JEQ Journal of Natural Resources and Life Sciences Education
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via ISI Web of Science (14)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Paul, J. F.
Right arrow Articles by Copeland, J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Paul, J. F.
Right arrow Articles by Copeland, J.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Paul, J. F.
Right arrow Articles by Copeland, J.
Related Collections
Right arrow Spatial Distribution
Right arrow Water Quality
Right arrow Watershed and Landscape Processes
Right arrow Water Pollution
Journal of Environmental Quality 31:836-845 (2002)
© 2002 American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America

TECHNICAL REPORTS
Landscape and Watershed Processes

Landscape Metrics and Estuarine Sediment Contamination in the Mid-Atlantic and Southern New England Regions

John F. Paul*,a, Randy L. Comeleob and Jane Copelandc

a USEPA, 27 Tarzwell Drive, Narragansett, RI 02882
b Indus Corp., Corvallis, OR 97333
c Computer Sciences Corp., Narragansett, RI 02882

* Corresponding author (paul.john{at}epa.gov)

Received for publication January 8, 2001.

   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 METHODS
 RESULTS
 DISCUSSION
 SUMMARY
 REFERENCES
 
In a previously published study, quantitative relationships were developed between landscape metrics and sediment contamination for 25 small estuarine systems within Chesapeake Bay. These analyses have been extended to include 75 small estuarine systems across the mid-Atlantic and southern New England regions of the USA. Because of the different characteristics and dynamics of the estuaries across these regions, adjustment for differing hydrology, sediment characteristics, and sediment origins were included in the analysis. Multiple linear regression with stepwise selection was used to develop statistical models for sediment metals, organics, and total polycyclic aromatic hydrocarbons (PAHs). The landscape metrics important for explaining the variation in sediment metals levels (R2 = 0.72) were the percent area of nonforested wetlands (negative contribution), percent area of urban land, and point source effluent volume and metals input (positive contributions). The metrics important for sediment organics levels (R2 = 0.5) and total PAHs (R2 = 0.46) were percent area of urban land (positive contribution) and percent area of nonforested wetlands (negative contribution). These models included silt–clay content (metals) or total organic C (organics, total PAHs) of sediments and grouping by estuarine hydrology, suggesting the importance of sediment characteristics and hydrology in mitigating the influence of the landscape metrics on sediment contamination levels. The overall results from this study are indicative of how statistical models can be developed relating landscape metrics to estuarine sediment contamination for distributions of land cover and point source discharges.

Abbreviations: ANOVA, analysis of variance • CV, coefficient of variation • DEM, digital elevation model • EMAP, Environmental Monitoring and Assessment Program • GIS, geographic information system • LUDA, land use data analysis • MLRSS, multiple linear regression with stepwise selection • NOAA, National Oceanic and Atmospheric Administration • PAH, polycyclic aromatic hydrocarbon • PC1, first principal component • PC2, second principal component • PCA, principal component analysis • PCB, polychlorinated biphenyl • TOC, total organic carbon • USEPA, U.S. Environmental Protection Agency • USGS, U.S. Geological Survey • VIF, variance inflation factor


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
METHODS
 RESULTS
 DISCUSSION
 SUMMARY
 REFERENCES
 
THE EMPHASIS ON CONTROL OF pollution sources over the last decade in the USA has shifted from point to nonpoint sources, the result of Congressional amendments to the Clean Water Act in 1987 (Navigation and Navigable Waters Act, 1994; USEPA, 1991). Runoff is the major source of nonpoint pollution of surface water, with the magnitude of the polluted runoff dependent upon the land cover, land use, and management practices employed (Basnyat et al., 1999). The interest in pollution control is for the protection and maintenance of aquatic resources within proximity of these land areas. A recent National Water Quality Inventory (USEPA, 1995) indicated that runoff from urban areas was the largest source of water quality impairments to estuaries.

