Spatial Analysis of Cesium in Sediments of Watts Bar Reservoir
Estimates of the spatial distribution of sediment type, and the resulting distribution of 137Cs in the contaminated sediments of Watts Bar Reservoir, was developed as a function of reservoir depth, local topography of the reservoir bottom, distance from the reservoir shore and river channel, and distance from the confluence of the Clinch and Tennessee River. The quantitative methods developed for this analysis are based on over 1400 samples of sediment type, previous analysis of 137Cs concentrations (Olsen et al. 1992), and new data sets that describe the bathymetry of Watts Bar. The quantitative methods used to characterize the distribution of 137Cs also provide a means to evaluate prediction errors and the usefulness of additional data.
A model for the bathymetry of Watts Bar Reservoir was developed from four 7.5 minute USGS quadrangles (Rockwood, Bacon Gap, Spring City and Ten Mile) obtained from the Tennessee Valley Authority (TVA) mapping services. These maps were printed in 1940, before the reservoir was impounded, and show contour lines surveyed at 10 ft (3.05 m) intervals within the reservoir boundary (the 741 ft [225.9 m] elevation). Because the estimated siltation rate is 0.63 cm/yr (Brenkert et al. 1992), these pre-impoundment maps provide a resonably accurate estimate of reservoir bathymetry. The maps were scanned at high resolution, adjacent maps were edge-matched, and a GIS vector map of bottom contours was created in ARC-INFO format. All vectors below the 741 ft (225.9 m) elevation were identified and converted into a GRASS (1993) site file. Elevations between these known data locations were estimated by interpolating with a set of spline programs specifically developed for GRASS (Mitas and Mitasova 1988, Mitasova and Mitas 1993). The interpolation resulted in a raster map with a resolution of 5 m (25 m2) per cell. The areal extent of the reservoir basin from the confluence of the Tennessee and Clinch Rivers to the Watts Bar Dam is 10.8 x 107 m2 (a total of 43,200,000 cells).
The original topographic maps (and therefore the scanned contour vectors) did not have elevation data for the area within the original pre-impoundment boundaries of the Clinch River, since these areas were submerged even before construction of the dam. The missing elevation data within the old river channel were estimated from the most recent survey data from 41 TVA transects across the Clinch River and Watts Bar Reservoir (McCain 1957). The 1991 data for the 11 transects within the river channel of Watts Bar Reservoir (Brenkert et al. 1992) were used to obtain channel elevation profiles. UTM coordinates were estimated for permanent monuments identifying the end points of each transect, and elevation data were associated with each site along each tansect. A separate spline was performed for the original river channel bottom, resulting in estimates of depths for every cell in the channel. The result was a single, smooth, downhill flowing plane whose depths corresponded to TVA's survey data. The estimated elevation values for cells within the river channel were then added to the Watts Bar digital elevation map to create a complete model of the bathymetry for the reservoir.
137Cs is particle-reactive, and is strongly adsorbed to suspended solids within the water column (Olsen et al. 1992). Adsorbed to particles, the contaminant may precipitate and accumulate in the bottom sediments. Although the removal, burial, and remobilization of sediments is a complex process, sampling has shown that the concentration of 137Cs is highly correlated with the particle size of the bottom sediments (Olsen et al. 1992). Therefore, a complete and accurate description of the spatial distribution of sediments within the Watts Bar Reservoir is critical for determining the spatial distribution and total amount of 137Cs remaining in the reservoir.
