Do Not Adjust the Vertical…

June 12th, 2013 by Jesse Allen

One of the wonderful things about working for the Earth Observatory is that we often get first crack at examining imagery from satellites new and old. It’s been especially exciting to look at data from Landsat 8, a joint U.S. Geological Survey and NASA mission launched in February 2013.

But with new things comes new challenges. We’ve had some odd problems with the very intense memory demands of Landsat 8 imagery, for example. And when I saw the image below, I thought for sure I had stumbled on a processing error.

elberiver_oli_2013126

This is a natural-color, pan-sharpened image of the Elbe River near Wittenberg, Germany, obtained by Landsat 8 on May 6, 2013. I had obtained this to compare to a new acquisition from June 7, 2013, which showed major flooding in the Elbe.

Oh dear. Look at that ripple pattern along the river banks. Superficially, it looks a lot like a software processing error. New code I wrote: my error, right?  In fact, at first glance, it looked a lot like Landsat data of a decade or so ago when the source files were being distributed with nearest-neighbor resampling–a technique used in remapping and resizing data which limits interactions between adjacent measures, something often useful in science measurements, but which causes jagged-looking edges.  Since this was not the first time my code had done something unexpected, it was the obvious first place to look for the cause.  The software failed me!  Again!

However, a quick glance through the data files showed that, whatever was going on, it was coming from the source data: the same rippling showed up in all the bands. Ha!  Someone else’s software had failed!

Because Landsat 8 is so new, it is easy to assume maybe I was not the only one having occasional processing problems with old software on new data. There was one more check I should have done before contacting customer service at USGS, but…

elberiver_google

…I didn’t think of it. If you see something odd in imagery, it is always good to check reality. In this case, a quick zoomed-in view in Google Earth (as shown here) would have informed me that the jagged edges along the banks of the river in the imagery are real jagged edges along the banks of the river.

In hindsight, there were other clues. Notice that the jagged features are present in some places and not others.  And notice that the rippled pattern along the banks bends and curves with the flow of the river. A processing artifact might only show up on very strongly contrasting features (the boundary between land and water here, for example), but would most likely be aligned consistently through the image.  It wouldn’t appear and disappear like it does here, and it would probably be more regular.  It would probably distort in the same direction every time it happened.

In the end, it turns out that all the new systems were working just fine and there really is a very oddly shaped series of features along the banks of the Elbe River near Wittenberg, presumably to stablize the banks of the river and control sediment flow.

elberiver_oli_2013158

But there’s not much they can do in the face of severe flooding.

Tags: , ,

3 Responses to “Do Not Adjust the Vertical…”

  1. mogesh says:

    “Ha! Someone else’s software had failed!”.
    I love those moments!

    • Jesse Allen says:

      I don’t love those moments when my software fails. Usually it means I’ve done something wrong,
      not the software. Still, the new Landsat 8 data comes in a higher dynamic range (data is stored as 16 bits per channel, so each Landsat band takes up twice as much memory to load), more bands (so if you load them all, there’s that much more in memory too), and with the pan-sharpening… well, it pushes the software and the computer in some ways it hasn’t done before, and occasionally we get some odd results that are not actually problems with the operator, but fundamentally with the code itself. Because Landsat 8 is new and we’ve already found a few places where the code acts up, it was easy to assume a software error when there was something odd looking in the results.

  2. Dave says:

    Those structures, built motr or less perpendicular to thr direction of flow, are called wing dams, dikes, or weirs. Their extension into the river increases flow in the center of the river, deepening the channel, and slowing the flow near the banks. They are used for erosion control, but they also ensure the river is navigable for barges and other riverine transportation; and they reduce the need for dredgin operations.