From several perspectives, the development of relationships between point sources of pollution and condition of aquatic resources is relatively straightforward compared with the development of comparable relationships for nonpoint sources. First of all, by definition, nonpoint pollution is more diffuse, and thus harder to quantify. It is widespread over the land surface, for example, by atmospheric deposition or application of agricultural pesticides (Patty et al., 1999; Mason et al., 1999; Sherrell and Ross, 1999). Secondly, the land, and the cover on it, acts as a filter, retarding the overland flow of pollutants, diffusing them even further, and providing a medium for geochemical transformations (Correll et al., 1992; Schueler, 1994; Lajtha et al., 1995; Munn and Gruber, 1997; Wahl et al., 1997; Thierfelder, 1998). Finally, the pollution can be directed downward into the ground water, providing for even more complex pathways (Hughes et al., 1998; Hussain et al., 1999; Li et al., 2000).

Measures of the characteristics of land cover, land use, and management practices have been of significant interest in recent literature (O'Neill et al., 1988; McGarigal and Marks, 1995; Wickham and Riitters, 1995; Jones et al., 1996, 1997; Riitters et al., 1995; Thierfelder, 1998; O'Neill et al., 1999). As a result, many different landscape metrics have been developed. A relevant question is how does one determine "which, if any, of these landscape metrics are appropriate for predicting the impact of nonpoint pollution on estuarine receiving waters?"

Comeleo et al. (1996) reported a study of 25 small estuarine systems in Chesapeake Bay to assess possible relationships between estuarine sediment contaminant levels and landscape metrics in the watershed. The area of developed land located in the watershed within 10 km of the sediment sampling station was a major contributing factor in the sediment concentrations of both metals and organic contaminants. Limitations to this pilot study included the absence of a wetlands class for the land cover data set used for development of the landscape metrics (USEPA, 1994), the limited number of watersheds (25), and the unknown transferability of results beyond the relatively low-relief, coastal plain watersheds investigated. The purposes of the study reported in this paper were to (i) address these limitations so that the resulting conclusions for estuarine sediment contamination would be applicable to other geographic areas and (ii) develop approaches that would have predictive capabilities, for use with other data sets.


   METHODS
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
RESULTS
 DISCUSSION
 SUMMARY
 REFERENCES
 
Study Area
Small estuarine systems (<260 km2 in estuarine surface area) in the Virginian Biogeographic Province (Cape Cod to mouth of Chesapeake Bay on the east coast of the USA; Holland, 1990) and their associated watersheds were used for the study (Fig. 1) . This province covers the mid-Atlantic and southern New England regions of the USA. Small estuarine systems were chosen because their physical characteristics (small in areal extent, semi-enclosed, relatively shallow, in close proximity to land-based stressors) enhance the possibility of coupling sediment condition to surrounding watershed conditions. In analysis of benthic community data collected in the summers of 1990–1993 across the estuarine waters of the Virginian Biogeographic Province, the primary stressor influencing benthic community conditions in small estuarine systems was found to be sediment contamination (Paul et al., 1999). The number of estuarine systems was 75 out of a possible 144 systems because of available quality-assured sediment contaminant data for both metals and organics and due to the inability to accurately delineate watersheds on some of the relatively flat coastal areas (see below under Watershed Delineation). The 75 watersheds are distributed throughout the Province (Fig. 1). These included the 25 systems in Chesapeake Bay used in Comeleo et al. (1996). Thirty-two were in Chesapeake Bay, with the rest distributed northward to Cape Cod.



View larger version (37K):
[in this window]
[in a new window]
  		Fig. 1. Distribution of estuarine watersheds in the mid-Atlantic and southern New England regions used in this study.

 
Data Sources
Sediment contaminant data were collected in the small estuarine systems of the Virginian Province during the summers of 1990–1993. One hundred forty-four small estuarine systems (<260 km2 in estuarine surface area) were selected for sampling, with probability survey designs used to sample all of the estuarine waters in the province (Holland, 1990; Strobel et al., 1995). For sample collection within each small system, a single sampling site was randomly selected. The top 2 cm of sediments were composited from multiple samples collected from each site with a grab sampler (440 cm2 surface area) and subsequently analyzed for a suite of organic and inorganic constituents. Sediment samples were analyzed for 24 polycyclic aromatic hydrocarbons (PAHs), 18 polychlorinated biphenyl (PCB) congeners, DDTs, 11 chlorinated pesticides, tributyl tins, and 15 metals (Strobel et al., 1995, 1999). Of these, 9 metals and 13 organic compounds shown to cause measurable effects on benthic organisms (Long et al., 1995) were selected for further consideration as dependent variables (Table 1) . In addition, total PAHs, the combination of the 24 measured individual PAHs (NOAA, 1991), was used in the analyses. Nondetectable values were assigned zeros.


