Forestry Applications Put Historical Digital Orthoimagery to Use
By F. Thomas Lloyd, Rose M. Fletcher and N. Jane Thompson

Introduction
Geographic information and remote sensing technology deals with issues of measurement in new and powerful ways. Forestry applications depend on taking the additional step of investigating biological and spatial relationships. As a result, a major science task for forestry research is to identify, quantify, and model the spatial relationships that biological processes have with their physical environment. The fact that such relationships exist is not new knowledge, but efforts to model them are dramatically lagging. Our past experience in the forestry profession of viewing spatial data primarily as maps, limits our view of spatial analysis and has caught us unprepared to fully respond to the rapid expansion in availability and capability of geographic information and remote sensing hardware and software.
      The long time frames of forestry make the usefulness of historical data intuitively obvious. This article discusses our approach, and some of the technical problems we encountered, in identifying and quantifying a biological/spatial linkage. We do it by showing how we built an historical data layer and how we incorporated the spatial information into an investigation of plant community composition and structure.

Study Area
The study is located in a National Environmental Research Park on the Savannah River Site (SRS). It is a 200,000-acre tract of land that was purchased by the Atomic Energy Commission in the early 1950's for the purpose of producing nuclear materials for the Department of Defense. The SRS is located in the Hilly Upper Coastal Plain region in South Carolina, bordered on the southwest by the Savannah River. Today it is administered by the Department of Energy, and operated by Westinghouse Corp. (the production facilities) and the USDA Forest Service (the natural resources). Dramatic land use changes have ensued since its creation 40-plus years ago.

Digital Orthoimagery
Aerial photographs have always been useful in making forest management decisions. In their original state, aerial photographs show a piece of the landscape, distorted and without locational information. The distortion results from perspective viewing of an irregular terrain from an unstable platform. Measures of the changing terrain in the form of digital elevation models (DEM's), camera parameters, ground control points (GCP's), and scanned imagery are inputs to photogrammetric software that generates relatively distortion-free products called orthoimages.

Changes of Historical Photography
The long-term nature of growing and managing forests suggests a high potential value of historical photography. However, the application of computer-based photogrammetric technology to historical photography presents some unique challenges. For example, the seemingly simple task of deciding which photographs to purchase from the provider (in this project, the USDA Agricultural Stabilization and Conservation Service (ASCS)) was difficult because there are usually very few clues in past imagery of modern political boundaries. After considerable investigation of land features in the study area, we settled on 251 panchromatic aerial photographs taken in May of 1951 at a 1:20,000 scale by Park Aerial Surveys under contract with the ASCS. For our application, a winter scene would have been better, but another limitation is having to work with what is available.
      Although the historical photographs were generally of good quality, they were made at a time when photographic systems and flight standards were poorer than they are today. Visual quality was marred by streaking in the film diapositives which was probably the result of early photographic processing techniques and storage on rollers. These images also exhibited a condition where the edge areas of the photographs received less light than the center, which produced challenges later in the mosaicking process. We also found that the fiducial marks were rough and rounded, increasing the difficulty of precisely registering the images.
     Precision in the orthorectification process depends not only on the quality of the source photography, but also on documentation of the camera. The camera calibration report for this 1951 flight was not available, but reports on the same camera were found for flights in 1945 and 1963. Unfortunately some of these parameters can change with time, such as the focal length calculation. In this case some differences in focal length were found, but fortunately, a sensitivity test suggested the differences to be of little consequence. We decided to use the 1963 values because the report had more information. Another problem with these old reports was that they did not provide the location of fiducial marks or the principle point in a camera coordinate system. This information, which was necessary for the soft-copy triangulation solution, had to be estimated from the diapositives we obtained from the ASCS. Unfortunately, we have no way of assessing the possible deviation of these estimates from the unknown true camera coordinates.
      One of our most challenging tasks in the rectification of historical photography was the acquisition of ground control from landscapes that had been changed vegetatively (see Figures 2a and 2b) and physically over time. The topographic surface had been significantly altered by construction of large man-made lakes, extensive grading for large production facilities, and massive road building for unusual transportation and security purposes. Through careful photo interpretation and field verification, we came up with 86 photo-identifiable points which to the best of our ability are temporally stable, the majority of which were small road intersections. Still, the possibility exists that some points could be inaccurate due to undetected changes in roadway alignments. Evaluating the accuracy of historical orthoimages will be an important step given these possible errors in the GCP's.

