LAND-COVER MAPPING FOR THE SECOND COLLEGE GRANT USING RECTIFIED AERIAL PHOTOGRAPHS AND AERIAL VIDEOGRAPHY
Personnel: Sean MacFaden, David Capen
Cooperators: Kevin Evans, Dartmouth College Woodlands Office
The Second College Grant is an unincorporated township in northern New Hamphire, owned and managed by Dartmouth College. In its continuing efforts to manage the Second College Grant for multiple economic
and ecological benefits, Dartmouth seeks to develop a comprehensive base map that
can be used in planning all management activities. The college's Woodlands Office
contracted with the Spatial Analysis Laboratory (SAL) to develop a land-cover map that would satisfy this purporse.
The SAL decided to use a method that relies on aerial photography as its fundamental data
source, with aerial videography and field visits as secondary sources.
True-color photographs of the Second College Grant area were obtained. These photographs were scanned into a digital format, geo-referenced, and ortho-rectified to compensate for relief displacement. The geo-referenced and rectified images were then used to digitize discernible land-cover boundaries.
Each polygon in the resulting land-cover map was then assigned a specific land-cover category by re-examining the aerial photographs. The land-cover categories were adapted from a land cover/use classification system by Anderson et al. (1976), which is a numerical and hierarchical labeling scheme. The following categories were used for the preliminary land-cover map:
Code Description 2 Developed/Agricultural 5 Open Water 41 Deciduous Upland 42 Coniferous Upland 43 Mixed Upland 49 Recently Logged 612 Scrub/Shrub Wetland 621 Emergent Wetland 6111 Deciduous Wetland 6112 Coniferous Wetland 6113 Mixed Wetland 6121 Alder Thicket
The RECENTLY LOGGED category was used in cases where logging activities had removed the canopy vegetation and the previous land-cover type was unclear. However, if the previous land-cover type could be reasonably inferred from the surrounding land cover, the logged area was not delineated (e.g., a small clearcut in a uniform northern hardwoods forest would not be digitized).
When a polygon's representative land-cover type could not be determined from the photographs, aerial videography was used to check the areas in question, where possible. To facilitate examination of the video, an ArcView project was created to simultaneously illustrate the tentative land-cover map and the video flight lines. When a flight line crossed an area in question, individual points in the flight line were queried to determine the approximate Global Positioning System (GPS) location of the area. Because the GPS location is recorded on every video frame, the video could then be forwarded to the appropriate sequence. The videography permitted a more detailed examination of land cover, and in some cases could even be used for identifying individual trees to the species level. If a flight line did not cross an area in question, the aerial photographs were re-examined and the best estimate of land-cover type was used for each polygon.
At this point, the aerial photographs and videography had been examined to the fullest extent possible, and a final map using these methods was produced:
Click on image for enlarged view
Although specific forest communities and wetland types were known to exist in the Second College Grant and indeed had been examined during the field visit, they were not included in final map because they could not be reliably and consistently mapped throughout the entire study area. Instead, only general forest and wetland classes (e.g., deciduous upland, coniferous wetland) were used.
The lone exception to the problem of identifying specific wetland types was alder thickets; these wetland areas could be reliably identified and mapped in the study area, especially in broad river valleys (e.g., Dead Diamond River valley). Thus, a land-cover category for alder thickets was added to the final map.
Use of aerial photographs as the fundamental data source in a land-cover mapping project is a labor-intensive process. The process of digitizing land-cover boundaries is the most laborious phase, requiring careful study of landscape features prior to actual boundary delineation. Digitizing itself is a slow, methodical process that is sometimes complicated by poor photograph quality (e.g., some photographs are deeply shadowed or distorted by the rectification process). Furthermore, in a complex landscape like the Second College Grant area, where two centuries of forest use have created a mosaic of diverse forest types and ages, some land-cover transitions are gradual and therefore quite difficult to delineate. In such cases, the photo-interpreter must decide how best to represent an indistinct land-cover transition, adding an element of subjectivity that is not present in other land-cover mapping techniques.
The advantage of this method, however, is the level of detail; more land-cover categories can be interpreted from aerial photographs than from other commonly-used data sources, including Thematic Mapper (TM) satellite data. For example, a land-cover map derived from TM data, produced by the SAL for the entire state of New Hampshire, has the following classes:
Land-Cover Categories in TM-derived Map Non-forest Agriculture Developed/Barren Forest-Coniferous Dominated Forest-Deciduous Dominated Forest-Mixed Open Water Non-forest Wetland
Fewer categories are represented in the TM-derived map (7 vs. 12), particularly wetland classes; only the NON-FOREST WETLAND class is present in the TM-derived map while the photograph-derived map has SCRUB/SHRUB, EMERGENT, DECIDUOUS WETLAND, CONIFEROUS WETLAND, MIXED WETLAND, and ALDER THICKET.
Most of the other land-cover categories are similar or identical between the photograph- and TM-derived maps, and the DEVELOPED/BARREN category is the only component of the TM-derived map that encapsulates more detail than the photograph-based map. Although an equivalent category could have been developed using the aerial photographs, the "agricultural/developed" class was used instead to provide a general description for all human- dominated areas. If fact, the distinction between agricultural areas and other types of anthropogenic disturbances were quite evident on the aerial photographs, so numerous other categories could have been added if desired (with a corresponding increase in digitizing labor, of course). Thus, a knowledgeable photo-interpreter can identify landscape features, particularly various wetland types, that often cannot be distinguished by other methods. For a relatively small area such as the Second College Grant (the entire study encompassed about 42,500 acres), aerial photographs can be effectively used to map dominant landscape features as well as unique, small-area land-cover categories.
Although forest age distinctions and specific stand compositions could not be consistently mapped throughout the entire study area, the resulting land-cover map provides an effective overview of primary landscape patterns in the Second College Grant. It shows how general land-cover categories are distributed across the landscape and and illustrates frequently occuring combination of specific categories. In addition, this map can serve as a base map for additional mapping and modeling efforts (e.g., forest stand delineation, natural communties modeling).
The photograph-based methods described here produced a detailed description of the primary land-cover features of the Second College Grant. The resulting land-cover map provided an effective overview of landscape pattern and indicated where specific natural communities and forest stand types are likely to occur. Additional modeling would be required to add stand boundaries or to comprehensively identify and map all representative natural communities in the study area, but such steps would likely be necessary using any set of mapping techniques.
The labor-intensity of these methods was their most serious disadvantage; each photograph required careful study followed by methodical digitization. However, the map produced here contained more land-cover categories than a similar map produced from TM satellite data; in particular, photographs provided a more detailed assessment of wetland areas.
In future mapping efforts, a hybrid method that uses both remote-sensing data and aerial photographs may be worth considering. The state-wide map produced from TM data could serve as a base map for all general analyses of land-cover, and aerial photographs could be used to add more detail for specific areas of interest. Rather than digitizing all categories from scratch, photographs could be used to add boundaries for fine-level features to a map already containing TM-based land-cover categories. This combination of methods would significantly reduce labor while potentially lending equivalent or even superior detail to the final land-cover product (i.e., more time could be spent eliciting rare or unique features).
Anderson, J.R., E.E. Hardy, J.T. Roach, and R.E. Witmer. 1976. A land use and
land cover classification system for use with remote sensor data. U.S.
Geol. Surv. Prof. Pap. 964. 28 pp.