You are here:
Sampling and inference in environmental surveys
Citation:
Marker, D. A. AND D. L. Stevens Jr. Sampling and inference in environmental surveys. Chapter 19, Danny Pfeffermann, C.R. Rao (ed.), Handbook of Statistics 29A: Sample Surveys: Design, Methods and Applications. North Holland, 29A:27 pages, (2009).
Impact/Purpose:
In this chapter, we focus on surveys of environmental resources, which we loosely define as the air, water, soil, and associated biota that sustain our environment.
Description:
In this chapter, we focus on surveys of environmental resources, which we loosely define as the air, water, soil, and associated biota that sustain our environment. The objective of the surveys we consider will generally be an assessment of status, condition, or extent of a resource. The target population of the survey may be discrete and finite, for example, small lakes or wetlands, with well-defined population units; or may be a one-, two-, or three-dimensional continuum, for example, a stream network, a forest, or the volume of water in a large lake. Each of these calls for different types of frames and sampling techniques. The survey may be a one-time assessment or may include a long-term monitoring objective to assess change or trend. Addressing both objectives requires a balance of revisiting sites to assess trend and adding new sites to assess status. Traditionally, the focus of sampling in the environmental sciences has been on relatively small and well-delimited systems, e.g., at the scale of a lake or watershed or forest stand. However, some current environmental issues, such as global warming, contamination of surface and ground water by pesticides and other pollutants, and extensive landscape alteration are not localized. Quantifying the extent of symptoms of widespread concerns requires large-scale study efforts which in turn needs environmental sampling techniques and methodology that are formulated to address regional, continental, and global environmental issues. Survey design is a well-developed and established area in the statistical literature. There are many textbooks that provide excellent accounts of the essential attributes of good survey design, such as the necessity of clear definitions of the population of interest, the sample units, the sample frame, and how the sample is to be drawn (Cassel et al., 1997; Cochran, 1977; Kish, 1987; Lohr, 1999; Sarndal et al., 1992; Thompson, 2002; Yates, 1981). However, designs for environmental sampling often present additional challenges which we identify below. These include the need for broad population description; spatial context of the population; availability of ancillary information; inadequate frames; difficult access; multiple objectives; including status and trend; evolving objectives; and the need to satisfy multiple stakeholders. The focus of most survey methodology is estimating the mean value or total of a population. In contrast, an environmental survey often has a more general object, such as estimating the distribution function or the proportion of the population in various classes, for example, the proportion of lakes that meet designated used criteria. There may be many environmental responses of interest that are interdependent. A common objective of environmental surveys is to characterize the status of some resource as well as the change or trend in that status. These two objectives have somewhat conflicting design criteria: status is generally best assessed by sampling as much of the resource as possible, whereas trends are generally best detected by observing the same resource locations over time. Frequently, a secondary objective is the evaluation of relationships between attributes, both measured at the site and available on the frame. A characteristic of overwhelming importance for environmental populations is that they exist in a spatial context. The response will have spatial pattern and structure. Sites near to one another will tend to have similar physical substrate and be subjected to similar stressors, both natural and anthropogenic. Response can be influenced by topography, hydrology, and metrology. All these influences will tend to induce spatial patterns in the response. Another important characteristic of environmental populations is that some ancillary information (in addition to location) is almost always available. Currently, there is a wealth of remotely sensed information available from satellites or aerial photography that may be used to structure the sampling design or used in analysis. Environmental resources are often expensive and time-consuming to sample. Logistics can be difficult; often, the population of interest includes sites in remote locations, for example, lakes in wilderness areas. There can be considerable time and money expended in traveling between sites. Laboratory costs for analyzing individual samples may be nontrivial. Some environmental metrics can be time-consuming to evaluate, for example, quantifying the species richness and abundance of a macroinvertebrate sample requires the services of a skilled benthic taxonomist. For a large program, it may be a year or more after data are collected in the field before laboratory analyses are available. These may also be subject to substantially more measurement error than routinely found in other types of surveys. Nonresponse in environmental sampling can be substantial for reasons such as ease of physical access, safety, or permission. A practical complication frequently encountered in environmental sampling is the difficulty in obtaining an accurate sampling frame. In many instances, available sampling frames include a substantial portion of nontarget elements or fail to cover the entire population. The frame problem is aggravated by the sheer difficulty of collecting and analyzing samples. Environmental sampling almost always occurs with the backdrop of political, economical, and societal considerations so that statistical considerations represent only one aspect of a sampling design. Furthermore, because environmental issues can impact human populations, there are often multiple groups, agencies, and organizations that have an interest in the products of the survey. The interests of the multiple stakeholders are not always perfectly aligned. Meeting the interests of multiple stakeholders, while maintaining a scientifically and statistically rigorous design, can be a challenge. In many instances, the need for an environmental sample will be driven by the need to assess the condition of an environmental resource because of concern over potential degradation. The design needs to address current environmental issues but that is not sufficient. The current issues will eventually be resolved, but new, presently unrecognized issues will emerge. These issues will manifest themselves in unforeseeable ways, and they will affect resources that cannot now be identified. An environmental assessment program with the dual objectives of status and trend must be able to accommodate regrouping, recombining, expansion and contraction of the sample to permit such emerging issues, and evolving objectives to be addressed. The issues of inadequate frames, nonresponse and missing data, and evolving objectives drive a need for sampling designs with the flexibility to add, remove, or reallocate samples. Below, we review some of the sampling methodology that has been developed to meet the challenges that sampling environmental populations present: focus on broad population description; spatial context; ancillary information; inadequate frames; difficult assess; evolving objectives; and the need to satisfy multiple objectives and stakeholders.
