Cartographic generalization uses a variety of methods to reduce map content and detail in a manner that legibly portrays desired features and conditions at a reduced map scale (Regnauld and McMaster 2007; Roth, Brewer, and Stryker 2011). Recent road generalization work within the scale range of 1:24,000 (24K) to 1:1,000,000 (1M) indicates that midrange scales (such as 1:100,000 to 1:300,000 (300K) are particularly difficult to represent through common display-based strategies, which eliminate road categories and use simpler, thinner line symbols for road representations (Brewer and Buttenfield 2010). This paper describes some recent efforts to improve road elimination and generalization when producing high-quality cartographic displays for multiscale topographic maps to be distributed through The National Map of the United States. Fundamental to this work is a general goal of fully automated and database-driven multiscale cartography; it is not an option for USGS cartographers to hand-select a few streets that run through the middle of each town to improve each of their maps, when the challenge is to map the whole country with limited staff and continual updates to national databases.
Various methods to thin road or hydrographic networks have been developed to support cartographic generalization, and research continues to refine these methods (Thomson and Brooks 2007; Stanislawski 2009; Savino et al. 2010; Touya 2010; Savino et al. 2011; Buttenfield, Stanislawski, and Brewer 2011; Gulgen and Gokgoz 2011). Invariably, the methods rely on database enrichment to establish selection criteria that quantify one or more characteristics deemed important to the thinning strategy. For instance, enrichment to define paths of best continuation (i.e., "strokes") or river courses have been used to establish selection hierarchies for network features (Thomson and Richardson 1999; Chaudhry and Mackaness 2005; Thomson and Brooks 2007; Touya 2007, 2010; Savino et al. 2011). Zhou and Li (2011) evaluated several stroke-building strategies. Using graph traversal techniques, hydrographic networks have been enriched with stream order, watershed area, or upstream drainage area values, which is subsequently applied for network thinning (Ai, Liu, and Chen 2006; Stanislawski 2009; Savino et al. 2011; Gutman 2012). Network thinning operations may be enabled or enhanced through enrichment computations of local line density (Stanislawski 2009; Buttenfield, Stanislawski, and Brewer 2011; Savino et al. 2011; Stanislawski et al. 2012), pattern (Heinzle, Anders, and Sester 2005; Heinzle, Anders, and Sester 2007; Touya 2007; Savino, Rumor, and Zanon 2011), and road network or block structure (Jiang and Claramunt 2004; Touya 2010; Gulgen and Gokgoz 2011).
The Thin Road Network tool is a relatively new generalization tool offered by Esri (Punt and Watkins 2010; Briat, Monnot, and Punt 2011; ArcGIS Resources 2012a). The tool applies a simulated annealing algorithm to filter road features based on the relative importance, significance, and density of the input features. Relative importance is determined from a hierarchical road classification field (e.g., Interstate, State Route, local road) that should be assigned for the features being processed. Significance is affected by the connectivity of the road network being processed, and is estimated from the number and length of possible itineraries in which each feature participates. An itinerary refers to alternative sequences of road segments one can use to traverse the road network efficiently.
A run of the tool populates a binary invisibility field to mark less significant road features that should not be displayed in the thinned network, while retaining visible features which maintain network connectivity, navigability, and the general morphology of road patterns. The degree of thinning is controlled by a minimum length parameter, which is an estimate of the shortest segment that is visually sensible to show for the desired scale. The itinerary length used for the visible/invisible decisions is approximately twice the minimum road length that the user provides during a tool run. Less important roads are thus removed from the display (but not from the database). Through trial and error, a user sets the minimum length that suits the scale of mapping to clear away short, tangled, and coalescing line segments, while retaining major thoroughfares for smaller scale representations. Esri offers guidelines for setting minimum lengths for specific mapping scales in its documentation (ArcGIS Resources 2012b), but these guidelines do not account for local variations in road density, as in a study area containing urban, suburban, and rural regions. The density of a group of road segments can be estimated as the ratio of the length of the segments to their associated area. Using the Thin Road Network tool with a single minimum length tends to homogenize road network density at smaller scales. Analogous to previous generalization work with hydrography (e.g., Stanislawski 2009; Buttenfield, Stanislawski, and Brewer 2011), we test the use of partitions for retaining differential road densities among areas in a map display.
Through generalization research on hydrography by the United State Geological Survey (USGS) Center of Excellence for Geospatial Information Science (CEGIS), pruning tools have been developed that enrich National Hydrography Dataset (NHD) flowlines (stream channels) with an upstream drainage area attribute (Stanislawski et al. 2007). The upstream drainage area attribute provides a criterion for pruning stream features to smaller-scale representations and for tapering stream symbols to improve map display. This paper applies a similar process to transportation data, working with minimum length parameters in the Thin Road Network tool to enrich road data with visibility attributes which support scale change and improve symbol hierarchies. Also, the paper details how road thinning and resulting invisibility attributes may be used to control the selection and prioritization of road labels.
Figure 1a shows the north half of the Atlanta, Georgia, road network dataset used in this research. The full dataset which covers sixty-one 7.5-minute, 1:24,000-scale (24K), quadrangle maps with 393,920 road segments. The dataset is part of the transportation layer in the USGS Best Practices (BP) data model (USGS 2006). The BP database includes two sources of transportation data that are used for map display: data compiled in 2011 by TomTom[R] North America, Inc., and data for US Department of Agriculture forest areas compiled by the US Forest Service. These BP transportation data are displayed on the US Topo 24K topographic maps and in The National Map viewer.
The Atlanta dataset was partitioned into three stratified road density classes (Stanislawski et al. 2012). To explain and demonstrate cartographic thinning and partitioning, a small subarea is selected (and shown by the rectangular subwindow with a bold outline; Figure la) north of the city center that overlaps with other processed data layers, including Level of Detail (LoD) generalized databases for hydrographic features, terrain shading and contours, land cover, labels for populated places, and emergency response structures. The subwindow covers approximately six quadrangles and contains 55,261 road features totaling 6043 kilometers (km) of roads. Figure 1a also shows three road density partitions calculated for the Atlanta region, which range through sparse (light gray), intermediate (medium gray), to dense (dark gray). A sample area shown in Figure 1b, used for many examples in this paper, is outlined by a white and black rectangle in Figure 1a and...