QUATERNARY PARK: RETRIEVAL OF LOST SATELLITE IMAGES FROM THE LATE 20TH CENTURY

Transcription

1 QUATERNARY PARK: RETRIEVAL OF LOST SATELLITE IMAGES FROM THE LATE 20TH CENTURY Grady Price Blount Department of Physical and Life Sciences Texas A & M University Corpus Christi, TX Thomas M. Holm U.S. Geological Survey EROS Data Center Sioux Falls, SD John L. Faundeen U.S. Geological Survey EROS Data Center Sioux Falls, SD ABSTRACT In a dramatic show of engineering expertise and data mining, researchers at the U.S. Geological Survey s (USGS) National Satellite Land Remote Sensing Data Archive have discovered and recovered more than 100,000 lost images from the earliest days of the Landsat Earth Observation Program. The data, which include images from every continent on Earth, dramatically extend the historical baseline for studies of land use patterns and environmental change. In many cases, the recovered data represent the very first time that some localities were subjected to multispectral and large-area analysis. The majority of the data, about 100,000 image sets, are from the 1980's to the present. About 3,000 image sets date to the 1970's. The discovery and recovery story involves how the Landsat Program was begun and how overall agency management changed from the National Aeronautical and Space Administration (NASA) to the National Oceanographic and Atmospheric Administration and, ultimately, to the USGS. Early technical limitations in data capture and operational procedures led to many image sets being essentially invisible to the research community until the 1990 s, when NASA and the USGS began to undertake data migration and, in some cases, data rescue activities on the Landsat image sets. Early procedures and policies from the Landsat program are described along with the systems that were developed to migrate and rescue these datasets.

2 INTRODUCTION The synoptic view of Earth's changing environment provided by satellite remote sensing devices has now been recognized as one of the most powerful and important tools of our time. With the recent declassification of much of the Corona imagery from the 1960's, the baseline for such orbital datasets now exceeds 40 years. The ability to access an unblinking digital archive of the Earth's surface from 20, 30 or 40 years ago is unprecedented in the history of science. The value of these data continues to compound. For digital remote sensing products, 2002 will mark the thirtieth anniversary of the continuous planetary land coverage gathered by the Landsat Multispectral Scanner (MSS) system. The scientific value of these earliest Earth-orbital digital datasets continues to grow as the environmental community develops new and novel processing algorithms to extract information never anticipated in the nascent days of satellite remote sensing. Most public awareness of developing environmental concerns such as tropical deforestation and the breakup of polar ice packs has come directly from the "then and now" imagery provided by the MSS, and later the Thematic Mapper (TM) instruments carried by successive generations of the Landsat series of Earth observation systems. To access these data, remote sensing professionals have long relied on the U.S. Geological Survey s Earth Resources Observation System Data Center (the EROS Data Center, or simply, EDC) in Sioux Falls, South Dakota.

3 In recognition of the irreplaceable value of this growing long-term dataset, EDC was designated by Congress to create the National Satellite Land Remote Sensing Data Archive (hereafter, the Archive). This is the federally mandated (Public Law ) archive site dedicated to providing long-term preservation and access for data from the Landsat and other land remote sensing systems, including formerly classified intelligence data such as that from the Corona series. DATA RESOURCES The database of Earth observations housed at the Archive is one of the oldest of it's kind. Numbering in the hundreds of thousands, MSS and TM image scenes have been cataloged, stored and analyzed there for over 25 years. Researchers looking for historical data for a particular area routinely perform database searches to find image data from the past. Until recently, it was assumed that the database was comprehensive. But there had been a standing question regarding the completeness of the database entries, specifically, had all of the images stored on early tape recording systems been processed and catalogued? The answer, we have discovered, was "no". During the 1970's, the Landsat Program was managed by the National Aeronautical and Space Administration (NASA) which employed a then-novel 28- track recording system utilizing tape media called Wide-Band Video Tape (WBVT) These 15-inch diameter reels of 1 inch wide tape were used to store Landsat image data as continuously varying analog video signals. Indeed, one of the first production analog to digital (AD) converters was built into this system. When a particular scene was requested, the analog WBVT recording would be passed

