Some features of this site are not compatible with your browser. Install Opera Mini to better experience this site.

Earth Matters

This post is republished from the Landsat science team page

In the giddy, early days following the flawless launch of Landsat 8, as the satellite commissioning was taking place, the calibration team noticed something strange. Light and dark stripes were showing up in images acquired by the satellite’s Thermal Infrared Sensor (TIRS).

Comparing coincident data collected by Landsat 8 and Landsat 7 — acquired as Landsat 8 flew under Landsat 7, on its way to its final orbit — showed that thermal data collected by Landsat 8 was off by several degrees.

This was a big deal. The TIRS sensor had been added to the Landsat 8 payload specifically because it had been deemed essential to a number of applications, especially water management in the U.S’s arid western states.

The TIRS error source was a mystery. The prelaunch TIRS testing in the lab had shown highly accurate data (to within 1 degree K); and on-orbit internal calibration measurements (measurements taken of an onboard light source with a known temperature) were just as good as they had been in the lab. But when TIRS radiance measurements were compared to ground-based measurements, errors were undeniably present. Everywhere TIRS was reporting temperatures that were warmer than they should have been, with the error at its worst in regions with extreme temperatures like Antarctica.

After a year-long investigation, the TIRS team found the problem. Stray light from outside the TIRS field-of-view was contaminating the image. The stray light was adding signal to the TIRS images that should not have been there—a “ghost signal” had been found.

The Ghostly Culprit

Scans of the Moon, together with ray tracing models created with a spare telescope by the TIRS instrument team, identified the stray light culprit. A metal alloy retaining ring mounted just above the third lens of the four-lens refractive TIRS telescope was bouncing out-of-field reflections onto the TIRS focal plane. The ghost-maker had been found.

Getting the Ghost Out—Landsat Exorcists in Action

With the source of the TIRS ghosts discovered, Matthew Montanaro and Aaron Gerace, two thermal imaging experts from the Rochester Institute of Technology, were tasked with getting rid of them.

Montanaro and Gerace had to first figure out how much energy or “noise” the ghost signals were adding to the TIRS measurements. To do this, a stray light optical model was created using reverse ray traces for each TIRS detector. This essentially gave Montanaro and Gerace a “map” of ghost signals. Because TIRS has 1,920 detectors, each in a slightly different position, it wasn’t just one ghost signal they had to deal with— it was a gaggle of ghost signals.

To calculate the ghost signal contamination for each detector, they compared TIRS radiance data to a known “correct” top-of-atmosphere radiance value (specifically, MODIS radiance measurements made during the Landsat 8 / Terra underflight period in March 2013).

Comparing the MODIS and TIRS measurements showed how much energy the ghost signal was adding to the TIRS radiance measurements. These actual ghost signal values were then compared to the model-based ghost signal values that Montanaro and Gerace had calculated using their stray light maps and out-of-field radiance values from TIRS interval data (data collected just above and below a given scene along the Landsat 8 orbital track).

Using the relationships established by these comparisons, Montanaro and Gerace came up with generic equations that could be used to calculate the ghost signal for each TIRS detector.

Once the ghost signal value is calculated for each pixel, that value can be subtracted from the measured radiance to get a stray-light corrected radiance, i.e. an accurate radiance. This algorithm has become known as the “TIRS-on-TIRS” correction. After performing this correction, the absolute error can be reduced from roughly 9 K to 1 K and the image banding, that visible vestige of the ghost signal, largely disappears.

“The stray light issue is very complex and it took years of investigation to determine a suitable solution,” Montanaro said.

This work paid off. Their correction—hailed as “innovative” by the Landsat 8 Project Scientist, Jim Irons—has withstood the scrutiny of the Landsat Science Team. And Montanaro and Gerace’s “exorcism” has now placed the Landsat 8 thermal bands in-line with the accuracy of the previous (ghost-free) Landsat thermal instruments.

USGS EROS has now implemented the software fix developed by these “Landsat Ghostbusters” as part of the Landsat Collection 1 data product. Savvy programmers at USGS, led by Tim Beckmann, made it possible to turn the complex de-ghosting calculations into a computationally reasonable fix that can be done for the 700+ scenes collected by Landsat 8 each day.

“EROS was able to streamline the process so that although there are many calculations, the overall additional processing time is negligible for each Landsat scene,” Montanaro explained.

