Earthshots comes from EROS, a U.S. Geological Survey facility outside Sioux Falls, South Dakota. EROS was placed in the center of the continent to receive data from satellites coast-to-coast, and it was placed out of town to avoid radio interference.
EROS is the National Satellite Land Remote Sensing Data Archive-- the official library of Landsat images, aerial photographs, AVHRR data, Corona and Argon satellite photographs, and other such land-related data from satellites and airplanes.
When Landsat 1 was launched on 23 July 1972, workers were still constructing the EROS building which would store the Landsat images. In 1973, EROS was completed and about 70 employees moved in from their temporary quarters in Sioux Falls. By 1997, EROS was the world's largest archive of earth science data and employed over 500 people (including 17 of the original 70 "downtowners").
EROS Sioux Falls
staff population
1973: 70 75,375
1984: 342 92,225
1997: 515 115,000
As these numbers show, EROS and Sioux Falls have both grown.1 Sioux Falls formerly lay in the crooked U of the Big Sioux River, with Skunk Creek joining on the west and a diversion channel on the north. Between 1973 and 1997 you can see new residential areas wrapping all around the south, and the mall district appearing just southwest of the U.
EROS's growth requires closer looking. Compare the 1997 Landsat image to the oblique aerial photograph. In our half section (1 x .5 miles) you can make out the driveway, main building, warehouse, parking lots, and "Lake EROS," a dammed pond which is fed by our water treatment system and fished for bass and bluegill. About 100 acres of EROS land is farmed by the Future Farmers of America. Bare construction soil, excavated as part of our expansion, is visible southeast of the building, as bright bluish white.
The 1997 hailstorm
An extraordinary late-afternoon thunderstorm pounded EROS with 20 minutes of baseball- to softball-sized hail on Sunday, 13 July 1997.2 Radar images from the National Weather Service show the falling hail as red cells tracking northwest of Sioux Falls. Landsat 5 passed overhead three days later and documented the aftermath in that same storm path, appearing in the Landsat images as bright reflections like the city. The dominant red pattern represents fields of healthy crops, laid out among square section lines and interspersed with a few plots of reflective bare ground. (The pre-hail image also shows some clouds and their shadows.) The post-hail image shows where hail effectively converted cropland into bare soil.
The storm cost EROS over $1.2 million in damages. As described on local television, vehicles were smashed, branches broken, foliage shredded, windows broken, and ground pockmarked. The hail destroyed an array of 512 solar panels which had heated 60% of the photo lab's water. Many skylight panels were broken, and the roof sprang leaks. The hail broke concrete paving stones on the roof. The AVHRR antenna dish was damaged but continued to work. The big antenna for Landsat 7 was only a month old; the electronics were smashed and the dish had over 2,000 dents. Luckily, no one was injured.
The principal components image
We processed the post-hail TM scene into a principal components analysis (PCA) image. The PCA image presents more information, by compressing the information in six TM bands into three synthetic "bands."
In any satellite image there is some redundancy between bands. For example, look at bands 4, 3 and 2 of the post-hail TM image, shown as red green and blue in the standard false-color composite. Bands 3 and 2 (representing visible red and visible green) tell almost the same story. And since we can only show three channels at one time, there is no room in the image for all that information in bands 1, 5 and 7 (we left out band 6, the long-wavelength thermal IR band). To get around this obstacle we have a computer "chart" the pixels in imaginary 6-dimension space, and then determine the three principal components-- that is, the three most salient trend-lines, or important dimensions of variability. (It's easier to understand in a two-dimension scatterplot, where one principal component is derived from two bands.) We could have made up to 6 principal components, but the first 3 explain about 98% of the variance seen in all 6 original bands.3
Our principal components, shown as red green and blue, contain little redundancy. The red component corresponds to variation in vegetation. Likewise green means reflective (hard) surfaces, and blue means moisture. The image was enhanced to make the cropland, the hail swath, and the urban areas distinct.
Question
If you are the type to have read this far, you may also have noticed that Earthshots shows AVHRR images with red-green-blue representing bands 2-1-1. Didn't that seem odd to you? Why would we show band 1 twice, and why would this work? (See the answer below.)
