.|  Baltimore Ecosystem Study
Characterizing Fine-Scale Spatial Structure in the Gwynns Falls watershed - Relating Land Use and Land Cover Information to Landscape Position
 
(Much of this text is taken from: Tenenbaum, D.E., M.L. Cadenasso, L.E. Band, and S.T.A. Pickett. 2006. Using Transects to Sample Digital Orthophotography of Urbanizing Catchments to Provide Landscape Position Descriptions. GIScience and Remote Sensing. 43(4): 1-29.)
 
Introduction
A popular approach for describing the urban landscape is in terms of a land use/land cover classification, often generated through automated, per-pixel processing of remotely sensed data. This approach is sometimes difficult to operationalize, however, due to the diverse set of materials and objects that make up the urban landscape. That is, within a presumably homogeneous land use/land cover class, there is often a wide collection of different sorts of objects and materials that make up that class. For example, in a residential neighborhood we might find trees, lawns, sidewalks, roadways, cars, and rooftops. Furthermore, we might find these various land covers in any number of proportions and spatial arrangements. In order to better understand how urban areas function as ecological systems, the high-frequency variation and spatial structure in the urban landscape needs to be explored, particularly with reference to the landscape position of different land covers. Characterizing the fine-scale spatial structure of the urbanizing landscape will be helpful in better understanding its ecology, and is a necessary step in the process of building useful simulation models to further study these landscapes.
 

Figure 1. Example of a transect.
Rather than attempting to produce a per-pixel land use/land cover classification from digital orthophotographs, a description of the finer spatial structure can be obtained through a sampling approach. Specifically, the sampling can be conducted through the use of randomly oriented transects placed on the orthophotographs. The transects are then visually interpreted, subdividing them into segments that describe the extent of a certain land cover along the transect (Figure 1). This approach can provide information on proportion and adjacency of land cover materials. Furthermore, this approach will be compatible with various methods of generating quantitative descriptions and statistical interpretations, including those designed to address questions of edge and adjacency of land cover materials, as well as the generation of statistics that relate the position of particular segments to available co-registered raster datasets.
 
In order to facilitate the transect sampling of digital orthophotography, a geographic information tool is needed. We have developed such a tool, ArcTrCS, in an extensible GIS, and have applied this tool to characterize two urbanizing catchments in suburban Maryland. While no available GIS provides the capability to perform this sort of sampling using its built-in functionality, several are suited for the development of such a capability through scripting languages, which provide for the creation of custom applications.
 
Study Sites

Figure 2. Study sites.

Figure 3. Glyndon Catchment.

Figure 4. Upper Baisman Run Catchment.

 
Gwynns Falls provides a broad cross-section of forms of urbanization, such that the full range of forms of development found in the metropolis can be studied in the context of a single watershed (Figure 2). Located at the headwaters of Gwynns Falls is the Glyndon Catchment. Glyndon is an urbanizing catchment with an area of approximately 81 hectares (Figure 3). Development is primarily in the form of low- to medium-density residential development, with some other land uses in small proportions. The Upper Baisman Run catchment is a 63.36-hectare basin located approximately 10 kilometers north of Baltimore (Figure 4). The catchment’s land cover includes both low-density residential development in the upper portions of the catchment, along with deciduous forest that developed subsequent to agricultural use in the early parts of the 20th century in the lower reaches. While the Upper Baisman Run does feature some development, the intensity of that development is considerably lower than that found at Glyndon.
 
Results
Examining the proportions of land covers as determined by transect sampling in the two catchments confirms what can be seen by observing the RESAC LULC map or from cursory examination of the digital orthophotography: Both catchments have a significant amount of woody vegetation (about 45% of the catchments’ areas), and a sizable proportion of herbaceous vegetation as well (about 30% in Glyndon and 40% in Upper Baisman Run), although Upper Baisman Run has about 10% more herbaceous vegetation than Glyndon. Upper Baisman Run has extensive private lawns because it is contains very low-density residential development. Another difference between the catchments can be seen in the proportions of road and pavement found in each of them: Glyndon’s higher intensity development is reflected here as well, with roads comprising nearly 6% of the catchment, whereas in Upper Baisman Run this class only is about 3% of the catchment. Proportions in the pavement class would be comparable, except that Glyndon contains parking lots that are almost 7% of the catchment themselves. Glyndon does not contain only residential development; it contains commercial and other institutional land-uses, whereas Upper Baisman Run does not. Glyndon also contains a larger proportion of the catchment area as structures (about 8%) for the same reason, whereas Upper Baisman Run, with its very low density residential development only has about 3% of its area in the structure class.
 
Examining the cumulative frequency histograms for the terrain statistics of each primary class in the two catchments provides an opportunity to identify tendencies in placement of certain land covers with regard to their landscape positions in the catchments. These two catchments are extremely flat, with the majority of cells having D∞ slope angles of less than 1°. It is only through the use of LIDAR spot elevation data, collected at extremely high vertical and horizontal resolutions, that it is even possible to accurately generate such shallow slope angle values such that these histograms can be produced. In Glyndon, the shape of the histograms for each primary class is more or less indistinguishable. There appears to be no tendency in the placement of different land covers according to slope angle, although this is not surprising given how shallow a range of slopes is present here. However, in Upper Baisman Run, we can distinguish between classes to some extent, with roads and lawns notably being located in flatter locations than other classes. Histograms showing the shape of the D∞ specific catchment area distribution (log-transformed so as to linearize the range of values to some extent) show that structures in Glyndon tend to be located in cells that receive more flow than other classes in Glyndon, whereas in Upper Baisman Run, a greater proportion of the structure cells are at lower specific catchment area values, meaning that structures in this catchment have been placed in locations that do not receive as much drainage as other classes’ cells. This difference is likely a reflection of the comparative development densities in these two catchments: Upper Baisman Run has been developed in a low-density residential pattern, with the houses located quite deliberately in upland locations, whereas Glyndon is developed at a considerably higher density, with the structures not located on extensive lots where there is ample opportunity to select the house site in this fashion. This may be a reflection of differing regulatory conditions that were in place at the time of development within each catchment (Upper Baisman Run was developed considerably more recently), or simply a product of the fact that Glyndon features a pattern of development that one would expect of a location closer to the metropolitan area (higher density). On the other hand, Upper Baisman Run is located farther from the city, and features considerably larger lot sizes as is often the case in locations farther into the country. Histograms of the D∞ topographic moisture index, a product generated from the slope and specific catchment area, have the same form as the specific catchment-area plots.
 
Conclusions
Transect sampling of digital orthophotography of two urbanizing catchments in suburban Maryland has provided a great deal of useful information about land-use/land-cover configuration in these catchments, specifically with respect to the landscape position of certain land-cover materials and associated implications of the sequence of those land-cover materials along hydrologic flow paths. The sampling density and transect length selected produced stable estimates of land cover proportions when viewed in terms of the proportions of primary classes.
 
Publications
Tenenbaum, D.E., Cadenasso, M.L., Band, L.E., and S.T.A. Pickett. 2006. Using Transects to Sample Digital Orthophotography of Urbanizing Catchments to Provide Landscape Position Descriptions. GIScience and Remote Sensing. 43(4):323-351.