View this table:
[in this window]
[in a new window]
  		Table 1. Sediment contaminants used in analyses as possible dependent variables.

 
Two landscape metric categories were used for analysis: land cover pattern and point source pollution input. Land cover pattern, for this study, refers to the composition of land cover types within a watershed. We assumed that land cover pattern is a suitable surrogate for nonpoint-source pollution information, as it has been shown that the severity of nonpoint-source pollution is closely related to land cover (Kim and Ventura, 1993). Land cover data were from the U.S. Geological Survey Land Use Data Analysis (LUDA) data set (Fegeas et al., 1983), which was interpreted from 1970s aerial photos. This data set was the only available consistent land cover classification for the entire Virginian Province estuarine watershed at the initiation of this study. The 37 LUDA land cover classes were aggregated to be compatible with the land cover classes used in Comeleo et al. (1996): urban (comparable to developed land), forested, agricultural, forested wetlands, and nonforested wetlands (comparable with herbaceous land).

Point source pollution data were obtained from the NOAA National Coastal Pollutant Discharge Inventory for the estuarine drainage areas of the Virginian Province (Pacheco, 1993). The location of active major point source discharge sites as well as 1991 estimates of annual pollutant input (kg/yr) and annual outflow (106 m3/h) at each discharge site were used for analysis. Estimates were made from effluent monitoring data available in monthly Discharge Monitoring Reports completed as part of each facility's National Pollutant Discharge Elimination System Compliance Monitoring Program (Pacheco, 1993). Effluent data were available for seven (As, Cd, Cr, Cu, Pb, Hg, and Zn) of the nine metals in Table 1. These seven metals are present in the environment from natural and anthropogenic sources and are among the most frequently measured pollutants at monitored discharge sites (Pacheco, 1993). Effluent data for organics were sparse and were not considered for further analysis.

Response of estuarine systems to the watershed stressors is expected to be mitigated by the complex dynamics occurring within the individual systems. Three ways to account for the dynamics were explored: (i) including grain size and total organic C content of bottom sediments as possible variables in the statistical models; (ii) grouping systems by whether or not they were glaciated; and (iii) grouping systems on the basis of estuarine hydrology.

Sediment grain size (SICL) and total organic carbon (TOC) were separately included as possible independent variables to account for sediment charactristics in the statistical models. Grain size, expressed as percent silt–clay content, and TOC (percent of sediment) were analyzed on samples collected with those for contaminant analyses. Silt–clay content and TOC are indicative of the sedimentary environment of the system and result from the interaction of loadings, circulation, and geomorphology (Nichols and Biggs, 1985). Total organic C is also indicative of biological processes (eutrophication) in the system. Contaminants tend to partition preferentially on fine grain particles, particularly onto the organic fraction (Landrum and Robbins 1990; Bierman, 1994).

Estuarine sediments north of the southern extent of the Wisconsin glaciation are composed of stratified proglacial deposits of outwash plains (Golet et al., 1993). South of the glaciated area, sediments are the result of continental surface erosion, transport, and deposition. The transport and fate of particle-bound contaminants can be influenced by the origin of the sediments (Bonner et al., 1994). A categorical variable (GLACIAL) was used to separate glacial (n = 30) from nonglacial (n = 46) estuarine systems.

