Automation of GIS-Based Population Data-Collection for Transportation Risk Analysis

Estimation of the potential radiological risks associated with highway transport of radioactive materials (RAM) requires input data describing population densities adjacent to all portions of the route to be traveled. Previously, aggregated risks for entire multi-state routes were adequately estimated from population data with low geographic resolution. Current demands for geographically-specific risk estimates require similar increases in resolution of population density adjacent to route segments. With the advent of commercial geographic information systems (GISs) and databases describing highways, U.S. Census Blocks, and other information that is geographically distributed, it became feasible to determine and tabulate population characteristics along transportation routes with 1-kilometer resolution. This report describes an automated method of collecting population data adjacent to route segments (for calculation of incident-free doses) based on a commercial GIS. It also describes a statistical method of resolving remaining resolution issues, and an adaptation of the automation method to collection of data on population under a hypothetical plume of contamination resulting from a potential transportation accident. Introduction Estimation of the potential radiological risks associated with highway transport of radioactive materials (RAM) by use of the RADTRAN computer code [1] requires input data describing population densities adjacent to all portions of the route to be traveled. Typically, population data have been obtained from the HIGHWAY routing code [2], which provides distanceweighted-average population densities in three categories (Rural, Suburban and Urban) along a route or route segment of interest. Population densities in HIGHWAY were derived from U.S. Census Tract data to describe the population within 1⁄2 mile (0.8 km ) of the route centerline for incident-free risk analyses. These population densities provide adequate accuracy and detail for analysis of an entire route of typical length (hundreds of kilometers or more). However, “stakeholders” often focus on particular points of interest or short route segments (e.g., through cities, towns or other population concentrations) of critical concern to them. In addition, Executive Order 12898 now imposes a requirement on risk assessments that the “environmental justice or equity” of a proposed action be assessed. The underlying structure of the population database employed in the HIGHWAY code does not permit resolution of small-scale populationdensity variations. Also, the fundamental purpose of HIGHWAY (routing) requires that routesegment definition be based on highway intersections, not population variations. A further concern regarding the population model for accident-risk analysis is that the analysis of risks associated with potential transportation accidents which might involve RAM packaging and result in the release of a plume of contamination, as it is implemented in RADTRAN, calls for entry of population densities under the plume “footprint.” The same population data (i.e., the population densities in the immediate vicinity of a route) have been used in most analyses. Since a plume may travel as far as 50 miles (80 km) from the point of an accident, the current practice, while conservative, does not directly model the downwind population and has been criticized, therefore, as inadequate. With the advent of commercial geographic information systems (GISs) and databases describing highways, U.S. Census Blocks (identified as “block(s)” in this report), and other information that is geographically distributed, it became feasible to determine and tabulate population characteristics along transportation routes with 1-kilometer resolution and to tabulate any population-related variable included in the block data. A preliminary study of population densities along the proposed WIPP routes in New Mexico revealed that the desired population densities could in fact be tabulated along any major highway and most minor highways on a kilometer-by-kilometer basis. However, the process was labor intensive (though much less so than creating the HIGHWAY databases). Also, in rural areas and most suburban areas, many blocks extend over distances which were large in comparison to the 1-km scale of interest in collecting population density data. This report describes an ancillary code [3] for use with a GIS to automatically tabulate kilometer-by-kilometer population densities along any selected route segment. Second, it describes a statistical method of estimating population densities near the route segment from the average densities in blocks (or groups of blocks) extending significantly more than 1 km from the route segment. Third, it presents a method of describing population density under a hypothetical plume as a function of downwind radioactive-aerosol concentration that is based on an adaptation of the same ancillary code. Geographic Information System The particular GIS employed in this study is ArcView which embodies a graphic user interface and a subset of the functionalities of ARC/INFO; both are products of Environmental Systems Research Institute, Inc. (ESRI). The capabilities inherent in the ArcView system include display of multiple maps of features and data that are geographically distributed (e.g., highway maps, block boundaries, household locations, county boundaries, etc.), selection of one set of features which is in a specified geographic relationship to another set of features (e.g., blocks within 0.8 km of a selected highway, or blocks intersected by a graphic figure), and tabulation of the characteristics of identified features (e.g., population count, population density, household count and area within identified groups of blocks). Functionality may be expanded through incorporation of ancillary codes (scripts) supplied by users or by ESRI. Correction for Large Census Blocks As mentioned earlier, a preliminary application of the GIS-based method of population-data acquisition to the proposed WIPP route in New Mexico (primarily I25 from Colorado south) revealed many instances of blocks or groups of blocks intersected by a 1.6-km-rectangle (1.0 along the route and 1.6 km wide, centered on the route) with areas much larger than 1.6 km. A single block of large area was not likely to be populated uniformly, and the geographic distribution of population within a block was not specified in block data. Thus, precise determination of the population within 0.8 km of the route centerline was not possible. Simple approximations, such as (1) the average population density over a block or group of blocks, or (2) the assumption that all of the population in selected block(s) lies within the 1.6-km bandwidth were expected to yield either underestimates or gross overestimates (by factors of 10 to 100 according to data described below), respectively. Underestimates are unacceptable because they do not yield conservative risk estimates and large over-estimates are unacceptable because the majorities (∼90%) of most routes are Rural or Suburban in character. There are several means of accurately determining population distribution within an area of interest: direct surveys, aerial photographic surveys and special U.S. Census databases. All three expensive and the U.S. Census databases are not always publicly available. The most economical approach for the present need was to acquire ArcView-compatible databases of the coordinates (locations) of households. An initial database of household locations for the 911 emergency system in McKinley County, NM, was the first such database used in this study. It was possible to demonstrate with this database that a distribution of population-density ratios, 1.6PD/AvgPD, could be developed. The “1.6PD” population density was computed by using the GIS to identify the number of households within the 1.6 km area of a selected rectangle and multiplying this number by the number of persons per household, as determined from the block data. Dividing this number by 1.6 km yielded an estimate, with acceptable accuracy, of the population density within 0.8 km of the route for the selected kilometer. The average population-density value (AvgPD) was derived from the total number of persons and total area of the block(s) intersected by the rectangle. Because the number of 1.6PD/AvgPD values obtained from this limited data set was small and related to a particularly rural area, more extensive and varied data sets were sought. A suitably large database of residence locations was obtained from the Houston-Galveston Area Council (HGAC) in Texas, which provided properly configured coordinates of customers within the Houston Lighting & Power Service Area (covering parts or all of 12 counties). Because these customers were approximately 80% residential and located along Interstate, US, and State highways, their locations provided the data needed for definition of a credible 1.6PD/AvgPD distribution function. Fifteen highway segments of varying length (11 to 59 km) were analyzed to provide 498 values of 1.6PD/AvgPD. Figure 1 shows 15 cumulative distributions of these ratios, derived from histograms calculated for each of the 15 data sets, plus a lognormal distribution function which was fitted visually to the 15 plots. These 15 data sets are distinguished by road type and county; each may traverse Rural, Suburban or both categories of population density. 0.00 0.20 0.40 0.60 0.80 1.00 1.20