your metric and/or to evaluate the variation introduced by your response design.

The following considerations are important for optimizing this allocation of sampling effort:

  •  Degree of certainty - The level of confidence that you must have in the results of your monitoring program plays a significant role in determining the appropriate design.  In general, the degree of certainty in monitoring results is lowest for opportunistic designs, intermediate for model-based and survey designs, and highest for census designs.  It is lowest for opportunistic designs because it is difficult or often impossible to assess how well the chosen sample sites represent the domain for which inferences are intended.  Because of the non-statistical nature of sample site selection, it is often impossible to assess the degree of certainty of results from opportunistic sample sites because you cannot determine the precision or bias associated with inferences to entire populations obtained from data collected at opportunistic sample sites.  The degree of certainty is intermediate for model-based and survey based spatial designs because, although they do use selection methods that allow for the sampled sites to be representative, with an estimated precision and bias, the uncertainty around estimates derived from model or survey based sites is inflated because of statistical sampling error.  In addition, model-based designs can be subject to unknown uncertainties associated with model assumptions.  The degree of certainty is highest for census designs because the sampling of all members of the target population (either via a fixed counting station or by sampling at all sites in the domain) results in no statistical sampling error or faulty assumptions about the representativeness of selected sites.
  • Cost - The cost of designs generally varies inversely in relation to their degree of certainty.  While the high degree of certainty provided by a complete census may be attractive, in many cases the cost associated with conducting a census over a large geographic area or for the entire study period will be prohibitive.  Because of the myriad of factors that are associated with estimating costs for different types of monitoring designs, we do not attempt to provide explicit guidance on the cost of different designs here.  However, it is important that you keep in mind the need to adopt a design that is within the available budget.  This may mean that you will have to take a hard look and potentially revise your objectives for the degree of certainty you will obtain from your monitoring program, given the spatial, temporal, and response designs that fall within your budget.   
  • Feasibility - Adopting a design that achieves your desired degree of certainty and that is within your budget may result in an inappropriate design if you have not considered other factors that may affect the feasibilty of implementing the design.  For example, census or survey-based designs that require permission from landowners to access sites will not be appropriate if you are denied access to their lands.  As a result, you must realistically evaluate the feasibilty of implementing a particular design given the access constraints specific to your monitoring area.
  • Existence of a verified model - Choosing a model-based design will obviously not be an option if you lack an appropriate model that can guide your site selection process. 
  • Flexibility - It is a common occurrence that over the life of a monitoring program, there may be changes in the goals and objectives, monitoring technologies, allocated budgets, or other constraints.  Some designs are more amenable than others to the modification that may be necessary to meet these new challenges.  For example, an initial objective that requires an abundance estimate over a prescribed monitoring region might be changed to an objective that requires abundance estimates for specific populations within that region.  A spatial/temporal design that allows you to add or subtract sites without biasing your results is more desirable than one that requires an entirely new design.


A framework that can be used to balance the various competing choices in designing a monitoring program consists of : 

The following citations provide a good foundation underlying the concept of power (the chance of detecting an effect if the effect is present) and how it has been applied natural resource monitoring:

Fairweather, P.G.  1991.  Statistical power and design requirements for environmental monitoring.  Aust. J. Mar. Freshw.Res.  42:555-567.
 

Gerow, K. G. 2007. Power and sample size estimation techniques for fisheries management: assessment and a new computational tool. North American Journal of Fisheries Management 27:397–404.

Hatch, S.A.  2003.  Statistical power for detecting trends with applications to seabird monitoring.  Biol. Conserv. 111:317-329.

Link, W. A., and J. S. Hatfield. 1990. Power calculations and model selection for trend analysis: a comment. Ecology 71:1217–1220.

Peterman, R. M. 1990. Statistical power analysis can improve fisheries research and management. Canadian Journal of Fisheries and Aquatic Sciences 47:2–15. 

Our guidance to help you decide on a temporal design for your monitoring program revolves around the differences between temporal designs involving sampling at either single or multiple sites.

Single Site Temporal Design

If your objectives concern sampling at a single site, then your temporal design choices focus on the allocation of sampling effort within and across temporal units (e.g. years).  How you allocate effort between these two temporal scales should be based on the influence that these two factors have on the precision and bias of your results balanced against the cost of implementation.  

For example, if your objective is to obtain yearly estimates of abundance, then your choices are defined by how much sampling effort should be devoted to within year sampling (to get a good estimate of that year's abundance) and whether you need to sample every year.  If an estimate of abundance is needed every year (and interpolation is unacceptable to fill in estimates for unsampled years), then your choice is clear: sample every year.  Your decision then focuses on how often you need to sample within the year to obtain an adequate estimate of each year's abundance.  If interpolation is acceptable (with potential loss of precision), then a design that doesn't require measurements each year might be used to achieve your objectives.

If your primary objective is trend evaluation, a design that excludes sampling some years may achieve approximately the same power as an annual sampling effort, but with substantially less cost.  In this case, temporal design options could include:

  • Sample every year
  • Select years randomly or systematically
  • Select years opportunistically   

Determining the optimum approach will require knowledge about the expected pattern of temporal variation during the study period.  Numerous single site change/trend detection methods are available (e.g., see Gerrodette, T. 1987;  Gerrodette, T. 1991, and Gerrodette, T. 1993) as are on line resources.   For example,  TRENDS is a program designed to carry out a linear regression power analysis under differing design options available from  NOAA's Southwest Fisheries Science Center. 

Multiple Sites Temporal Design

Your temporal design choice is a little easier to think about when our monitoring objectives require sampling a set of sites during the study period.  Again, thinking about year as the temporal unit, we could sample all sites every year.  However, for some objectives, it might be just as efficient to sample some sites annually, and some sites on a periodic cycle like every "n" years (where n could by 2, 3, 4...years).  These panel designs allow investigators the opportunity to investigate a greater number of sites during the study period than might be possible with a "sample every site every year" design without a loss of precision for status estimates (e.g., population abundance estimates).  In addition, panel designs afford approximately the same mean trend detection power after three monitoring cycles have passed (Urquhart and Kincaid, 1999).  For example, a design that includes an annual panel (25 sites), along with three 3-year panels (25 sites each panel, beginning in year 1, year 2, and year 3) will achieve very closely the same power after nine years, as a design with the same 50 sites visited every year.  The advantage of the multi-panel design is that a total of 100 sites will have been sampled, compared with 50 for the every site-every year design (see Urquhart and Kincaid, 1999).

Allowing flexibility in the choice of sampling patterns over years offers a variety of possibilities when your monitoring program includes multiple sites.  However, a thorough evaluation of selecting a temporal sampling pattern cannot be done without simultaneously considering both the spatial and response designs.  You will have a relatively fixed budget available that you will have to allocate to achieve multiple objectives.  This optimization process will require a knowledge about the spatial, temporal, and "residual" variation (including how well your response design estimates the site's metric score), and costs for collecting the relevant data at a site (including travel cost between sites).  Unlike the availability of single site trend detection tools, we are unaware of analogous "off the shelf" tools for multiple site trend detection.  However, with an estimate of the important components of variation, the use of simulation tools (along with the computational power of current desktop computers) allows an evaluation the relative merit of different design choices.   Several recent publications describe a variety of simulation examples that illustrate how this problem can be addressed (see Wagner, et al., 2007; Dauwater, et al., 2009, Jacobs, et al., 2009).

 

Next: Results and next steps | Go Back

Document Actions