Hydrology was represented by the ratio of estimated freshwater inflow to estimated tidal volume (Kennish, 1986). U.S. Geological Survey stream flow records were used to quantify inflows, with ungaged flow estimated using watershed area factors. We defined hydrology classes as follows for our analyses. Inflow was greater than tidal volume in Class 1 estuaries, approximately equal to tidal volume for Class 2 estuaries, and less than tidal volume for Class 3 estuaries. Class 4 estuaries were shallow, heavily influenced by surface processes, and included barrier beach/lagoonal systems. The estuarine circulation dynamics control the transport and ultimate dilution of material entering the systems (Fischer et al., 1979). A categorical variable (HYDROLOGY) was used to separate Class 1/Class 2 estuarine systems (n = 51) from Class 3/Class 4 systems (n = 24).

Watershed Delineation
A watershed approach (Comeleo et al., 1996) was used to analyze the landscape surrounding the sampling site in each estuary. Watershed delineation was attempted for each small estuarine system using one-degree (1:250000 scale) U.S. Geological Survey (USGS) Digital Elevation Model (DEM) data processed with the GRID module of the Environmental System Research Institute (ESRI) ARC/INFO GIS software package. The accuracy of computer delineation for many of the low relief Coastal Plain watersheds was questionable because the horizontal and vertical resolution of the elevation data (93 and 1 m, respectively) were inadequate to discern low topographic relief. The computer-derived watersheds were reviewed and enhanced through on-screen interpretation using DEM-derived shaded relief maps and hydrography (1:100000 USGS Digital Line Graph) as backdrops. In areas of low topographic relief where the accuracy of computer-interpreted watershed boundaries was questionable, divisions between drainage systems were manually located by interpolation. The delineated watersheds were combined with available sediment contaminant data. This resulted in 75 nonnested estuarine/watershed systems for analysis.

Landscape Analysis
The vector module of ARC/INFO GIS software was used to determine the area of each land cover class found within each watershed and the total annual outflow and metals input from all point discharge sites located within each watershed. A total of 14 landscape metrics were calculated: area of watershed (AREA), percent area of urban land (PURB), percent area of agricultural land (PAG), percent area of forested land (PFOR), percent area of forested wetlands (PFWET), percent area of nonforested wetlands (PNFWET), and the annual volume (FLOW) and metals input (LAS, LCD, LCR, LCU, LPB, LHG, LZN) from point discharge sites.

Statistical Analysis
Principal component analysis (PCA) was used to reduce the number of variables for statistical modeling. Principal component analysis was performed on the correlation matrix of the sediment contaminant and point source input variables. Principal component analysis was performed separately on the group of nine metals and the group of 13 organic compounds in the sediments. Principal component analysis also was performed on the seven metals input rates.

Pairwise correlations were calculated as measures of association between each of the independent variables and the dependent variables used to develop the statistical models. Multiple linear regression with stepwise selection (MLRSS, a.k.a. stepwise MLR) was conducted to develop statistical models between the dependent variables (the first principal component for sediment metals and organics, and total PAHs) and independent variables (SICL, TOC, GLACIAL, HYDROLOGY, and landscape metrics: AREA, PURB, PAG, PFOR, PFWET, PNFWET, FLOW, and the first principal component for annual metals input). Refer to Table 2 for summary of the independent variables. GLACIAL was a categorical variable (0/1) to separate glacial from nonglacial systems, respectively. HYDROLOGY was a categorical variable (0/1) to separate Class 1 and 2 from Class 3 and 4 systems, respectively. A significance level of p < 0.15 was used for independent variables to enter the model. Two basic forms of statistical models were investigated with MLRSS:

[1]

[2]
where y is the dependent variable, xi are independent variables, {alpha} and ßi are regression coefficients, offset is a constant, and {epsilon} is the residual error. An offset was used in Eq. [2] with natural log transform of the dependent variable to accommodate zeros (total PAHs) and negative values (principal components).


View this table:
[in this window]
[in a new window]
  		Table 2. Combinations of independent variables considered in regression analysis. Abbreviations for variables in parentheses.