A Ghost-Free Future

Gerace is now determining if an atmospheric correction based on measurements made by the two TIRS bands, a technique known as a split window atmospheric correction, can be developed with the corrected TIRS data.

Meanwhile, Montanaro has been asked to support the instrument team building the Thermal Infrared Sensor 2 that will fly on Landsat 9. A hardware fix for TIRS-2 is planned. Baffles will be placed within the telescope to block the stray light that haunted the Landsat 8 TIRS.

The Landsat future is looking ghost-free.

Related Reading:
+ RIT University News
+
TIRS Stray Light Correction Implemented in Collection 1 Processing, USGS Landsat Headline
+ Landsat Level-1 Collection 1 Processing, USGS Landsat Update Vol. 11 Issue 1 2017
+ Landsat Data Users Handbook, Appendix A – Known Issues

References:
Montanaro, M., Gerace, A., Lunsford, A., & Reuter, D. (2014). Stray light artifacts in imagery from the Landsat 8 Thermal Infrared Sensor. Remote Sensing, 6(11), 10435-10456. doi:10.3390/rs61110435

Montanaro, M., Gerace, A., & Rohrbach, S. (2015). Toward an operational stray light correction for the Landsat 8 Thermal Infrared Sensor. Applied Optics, 54(13), 3963-3978. doi: 10.1364/AO.54.003963 (https://www.osapublishing.org/ao/abstract.cfm?uri=ao-54-13-3963)

Barsi JA, Schott JR, Hook SJ, Raqueno NG, Markham BL, Radocinski RG. (2014) Landsat-8 Thermal Infrared Sensor (TIRS) Vicarious Radiometric Calibration. Remote Sensing, 6(11), 11607-11626.

Montanaro, M., Levy, R., & Markham, B. (2014). On-orbit radiometric performance of the Landsat 8 Thermal Infrared Sensor. Remote Sensing, 6(12), 11753-11769. doi: 10.3390/rs61211753

Gerace, A., & Montanaro, M. (2017). Derivation and validation of the stray light correction algorithm for the Thermal Infrared Sensor onboard Landsat 8. Remote Sensing of Environment, 191, 246-257. doi: 10.1016/j.rse.2017.01.029

Gerace, A. D., Montanaro, M., Connal, R. (2017). Leveraging intercalibration techniques to support stray-light removal from Landsat 8 Thermal Infrared Sensor data. Journal of Applied Remote Sensing, Accepted for Publication.

How to Pan-sharpen Landsat Imagery

June 13th, 2017 by Joshua Stevens

Landsat imagery has become a staple of the Earth observation community. By acquiring as many as 1,400 new images a day, Landsat 7 and 8 have given us an unprecedented look at our planet. But it’s not just the quantity of the imagery that is so exciting. Within each scene, a trove of data awaits those eager to explore the planet, pixel-by-pixel, from the vantage point of space.

A true-color Landsat 8 image of Mount Jefferson, Oregon, at a resolution of 15 meters per pixel.

It’s almost like having your own camera in orbit, 700 kilometers (430 miles) above the Earth. But Landsat is quite different from the camera in your mobile phone or dSLR. The sensors onboard Landsat capture imagery in several different “bands.” Each band represents a unique chunk of the electromagnetic spectrum, defined by wavelength. This allows for multispectral imagery that goes beyond the visible light our eyes can see. It also enables us to combine these bands in order to see the unseen, make important calculations, or assemble just the data from visible light to produce a natural-color image.

But there’s another great feature of the different bands in Landsat 7 and 8 data: one of them can be used to effectively double the resolution of your image. Yes, you read that correctly!

Panchromatic sharpening (pan-sharpening) is a technique that combines the high-resolution detail from a panchromatic band with the lower-resolution color information of other bands (usually only the visible bands).

The spectral bands of Landsat 7 and 8 have a resolution of 30 meters per pixel, while the panchromatic band has a resolution of 15 meters per pixel. By itself, the panchromatic band would look somewhat uninteresting; like other individual bands, it appears black and white. But unlike other bands, it captures a much wider range of light, allowing it to be significantly sharper. As a result, panchromatic imagery is twice as detailed as the individual spectral bands.

