Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 89
Chemical and Biological Methods INTRODUCTION The purpose of this section is to set forth general pr inciples in study design and in chemical and biological methodology. The specific examples cited should aid the reader in critically evaluating oil pollution research literature and present a general guide to the present state of the art in this area of research. We strongly advocate that scientists be properly educated and trained in the fundamentals of the various disciplines of biology, chemistry, ~ . or marine sciences before engaging in the difficult research problems associated with petroleum pollution in the mar ine environment. Thus, it is not the intent of this section to provide a detailed account or "cookbook " approach to par ticular measurements . General Strateg ies of Study Design · In the east 10 vear s a number or new scat 1S1: coal · _ _ A discussion of statistical techniques used in studies of the fate and effects of oil in the environment would need to cover most of the areas in modern statistics. The more important general problems that have arisen in oil spill research will be br lefty discussed and areas of study identified where recent developments in statistics may contr ibute to their solution. For the most part, discussions of the role of sampling and statistical design in oil spill studies have been confined to the use of classical statistical methods Cox et al., 1977) . _ , _ , _ _ _ methods have been developed for a wide range of environmental problems . These methods, used in water qual ity research, air pollution, and environmental sampl ing, have obvious appl ication to the study of oil in the mar ine environment . _ ~, d iscussed and see, for instance, G .V . In comparison with the esoteric techniques of organic chemistry, f luid dynamics , and faunal analysis employed in studies of the effect of oil in the environment, the basic concepts of statistical design and analysis are relatively straightforward. Yet, statistical problems have often been cited as the cause of many of the def iciencies in oil pollution studies. The main issue is to design an experiment, set of exper iments, or sampl ing program to answer a specif ic set of questions 89
OCR for page 90
go within a specified set of cost, manpower, and time constraints. It is assumed that invest igator s can agree on a general model for a g iven system. When the questions, constraints, and model have been specif led, the next step is to allocate samples and experimental treatments optimally so that the scientific and regulatory questions being asked can be answered clear ly. I f a problem is less sharply defined, the Dual ity and efficiency of the statistical design decrease. Studies of oil pollution in the mar ine environment are particularly sub ject to this effect. In the past, but less frequently now, fortunately, several large environmental research programs were constructed by committees in which scientific questions, cost constraints, model processes, and statistical des ~ gn were decided in a more or less haphazard and illog i- cal order . Thus, rather than being the cause of the problem, improper statistical design and data analysis were more often symptoms of fundamental cliff iculties within the research plan. The nature of pollution studies is such that some of the work must be done in areas where pollution has occurred, in what might be called an unplanned experiment. The objective, in most cases, is to assign a causal relationship between a particular event and some effect in the area. There is also an interest in drawing conclusions that could be appl fed in other r elated cases. In a classical randomized experiment, a set of areas is selected and then oil treatments are randomly assigned to half of the areas. In most field studies of oil in the environment, the treatment, for example, of the concentration of oil in the sediment, has not been assigned by the investigator as part of the experiment design, but rather is a consequence of human error and natural forces that dispersed the oil unevenly in different areas. This does not preclude making reasonable inferences about the causal nature of the event observed; however, one must make stronger assumptions about the underlying nature of the data and use rather different analyses. There is a large literature in statistics on the problems associated with observational studies. Recently McKinlay (1975) has reviewed the literature on observation studies, particularly those influencing human populations. Cochran (1951, 1968) was an early contributor in this area and still offers the definitive work. R.H. Green (1979) gives a relatively clear analysis of this problem in an environmental setting. A s imple example serves to illustrate the problem of observational data. The following Is a f ictitious data set used to illustrate Simpson's paradox" (Lindley and Norick, 1981) in an environmental setting. Table 3-1 describes the results of a large, well-designed, random survey of an oil spill area. The 200 samples were classified as to whether they fell above or below the median oil concentration and whether the biomass of the benthic community fell above or below the median. Table 3-1 shows clearly that there is a positive correlation between high oil concentration and high biomass concentration. Suppose, however, we again divide the samples into two subareas, those below 5 m depth and those above 5 m depth, thus spl itting the data into two separate tables. We see that the correlation in Table 3-1 is reversed when we look at the two subpopulations separately. The explanation for this is obvious: although the sampling program was random, the d istr ibution of oil in the sediment (the treatment) was not; heavy
OCR for page 91