Footnotes
1. Population figures are from the Sioux Falls Planning Office. Source U.S. Census: 1950 52161, 1960 65466, 1970 72488, 1975 77300. Source S.F. Planning Office: 1984 92225, 1 January 1997 115000, 1 January 1998 117500. 1956 and 1974 figures are linear interpolations.
2. Sioux Falls Argus Leader, 14 July 1997, p. 1.
3. Lillesand, Thomas M., 1994, Remote sensing and image interpretation, 3rd ed.: New York, Wiley & Sons, p. 572.
Other references
Thanks to the National Weather Service, Sioux Falls, for the radar images. "The EROS Hailstorm, July 13, 1997" by Ron Holmes, NWS.
Thanks to KDLT TV, Sioux Falls, for the use of the news footage.
Satellite images
LM1031030007324090 (Landsat 1 MSS, 28 August 1973)
LM502903000842420 (Landsat 5 MSS, 29 August 1984)
LT5029030009718110 (Landsat 5 TM, 30 June 1997)
LT5029030009719710 (Landsat 5 TM, 16 July 1997)
Special Projects Images
A 15 August 1972 false-color Landsat image of southeastern South Dakota is available as Special Projects Image 402. A 1984 NHAP color-infrared aerial photograph of Sioux Falls is available as Special Projects Image 1529. Several black-and-white digital aerial images of Sioux Falls from 1937 to 1990 are also available as Special Projects Images. Contact EROS Customer Services for information and ordering. The USGS EROS no longer offers hardcopy prints.
Maps
U.S. Geological Survey, 1980 (1955, revision 1980), Sioux Falls: scale 1:250,000.
U.S. Geological Survey, 1984 (compiled 1964, edition of 1984), State of South Dakota: scale 1:500,000.
Other images
The 2 September 1956 Soil Conservation Service aerial photograph is from the USDA Aerial Photography Field Office, 2222 West 2300 South Salt Lake City UT 84119, phone 801-975-3503. Frame "9-2-56 VM-1R-94".
The photographs of the hailstones and the Landsat 7 antenna were taken by Mike Austad, USGS EROS, on 14 July 1997.
The photographs of the damaged bean field (SE1/4 sec18 T103N R48W) and undamaged bean field (SE1/4 sec14 T103N R49W) were taken by Jim Vogelmann, EROS, on 16 July 1997.
The aerial photograph of EROS was flown on 2 July 1996 by Horizons, Inc., on color negative film, which is now stored at at HAS Images Inc. in Dayton, OH.
The radar images are courtesy of the National Weather Service, Sioux Falls, S.D. These are NWS doppler radar refectivity images, measured in decibels, with the radar pointed one-half-degree upward. "ND" = no data.
The television video is courtesy of KDLT TV, Sioux Falls, S.D. This clip was edited from the 5 p.m. newscast of 14 July 1997.
USGS uses Landsat images to examine land. NOAA (the National Oceanic and Atmospheric Administration, which includes the National Weather Service) uses AVHRR images (from the Advanced Very High Resolution Radiometer on NOAA satellites) to look at everything else-- clouds and water. So the Landsat bands were designed for land applications, while the AVHRR bands were designed for weather applications.
But it turned out that two of the AVHRR bands were very useful for looking at large areas of land. These bands were designed for distinguishing land from the clouds and water, but we can also use them to measure the land's vegetation and other properties. AVHRR sees the whole world every day, and its low resolution (about 1 km) yields a nice, manageable amount of data. So land scientists are very interested in AVHRR data, and these weather sensors are now operated for land study as well.
But what about the images? All these sensors have a near-infrared band, so that is comparable. But in the visible range, Landsats have separate red, green and blue bands, where AVHRR has only one cross-color band. But as we explained above, visible bands have high reundancy, so we can substitute the one visible AVHRR band for both visible red and visible green. The images come out looking very similar, even though one is a red-green-blue mix and the other really just red-cyan. All these band combinations are visually comparable:
Answer to the question above
shown as:
sensor R G B
--------------------
AVHRR 2 1 1
MSS 4 2 1
MSS 4 2 2
TM 4 3 2
TM 4 3 3
(See the Welcome / Help article for more on how RGB represents EMR.)