Land Cover Mapping of the Chesapeake Bay Watershed

Background

The Regional Earth Science Applications Center (RESAC) at the University of Maryland, College Park facilitates interactions between researchers and a broad base of end-users working on resource management issues in the mid-Atlantic region using geospatial data and technologies (Goetz et al. 2001). The RESAC has developed new land cover and land use maps using remotely sensed data and new mapping technologies for the Chesapeake Bay watershed and the intersecting counties (Fig.1).

Methodology

In order to classify the multitemporal and multispectral data, the RESAC uses decision tree algorithms (Breiman et al., 1984; Quinlan 1993), which have several advantages over traditional classification techniques. There is an increasing recognition of the power of these advanced machine-learning techniques for land cover classification. The techniques used by the RESAC are under constant review. In cooperation with the United States Geological Survey (USGS) National Land Cover Data (NLCD) team, the RESAC is working towards improving land cover information in the Mid-Atlantic Region and advancing the capabilities of the Landsat archive for this application. The current version has been adopted by NLCD to replace the MRLC land cover map, (http://landcover.usgs.gov/).

The RESAC has selected two eras for land cover mapping; one centered on 2000 and the other on 1990. Eras are used rather than specific years because adequate data are not always available for the target year. The increased rate of acquisition of Landsat 7 ETM+ data have made it possible to use multi-temporal imagery, which makes the use of seasonal phenology to distinguish cover types feasible, in additional to the spectral radiances. Landsat 7 has provided adequate repetitive coverage to prepare mosaics for three seasons of the year. Landsat 4 and 5 were the satellites that provided data in the 1990 era and fewer scenes were acquired, so it is only feasible to develop consistent mosaics for two seasons. Furthermore, because of the lower acquisition rate of Landsat 4 and 5, a wider spread of years must be used to obtain completely cloud-free coverage for two seasons.

Figure 1. Landsat scenes used for the Chesapeake Bay Watershed (green) and intersecting counties (yellow).

Three map types are being prepared, a land cover map, a tree cover map and an impervious surface map for each era, six maps in all. All maps have a grid cell resolution of 30m and a minimum mapping unit of 1ha. All are developed from the same Landsat data archive. In order to undertake mapping of a large area such as the Chesapeake Bay Watershed, multiple Landsat scenes are acquired (Fig.1b) and an intensive pre-processing is undertaken to orthorectify the data to UTM coordinates and remove disturbing effects such as clouds, cloud shadows, the effects of steep slopes and radiometric variation between acquisition dates.

The land cover map

The land cover classification is modified from Anderson Level II (Anderson et al., 1976). A decision-tree machine leaning algorithm is used for classification that produces an explicit, hierarchical tree that can be used to classify additional data that have similar properties to those for which it was constructed. The methodology developed by the RESAC is described in Varlyguin et al. (2001).

Figure 2. RESAC land cover map of Chesapeake Bay Watershed showing the classes mapped, and an enlarged segment over Washington, D.C.

The impervious surface map

The RESAC team has advanced the capabilities of the Landsat series of satellites to measure the amount of impervious surface with in a 30m pixel. Impervious surfaces include all surfaces (man-made or natural) that inhibit infiltration by rainfall. The sub-pixel classification technique used by the RESAC assigns a percentage value (between 0 and 100%) to each location based on the spectral measurements of the ETM+ sensor. The impervious map is new, and is a significant improvement on older, per pixel maps in which each pixel was either impervious or not. It is also superior to maps in which a land cover map is used with an arbitrary, fixed value of imperviousness assigned to each class, and the same value is used for each class throughout the map. The new maps have found applications not only in the study of surface water redistribution, runoff and pollution (Goetz et al. 2003), but also in monitoring development, and are used to measure and model sprawl (Jantz et al. 2003).

a.	b.
Figure 3. Impervious surface map of (a) the Chesapeake Bay Watershed and (b) an enlarged view of Richmond, VA . Color scale; dark red (high proportion impervious), yellow (low). Each 30m pixel has a percent imperviousness value associated with it.

a.	b.
Figure 4. (a) Map of New York City. (b) Four years NVA/SMD.

The tree cover density map

The RESAC team has developed a new technique for mapping tree density using information gathered by the Landsat 7 satellite. The tree cover map shares some characteristics with the impervious surface map in that measurements are taken directly from the sensor and processed through a regression tree algorithm to produce estimates of tree density on a scale from 0-100%. The map is sensitive to features that are overlooked in traditional land cover maps, in which some forested land cover classes are placed in non-forest classes and are therefore not incorporated into estimates of forest cover. This ability to discriminate small patches of trees and partially tree-covered pixels greatly improves the discrimination of forests. The new maps have been used in applications related to connectivity of resource lands, and in preliminary assessments of standing above-ground biomass for carbon accounting.

a.	b.
Figure 5. (a) Map of tree cover in the Chesapeake Bay Watershed and intersecting counties. (b) An enlargement of the Shenandoah Valley and Harrisonburg showing agricultural and urban areas in black and tree cover in shades of green.

a.	b.
Figure 6. Harrisburg PA. The impervious surface and the tree cover density maps can be combined (a) to characterize the complex cover types created by variation in density of development (imperviousness) and tree cover. The tree cover map alone (b) shows how land cover classes in and around complex, fragmented environments may, nevertheless, form continuous corridors of forest cover.

References

Breiman, L., J. Freidman, R. Olshend, and C. Stone (1984). Classification and regression trees. Monterey, CA: Wadsworth.

Goetz, S.J., S.D. Prince, M.M. Thawley, A.J. Smith, and R. Wright (2000). The Mid-Atlantic Regional Earth Science Applications Center (RESAC): an overview. Available at www.geog.umd.edu/resac and on ASPRS CD-ROM in American Society for Photogrammetry and Remote Sensing (ASPRS) Conference Proceedings, Washington DC.

Goetz, S. J., R. Wright, A. J. Smith, E. Zinecker, and E. Schaub. 2003. Ikonos imagery for resource management: tree cover, impervious surfaces and riparian buffer analyses in the mid-Atlantic region. Remote Sensing of Environment (in press).

Jantz, C.J, S J Goetz, A.J. Smith, M. Shelly (2003). Using the SLEUTH Urban Growth Model to Simulate the Impacts of Future Policy Scenarios on Land Use in the Baltimore-Washington Metropolitan Area, Environment and Planning (B) (in press).

Smith, A. J., S. J. Goetz, S. D. Prince, R. Wright, B. Melchoir, E. M. Mazzacato, and C. Jantz. 2003. Estimation of sub-pixel impervious surface area using a decision tree approach, Ikonos and Landsat imagery. Remote Sensing of Environment (forthcoming).

Varlyguin, D., R K Wright, S J Goetz, S D Prince (2001). Advances in land cover classification from applications research: a case study from the mid-Atlantic RESAC. Available at www.geog.umd.edu/resac and on ASPRS CD-ROM in American Society for Photogrammetry and Remote Sensing (ASPRS) Conference Proceedings, Washington DC.