 
Potential models relating landscape metrics to sediment contamination were investigated with MLRSS for the following combinations: three dependent variables (PC1-metals, PC1-organics, total PAHs), two functional forms [linear and exponential (a.k.a. log transform of dependent variable)], and seven possible combinations of independent variables. The independent variable combinations are indicated in Table 2. Combination A treats the two categorical variables as potential additive variables in the regression (i.e., additional xi in Eq. [2] with values of 0 or 1), resulting in a single-equation model. For this model, the coefficients for each of the independent variables are not a function of the categorical variables. Combinations B and C exclude variables for sediment origin (GLACIAL) and estuarine hydrology (HYDROLOGY) from the regression, and result in single-equation models. For combinations D through G, two-equation models (or nested regressions) were developed, with a different equation for each value of the categorical variable GLACIAL (D and F) and HYDROLOGY (E and G). These combinations treat GLACIAL and HYDROLOGY as grouping variables (not as additional xi in Eq. [2]), that is, the coefficients for the independent variables are functions of the categorical variable [for example, {alpha} = function(GLACIAL) and ßi = function(GLACIAL)].

Each fitted model (3 x 2 x 7 possible combinations) was tested for validity of regression assumptions by examining the distribution of residuals (Q-Q normal probability plots and Shapiro-Wilk test for normality). Variance inflation factors (VIFs) were calculated to identify correlations among independent variables in the fitted models. Variance inflation factor is a measure of the amount that the variance in a regression coefficient is inflated due to multicollinearity of variables. A VIF greater than 10 for a variable indicates multicollinearity (Philippi, 1993), and an ambiguity in the interpretation of regression coefficients. For each dependent variable and functional form, the best overall model was determined by the coefficient of determination (R2) of the fitted models. The best overall model was compared with the other fitted models for the same dependent variable and functional form by testing for significant difference (ANOVA with p < 0.05). Statistical analyses were done using SAS (SAS Inst., 1989) and S-Plus 2000 (MathSoft, 1999) software.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
 RESULTS
DISCUSSION
 SUMMARY
 REFERENCES
 
Sediment Contaminants
The characteristics of the sediment contaminant data for the 75 systems are presented in Table 3 . The sediment metals were less variable across the systems compared with the sediment organics based on the coefficient of variation (CV). Two organic compounds (2-methylnaphthalene and naphthalene) had lower variability than the other compounds. These two PAHs have the lowest molecular weights. For total PAHs and the individual compounds used in the analysis, very strong correlations exist except for three compounds (fluorene, 2-methylnaphthalene, and naphthalene), two of which had lower variability across systems (CV in Table 3). Total PAHs was strongly correlated with the first principal component for the 13 organic compounds but not correlated with the second principal component. So PC1-ORG can be interpreted as a good measure for total PAHs.


View this table:
[in this window]
[in a new window]
  		Table 3. Summary statistics for sediment contaminants for systems (n = 75) used in analyses.

 
Landscape Characteristics
The characteristics of the 75 watersheds used in the analysis varied widely (Table 4). Watersheds ranged in size from 6 to 70800 km2. The percentage of urban land in watersheds ranged from 2 to 99% and was the most abundant land cover in 17 of the 75 watersheds. Forested land was the most abundant land cover in 43 and agricultural land in 15 watersheds. Point source outflow and metals input also varied widely among watersheds. The characterization of systems for glacial–nonglacial systems and Class 1 and 2/Class 3 and 4 estuaries are also presented in Table 4. Comparing mean values, the Class 3 and 4 system watersheds were much smaller in area, were more urbanized, had less agricultural lands, and had lower point source input. These watersheds received much less metals loadings, with chromium input being the exception. The glaciated watersheds were smaller, more urbanized, had less agricultural lands, and had higher point source flows and metals input.


View this table:
[in this window]
[in a new window]
  		Table 4. Summary statistics for landscape metrics for grouping of systems (Virginian Province, glaciated/nonglaciated, and hydrology class) used in analyses.{dagger}

 
Statistical Results
The first principal component for each of the sediment metals and organics showed strong correlations between principal component and sediment concentration for nearly all contaminants in the 75 systems (Table 5). Arsenic, which was weakly correlated with PC1, was the only metal strongly correlated with PC2. The three organic compounds weakly correlated with total PAHs (fluorene, 2-methylnaphthalene, and naphthalene) were weakly correlated with PC1, but the two with low variability (low CV) were strongly correlated with PC2 (negatively). We interpret PC1 to reflect the overall amount of metals or organic contaminants: the larger PC1, the greater the magnitude of sediment contamination. The first principal components for metals (PC1-MET) and organics (PC1-ORG) were used to represent the level of sediment contamination in the regression analyses in addition to total PAHs, with 63 and 83% of standardized variance explained, respectively.


