Jeffrey L. Myers
FIELDS Research Associate
United States Environmental Protection Agency

8-27-03

Poke Locations
Sediment Thickness Estimates

The Fields team has used the Sediment Thickness Estimate for Clear Creek to conclude that there is approximately 69,292 cubic yards of sediment in the portion of clear creek that was sampled by the FIELDS team between August 11, 2003 and August 15, 2003. There was 11,701 cubic yards west of old 37 and 57,591 cubic yards east of old 37. The volume was determined by placing 674 sediment poke's GPS locations and their depth into a GIS. The GIS was then used to do various interpolations on the sediment pokes in order to determine an overall volume.

The interpolations done were natural neighbor, IDW power of 2 and neighbor of 2, IDW power of 6 and neighbor of 6, and IDW power of 2 and neighbor of 8. The sediment volume values of each can be seen in Table I. We chose to use the interpolated volume determined by the interpolation using IDW power of 2 and neighbor of 8 because it yielded a reasonably low Estimation Error and it had the lowest cross validation. The cross validation results can be seen in figure 1.

The volume was determined by multiplying the floating-point grid by 100 then converting it to an integer grid. The integer grid value was then divided by 100 so that 2 decimal places were preserved. The new value is equal to the depth in inches. The inches for each interpolated grid cell were converted into meters by dividing them by 39.37. The depth in meters multiplied by the area of each cell determined the volume for each cell. Finally the sum of the volume for each cell was used to determine the overall volume for Clear Creek. A picture showing the procedure can be seen in figure 2.

The IDW interpolation is a weighted, moving average technique. IDW assumes that each input point has a local influence that diminishes with distance. Hence, the interpolated points (the new surface) will be more influenced by nearby points than more distant points. The power parameter controls the influence of neighboring data points on the estimated data value. A higher power results in less influence from distant neighboring data points. Conversely, a lower power results in more influence from distant neighboring data points. The neighbors option determines the number of neighboring data points used to estimate the value at an unsampled location.

Cross validation is a procedure that compares estimated and true values by removing original data one value at a time and re-estimating that value from the remaining neighboring data (Isaaks and Srivastava, 1989). The difference between the estimated and true value is called the cross validated residual. The Cross Validation procedure calculates a root mean square error (RMSE) of these residual values. Although empirically derived, the techniques used here are influenced by several factors and may not always yield the best parameters for a given data set. Cross validation is simply a means of assessing the performance of the interpolation parameters. It is not a method that finds the "best" interpolation parameters in the sense that these parameters will more accurately interpolate the data.

The purpose of the estimation error function is to compare each original data value to the interpolated (estimated) value. The output from the estimation error function is a table sorted by the greatest difference (percent estimation error) between original values and estimated values. The information from the table can be used to identify those original data values that are the most poorly estimated by the interpolation method. Often these data values are extremely large and/or occur at the edges of the interpolated area.

The sediment volume may be overestimated since the boundary of Clear Creek w as never established. The boundary of the creek was determined by using aerial photography along with the edge sample points taken on August 11 through August 15. Since the creek was low during our sampling it was not as wide as what it would be during high flow. This was taken into account when processing it on the GIS. The boundary of the creek was set wider than the edge sample points to ensure that all points were included within the creek polygon, thus the area of the Clear Creek polygon used in our interpolation may be larger than in actuality.