My 10-Year Challenge: Glacier National Park

How I used a meme and satellite imagery to show the devastation of climate change — and you can too.

Glacier National Park difference from 2008 to 2018, using a SWIR/NIR/Red visualization of Landsat imagery. Cyan shows a glacier extent. Displayed using the Radiant Earth Foundation platform; slider on the right is directly from the platform.

The effects of climate change aren’t always big shocks — often, it takes place over the years. What better way is there to show what we have lost than using a fun internet meme like the “10-year challenge?” I used the Radiant Earth Foundation platform and historical US Geological Survey (USGS) Landsat satellite imagery to show how glaciers have disappeared from Glacier National Park — and, following my process, you can produce an image just like it.


Vanishing Glaciers

Glacier National Park is a beautiful rocky mountain destination located near Whitefish, Montana. Its valleys and mountains have been cut and smoothed by ancient giant glaciers, dating back thousands of years. The remaining glaciers, however, are under severe threat.

USGS estimates that there were 150 glaciers in 1850. Now, of the 37 glaciers identified in recent years, only 25 remained active in 2015. As the glaciers continue to recede, experts expect glaciers to disappear from the park by 2030.

Jackson Glacier from the same viewpoint in 1911 (top) and 2009 (bottom) Credit: M. Elrod, University of Montana | USGS/Lisa McKeon

Though photographic and other direct evidence of receding glaciers is available, historic satellite imagery provides some of the starkest evidence of how quickly these glaciers are disappearing.


Satellites on Ice

In contrast with aerial and land-based monitoring, satellite imagery has the distinct advantage of serving as an archive of standardized images with a consistent revisit time, meaning that nearly identical images are captured of the same location every few weeks (or in recent years, every few days). While there are some drawbacks to satellite images — they have a limited history and are often limited by cloud cover, for example — the USGS Landsat program allows us to see back in time at regular intervals since the 1980s.

In addition, the multispectral nature of Landsat imagery — in contrast with typical Red/Green/Blue (RGB) aerial photography — allows for a more fine-tuned display of glacier ice. Whereas in RGB photography rock and sand can appear very similar, Landsat imagery includes near-infrared (NIR) and shortwave infrared (SWIR) bands that can be used to delineate areas covered by ice.

Ice has a high reflection in the visible bands (Red, Green, Blue); therefore, it appears bright white to human eyes. However, at lower frequencies, particularly SWIR, light is absorbed almost entirely by the structure of water molecules within ice.

The red line shows how glacier ice reflects visible light but not NIR or SWIR. Credit: Andreas Kaab, University of Oslo, “Optical properties of ice and snow,” ESA.

This spectral signature looks fairly unusual when using a visualization that shows SWIR/NIR/Red instead of the typical RGB. Vegetation, which reflects high in NIR, appears green; water, which absorbs most light and reflects only a little visible light, appears very dark blue; and mountains and other bare soil typically appear red, since they reflect highest in the SWIR part of the spectrum.

The 2018 image is shown as an RGB or ‘natural color’ image (right); the same image is shown SWIR/NIR/Red image (left). Note that vegetation appears bright green, water dark blue, bare ground as brown, and ice as cyan.

Ice, however, appears cyan, reflecting very high in the visible range (displayed as blue) and moderately high in the NIR range (shown as green). It makes ice very easy to identify in contrast with an otherwise bright rock. Water is also much easier to differentiate from shadows.


Assembling the Meme

Using the Radiant Earth Foundation platform, constructing a 10-year challenge image like this is straightforward:

  • Start by logging in at app.radiant.earth, which is available for free to anyone.
  • Create a project with a name and date. Note that each date should have its own project. You can read more about creating projects in Radiant Earth’s help guide.

I called mine “Glacier Park 2018.”

  • Zoom into your area of interest and browse for imagery, filtering by sensor or satellite, time, and location. When you find an image without cloud cover for your date of choice, click the ‘+’ button. (Note that scenes can take some time to ingest into the system; the gray button will transition to yellow while processing and will turn blue button once fully ingested. Once a scene has been ingested by one user, it is available instantaneously for everyone).

The Landsat program is the only mission that has regular global coverage going back ten years with good spatial resolution to detect glaciers. As a result, I filtered to only Landsat 8 images, available within the Radiant Earth Repository (that’s the default option for the dropdown menu at the top). I zoomed into the area of interest within the national park and set the date range to September (typically late summer is used for monitoring glaciers, as all non-glacial ice will have melted by then), found a suitably cloud-free image, and added the scene to my project.