View this table:
[in this window]
[in a new window]
  		Table 5. Loadings (for absolute values > 0.1) and correlation coefficients (r) for the first (PC1) and second (PC2) principal components for sediment metals, sediment organic contaminants, and metals input from point sources with individual contaminant concentrations and metal loadings.

 
The result of PCA with metals input variables indicated strong correlations for all loading variables in PC1 except Zn, which was the only loading in PC2 (Table 5). The PC1-INPUT explained 68% of the standardized variance in metals input. The PC1-INPUT was used to represent the metals input in the regression analyses.

The pairwise correlations between PC1-MET, PC1-ORG, and total PAHs and the independent variables can be summarized as follows. Sediment metals (PC1-MET) significantly correlated with sediment characteristics (SICL and TOC), nonpoint sources (positive for PURB and negative for PFOR and PNFWET), and point sources (FLOW and PC1-INPUT). Sediment organics (PC1-ORG) was correlated with sediment characteristics (primarily TOC) and nonpoint sources (positive for PURB and negative for PFOR). Total PAH correlations were similar to organics except PAG was substituted for PFOR. In almost all cases, urban land exhibited the strongest correlation with sediment contamination.

The multiple linear regression with stepwise selection results for statistical models with Eq. [2] were consistently found to have normally distributed residuals. The residuals for all models with Eq. [1] were nonnormal. These results are summarized in Table 6. Except for one case, the VIFs for the fitted models were much less than 10, indicating little concern for multicollinearity of variables. The exception exhibited strong correlation (|r| > 0.9) among PURB, PAG, and PFOR for nonglacial systems. In addition to the R2 for each model and dependent variable, those models shown to be not statistically different from that with largest R2 (ANOVA with p <= 0.05) are indicated. A summary of these latter models is shown in Table 7 with coefficients that entered the regression with p <= 0.15.


View this table:
[in this window]
[in a new window]
  		Table 6. Coefficients of determinations (R2) for models with natural log transform of dependent variable (Eq. [2]) developed from multiple linear regression with stepwise selection on first principal component for metals (PC1-MET) and organics (PC1-ORG) and total PAHs as dependent variable. Results shown only for those models with normally distributed residuals.

 

View this table:
[in this window]
[in a new window]
  		Table 7. Coefficients for statistical models with natural log transform of dependent variable (Eq. [2]) developed using multiple linear regression with stepwise selection with first principal component for metals (PC1-MET) and organics (PC1-ORG) and total PAHs as dependent variables (models not significantly different, ANOVA at p <= 0.05).

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
 RESULTS
 DISCUSSION
SUMMARY
 REFERENCES
 
The results of this study have shown a quantitative linkage between land use and sediment contamination for small estuarine systems. The sediment contaminant levels in small estuarine systems increase with increasing percent of the watershed in urban land. This is consistent with the conclusion of a recent National Water Quality Inventory (USEPA, 1995) that runoff from urban area is the prime source of water quality impairments in estuaries. Runoff from urbanized areas is a major source of nonpoint pollution (Wang et al., 1997). The amount of impervious surface resulting from urbanization influences the condition of receiving waters (Boward et al., 1999). Dauer et al. (2000) observed strong correlations between sediment contamination and percent urban area in the watersheds of 10 major tributaries of Chesapeake Bay.

Sediment contaminant levels decrease with increasing percent of the watershed in nonforested wetlands. These wetlands are generally located in the lower parts of the watershed, closer to the shoreline, and function as a buffer between the urban areas and receiving waters. Herbaceous land within 10 km of the sampling site was a surrogate for nonforested wetlands in Comeleo et al. (1996). Forested wetlands are generally further up in the watershed and are less significant in explaining the variability in sediment contaminant levels. The ability to delineate between forested wetlands and forests is a possible limitation with the land cover data set. However, forests only entered one model (organics, combination F), and this was not a best model.