Development of Orthoimage Mosaics
Film diapositives were scanned on a drum-based scanner to a ground resolution of approximately 0.5 meters. GCP coordinates were collected to accuracies within plus or minus 0.25 meters using Trimble 4000 survey-grade Global Positioning System (GPS) receivers linked to the National Geodetic network. These GPS coordinates, based on the GRS 80 ellipsoid, were converted to our required output reference system and orthometric heights using the conversion programs Corpscom, Vertcon, and Geoid 93. Standard 7.5 minute U.S. Geological Survey Level 2 DEM's, edge-matched with an averaging algorithm, constituted the final input into the orthorectification process. O
      Orthorectification was performed on SUN Sparc 10 and 20 workstations using Imagine OrthoMAX software developed by Autometric Inc. and marketed by ERDAS. The mosaic shown in Figures 2a and 2b was triangulated in two blocks of 52 and 72 photographs. Although a more comprehensive test will be performed when this project is finished, an abbreviated test (using only five surveyed control points set aside for testing) resulted in four being less than 3.5 meters apart. The fifth was off by 8 meters in a northing direction. Despite the many uncertainties in working with historical photography, we think the finished product will meet National Accuracy Standards at the 1:12,000 scale.
      Imagine also served as our image processing software. The central portion of each orthoimage in the stereo coverage was cut out in an irregular fashion along the edges of fields and roads. The photo chips were enhanced with modified linear stretches and mosaicked with a filtering algorithm. Our final products will be a set of 7.5 minute Digital Orthoimage Quads (DOQ's) with 2 meter resolution and header information in conformance with National Spatial Data Transfer Standards, a set of digital orthoimages with original brightness values, and a set of raw scanned images for stereo-viewing applications. A Data Users Guide will provide details on development methods used.

Biological Response Study Using Historical DOQs
Our biological interest was generally on the effect that past land use had on present-day vegetative composition (plant species) and structure (the size distribution of plants). The SRS had a wide range of land use (or disturbance categories) in 1951, ranging from active agriculture, abandoned agriculture, pastures, forested areas that had been recently harvested, areas with sparse timber, and fully forested areas. Immediately after purchase, a massive pine planting program began on the fields, pastures, and cut-over forested areas (the largest ever done at that time in the South), resulting in establishment of tens of thousands of acres of pine plantations in just a few years over a range of disturbance conditions. This set of circumstances provides a unique, large-scale experimental opportunity to study the effects of prior land use after 40 years of forest growth and plant community succession.
      Our next step was to install a field study of vegetative composition and structure that was linked to the historical DOQ data. The approach was to install center points of circular vegetation plots at known coordinates using GPS technology to establish starting points from which plot centers were surveyed using a laser instrument. The grid points, located 60 meters apart, were arranged into irregular shaped groups at 10 forested sites (stands) on the SRS. See Figure 1 for a map of one of the sets of points laid over the orthoimage for that location. A total of 1040 plots were measured.
      The study had two objectives: (1) to eliminate the confounding effects of plant disturbance (via land use) on a test of the predictive ability of a GIS-based ecological classification (EC) model used to delimit ecologically equivalent land units (ecological land units are predicted from soil-type polygons), and (2) to test whether knowledge of historical land use could be used to predict the structure (that is, the size distribution) of the tree species developing in the understory of the pine plantations. Both objectives deal with the restoration of plant communities after the forest has been disturbed, and the historical DOQ provides spatial data on the degrees of disturbance at the time the forest was regenerated.
      The use of a plant-community-based EC model is a complete topic of its own, so a detailed discussion would be beyond the scope of this article. Suffice it to say that numerous biological processes, like growth rates, species site compatibility, response to management actions, etcetera, are linked to these spatially modeled land units. Using the historical orthoimage to control for past land use allows us to more accurately test the functioning of the EC model.
      Another planned use of this historical spatial data relates to using the historical DOQ data to predict the effect of land use on the structure (size distribution) of the hardwood (deciduous) trees that are developing in the understory. Specifically, we need to be able to spatially describe understory habitat structure for use in a GIS-based demographics model being developed for an endangered bird (the red cockaded woodpecker) found in small numbers on the SRS. More precisely, the goal is to model the interacting effects that timber harvesting have on the bird's contrasting need for large, old trees and no hardwood understory. We can not afford to field measure the hardwood understory for the thousands of stands on the 200,000-acre SRS forest, so our field study and the historical DOQ data will be used to predict structure, which will be field checked at a few sample locations.

Conclusion
We have tried to illustrate two points. First, creation of high quality spatial data layers for natural resource applications is time consuming and difficult. Second, we need to design spatial variables into our investigations of biological response if we expect to build highly useful spatial models for natural resource management. Each of these tasks is challenging. We believe that efforts to build useful spatial models will continue to lag if we avoid the challenges of building quality spatial data sets and spatially references experimental designs.

About the Authors:
F. Thomas Lloyd is a research forester for the USDA Forest Service, Southern Research Station, Bent Creek Research and Demonstration Forest. He may be reached at 704-667-5261 ext. 116. Rose M. Fletcher is geographer/ecologist for the USDA Forest Service, supported by funds from the Department of Energy (DOE) administered through the Savannah River Forest Station at the SRS and located at Clemson, S.C. She may be reached at 864-656-1290. N. Jane Thompson is a forester, formerly with the USDA Forest Service at Clemson, S.C. and was also supported by DOE funds administered through the SRS.

Acknowledgements
This research was funded by the U.S. Department of Energy in cooperation with the Savannah River Forest Station and the Southern Research Station of the USDA Forest Service. The regional office in Atlanta, Ga., of the National Forest System contributed technical advice and GPS survey equipment. The use of trade or firm names for product and service providers is for reader information and does not imply endorsement by the U.S. Department of Agriculture.

Back