Additional sediment grab samples were taken in the summer of 1993 to obtain a more systematic characterization of the spatial distribution of sediment types (i.e., finer versus coarser sediments) within the reservoir. The GRASS data set for reservoir bathymetry, as well as coarse maps of sediment types based on expert opinion (Olsen et al. 1992), were used to locate 18 sampling transects within the reservoir. 346 samples were collected from these transects reflecting a range of sediment types, depths, and distances from shore and river channel. Field crews spaced samples along each transect at 50 m intervals. The coordinates of each sample were determined by a geographic positioning system (GPS), the depth to nearest foot was measured by a depth finder, and a grab sample of the sediment was collected. The transects are visible in the bathymetry map as lines of plusses that traverse the river. Each sample was examined in the field and again in the laboratory, and was classified into 1 of 5 broad sediment types by a single technician who was involved in the original Clinch River sediment study (Olsen et al. 1992).
The sediment classification scheme (Olsen et al. 1992) is based on color, texture and cohesiveness of material and represents an ordinal scale from soft and fine to very coarse and hard material. The categories are: (1) soft mud, (2) cohesive mud, (3) sandy mud, (4) sand and gravel or (5) submerged soil. Comparison of the field assignment of sediment type with a second, independent classification in the lab showed inconsistencies in distinguishing between cohesive and soft mud categories, perhaps because of disruption of sample integrity during transport. Because these two categories also have similar concentrations of 137Cs, the final classification system combined categories 1 and 2 into a single ``soft mud'' category.
Local variation in the topography of the reservoir bottom results in ridges from which sediments may be scoured and depressions within which sediments are more likely to be deposited. A comparison of sediment type with local variations in topography indicated that coarser and harder sediments are often located on local ridges, while softer and finer sediments are located in local depressions. One of us (WWH) developed an adaptive spatial filter procedure for identifying these regions from the bathymetry data set. Four transects of 105 m (21 cells in each transect, 81 cells in all four transects) were centered over each cell in the four cardinal and subcardinal compass directions.
The average deviation in elevation of all cells from the central cell was calculated and weighted by 1/d2, where d is the distance of the cell from the outer cell. The natural logarithm (with sign preserved) of the weighted average difference of all cells on 4 transects was used as a relative index, e, of the local topography for the center cell. Large negative values indicate local depressions while positive values indicate local ridges. The process was repeated for all cells, producing a map of spatially distributed values of e.
Previous analyses have shown that the concentration of 137Cs varies significantly as a function of distance to shore and distance to river channel (Olsen et al. 1992). Buffer distance maps were generated using GRASS which contained the distance of each cell to the center channel and to the reservoir boundary (741 ft elevation line). Although the resolution of each cell is 5 m, the distances of each cell from the reservoir boundary and river channel was measured to the nearest 10 m.
Over 1400 sediment samples for which sediment type had been observed, shown as plusses on the bathymetry map, were located with respect to the synthetic map layers described above, and each observation was associated with sediment type, measured depth, reservoir bottom elevation, UTM coordinate (northing and easting), distance to river channel, distance to shore, and e. Logistic regressions were used to build a statistical model characterizing the probability of observing soft and cohesive mud, sandy mud, and sand and gravel. Submerged soil was assumed to occur where lake bottom elevations were >=735 foot (224 m), the winter drawdown water level controlled by TVA. The logistic regressions showed that sediment type is most strongly related to the distance of the site from shore, and the easting/northing UTM locations (i.e., the distance from the confluence of the Clinch and Tennessee Rivers). The latter represents a general trend in concentrations, with higher values nearer the confluence of the Clinch and Tennessee Rivers and lower concentrations near the Watts Bar Dam.
A sediment map was produced from these results by: (1) obtaining the environmental characteristics for each cell from the synthetic GIS maps described above, (2) estimating the probability of occurrence for fine-mud, sandy-mud and sand-gravel from the logistic regressions, and (3) randomly selecting the sediment type from a normalized cumulative frequency distribution of probabilities described by the regression equations.
