Case Study:

SPARTACUS

Methodology

GIS Raster Model

Overview

The modeling of locally important environmental processes and their social implications requires the exact spatial location of input information and output data. Raster cells or pixels can be used as a spatial reference framework for environmental and social impact models. Most available data and most urban models, however, are spatially aggregate and work within a zonal (polygon) framework. The Raster module disaggregates zone-level and network-level data into individual raster cells. In the SPARTACUS project, each cell is 100 meters (328 feet) square. This dimension was selected to allow a sufficiently detailed level of analysis but not to overwhelm the memory and speed of the computer.

For a typical U.S. application, disagreggation of data can be done using tools such as ESRI's ArcView Spatial Analyst or MapInfo Professional and SpatialWare software. While disaggregation can be performed with a simple command, some data processing may be required, depending on the form of the data. For example, zonal population data must be converted into densities by dividing by the area of each zone, then disaggregated to assign the densities to grid cells. Finally, the density of each grid cell is multiplied by the area of the grid cell to obtain the total population in each grid cell. The SPARTACUS study used more sophisticated simulation techniques to disaggregate land use data, as described below.

The Raster representation of households and employment data has a number of functions in the SPARTACUS modeling system, including:

• The localization of intrazonal and access trips;

• Calculation of land coverage;

• As a proxy for barriers in the dispersion models; and

• For consideration of households as recipients of pollutants and noise in the exposure models.

To identify equity impacts, households can also be disaggregated by socioeconomic group. In the SPARTACUS study, three socioeconomic groups were defined, based on occupation. The definition of socioeconomic groups could also rely on other characteristics, such as income, race, or mobility limitations.

The relationships among the databases, MEPLAN output, Raster module, and calculation of environmental and social indicators are shown in Figure 3.

Figure 3. Raster Structure

Disaggregation of Households and Employment

The disaggregation process begins with zone-level household and employment data for the base year and for the forecast year for all policy scenarios tested. If no information on the distribution of population and employment within the zone is available, data can be allocated equally to each grid cell within the zone (represented in the GIS as a polygon). If some information on the distribution within the zone is available, the disaggregation can be performed with greater accuracy. This is done by assigning a weight to each grid cell, which is used to allocate total population or employment proportionally (Figure 4). If sufficient data are available, population can be given different weights by socioeconomic group.

Disaggregation of data in the SPARTACUS was performed using microsimulation techniques. Each single activity (such as place of residence or work) was assigned a raster cell by applying Monte Carlo Simulation. The simulation technique avoids the problem of small numbers, i.e., how to allocate 50 households to 300 raster cells. Different densities within each zone were also assumed, using information on land use types.

In the U.S., sources of information on the distribution of activity within the zone might include:

• Population at the census block group or block level, which may provide a finer level of detail than is available at the TAZ level. The population of each block or block group can be equally allocated to the raster cells within that block or block group.
• Parcel-level land use databases that identify the square footage of development on each parcel by land use type. Some U.S. cities, including Buffalo, Orlando, and Portland, OR, have developed area-wide parcel-level databases. The square footage or number of units of development on the parcels (or portion of a parcel) within each raster cell can be used as the weight for allocating population or employment to the cells.
• A GIS-based zoning map, which can be used to estimate the distribution of land uses even if actual development is not known. For example, different weights for population could be assigned to areas that are zoned multi-family, single-family, or industrial. An example of the use of GIS-based land use data to refine population allocations is provided in the Orange County Case Study.

For the forecast years, the distribution of development within the zone may change. If additional development (as measured in floorspace) is forecast in the zone, it is allocated to both developed and undeveloped cells. In zones where the amount of floorspace decreases, the allocation weights remain the same and overall activity is reduced. If specific information is available on the locations of development planned for construction or elimination, this can also be used.

A uniform disaggregation of zonal data to grid cells can be done with a minimum amount of effort. As the number of judgments to be made in the disaggregation scheme increases (e.g., assigning weights by zoning classification), the level of effort will increase as well. The amount of time required for non-uniform disaggregation will also vary considerably depending upon the format and quality of the GIS databases used for the disaggregation and the amount of data processing involved.

Figure 4. Disaggregation of Polygon Data to Grid Cell Level

Source: Commission of the European Communities, 1998.

Disaggregation of Networks

Network data are disaggregated by linking MEPLAN output with the information in the GIS database. For each link in the MEPLAN network, the number of cars, trucks, and buses, the link type, and average speed are transferred from the model. The GIS database contains the model node numbers and the alignment for each link; the node numbers are used to merge the GIS information and the model output. If the network is not contained in a GIS format, it can be approximated by considering links as straight lines between the network nodes.

Once the information is transferred, the raster grid is laid over the network. Each cell touched by a network link receives the information assigned to the link; if the cell is touched by more than one link, the loads are summed (Figure 5).

Figure 5. Conversion of Network Data to Grid Cell Data

Source: Commission of the European Communities, 1998.

