Appendix 1. Modules used in the Regional Simulatior.

A. Modeling land-cover change in the Regional Simulator

A.1 Modeling urbanization in the Regional Simulator

The Regional Simulator (RSim) simulates changes to urban pixels for land cover maps for the five-county region around Fort Benning, Georgia, USA. Urban growth rules are applied at each iteration of RSim to create new urban land cover. The subsequent RSim modeling step then operates off a new map of land cover for the five-county region. The computer code (written in Java) has been built from the spontaneous, spread-center, and edge-growth rules of the urban growth model from SLEUTH (Clarke et al. 1997, Clarke and Gaydos 1998, Candau 2002;

The urban growth submodel in RSim includes both spontaneous growth of new urban areas and patch growth (growth of preexisting urban patches). RSim generates low-intensity urban areas (e.g., single-family-dwelling residential areas, schools, city parks, cemeteries, playing fields, and campus-like institutions) and high-intensity urban areas. Three sources of growth of low-intensity urban pixels are modeled: spontaneous, new spreading center, and edge growth. First, an exclusion layer is referenced to determine the pixels not suitable for urbanization. The exclusion layer includes transportation routes, open water, the Fort Benning base itself, state parks, and a large private recreational resort (Callaway Gardens). Spontaneous growth is initiated by the selection of n pixels at random, where n is a predetermined coefficient. These pixels will be urbanized if they do not fall within areas defined by the exclusion layer. New spreading center growth occurs by selecting a random number of the pixels chosen by spontaneous growth and urbanizing any two neighboring pixels. Edge growth pixels arise from a random number of nonurban pixels with at least three urbanized neighboring pixels.

Low-intensity urban pixels become high-intensity urban pixels according to different rules for two types of desired high-intensity urban pixels: (1) central business districts, commercial facilities, and highly impervious surface areas (e.g., parking lots) of institutional facilities that are created within existing areas with a concentration of low-intensity urban pixels; and (2) industrial facilities and commercial facilities (malls) that are created at the edge of the existing clumped areas of mostly low-intensity urban pixels or along four-lane roads.

For the first high-intensity category, land-cover changes occur in a manner similar to changes in low-intensity growth, as described above: a spontaneous-growth algorithm converts random low-intensity pixels to high-intensity pixels, and an edge-growth algorithm converts random low-intensity urban pixels with high-intensity urban neighbors to high-intensity pixels. The second type of conversion from low-intensity to high-intensity urban land use is road-influenced growth and is described in the next section.

The user can influence the pattern and rate of urban growth via changes to several parameters:
  • Dispersion (low): This parameter influences the number of randomly selected pixels for possible low-intensity urbanization. For dispersion (low) coefficient dL, a new DL value is computed as DL = (dL × 0.005) × sqrt(r² + c²), where r and c are the number of rows and columns in the land-cover map, respectively. During each time step, DL pixels are selected at random for attempted low-intensity urbanization. For this and all other rules defining the creation of new low-intensity urban pixels, only previously nonurban pixels lying outside urban exclusion zones may be changed to low-intensity urban;
  • Dispersion (high): This parameter influences the number of randomly selected pixels for possible high-intensity urbanization. For dispersion (high) coefficient dH, a new DH value is computed as DH = (dH × 0.005) × sqrt(r2 + c2), where r and c are the number of rows and columns in the land-cover map, respectively. During each time step, DH pixels will be selected at random for attempted high-intensity urbanization. For this and all other rules defining the creation of new high-intensity urban pixels, only previously low-intensity urban pixels lying outside urban exclusion zones may be changed to high-intensity urban;
  • Breed (spread): This parameter indicates the probability that a spontaneously created (by the above dispersion rule) low-intensity urban pixel is chosen to become a potential new spreading center. For each such pixel, two of its neighboring pixels are randomly selected for new low-intensity urbanization, if possible. A patch of three or more urban pixels is considered a spreading center and is eligible for edge growth, as described below;
  • Spread (low): This parameter indicates the probability that a low-intensity urban pixel within a spreading center will spawn a new neighboring low-intensity urban pixel during any time step. Such growth is also termed edge growth;
  • Spread (high): This parameter indicates the probability that a high-intensity urban pixel within a spreading center will spawn a new neighboring high-intensity urban pixel during any time step.

The SLEUTH model has been applied to more than 32 urban areas around the world. The parameters for these model runs are stored at the application’s website ( RSim was calibrated to the five-county region surrounding and including Fort Benning by running the model in a hind-cast mode and comparing projections to U.S. census data.