The sediment core data taken during the Olsen et al. (1992) study for lower Watts Bar were analyzed with regression analysis to produce statistical models describing variation in 137Cs inventory based on the following environmental characteristics: sediment type, lake bottom elevations (water depth), e, distance from old river channel, distance from shore, and UTM coordinates (northing and easting). The results showed a significant relationship between 137Cs inventory and both lake bottom elevation and UTM easting and northing. 137Cs inventory in submerged soil was influenced by UTM northing. Because there were relatively few actual core samples comprised of Sandy-mud and sand-gravel, no significant regressions with the environmental variables could be produced for these sediment categories. Therefore, 137Cs inventory for these two sediment types was assumed to be described by the mean values for the respective sediment types and to be unaffected by other environmental characteristics. Using these regression relationships, 137Cs inventories were estimated for each cell, and a map of 137Cs inventory in Lower Watts Bar Reservoir was produced. Units for this map are pCi/cm2. 137Cs inventory within each river reach section was also calculated.
The collection of additional sediment samples and the analysis of sediment types showed that there was more mud present in the reservoir than originally estimated (Olsen et al. 1992). Because 137Cs selectively adsorbs to the fine particles that constitute the ``mud'' sediment type, the presence of more mud within the lake makes it possible to have a greater inventory of 137Cs than originally estimated. Nevertheless, the gradient in concentration from the confluence of the two rivers to the dam (see 137Cs concentrations by river reach) results in a total inventory of 153.8 Ci. This total inventory is approximately half of the 304 Ci originally estimated in the Olsen et al. (1992) report.
Although the stochastic elements of the simulation produce slight differences between individual realizations of the map, meso- and macro-scale trends and ``contamination features'' remain unchanged. Four areas, shown within black boxes on the cesium map were consistently predicted by the model to have high concentrations of 137Cs. These areas of high concentration are associated with deeper parts of the river channel that also contain local depressions which tend to collect sediments. Click on the black boxes within the cesium map to see enlarged 3D perspective views of these predicted ``hot spots'', or click here to see hotspot one, two, three, or four. The color scale is the same on these perspective views as on the inventory map, and all maps are in units of pCi/cm2. It is from these locations within the reservoir that additional sampling would be most valuable to: (1) verify the methods presented here and (2) provide upper bounds on possible 137Cs concentrations.
Although the total amounts of 137Cs are less than originally projected, the general conclusions of the Olsen et al. (1992) are verified by this analysis. That is: (1) the highest 137Cs concentrations are within the channel of the old river bed, (2) there is little or no 137Cs present where river scouring occurs, and (3) sediments in near-shore coves generally have marginal amounts of 137Cs.
GRASS 4.1 Reference Manual. 1993. U. S. Army Corps of Engineers, Construction Engineering Laboratories, Champaign, Illinois, p. 422-425.
Mitas, and L., and H. Mitasova. 1988. General variational approach to the interpolation problem. Comp. Math. Appl. 16: 983-992.
Mitasova, H., and L. Mitas. 1993. Interpolation by regularized spline with tension: I. Theory and Implementation. Mathematical Geology, 25: 641-655.
Olsen, C. R., I. L. Larson, P. D. Lowry, C. R. Moriones, C. J. Ford, K. C. Dearstone, R. R. Turner, B. L. Kimmel, and C. C. Brandt. 1992. Transport and accumulation of cesium-137 and mercury in the clinch River and Watts Bar Reservoir system. Oak Ridge National Laboratory Technical Publication ORNL/ER-7, Oak Ridge National Laboratory, Oak Ridge, TN 37831, USA.
Brenkert, A. L., C. C. Brandt, K. A. Rose, R. B. Cook, and M. A. Wood. 1992. A comparison of two methods for estimating spatial patterns of sediment accumulation in the Clinch River-Watts Bar Reservoir system. pp 78-81 in Fifth Tenneessee Water Resources Symposium, F. Quinones and K. L. Hoadley (eds.), American Water Resources Association, Nashville, TN.
McCain, E. H. 1957. Measurement of sedimentation in TVA reservoirs. American Society of Civil Engineering, J. of Hydrology Div., vol. 83.
For additional information contact:
(865) 574-1902