An image that is twice as detailed is a lot more valuable and can reveal features or phenomena that would otherwise be difficult to see:

Left: a natural-color Landsat 8 image of Mt. Jefferson, by Robert Simmon. Right: A pan-sharpened version of the same data, by Joshua Stevens.

Let’s make a pan-sharpened image with Landsat 8

The first step in creating a pan-sharpened image is to acquire the data. If you need help, we have an Earth Explorer tutorial that will guide you through the process. You may also use this sample data of Mt. Jefferson (960 MB zip archive).

The following steps assume you have acquired and unzipped the necessary data. You will also need an image editor capable of working with 16-bit data, the ability to combine and edit color channels, and support for the Lab color space. Adobe Photoshop and GIMP, a free and open source editor, are both great tools that meet this criteria.

I use Photoshop, and the examples below will reflect the Photoshop tools and interface.

First things first: create a natural color RGB composite

Follow the first part of Rob Simmon’s tutorial, How To Make a True-Color Landsat 8 Image, but stop after the initial RGB image is created. Do not color correct the image!

Pan-sharpen the RGB image

1. At this point, you should have a single, uncorrected RGB composite. Mine looks like this:

This unsharpened image will have a height and width in the ballpark of 7000 to 8000 pixels.

2. To prepare the image to be twice as sharp, we first need to double its size. In the Photoshop menu, go to Image and then Image Size. In the dialog, set the height and width to 200% and ensure that resampling is set to bilinear (this will prevent any unwanted, extra sharpening from occurring).

Your image will now be twice as large (15000-16000 pixels wide and tall). But you will notice it is not very sharp; in fact, it will look quite fuzzy. Let’s change that.

3. Recall that the panchromatic band has a resolution of 15 m in black and white. What if we could use the color from this fuzzy image, but add detail from the lightness and darkness of the panchromatic data? That is exactly what we are going to do.

Currently, our image is composed of three channels: red, green, and blue. We want to change that so one of the channels reflects lightness.

That is where the Lab color space comes in. Lab color space depicts color in terms of lightness and two other channels known as a and b (which control the red-green and yellow-blue of an image, respectively).

In the Photoshop menu, go to Image –> Mode -> Lab Color. You won’t see the image change, but if you look at your image channels, you will see that you now have Lightness, a, and b.

4. We need to replace the image in the Lightness channel with the panchromatic data. Open your panchromatic band (the filename should end with _B8.tif in your data).

5. With the panchromatic image open, select everything (Photoshop menu Select -> All). Then copy the image (cmd-c on Mac, ctrl-c on Windows).

Go back to your RGB composite image. In the Channels window click on Lightness to isolate that channel. Your image should change to black and white to indicate that only the selected channel is active:

Paste the panchromatic data (cmd-v on Mac, ctrl-v on Windows). You should see the black and white image become significantly sharper as the higher resolution imagery replaces the previous data:

6. Convert the image back to RGB space. In Photoshop go to Image -> Mode -> RGB Color.

Color has returned and you now have a pan-sharpened, RGB composite!

You can now color correct the image and apply any additional sharpening desired.

Return to How To Make a True-Color Landsat 8 Image to learn how to color correct your pan-sharpened imagery.

Additional notes and advice

Pan-sharpening is a common technique, and there’s more than one way to do it. The method shown here is called HSV Sharpening. It is the simplest of the techniques and relies only on swapping the value (the ‘v’ of HSV) data with a higher resolution counterpart. The hue and saturation (the ‘h’ and ‘s’) provide the color from the lower resolution imagery.

Another more sophisticated method known as Gram-Schmidt pan-sharpening is frequently used in GIS software. Gram-Schmidt generates a closer representation of the panchromatic data by simulating it from the other spectral bands.

On Landsat 8, the panchromatic band doesn’t go as far into the blue wavelengths as the blue band, and it goes slightly beyond red. If you followed this tutorial and noticed a slight change in tone when you pasted in the panchromatic data, this is why. Panchromatic data from Landsat 7 is even more extreme and includes more near infrared light, so it will change the tone of the image even more.

A third technique, known as a Brovey transform, or color normalization, attempts to balance the color information with the panchromatic data. It can be useful for pan-sharpening false color imagery, the success of which depends on the wavelengths captured by the panchromatic data.

Although this tutorial covered the process on a single image manually, much of this can be automated from the command line with relative ease. A few tools I recommend:

There you have it. Happy sharpening!