USGS Multimedia Gallery: (Video)--Positional Accuracy of the National Hydrography Dataset

Hi, I’m Jeff Simley of the U.S. Geological Survey. I’m going to be talking about positional accuracy of the National Hydrography Dataset. This is a study that was done by myself and Keven Roth also of the U.S. Geological Survey.

Our objective is to talk about positional accuracy of the NHD
One of the problems in analyzing positional accuracy of hydrography is that there are no well-defined points and it is difficult to find a well-defined point in the landscape and measure it’s position against another well-defined point as a control point. So what we do is to analyze a stretch of river and measure the offsets. In this image we can see that there is a river running down through it. There are little yellow lines that are offsets. We measure these offsets at one-hundred foot intervals and measure a one-thousand or five-thousand foot stretch of river and measure the offsets and collect statistics. By collecting many offsets we are able to cancel out some of the errors and come up with statistics that are meaningful on positional accuracy.

The first thing we’re going to look at is the Middle Fork of the Platte River headwaters in Colorado. This is high in the mountains at an elevation of 11,000 feet. We are looking at a stream that is about one cubic foot per second or less high in the mountains in a broad valley.
Of the one hundred samples we have measured here we have a mean of 27 feet of lateral offset error, a standard deviation of 20 feet, and a 90th percentile of 53 feet meaning that 90-percent of our measurements are within 53 feet. So what we are doing is comparing the stream in the NHD, which is the blue line, to the stream that we see in the imagery.
Also, of those 100 measurements, the maximum measurement was 92 feet.

Further downstream we’re looking at a stream in a broad U-shaped glacial valley. The stream has picked up a little more water at this point.
Another one hundred samples measuring at a mean of 25 feet, a standard deviation of 26 feet, a 90th percentile of 45 feet, and a maximum deviation measurement of 70 feet.

Here we’re further downstream and the stream has become a double-line stream, meaning that the stream is more than fifty feet wide, and we are comparing the centerline of that fifty-feet wide NHD stream to the stream in the imagery. Every one hundred feet we’re measuring the lateral offset between the two. Here at one hundred samples we have a mean of 20 feet, a standard deviation of 14 feet and a 90th percentile of 39 feet, and a maximum deviation of 70 feet. Here the river is meandering quite a bit and we find these large maximum deviations where these meanders are. What we’re seeing in the NHD is where streams are meandering in a floodplain the meandering is changing quite a bit since the NHD was compiled. And so this is typical of the meandering changes we find. Here we have a maximum of 70 feet offset, probably at one of these meanders.

Here we are further downstream now the stream is probably running at about 50 cubic feet per second. There are 100 samples, the mean is 24 feet, the standard deviation is 17 feet, the 90th percentile is 45 feet, the maximum is 104 feet. This is a floodplain.

Here we are looking at the same stream running through a canyon. Here the stream is running on the order of 150 cubic feet per second. Samples are 100, the mean is 15 feet, the standard deviation is 12 feet, the 90th percentile is 31 feet, and the maximum is 66 feet. You can see these statistics are quite a bit tighter because the river is being confined by a canyon of granite rock and is no longer meandering through a floodplain, but is more-or-less defined by the canyon walls, and so the deviations are quite a bit tighter.

Here we are even further downstream, maybe looking at 200 cubic feet per second. One hundred samples, a mean of 20 feet, a standard deviation of 20 feet, a 90th percentile of 50 feet, and a maximum deviation of 91 feet. So you notice that in the past several measurements that we have looked at in this river here the means are pretty consistently around 20 to 25 feet, the standard deviations around 20 to 25 feet, meaning a rather tight distribution of the lateral offsets, a 90th percentile of around 50 feet, which is pretty consistent, and a maximum of 91 feet. So these numbers are more or less what we are going to see across the nation.

Let’s look at some other statistics here. This is the frequency distribution of those measurements that we made. 500 measured points on the South Platte River. We can see the distribution mainly bunching up between 10 and 25 feet. Then the outliers maximum displacement of 105 feet, very few outlaying measurements, mostly bunched up from 10 to 30 feet or so. So this is also a very common frequency distribution that we find across the nation. It is an indication of what we will see. So our hypothesis is that this is what our frequency distribution will look like for the lateral offsets on a stream or small river in the United States.

Here is looking a Black Earth Creek in Wisconsin just west of Madison. This is very typical of a meandering stream. Notice that the NHD, the light blue line, and the contemporary imagery, which shows that the stream has move quite a bit in the last 30 years or so.

Let’s look at some statistics here. Here is 100 measurements. You can see the little yellow lines that are the lateral offsets that we are measuring the NHD stream against the stream in the imagery. 100 samples, a mean of 18 feet, a standard deviation of 11 feet, a 90th percentile of 34 feet, a maximum of 61 feet.

Further downstream, another 100 samples, 24 feet mean, a 15 foot standard deviation, a 44 foot in the 90th percentile, and a maximum of 86 feet.

Let’s look further downstream. Another 100 samples, a mean of 26 feet, a standard deviation of 23 feet, a 90th percentile of 48 feet, a maximum of 130 feet. If you look at this closely a good example once again of a stream. This is a single line stream. You can see how it matches up against the imagery and how the meanders are changing.

So let’s look at the frequency distribution. 300 samples. A mean of 23 feet in those samples. A standard deviation of 17 feet. A 90th percentile of 44 feet. A maximum of 130 feet. So once again, these numbers in this frequency distribution seems to be pretty consistent around the country as you look at more and more data. Here the frequency distribution between 10 and 30 feet or so, and a few outliers up to 130 feet, but very few. So this is what we think is going to be pretty consistent across the country.