Sediment metals concentrations increase also with increasing inputs from point source discharges in the watershed. Data were not available for point source input of organic contaminants. It is expected that point source inputs of organics would be important. Not including these inputs in the regressions may partially explain the lower overall R2 compared with the metals model. Other reasons include the high variability in the organics values and the large number of concentrations below detectable levels.

Because of the different sediment characteristics and dynamics of estuarine systems across the mid-Atlantic and southern New England regions, watershed–estuarine processes were shown to be important in describing the quantitative linkages. These processes mitigate the effect of the watershed stressors on sediment contaminant concentrations. In every case the best models included at least one of the following: sediment grain size, TOC concentration, and estuarine hydrology. Results indicated that it was more important (based on R2) to include estuarine hydrology as a grouping variable (two-equation model, i.e., different coefficients for the independent variables based on the grouping) than as an additional additive independent variable (single-equation model). This implies that the estuarine dynamics quantitatively changes the linkage between the watershed stressors and sediment contaminant levels rather than just provide an additive factor over and above that provided by the stressors.

The importance in accounting for sediment characteristics and estuarine dynamics for reasonable predictions of sediment contaminant levels provides for a possible approach to answering the question posed in the title of the paper by Rose (2000), "Why are quantitative relationships between environmental quality and fish populations so elusive?" First of all, the intervening processes (such as surface water dynamics) need to be taken into account. Secondly, the distinction needs to be explicitly made between local factors (such as water and sediment characteristics) and remote factors (for example, land use and point sources) in exploring the development of quantitative relationships.

The overall best model (largest R2) was for sediment metals. This model included grain size and grouping by estuarine hydrology. In contrast, the best models for organics and total PAHs included TOC instead of grain size. This suggests that if linkages were to be explored with benthic community conditions, then it may be necessary to normalize the organic contaminants by the organic C content of the sediment. This would account for the partitioning to the organic C and the effective exposure to the organisms.

The land cover data set used for the analysis was based on 1973 aerial photos. Two limitations to use of this data set are (i) areas of rapid changes since 1973 and (ii) quality of wetlands delineation. Some coastal areas, particularly in New Jersey and Delaware, have experienced major land use changes between the acquisition of the aerial photos and the collection of the sediment data. There is no statistical procedure to deal with this, and it remains as a possible limitation to the study. Since the analysis in this study has been completed, a more recent land cover data set for the entire region has become available (Vogelmann et al., 1998). The statistical models developed in this study should be evaluated against the more recent land cover data set, and is the subject of ongoing work. The quality of the wetlands delineation was uneven across states because of the different photo interpreters. Percent area of nonforested wetlands was an important component of the models. However, the delineation of the nonforested wetlands in the aerial photography could be expected to be less variable than for forested wetlands. This limitation can also be addressed by the ongoing work with the new land cover data set.

Classical parametric methods (i.e., multiple linear regression with stepwise selection) were used in this study to develop the statistical models. A limitation in the application of the classical methods for developing the statistical models is the possible transformation of variables that can be used. One approach for expanding beyond the functional forms used with the classical parametric methods, and which are the subject of further research, are deterministic–statistical models. These models involve specification of a general mathematical form relating independent to dependent variables (typically nonlinear), derived from separate studies or available in the literature. The parameters in the model are solved by methods such as nonlinear least squares. An example of this approach is the work by Smith et al. (1997) and Preston and Brakehill (1999) in development of a set of spatially referenced regression models for the evaluation of nutrient loading in watersheds. The functional form they used was a sum of exponentials, with nonlinear regression analysis used to solve for the parameters. Results from the classical methods do provide direction for those processes that should be explored with these more detailed models.

