Common objectives for cartographic generalization include preservation of cartographic and geographic logic. Cartographic logic refers to the condition that the smaller scale data version retains levels of detail which meet visual expectations. Essentially this means that the simplified data "looks right" in the context of other map information. Geographic logic is retained when generalized data versions reflect evidence of their site and situation in the landscape. For example, a smaller scale representation of an arid landscape must preserve ephemeral stream characteristics such as channel discontinuity and the presence of playas and washes. A depositional coastline processed for smaller scales should preserve the regular periodicity and scalloped character of barrier beaches. Generalization processing to meet these objectives often involves modifying data geometry, symbolization, or both. Satisfactory results can be achieved for data spanning small scales or example landscapes with uniform characteristics, as evidenced in example illustrations found in many American cartographic textbooks (Slocum et al. 2009; Dent 1999; Robinson et al. 1995). Larger regions with diversified landscape characteristics present more of a challenge for a national mapping agency such as the United States Geological Survey (USGS) in developing a generalization strategy.
The premise of this research is that a single automated generalization sequence with uniform tolerance parameters cannot create adequate reduced scale representations in all types of heterogeneous landscapes encountered across the United States. There are important implications in this approach for data production and for cartographic design at multiple scales.
Automated generalization processing and data modeling will reduce workloads and improve consistency of results, but may require special expertise. Some data layers (terrain and hydrography) are more sensitive to scale change than others (transportation and settlement) and must be generalized at more frequent scale intervals to produce useful data products and readable maps. Analytical uses of reduced scale data carry additional needs and requirements, to support reliable data measurements, and to ensure that features integrate horizontally (within layers) as well as vertically (between layers) (Bobzien et al. 2008; Buttenfield and Frye 2006; Spaccapietria et al. 2000). Consistent data modeling mandates metric assessment of generalized data versions, to ensure reliability of measured geometric characteristics at all levels of resolution.
This paper reports on generalization and data modeling to create reduced scale versions of hydrographic data for The National Map (http://nationalmap.gov) of the USGS. The work draws upon several years of stepwise efforts by the authors to estimate upstream drainage area (UDA) for every stream reach between confluences (Stanislawski et al. 2007), to automate stream pruning on the basis of local density (Stanislawski 2009), to quantify reliability of generalization results (Stanislawski et al. 2010a; Buttenfield et al. 2010), as well as for visual evaluation of mapped hydrography (Brewer et al. 2009). The paper demonstrates that generalization processing can be varied to preserve local or regional differences in hydrographic characteristics that reflect natural variations in landscape type. Specifically tailored processing sequences generalize data compiled for the National Hydrography Dataset (NHD) at 1:24,000 (24K) scale. Results are evaluated metrically against benchmark NHD data compiled for l:100,000 (100K) scale subbasins. Terminology and concepts common to United States hydrographic data such as towlines, reaches, and subbasins may be reviewed at the NHD website (http://nhd.usgs.gov/documentation. html), with a particularly helpful overview in the chapter called Concepts and Content (http://nhd.usgs. gov/chapter1/chp1_data_users_guide.pdf).
Hydrographic data is chosen for a number of reasons. It comprises the vector data layer most sensitive to changing spatial resolution. It is characterized by having the most stringent requirements for vertical integration with terrain, so that streams run along valley bottoms and not up the sides of ridges for example. As such, hydrography is expected to manifest the most difficult data modeling problems in generalizing vector data. In addition, hydrography is commonly utilized in topographic base mapping at every scale, and will be in high demand by users of The National Map.
Establishing a Reliable Physiographic Context
The United States is large, and comprises diverse physiographic regions (Figure 1a). Initial results by the authors of this paper (Buttenfield et al. 2010; Stanislawski et al. 2009; Brewer et al. 2009) led to the argument proposed here that landscape differences which reflect local physiography and local climate require differing generalization sequences for effective multiscale representation. The traditional resource cited for defining United States physiographic regions is Fenneman and Johnson (1946), whose divisions were created manually and at a relatively coarse resolution. Relying solely on the Fenneman and Johnson physiographic divisions however does not reflect enough spatial variability at the subbasin level to model realistic transitions for generalization strategies across the range of conditions in the country. Consequently, alternative landscape delineation approaches are needed for hydrographic generalization of the United States.
Touya and others (Touya 2008; Touya et al. 2010) have proposed an implementation of context-specific processing applied to subjectively determined urban landscape delineations. Their solution is based on terrain and transportation characteristics. Other research has been completed on automatic delineation of landscape partitions with specific characteristics, to help orchestrate choices among a set of automated generalization operations for large and/or varied datasets (van Oosterom and Schenkelaars 1995; Bobzien et al. 2008; Chaudhry and Mackaness 2008a, 2008b; Fathi and Krumm 2010). Progress in multiscale morphometric approaches are also reviewed by Deng (2008) with an emphasis on environmental modeling goals. But none of these approaches can account comprehensively for the wide range of terrain and climate conditions within the United States, which form diverse hydrographic conditions.
Chaudhry and Mackaness (2008c) apply morphometric analysis to build extents of mountain ranges from individual peak locations. Their objective is to derive "morphostructural regions" suited to smaller scale mapping. Multiscale morphometric analysis approaches typically encompass landscape variation across distances ranging from tens of meters to approximately one kilometer (e.g., Schmidt and Andrew 2005) and with differences resulting from focal windows ranging from 3x3 pixels to 75x75 pixels (which covers 3,700 ground meters at the working scale of Fisher et al. 2004). In these contexts, scale change refers to DEM resolution change, but is still focused on automatic identification of parts of landscapes such as peaks, ridges, passes, plains, channels, and pits (Wood 1996). This level of detail is much finer than the subbasin-based approach applied in this paper, and much too fine to process hydrography for the entire United States in a manageable way. The aim in this research is not, for example, to differentially generalize the opposing sides of every ridge and valley in the United States at the resolution of individual formations. Such a data processing task could not be completed within a reasonable update cycle.
[FIGURE 1 OMITTED]
Regional classification based on hydrography remains a challenging problem, because water channels are quite sensitive to terrain roughness, precipitation and other factors (Carlston 1963; Montgomery and Deitrich 1989; Tarboton et al. 1991; Tucker and Bras 1998). The premise of the research reported here is that differences in hydrographic pattern cannot be preserved across all variations evident in the national landscape using a single uniform processing sequence. Tailoring individualized generalization sequences to each subbasin would be unmanageable, of course. The middle path is to establish a set of terrain and climate characteristics that reflect the primary hydrographic patterns, and use these to regionalize the national...