  • Use color mode to change the visualization if needed.

I changed the visualization from natural color to the custom 4/5/7 visualization, as shown in the picture below, so that the glacial ice would show up as cyan, as discussed above.

Custom color visualization, with Band 4: Red as blue, Band 5: Near Infrared as green, and Band 7: Shortwave Infrared 2 as red.
  • Create the second project for the 2008 image.

I have similarly called it, “Glacier Park 2018”.

  • Find appropriate archival imagery and add it to your project. Historical Landsat imagery (from Landsat 4, 5, and 7) is available through the NASA CMR repository rather than the default Radiant Earth Foundation repository, found in the dropdown menu at the top.

I used Landsat 4/5, because Landsat 7 has had a scan line corrector issue since 2003. Landsat 5 was on orbit through 2013 (although it was not providing as much imagery towards the end of its service) and Landsat 8 was on orbit starting in 2013.

Once I switched to the NASA CMR repository and selected Landsat 4/5, I limited the search to September 2008 and chose the scene with the least cloud cover.

  • Add the visualization options to the project.

Landsat 5 has slightly different bands captured, so it’s not an exact match; the most similar visualization uses the designated Red, NIR, and SWIR2 bands (Bands 3, 4, and 7).

  • Use the lab to display the two images side-by-side in a slider.

More details are given in the section below how to do this.

SWIR visualizations of Glacier National Park from 2008 and 2018 using the Radiant Earth platform Lab.

Some Light Math

To compare, I used the Normalized Difference Snow Index (NDSI). NDSI compares visual and SWIR reflectance, as ice reflects lots of visible light and almost no SWIR light, in contrast with other types of land cover. NDSI uses the formula (VIS — SWIR)/(VIS + SWIR), which compares the visual reflectance to SWIR and then normalizes that value over the total amount of visible and SWIR reflectance. That normalization allows for comparison across time, meaning that we can do a change detection between the new and old images, using this formula:

Normalized Difference Snow Index Change Detection formula

Any NDSI value over 0 is considered typically considered snow, although, as seen below, certain environmental features can be captured. When subtracting the old NDSI value from the new NDSI value, any positive value means that there is added snow/ice cover and any negative value means that there is lost ice cover. In this case, negatives indicate lost glacier ice.

To do this comparison, I used the Lab section of the platform, using the following process:

  • Go the Lab on the platform.
  • Go to Templates and find NDSI, and click Use Template.

If NDSI is not an available template, you may have to create a new template. I entered the formula ((REDnew-SWIRnew)/(REDnew+SWIRnew))-((REDold-SWIRold)/(REDold+SWIRold)) .

  • Add a name to the analysis and click Create Analysis.
  • Select the projects by clicking Change and then the appropriate dropdown (i.e. the 2018 project and the red band for REDnew)
  • When all bands are selected, click the map button (first button on the left) to display it.

You can also change the colors using the histogram (bar chart) or display two side-by-side by clicking the split icon in the top right and then clicking two modules.

For better comparison of what these two calculated values look like, I took snapshots of the two NDSI images.

Normalized Snow Difference Index (NDSI) of part of Glacier National Park, 2008 (left) and 2018 (right). Orange indicates no ice; light blue indicates shadow/some ice; dark blue indicates glacial ice.

In this case, the darkest blue indicates significant ice presence. Notably, some the water and shadows are partly registering as ice in the 2018 image, appearing as the lighter blue. This is because, the difference in brightness overall is minimized in these areas, so the SWIR and Red parts of the spectrum are reflecting minimally.

However, only the glacial ice appears dark blue. While some of the lighter regions appear to expand due to the added sensitivity of Landsat 8 in comparison with Landsat 5, when we focus on the dark blue areas, especially in the northern part of the image, significant ice areas have disappeared.

This disappearance is happening only over the course of 10 years. The total disappearance of these glaciers may be only a few years down the road.

Fortunately, while the interpretation can sometimes be difficult, the Radiant Earth Foundation platform, which is open to anyone, provides tools to make the processing much easier. With only a small amount of effort, anyone can make this kind of change detection.


A Challenge For You

I would love to see more people creating similar 10-year challenges, demonstrating how climate change and other anthropogenic forces are transforming the Earth and taking advantage of historical Landsat imagery. Please share on Twitter by tagging @OurRadiantEarth or on Facebook by tagging Radiant Earth Foundation!