1 Introduction

Visibility research, calculating lines-of-sight and viewsheds, on terrain databases, is an established, and important, field in GIS. Nevertheless, progress is still possible, since both more powerful Unix workstation hardware, with tens of megabytes of memory and software environments, and much larger amounts of input data are available. The research reported here would not have been feasible on mainframes nor possible on personal computers. Ninety meter resolution coverage of the US is available from the USGS, and of other regions for appropriate purposes from DOD, in the Digital Elevation Model (DEM), or equivalently, the Digital Terrain Elevation Data (DTED) format. Higher resolutions can be available for specific purposes. Synthetic aperture radar can produce elevation data at a 1 meter resolution; about the biggest problem here is how to store it. Elevation data of other planets, such as Mars, is available.

Some notable earlier work is as follows. [Wood and Fisher, 1993] consider the problem of elevation data accuracy. [Fisher, 1992,Fisher, 1993] discusses viewshed uncertainty, while [Lee et al., 1993] describe what data errors do to feature extraction. [Lee, 1992] considers how high visibility points relate to topographic features. For a definitive survey listing older references, see [Nagy, 1994]. [De Floriani et al., 1986] give some of the relevant issues in using Triangulated Irregular Network data structures, instead of the raster databases used here.

The long-term goal of this work is to facilitate development of an automated procedure to place facilities, given information on terrain and the area to be defended for the US Army Topographic Engineering Center (formerly called the Engineer Topographic Labs), part of the Corps of Engineers, Ft. Belvoir, Virginia. Army engineers need to know where to place facilities, or alternatively, where to hide. In fact, students at West Point are taught how to calculate this approximately by hand on a contour map. Some techniques are described in [Shannon and Ignizio, 1971].

If this is used to locate facilities, then, after calculation, then the fraction of the potential viewshed which is visible from the observer, or the visibility index, may be postprocessed to select points that also have other desirable properties, e.g., that are accessible on the ground.

However, many other applications are possible for this work.

1. Cellular telephone companies, and other radio transmitter operators, wish to optimize their antennas' locations. This application also illustrates the utility of having the observer and target heights being different. Eventually, these Line of Sight algorithms might be modified to account for attenuation effects as described in [Conrad, 1993].

2. Natural resource extractors may wish to site visual nuisances, such as clearcut forests and openpit mines, where they cannot be seen from public roads. Also, zoning laws in some regions, such as the Adirondack Park of New York State, may prohibit new buildings that can be seen from a public lake.

3. Finally, if other planets, such as Mars, are being explored by semi-autonomous roving vehicles communicating with a fixed ground base, then we need to know where to put that base, and where it is safe to rove around it. [Gennery, 1989] describes matching a local terrain description obtained from a Mars rover with a more global terrain description from an orbiting probe. Since the terrain points of higher visibility help to characterize a surface, matching these particular points would make the correlation-based approach even more effective.

The short term goal of this work is extracting key visibility information from the mass of terrain elevation data in a concise form that would facilitate the use of any siting model, that is, calculating a viewshed.

We are attacking a narrow technical problem here, and ignore many important issues that must be considered in an actual useful system. This includes data reliability, the effect of vegetation, light refraction, etc. Our strategy is to study one facet deeply first. Nevertheless, this does affect other facets, e.g., we have observed that points with the highest visibility seem to be least affected by uncorrelated errors in the elevation data.

Much of the work reported here is from [Ray, 1994]. This is a preliminary paper; some conclusions might be modified as we proceed.