Database Structure and Development (con't)
Problems in Compiling Heterogeneous Data The data in this study have been acquired from historical sources as well as from ongoing field programs. It is not feasible to apply the standard quality-control (QA/QC) protocols that check on the details of sampling and analytical methodology (see Baker and Kravitz, 1992) to heterogeneous data. For this reason, until recently, data from heterogeneous sources were rarely compiled. Yet, historical data may contain considerably useful information. To rescue data while minimizing problems associated with data comparability, special batch screening techniques were used to identify and call attention to data that have unresolved problems. Although such tests do not necessarily prove that the data are in error, they alert users to data that should be reevaluated or confirmed before use in environmental characterization. It is emphasized here that data anomalies are a normal and virtually inescapable part of any analytical program – especially those that deal with constituents in trace quantities. In recent years, increasing attention to data quality has greatly improved the reliability of analyses, especially from major federal agency programs and their contractors. However, as will be seen in this chapter, comparability checks are necessary even for data from well-established laboratories. Goal - Maximizing Effective Data Resources; Caveats Inaccuracies in one or more constituents in a data set do not necessarily apply to other constituents from that set. Data screening thus has as goals maximizing both the spatial extent and reliability of the total data set to users. Although reasonable care has been taken in reporting and quality-screening data, the approaches used can only detect larger, relatively persistent and environmentally significant data quality problems. The responsibility for setting standards of quality for specific uses of the data cited here rests with the user. The designation, "W", in quality-control fields placed adjacent to data fields in the full database (Appendix D) warns that the data in question appear to show unresolvable anomalies. The designation "N" before remarks in the quality-control fields indicates that the data in question were not quality-screened. "0" in a concentration field means that the constituent was below detection limit for the method utilized. Other remarks in the quality control fields provide comments relating to the data that may be helpful in interpreting or further evaluating the data. Appendix D shows quality codes, other interpretive notes and EPA station class codes for further explanation. Data Screening - Discriminant Functions Much is now known about the chemical composition and behavior of natural and contaminated sediments in estuarine ecosystems (e.g. Windom and others, 1989; Horowitz, 1991; Schwarzenbach and others, 1993). Expected internal chemical relationships and comparisons among data from different sources for given geographic areas help evaluate the quality of data sets. We used a series of operations and screening steps to explore heterogeneous historical data. These techniques were developed for use in studies of sediments from the Boston Harbor - Massachusetts Bay area (Manheim and others, 1998). In the present approach batches of data are screened for completeness, internal consistency, and reasonableness in terms of geochemical parameters widely applicable to sediments. One problem is incorrect or uncertain sample and station locations. For example, some listed latitudes and longitudes for stations near New Orleans were found to plot on land, rather than offshore. Fortunately, supplementary site descriptions available in the original reference made better site identification possible. Data sorting and concentration range assessment A preliminary screening of data utilized sorting analytical parameters by concentration in descending order, delineating (color-coding) data that fell into given concentration categories. A set of concentration range tables applicable to estuarine sediments is provided in Appendix B. Of course, high concentrations may be real and due to influence of specific contaminants – i.e. one of the objects of inquiry. More sensitive discriminant functions relating to data quality utilize batch plotting of data on log/log scales, as may be seen in Figures 7, 8, 9, and 10. Aluminum, organic carbon, and grain size (e.g. "fines") have been used as normalizing constituents to attempt to reduce the effect of sediment variability. However, none of these are suitable for sediment quality control in the current study. Instead, zinc, which is included in virtually all data sets and has shown special attributes for this purpose (Manheim and others, 1998) has been found most suitable as a comparator element in discriminant plots for inorganic constituents (mainly metals). Phenanthrene and other widely-reported organic components were used for similar discriminant plotting purposes for organic constituents. Figure 11 shows the linear relationship between iron and zinc for two data sets, one performed by hot nitric acid extraction, and the other (Lab. 2) by total dissolution. Divergence of the latter to higher iron values is expected because much insoluble iron is present in clay minerals. Scattering is attributed to local inhomogeneities and sample variability. A third laboratory's data had no iron values. The higher Zn values are plotted in Fig. 10 at approximate Fe ranges inferred for the types of sediments in question. Most of the high values as well as other datapoints showing anomalous Zn/Fe ratios have as a common feature proximity to the New Orleans shoreline region where samples close to discharge canals were earlier documented to have enhanced contaminant concentrations (Overton and others, 1986). The preponderance of the information indicates that the anomalous Zn values represent contaminant input rather than data quality problems for this area. Plots of data sets including aluminum (Figure 7) show more drastic influence of different analytical methods, since the greater part of aluminum in silicate minerals like clay and feldspars is not extracted by nitric acid. All the data shown in Figure 6 were performed by a USEPA leachate methods utilizing standard hot nitric acid extraction, followed by spectrochemical end detection methods (USEPA, 1983). The datapoints plot well below the regression line based on a national set of analyses (NST, 1994) based on total breakdown of the aluminosilicate mineral structure. Differences between alternative methods are smaller but clearly discernible for chromium (Figure 8). The differences are still smaller for heavy metals such as Cu (Figure 9) and are not significantly different for samples in which contaminants are a significant component. The purpose of leachate analyses is, in fact, discrimination against naturally-occurring metals in clay and iron oxide matrices, thereby enhancing the relative contaminant signal. There are advantages and disadvantages to both the leachate and total dissolution approaches. Examination of data for arsenic/ zinc relationships (Figure 10) reveals wide divergence, when compared with the national NS&T data set. One set (Lab. 1) showed consistency within expected local scattering whereas data from Lab 4 displayed distributions that could not be reconciled with the analytical methodologies employed, or with data from well-controlled data from independent sources for common local areas and sediment types. Arsenic data from this source were therefore assigned "W" designations. Samples from Lab 2 and Lab 3, utilized leaching techniques. Because these data are on the low, rather than the high side for As, they may reflect a deliberately weak attack on naturally-occurring As concentrations in clay minerals. However, such data should not be combined with other data to form regional means without further data qualification. Notes on these observations were placed in the Quality column for As.
After anomalies were isolated by the methods indicated above, the data were checked against data from geographically-controlled independent sources. "Standard data sets" for which a high degree of data quality control was available were mainly used for the comparisons. The purpose was to distinguish between valid (higher values) contaminant anomalies that might be due to local contaminant sources from anomalous data of unresolved origin. An efficient method for constraining local area in comparisons of discriminant functions was sorting data according to geographical grid numbers. (Fig. 6). The data were first queried or sorted or in descending order by the parameter in question to eliminate nondetects. Then a consolidated table containing key station data with source information and a limited number of analytical parameters (texture and heavy metals, for example) were sorted by grid number followed by the element in question and the discriminant ratio (e.g. Zn/Cu). Discriminant ratios differing from those of well-qualified data by a factor of greater than 2.0 within grid squares were often a helpful indicator of significant anomalies. Where no problem resolution was available, data sets with a significant number of anomalous data were flagged with warning designations and excluded from interpretive output like histograms, statistical computations or area plots. Organic constituents were analyzed in the Louisiana Department of Environmental Quality resource studies of 1983-84 (Overton, 1984, Schurtz and St. Pe, 1984). Except for samples close to the New Orleans shore these were mostly below detection limits. The USEPA EMAP studies of 1991-1994, used more recent methodologies with greater sensitivities. These data yield much more quantitative data on the many related congeners or organic species. Since these data were entered into the database just prior to this publication, they were not reviewed for quality nor presented in interpretive diagrams as were the metals. USEPA and its consultant firms employ carefully controlled procedures and these data are expected to maintain a high standard.
|
||||||||||||||||