The log transformation of the dependent variables provides statistical models that can be used for applications that involve predictions with new sets of independent variables. As an example, the models could be applied to the independent data set collected in small estuarine systems in 1997–1998 by the mid-Atlantic Integrated Assessment estuarine program (Strobel et al., 2000). These data would be used in a validation mode to test the statistical models developed in this study against an independent data set. Another example would be to use the models in a forecasting mode for evaluating possible management actions, under the assumption that the relationships would be valid in the future. Scenarios for future distributions of land cover for different land management options could be used to predict what sediment contamination might result from the management actions.


   SUMMARY
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
 RESULTS
 DISCUSSION
 SUMMARY
REFERENCES
 
The question posed at the beginning of this paper was how does one determine "which, if any, of these landscape metrics are appropriate for predicting the impact of nonpoint pollution on estuarine receiving waters?" We chose to address this question by developing statistical models that relate the landscape metrics to estuarine sediment contamination levels. This approach was similar to that introduced in Comeleo et al. (1996). We addressed the identified limitations in that study by (i) using a consistent land cover classification data set for the entire geographic region that included a wetlands class; (ii) using triple the number of independent estuarine systems for the analysis (n = 75); and (iii) extending the results from the relatively, low-relief coastal plain watersheds to those across a broad, diverse geographic region.

Results suggest that variation in sediment contamination levels across small estuarine systems in the mid-Atlantic and southern New England regions was related to the surrounding land cover composition and point source discharges of pollutants. The influence of these variables was mitigated by sediment characteristics and estuarine dynamics. The landscape metrics important for explaining the variation in sediment metals levels (R2 = 0.72) were the percent area of nonforested wetlands (negative contribution), percent area of urban land, and point source effluent volume and metals input (positive contributions). The metrics important for sediment organics levels (R2 = 0.5) and total PAHs (R2 = 0.46) were percent area of urban land (positive contribution) and percent area of nonforested wetlands (negative contribution). Incorporation of estuarine hydrology and sediment characteristics into the statistical models was necessary to generate reasonable R2. The result that percent area of urban land is the major correlate with sediment contamination is consistent with the conclusion of a recent National Water Quality Inventory (USEPA, 1995) that runoff from urban area is the prime source of water quality impairments in estuaries. The overall results from this study are indicative of how statistical models can be developed relating landscape metrics to estuarine sediment contamination.


   ACKNOWLEDGMENTS
 
Thanks to Cathleen Wigand, Rick McKinney, Anne Kuhn, Bruce Jones, Maliha Nash, and three anonymous reviewers for their reviews and comments on this paper. Special thanks to Carol Baker for assistance in the watershed delineation, Wayne Munns for suggesting glaciation and estuarine hydrology as categorical variables, Peter August for insights into landscapes, and Jim Heltshe for discussions on statistical issues. The USEPA has funded, wholly or in part, the research described in this paper. This paper has not been subjected to Agency review. Therefore, it does not necessarily reflect the views of the Agency. Mention of trade names or commercial products does not constitute endorsement or recommendation for use. Authors R.L. Comeleo and J. Copeland were supported through USEPA contract no. 68-W5-0065, Lawrence Rossner Delivery Order Project Officer. This is contribution no. AED-00-085 of the Atlantic Ecology Division, National Health and Environmental Effects Research Laboratory.


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
 RESULTS
 DISCUSSION
 SUMMARY
 REFERENCES
 




This article has been cited by other articles:


Home page
 J. Environ. Qual.Home page
J. W. Hollister, P. V. August, J. F. Paul, and H. A. Walker
Predicting Estuarine Sediment Metal Concentrations and Inferred Ecological Conditions: An Information Theoretic Approach
J. Environ. Qual., January 4, 2008; 37(1): 234 - 244.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via ISI Web of Science (14)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Paul, J. F.
Right arrow Articles by Copeland, J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Paul, J. F.
Right arrow Articles by Copeland, J.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Paul, J. F.
Right arrow Articles by Copeland, J.
Related Collections
Right arrow Spatial Distribution
Right arrow Water Quality
Right arrow Watershed and Landscape Processes
Right arrow Water Pollution

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Agronomy Journal Crop Science
Journal of Natural Resources
and Life Sciences Education		 Vadose Zone Journal
Soil Science Society of America Journal Journal of Plant Registrations The Plant Genome