Error analysis procedure

Table 8. Elevation range for each multitemporal class.

Figure 4. 1993 multitemporal water classification with ground suvey profile points overlaid

SR99_06
Page Navigation
9

the lowest layer plus the volume of each of the in-

were shallow water or mudflats, and how they

cremental volume layers.

should be counted in the water surface acreage tal-

By knowing the volume for each incremental

ly. The error bars were determined by including

volume layer for each of two years (in this case,

and excluding these pixels from the area estimate.

1985 and 1993), it can be determined how much

Rectification (#2)

volume was lost per layer over that time period by

subtracting the incremental volumes. Part of any

There is some uncertainty inherent in the rectifi-

one layer lies over water (the next lower water lay-

cation procedure, caused by error in the DLG data

er), part over land (the incremental area increase

and in selection of control points. The error esti-

ring for that elevation increment). The part of the

mate in this case was derived from an estimate of

layer lying over water cannot have lost any vol-

the error in the DLG data, because that error was

ume to sedimentation because the sediment would

estimated as larger than the control point error.

keep sinking through the water below. This means

Wind setup (#3)

that the volume lost in this layer has to be assigned

to the part of the layer lying over land, which cor-

To derive capacity estimates, it was assumed

responds exactly to the incremental area ring for

that the reservoir was level. If there is wind at the

that elevation increment. The average depth of

time of the Landsat overpass, the water surface

sediment in that incremental area ring can be de-

may not be level, and the area reported for the ele-

termined by dividing the volume lost in an incre-

vation given at the dam may be higher (west wind)

mental volume layer by the area of the incremental

or lower (east wind) than for level water. Error bars

area ring. This procedure was used to compute the

were computed making a number of assumptions,

average sediment depth for the incremental area

but in particular that all scenes had a half-gale-

rings associated with the 11 1993 Landsat scenes.

force wind, either east or west.

Classification threshold (#4)

Error analysis procedure

A number of assumptions had to be made in this

In the Band 5 threshold classification procedure,

procedure for estimating the volume of the reser-

it was a judgment call as to where to put the divid-

voir using Landsat data, and a number of uncer-

ing line between water and shoreline, and between

tainties exist in the methods used. An attempt was

shoreline and land. These error bars give what the

made to identify and quantify the sources of uncer-

difference in water area would be if the thresholds

tainty. These are not formal error estimates, but

for all the scenes were moved either up or down

should give a feeling for the magnitude of the un-

one value.

certainty. For each case, a reasonable maximum er-

Masking (#5)

ror was estimated. Because there were no probabil-

ities associated with these error estimates, formal

It was a judgment call as to where to draw the

statistics were not possible.

line between the reservoir pool and the flowing

For error sources #1 through #5 below, the pro-

water at the upstream end of the reservoir. The

cedure used to estimate the error bars for storage

masking error was figured by outlining an alter-

capacity was to estimate the error in water surface

nate "reasonable maximum" and "reasonable min-

area, and then recompute the elevation vs. storage

imum" water area that might have been included.

capacity curve, as outlined in the Storage capacity

Depth reading (#6)

procedure section above, using areas plus and mi-

nus the estimated error.

The reservoir elevation readings at the dam pre-

Sources of error are described below. Undoubt-

sumably have some uncertainty. To compute the

edly there are other sources of error as well, includ-

error bars, it was assumed that the elevation read-

ing those involved in the technique used to com-

ings at the time of the overpasses were all either 0.1

pute capacity from area. More detail on the error

ft high or 0.1 ft low. At the exact elevations of the

analysis can be found in Appendix D.

Landsat passes, the effect on the capacity is much

greater (thousands vs. hundreds of acre-ft) and in

Mudflats (#1)

the opposite direction than at the intermediate eleva-

For some scenes and in certain locations, the in-

tions.

termediate "shoreline" classification category cov-

Lowest levels (#7)

ered rather extensive areas, instead of a one-pixel-

wide area. It was uncertain whether these areas

Because the lowest elevation represented in the

Back to contents

Index

Privacy Statement
Press Release
Contact