91 TABLE 3-1 Matrix of Oil Concentrations in Sediments and Correlations With Benthic Organism Biomass Oil Concentration Groups (~) Below Median Concentration Biomass Concentration Groups Above Median Concentration c-~ i men t a t A 1 1 Ma to r n-lath A; ; n .R t~dv A r ea Below median concentration Above median concentration Sediment Above and Below 5 m Water Depth Above 5-m water depth Below median concentration Above median concentration Below 5-m water depth Below median concentration Above median concentration 70 30 70 10 o 20 30 70 30 o 10 70 NOTE: Hypothetical case (see text). in entire data set. Percentages are of total samples concentrations of oil appear in the shallower areas exactly where the highest biomass is. Inference under these conditions has been studied by Lindley and Novick (1981) and by Blyth (1972~. The reader is referred to these articles for various proposed solutions to this problem. Needless to say, this situation can arise in more subtle ways than that presented above; for instance, chronic oil pollution is usually associated with other industrial and urban activity which could also have an impact on the system. Sampling Procedures and Equipment The requirements of specialized sampling procedures and equipment are discussed in the following sections on chemical and biological methods. Statistical Design of Analytic Procedures in the Laboratory Often the complexities of environmental studies lead one to overlook the importance of careful design of laboratory analytical procedures. Proper design can prevent confounding of laboratory effects with natural processes that we wish to study. For example, where the same analysis is undertaken by two laboratories, samples should be assigned
OCR for page 92
92 so that possible differences in analytic techniques are not confounded with natural variation. For instance, assigning all samples from area A to laboratory X and all samples from area B to laboratory Y will certainly confound the ef feet of analytic bias between laborator ies with true var. iations in concentration between the two areas. There is a lengthy and very useful literature on statistical design and quality control in laboratory situations (Tietjen and Beckman, 1974~. Presenting Results From Complex Oil Pollution Studies One of the more dif.icult problems in environmental oil pollution studies is to present the data and analysis in a comprehensible way. In most cases the data f rom a study consist of many different kinds of measurements made on samples distributed in both space and time. The data analysis problem is to decipher relationships among samples and variables and then to present these relationships in a compact way. Often, classical statistical analysis (analysis of variance, regression, etc.) used in reports and papers describing oil spill events provides little insight into the character of the processes and the data under investigation. Many of the techniques of exploratory data analysis can be used to investigate these relationships. J.W. Tukey and P. Tukey (personal communication, 1981) have described different techniques for displaying multivariate data. Cleveland and Kleiner (1975) and R.H. Green (1979) have given a number of methods for graphically presenting multivariate environmental data. This is one area where a great deal more work needs to be done. Designing Ecological Experiments to Study the Effects of Pollutants The discussion of problems one can encounter in observational studies should make clear the reasons for going to planned experiments to study the impact of oil or other marine pollutants. W. Smith et al. (1981) provide a general discussion of the design of ecosystem experiments. Large experiments, such as the No. 2 fuel oil experiment conducted at the University of Rhode Island Marine Ecosystems Research Laboratory (see Biological Methods section), present certain problems not encoun- tered in usual experimental design situations. First, since such a system is partially open to the environment, all conditions within the system cannot be precisely controlled. The second problem is that since each experimental unit replicated is costly to maintain and measure, the number of exper imental units repl icated is relatively small. These k inds of exper iments occur over t ime, and thus, par t of the exper imental design is to determine the treatment time course. Careful des ign of t ime ser ies exper iments can help to overcome the problems mentioned above. There is a growing literature in this area t~qulholland and Gowdy, 1977; Glass et al., 1975; Box and T'ao, 1975; Huitema, 1979; Brockway et al ., 1979 ~ .
OCR for page 93
93 An Oil Spill Survey Design Environmental sampling problems have been well studied from a statisti- cal design perspective. Sampling spatially and time-varying processes has been discussed by W. Smith (1979) and others. Moss and Tasker (1979) give a review of these methods in water quality research. In addition, composite sampling and other techniques for efficient sampling and analysis of environmental data have been investigated (Elder et al., 1981; R.H. Green, 1979; Smeach and Jernigan, 1977~. W. Smith (1979) developed a specific plan for the oil spill survey problem involving a two stage sampling scheme where, in the first stage, a large number of samples are collected in a systematic survey (grid or modified grid). Easy to record data, such as sediment type or if oil is present, are recorded in the first stage. In the second stage of the sampling program, the data from the first stage are used to stratify the samples. The original samples can then be subsampled for the more costly chemical and biological analysis. In oil spill work this procedure has the great advantage that the first stage of the program can be implemented on short notice. The cliff icult sampl ing question can be postponed to the second stage of the program when the legal engineer ing and scientific questions that the oil spill raises have been better defined. Summary The interweaving of the various facets of observational and experimental studies of inputs, fate, and effects of petroleum in the marine environ- ment demands careful attention to the combined requirements of many different types of sampling and measurement methods. It is not appro- priate in a report of this type to review critically all aspects of environmental and engineering measurements relevant to oil pollution studies. For example, meteorological, physical oceanographic, and sedimentological measurements are often of importance to these studies. The reader is referred to texts and reviews in these areas of study for description and evaluation of methodology. Physical parameters such as specific gravity, viscosity, and pour point of crude oils and petroleum products are not extensively discussed because they have been well defined in American Society for Testing and Materials (ASTM) protocols (R.C. Clark and Brown, 1977) and have not been the subject of controversy in oil pollution research and monitoring activities. Chemical methods for measur ing petroleum and petroleum compounds and biolog ical methods for measur ing var ious levels of ef feats of petroleum in the mar ine environment have been sub jects of cant inuing controversy and debate since an increase in studies of petroleum pollution in the late 1960s. For this reason chemical and biological methodology is descr ibed and evaluated in the following sections .
Representative terms from entire chapter: