Landsat Pathfinder Project
Advances Deforestation Mapping

By Walter Chomentowski, Bill Salas, and David Skole. Chomentowski and Salas are research scientists and Skole is an assistant professor with the University of New Hampshire Complex Systems Research Center, Institute for the Study of Earth, Oceans, and Space, Morse Hall, Durham, NH 03824-3525, USA.

The Landsat Pathfinder project is demonstrating the use of Landsat data for global change research and monitoring the world's forest resources. One of the five Pathfinder projects initiated by the U.S. National Aeronautics and Space Administration (NASA), Landsat Pathfinder is a first step toward establishing a global monitoring system using high-resolution satellite imagery. The imagery will be stored, managed and analyzed with a GIS.

Pathfinder is part of NASA's response to the requirements of the U.S. Global Change Research Program--the centerpiece of which is the Earth Observing System (EOS) satellite series. Scheduled to begin operation in 1998, EOS will transmit as much as two terabytes (2,000 gigabytes) of Earth data each day. Pathfinder's mission is to explore the data needs of scientists studying global change and prepare for the massive EOS data stream.

Some researchers estimate the global deforestation rate has more than doubled since the early 1980's; others say it has increased by only 50 percent. The Landsat Pathfinder project will help scientists better quantify estimates of land use/land cover change in the world's tropical forests. Above, the project's study area is shown in red. The Amazon/Orinoco river basins, central Africa and southeast Asia contain 75 percent of the world's tropical forest.

Pathfinder Particulars

Each Pathfinder project centers around currently available data form different satellites. Landsat is the largest Pathfinder in terms of data volume, consisting of approximately 3,000 Multispectral Scanner (MSS) and Thematic Mapper (TM) images. A three-year cooperative effort between the University of New Hampshire (UNH) and the University of Maryland (UMD), Landsat Pathfinder was started on year ago to develop a sub-kilometer-resolution data set for better quantitative estimates of land use/land cover change in the humid tropics--a relatively unknown, but significant part of global change.

Landsat Pathfinder will use GIS and remote sensing technologies to map tropical deforestation from the 1970s to the 1990s across an area 2.5 times the size of the continental United States. Three regions containing 75 percent of the world's tropical forests are being mapped; the Amazon/Orinoco river basins, central Africa and southeast Asia. Complete coverage of each region is being acquired for the early 1970s, 1980s and 1990s to document the extent and rate of deforestation. MSS will be used for mapping the '70s and '80s; TM for the '90s. UNH is processing data from the Brazilian Amazon basin and southeast Asia, while UMD works with the non-Brazilian Amazon/Orinoco basins and central Africa.

Landsat inagery is valuable for mapping deforestation in the tropics. Here, a 14-kilometer-square portion of a Multispectral Scanner band 4,2,1 false color composite of Ariquemes, Rondonia, Brazil, in 1972 (left) is contrasted with a Thematic Mapper band 4,2,1 false color composite of the same area in 1992 (right). The large green area in the middle of the image at right is the town center, reddish speckled areas are uncut forests and bright red areas are regrowing forest.

Data Set Components

Four components are essential when building a global tropical high-resolution data set with satellite imagery; image selection, processing, data management and archive management. Imagery must be selected and acquired before any of the other three components can be executed. Approximately 3.3 million MSS and TM scenes have bee collected in U.S and foreign archives since the Landsat series was launched in 1972. The global scope of Landsat Pathfinder required researchers to select imagery from archives in Brazil, Ecuador, Thailand, India and the United States.

Choosing and acquiring imagery using the traditional method of manually searching through pages of computer printouts is tedious when looking for a few scenes. It becomes overwhelming when large archives and expansive geographic areas are involved. Thus, a software tool was needed to assist in image selection. Currently, there is no database of world Landsat holdings or software to search and display those holdings. To solve that problem, UNH developed a point-and-click Information Management System (IMS) for image search and selection. The system initially was built for selecting imagery, but later it was expanded to include data management and archive tasks.

System Management

Most of the IMS is built with tools provided by the workstation version of ARC/INFO, GIS software developed by Environmental Systems Research Institute, Inc. Tools not provided were written in the C programming language or borrowed from the X window toolbox. The IMS look and feel is similar to window managers provided by major workstation vendors--complete with pull-down menus, buttons and dialog boxes. Everything is point-and-click; there is no command line keyboard input.

The IMS is composed of three distinct modules; query and browse, data manage- ment and archive management. The query and browse section enables a user to locally search both U.S. and foreign archive image metadata. (Metadata describe image quality and geographic position.) The data management module is a project accounting system used to track imagery through the processing stream until it is archived. The archive management system picks up where the data management leaves off and provides a data archive interface.

Query and Browse: To Search the metadata library with the query and browse tool, pull-down menus are used to define a query with constraints on geographical region, date, cloud cover and/or other image descriptions. The query result is displayed as one or many rectangular polygons outlining the image footprint. Other data layers can be displayed simultaneously such as a regional coastline vegetation and towns. If a more detailed view of a selected scene is desired, a compressed picture, called a "browse product," can be displayed by clicking on a footprint of interest. Browse products in the IMS only are available for imagery acquired by Landsat Pathfinder.

The query and browse section of the Pathfinder Information Management System enables a user to locally search both U.S. and foreign archive image metadata. This figure is a snapshot of data availability in the Amazon basin for 1986. Two samples of Multi Spectral Scanner scenes from the Pathfinder archive are shown in the upper right and left corners.

Data Management: The data management section is similar to package tracking systems used by express mail companies. But instead of tracking a package from origin to destination, the data manager tracks imagery through each phase of the processing stream and provides detailed information about individual scenes. Each image is described by 61 attributes. Image descrip- tions are stored in a database management system (DBMS) and contain information such as date ordered, processing status, processing parameters and Universal Transverse Mercator (UTM) zone. The DBMS is linked internally to rectangular polygons in ARC/INFO, representing the image boundary/footprint. The data manager can be queried to answer myriad questions, and answers are displayed graphically or in a tabular report.

The ability to sit at one's desk with the IMS and get current information on project status or an individual scene is necessary to efficiently allocate project resources. Receiving 3,000 scenes during a three-year period, then distributing, analyzing and archiving them would be disastrous without an accurate accounting system.

Archive Management: The raw imagery and finished products will compose an archive of 500 gigabytes that will be stored on a magneto-optical juke box. Manually searching through a large archive with system commands entered at the keyboard would be time-consuming and frustrating. Thus, an archive management section was built into the IMS to simplify access to the Pathfinder archive. It presents the user with a graphical, point-and-click interface to the archive, similar to the query and browse, and data management modules. Ground truth are archived with the other data and are accessible through a multimedia section of the archive manager. If visual inspection of the imagery is necessary, a browse product can be displayed by clicking on the area of interest. After a successful archive query, the desired data can be loaded on tape or magnetic disk.

Extensive ground truth data were collected in selected areas of the Amazon and southeast Asia. The field data consist of photographs taken on the ground with a 35mm camera, geographic locations obtained with a Global Positioning System (GPS) and tape-recorded voice descriptions. The three data sources are organized and stored in the archive manager and are accessible through a graphical interface. Photograph and audio information are linked to the GPS data points inside the GIS. When querying the ground truth database, the user selects an image from the archive with the archive manager. The ground truth module is activated next, and GPS points appear on top of the selected image. Click on a GPS point and one or more photographs taken on the ground will appear along with an audio description of the area selected. If ancillary information on the subject is desired, pull-down menus allow Wide Area Network (WAN) access to nationwide book libraries or locally stored scientific papers.

Data Processing

The data processing stream combines automated classification and visual photograph interpretation. The imagery is automatically classified, manually edited and, finally, merged into a single seamless database. After the necessary descriptor information is entered into the data manager, an image is transferred from tape to disk and rectified to a UTM projection. All features on the imagery then are classified as forest, deforestation, regrowing forest, water, clouds, cloud shadows and grassland. Classification output is a georeferenced, raster data file in which each cell has a value corresponding to one of the land cover classes.

The automated classification method is precise, but not always accurate. To correct inaccuracies, the raster data are vectorized and manually edited. The vectors from a classified image are plotted on transparent paper at 1:250,000 scale and overlaid on top of a color image photograph of the same scale. Three types of errors are encountered in the vector data: mislabeled polygons, incorrect polygons and omitted polygons. The three errors are corrected by changing the label, deleting polygons and adding polygons, respectively. Editing is an iterative process in which the data may be plotted two or more times until the thematic accuracy passes through quality control. When all errors have been removed, a shaded plot representing the seven land use classes is archived with the photo product.

The end result or the processing stream is a seamless digital map of deforesta- tion stored in a GIS. A seamless map is produced by merging individual images that have been classified and edited into a single database. Before all the pieces of this large puzzle will fit together, their edges must be thematically and positionally matched.

The imagery received for the project is system corrected, not precision geocoded; therefore, some imagery is positionally incorrect--up to five kilometers in extreme cases. Positional matching is done with a linear transformation.

Thematic classification is not consistent from scene to scene because of changing atmospheric conditions. One scene may be hazy from the smoke of burning forests, while the adjacent scene, acquired on a different day, is clear. Therefore, grassland sometimes is classified as forest and vice versa. There are few problems thematically matching the other five classes: defores- tation, regrowing forest, clouds, cloud shadows and water. After edge match- ing, the data are converted from vector to raster and merged into a seamless database. The conversion back to raster was chosen because of constraints on CPU speed and disk space. Merging the vector data took much more time and disk space.

Fruition or Failure

After completion, all Landsat Pathfinder imagery and land cover classifications will be available to the global change community at the EROS Data Center (EDC) in Sioux Falls, S.D., USA. Such information will help scientists understand more about how human-induced changes affect the world's ecosystems and aid in environmental planning on a global scale.