4 through an AD converter to create a digital dataset on 9-track Computer Compatible Tape (CCT) which was shipped to the customer. The highly specialized WBVTs were one of the first attempts at storing massive quantities of remote sensing data. The heart of the new discovery is the fact that the standard operating practice in the 1970's and 1980's was to decode and retrieve images from the Landsat WBVTs only on-demand. Only when these data were decoded from the original WBVTs (and after 1979, on High Density Tapes, or HDTs), did they became part of the image archive database. What was not recognized until the time of this discovery was that data, which was never requested, was never decoded! As such, it's existence never showed up in any database. As employees of NASA retired or took new positions elsewhere, these facts about operating procedures were lost, and with them, the unprocessed data languishing on the WBVTs and HDTs. By 1996, the early tapes had almost been forgotten. The specialized tape drives used to read the WBVTs had been decommissioned and dismantled years before as more modern data storage techniques came on-line. Likewise, the analog to digital decoding devices capable of translating the WBVT signals into digital data no longer existed. The physical inventory of WBVTs ended up stacked in the basement of the EDC. In a series of squat rows, the boxes of WBVTs formed a cube 50 feet on a side by 6 feet high. By that time, Archive staff knew that these stacked boxes contained the original data from the Landsat series but incorrectly assumed that all of the imagery represented by the WBVTs had been decoded.

5 Likewise, the HDTs containing post 1979 data, were thought to have been completely captured in databases describing the images. Both assumptions, it turns out, were incorrect. DATA RECOVERY Beginning in 1992, the USGS began a migration effort to refresh and copy data from the relatively newer HDT format (High Density Tapes) onto newer media. The HDT data includes the years following the use of the WBVT media post The data migration to newer media is a multiyear project, but it quickly became apparent that there was more data on the tapes than the databases predicted. As the migration activity proceeded, database discrepancies mounted; there was much more data being copied than had been expected. The migration project continues into the present. Thus far, over 100,000 previously unknown images have been added to the Archive. Figure 1 depicts the spatial distribution and frequency of images that have been discovered so far. Figure 2 is similar, but lists only those images with 30 percent or less cloud cover.

6 Figure 1. Spatial distribution and frequency of discovered HDT images. Figure 2. Spatial distribution and frequency of discovered HDT images with 30 percent or less cloud cover. The success of the HDT data recovery triggered a renewed urgency to recover the seemingly "lost" data on the older WBVTs. The physical condition of

7 the WBVTs, and the fact that a tape drive to read them no longer existed anywhere in the world presented obvious challenges to the USGS. The tapes were physically present but the data on them was unreadable. Realizing the value of these data, NASA provided seed money to design and build a new system capable of handling the decades-old tape reels. The reconstructed system became operational at the Archive in April of To date, more than 7,000 of the 20,000 original WBVT reels have been decoded and copied. In the process, over 10,000 previously unknown Landsat images from the 1970's have been recovered. Figure 3 depicts the spatial distribution and frequency of the rediscovered WBVT images. Figure 4 depicts the new images with less than 30 percent cloud cover. Figure 3. Spatial distribution and frequency of discovered WBVT images.

8 Figure 4. Spatial distribution and frequency of discovered WBVT images with 30 percent or less cloud cover. PRELIMINARY RESULTS The remote sensing community, and the Nation as a whole, will benefit from these data rescue and preservation efforts. Over 110,000 additional images have already been recovered. These data are now available for applications ranging from coastal management and geologic fault detection to flood monitoring and agribusiness. More new data will be found as the remaining 13,000 WBVT reels are processed. Figures 5 and 6 illustrate examples of some of the recovered data.

9 Figure 5. Landsat MSS image acquired May 12, 1977 over St. Louis, Missouri. Figure 6. Landsat MSS image acquired September 20, 1984 over the Nile Delta area.