Hydrographic Generalization Tailored to Dry Mountainous Regions

Article excerpt


A framework for automated cartographic generalization is being developed by the U.S. Geological Survey (USGS) to support multiscale display and delivery of The National Map products. The framework must be able to handle the wide variety of climate and terrain conditions that exist in the United States. For instance, within the conterminous United States, elevations range from about 85 meters below sea level in Death Valley, California, to greater than 4000 meters above sea level in some parts of the Rocky and Cascade-Sierra Mountain ranges (USGS 2001), and average annual precipitation between 1961 and 1990 ranges from 0 to 508 centimeters (cm) [0 to 200 inches (in)] (Daly and Taylor 2000). Variations in terrain elevation, terrain roughness, and availability of water through precipitation comprise three significant factors in characterization of hydrography (Strahler 2010). Given the range of conditions, various natural landscapes have evolved and must be adequately represented in the multiple-scales of USGS cartographic products through the generalization framework.

The Center of Excellence for Geospatial Information Science (CEGIS) of the USGS is collaborating with the University of Colorado-Boulder and the Pennsylvania State University to develop and test automated processes that enrich, prune, and simplify high-resolution [l:24,000-scale (24K) or larger] data from the National Hydrography Dataset (NHD) to produce l:50,000-scale Level of Detail (50K LoD) NHD data (Cecconi et al. 2002; Brewer et al. 2009; Stanislawski 2009; Stanislawski et al. 2009; Buttenfield et al. 2010). The LoD is produced because the USGS has an objective to provide data for mapping over a very large range of scale, and it is logistically impossible to produce generalization solutions for all possible scales. The intention is to distribute data for mapping that have not been available before. Hydrography is a vector data layer which is highly sensitive to scale, and when generalizing this layer, changes in content and geometry are required more often than needed for other data layers, for example, transportation or settlement. Generalization processing for hydrography is computationally intensive as well. For both reasons, it makes sense to pre-compute versions of simplified hydrography at selected scales to fill in the scale "gaps" for mapping water features.

Producing a LoD requires four types of processing steps, which include pruning (also referred to as elimination or selection, which is removing partial or entire features on the basis of size, shape, or prominence), simplification (reducing or modifying coordinates to improve clarity), enrichment (adding attributes to support other procedures; in this case pruning, simplification, and display), and validation (comparisons to benchmark datasets and summary statistics). These steps will be further described later in this paper.

Through symbol redesign and elimination operations, the 50K LoD can represent hydrographic features on topographic maps ranging in scale from 1:50,000 to 1:200,000 (Brewer et al. 2009). Sequences of enrichment, pruning, and simplification operations that produce 50K LoDs have been tailored for six NHD subbasins, which were manually selected from six different climate (dry, humid) and terrain (flat, hilly, mountainous) regimes that span much of the conterminous United States (Buttenfield et al. 2010).

Once a generalization knowledge base for the NHD is developed, subsequent research will investigate terrestrial classification systems that can inform the tailored generalization sequences to furnish a database of blended generalization parameters, constraints, and operations that produce 50K LoDs with smooth transitions over landscape boundaries. This strategy, follows that of Burghardt and Neun (2006), who proposed a collaborative filtering approach to predict the best sequence of generalization operations for features from a knowledge base of successfully generalized features. …