In addition to trips on the network contained in the urban model, neighborhood trips are also estimated for each grid cell. Trip-ends are assigned to each cell in proportion to the total activity in each cell; each trip is assumed to go straight to the nearest raster cell that belongs to a link (limited-access highways excluded). All raster cells touched by this trip increase their load by one vehicle. Households are used to assign trip origins, and employment is used to assign trip destinations. A similar technique is used to assign intrazonal trips.

The result is five raster layers representing urban traffic: the raster load for cars, trucks, and buses; dominant link type, and average speed for each cell. The cells are considered as point sources for emissions in the environmental models.

Land Use Cover

The proportion of land cover which is impermeable is an important environmental indicator for a number of reasons. Impermeable cover increases runoff in drainage and sewerage and can affect water quality in streams and rivers in the watershed, as pollutants such as oil and sediment are collected in the runoff. Impermeable cover also decreases the renewal rate for groundwater, increases the risk of floods, and indicates the area and space taken for human use at the expense of plant and animal species.

Within the Raster module, the percentage of impermeable cover in each cell can be estimated based on land use categories as well as population and employment. In the SPARTACUS project, functions relating impermeable land to land use type (residential, mixed-use, or industrial) and to total population plus employment density, based on recent German research, were used. Since actual land coverage factors vary considerably, this provides only a rough estimate, but one that should at least be suitable for assessing differences among alternative policies and land use scenarios.

Exposure To Air Pollution

The Raster module in the SPARTACUS study was set up to model the chain from emission to exposure using 1) emission functions for different vehicle types, 2) an air dispersion model, and 3) GIS to overlay concentration and population data.

Functions relating emissions of particulate matter (PM), oxides of nitrogen (NOx), and carbon monoxide (CO) to vehicles by speed and type are applied to the traffic characteristics of each raster cell. (These functions can be estimated for vehicle fleets in U.S. cities by running U.S. EPA or California Air Resources Board emission models under local conditions, for a variety of speeds.) The emission is then fed into a Gaussian air dispersion model, which was developed based on German technical guidance for air quality. The model includes meteorological parameters to describe a predominant wind speed and direction. A separate equation is applied to PM to account for PM sedimentation. The dispersion model computes the concentration (in milligrams per cubic meter) of pollutant at every other raster cell as a result of the emission cell. This calculation is repeated for all emission cells, and the results are summed to estimate the total concentration for each cell. A similar procedure, utilizing a dispersion model developed in the U.S., is described in the Envision Utah case study.

To measure population exposure, the total population is summed across the raster cells for which concentrations exceed air quality standards. For a variety of reasons, this does not measure actual exposure. People may spend varying levels of time in their residential zone; pollutant levels indoors and outdoors may differ; and pollutant levels and diffusion patterns may vary from day to day depending upon the weather. The measure also does not consider the exposure of people outside of their zone of residence; for example, at the workplace, or in other activity centers such as schools, hospitals, and shopping areas. The indicator does, however, provide a relative measure of exposure under different scenarios. In addition, the proportion of households from each socioeconomic group subject to unacceptable pollution levels can be compared to assess the equity of alternative policies.

Modeling of NO₂ levels in Helsinki (Figure 6) shows that in 2010 concentrations will exceed guidelines in only 4.9 percent of the metropolitan area, but that 10.6 percent of the total population and 12.6 percent of lower-income groups will live in these areas of exceedance.

Figure 6. NO2 Concentrations

Exposure to Noise

The impact of a traffic noise source ranges from a few meters to a few kilometers depending on the amount of noise being generated and on local circumstances. Most noise analysis models rely on spatially disaggregate information on topography, built form, and the distribution of population to estimate noise propagation and exposure. A simplified version of a German noise propagation model was applied in the SPARTACUS framework to make use of the available data.

Noise is treated similarly to pollutant emissions in the raster environment. The noise emission from each raster cell is treated as coming from a point source as a function of traffic volume, percentage of trucks, and a speed correction factor. The resulting noise level at a receptor cell is then calculated as a function of the segment length in the emission cell, distance from the emission cell, and absorption due to buildings. This last factor is estimated from land use categories and population and employment densities based on recent German research (Lee, 1998). The noise levels for each receptor cell are then aggregated.

To calculate social impacts of noise, two approaches can be taken. First, the number of households living in areas that have noise levels exceeding guidelines can be calculated. Alternatively, the percentage of households disturbed by noise levels can be assumed to increase as a function of the noise level. For example, Finnish researchers estimate that for a noise level between 55 and 64 dB(A), 33 percent of the population is disturbed; for levels between 65 and 69 dB(A), 50 percent is disturbed, and for levels of 70 dB(A) the entire population is assumed to be disturbed. As with pollutant emissions, the distribution of noise impacts across socioeconomic groups can also be calculated. For an approach to estimating noise impacts in the U.S., see the Waterloo case study.

[TOP]