Statistical Methods for Environmental Pollution Monitoring.

Abstract

Environmental pollution studies may be divided into the following broad and somewhat overlapping types.1. Monitoring. Data may be collected (a) to monitor or to characterize ambient concentrations in environmental media (air, water, soil, biota) or (b) to monitor concentrations in air and water effluents. The purpose may be to assess the adequacy of controls on the release or containment of pollutants, to detect long-tenn trends, unplanned releases, or accidents and their causes, to provide a spatial or temporal summary of average or extreme conditions, to demonstrate or enforce compliance with emission or ambient standards, to establish base-line data for future reference and long-range planning, to indicate whether and to what extent additional infonnation is required, or to assure the public that effluent releases or environmental levels are being adequately controlled. Research.Field and laboratory data may be collected (a) to study the transport of pollutants through the environment by means of food chains and aerial pathways to man and (b) to detennine and quantitate the cause-andeffect relationships that control the levels and variability of pollution concentrations over time and space.Many design and statistical analysis problems are common to monitoring and research studies. Environmental data sets also tend to have similar statistical characteristics. These problems and characteristics, discussed in the next section, motivate the topics discussed in this book. STATISTICAL DESIGN AND ANALYSIS PROBLEMSNumerous problems must be faced when applying statistical methods to environmental pollution studies. One problem is how to define the environmental "population" of interest. Unless the population is clearly defined and related to study objectives and field sampling procedures, the collected data may contain very little useful infonnation for the purpose at hand. Chapter 2 gives an approach for conceptualizing and defining populations that leads into the discussion of field sampling (survey) designs in Chapters 3-9. The important role that objectives play in detennining sampling designs is discussed in Chapter 3.Once data are in hand, the data analyst must be aware that many statistical procedures were originally developed for data sets presumed to have been drawn from a population having the symmetric, bell-shaped Gaussian ("nonnal") distribution. However, environmental data sets are frequently asymmetrical and skewed to the right-that is, with a long tail towards high concentrations, so the validity of classical procedures may be questioned. In this case, nonparametric (distribution-free) statistical procedures are often recommended. These procedures do not require the statistical distribution to be Gaussian. Alternatively, an asymmetrical statistical distribution such as the lognonnal may be shown or assumed to apply. Both of these approaches are illustrated in this book. Frequently, a right-skewed distribution can be transfonned to be approximately Gaussian by using a logarithmic or square-root transfon...

Assessing Risks to Biodiversity from Future Landscape Change

“…For the results reported here we used 1000 repetitions of the simulations and a standard deviation of 2.0 for the measurement errors. This value for the standard deviation corresponds to a coefficient of variation of about 7.3 (Gilbert 1987), or 730%, a substantial degree of variation.…”

Nitrous Oxide Emissions from Turfgrass Receiving Different Irrigation Amounts and Nitrogen Fertilizer Forms

“…Minimum positive and negative detection limits for each measurement date were determined using the method described by Parkin et al (2012). However, based on past research by Gilbert (1987), all measured flux values in this study were used even if they fell below the minimum detection limit (or within the detection limit band), which is the least biased course of action.…”

Trends of precipitation in Beijiang River Basin, Guangdong Province, China

“…For n < 10, jSj is compared directly to the theoretical distribution of S derived by Mann and Kendall (Gilbert, 1987). At a certain probability level, H 0 is rejected in favour of H 1 if jSj ½ S˛/ 2 , where S˛/ 2 is the smallest S which has a probability less than˛/2 to appear in the case of no trend.…”