A.2 Modeling the effects of roads on urban growth in RSim

The road-influenced urbanization module of RSim consists of growth in areas near existing and new roads by considering the proximity of major roads to newly urbanized areas. The new-road scenario makes use of the Governor’s Road Improvement Program (GRIP) data layers for new roads in the region. Upon each iteration (time step) of RSim, some number of nonurban pixels in a land-use–land-cover map are tested for suitability for urbanization according to spontaneous- and patch-growth constraints. For each pixel that is converted to urban land cover, an additional test is performed to determine whether a primary road is within a predefined distance from the newly urbanized pixel. This step is accomplished by searching successive concentric rings around the urbanized pixel until either a primary-road pixel is found or the coefficient for a road search distance is exceeded. If a road is not encountered, the attempt is aborted.

Assuming that the search produces a candidate road, a search is performed to seek out other potential pixels for urbanization. Beginning from the candidate road pixel, the search algorithm attempts to move a “walker” along the road in a randomly selected direction. If the chosen direction does not lead to another road pixel, the algorithm continues searching around the current pixel until another road pixel is found, aborting upon failure. Once a suitable direction has been chosen, the walker is advanced one pixel, and the direction selection process is repeated.

In an effort to reduce the possibility of producing a road trip that doubles back in the opposite direction, the algorithm attempts at each step of the trip to continue moving the walker in the same direction in which it arrived. In the event that such a direction leads to a nonroad pixel, the algorithm’s search pattern fans out clockwise and counterclockwise until a suitable direction has been found, aborting upon failure. Additionally, a list of road pixels already visited on the current trip is maintained, and the walker is not allowed to revisit these pixels.

The road-trip process continues until it must be aborted because of the lack of a suitable direction or because the distance traveled exceeds a predefined travel limit coefficient. The latter case is considered a successful road trip. To simulate the different costs of traveling along smaller two-lane roads and larger four-lane roads, each single-pixel advancement on a two-lane road contributes more toward the travel limit, allowing for longer trips to be taken along four-lane roads, such as the GRIP highways.

Upon the successful completion of a road trip, the algorithm tests the immediate neighbors of the final road pixel visited for potential urbanization. If a nonurban candidate pixel for urbanization is found, it is changed to a low-intensity urban pixel, and its immediate neighbors are also tested to find two more urban candidates. If successful, this process will create a new urban center that may result in spreading growth as determined by the edge-growth constraint.

Roads also influence the conversion of low-intensity urban land cover to high-intensity urban land cover. For the second high-intensity urban subcategory (industry and malls), the RSim code selects new potential high-intensity-urbanized pixels with a probability defined by a breed coefficient for each pixel. If a four-lane or wider road is found within a given maximal radius (5 km, which determines the road_gravity_coefficient) of the selected pixel, the pixels adjacent to the discovered four-lane or wider road pixel are examined. If suitable, one adjacent pixel is chosen for high-intensity urbanization. Hence, the new industry or mall can be located on the highway, within 5 km of an already high-intensity urbanized pixel.

A.3 Modeling changes in land cover types other than urban in RSim

Changes within land cover types other than urban in the RSim region can affect the potential for pixels to be urbanized. Therefore, a brief description of that change process is included here. The annual nonurban land-cover trend was determined by using change-detection procedures that identify changes from one land cover type to another. Changes to and from urban classes were not considered in the results because they were being dealt with using different growth rules. Based on the land-cover changes happening over a period of time, the annual rate of change was calculated. These nonurban changes were incorporated in the form of a transition matrix from which transition growth rules were derived. Because forest management activities differ between Fort Benning and the surrounding private lands, the transition rules were calculated separately for Fort Benning and for the area outside Fort Benning. Outside Fort Benning, National Land Cover Datasets (NLCD) of 1992 and 2001 were used. The 2001 data set covers only the northern part of the RSim study region. The data for the remaining regions is yet to be released. Hence, currently, the changes observed in the northern portion are assumed to be representative of changes in the whole five-county study region outside Fort Benning. Within Fort Benning, land-cover data sets from 2001 and 2003 were used to derive the annual transition rules for nonurban land-cover changes.

B. Modules for environmental effects in RSim

RSim was designed to focus explicitly on how changes in land cover affect and are affected by environmental conditions. As such, the following environmental interactions are an integral part of the RSim package.

B.1 Air quality module

The air-quality module (AQM) of RSim estimates how demographic and economic growth, technology advances, activity change, and land-cover transformations affect ground-level ozone concentrations in the Columbus–Fort Benning area. The AQM is largely based on air-quality computer modeling completed during the Fall Line Air Quality Study (1999–2004; Chang et al. 2004). Unlike the Fall Line Air Quality Study models, though, the design of the AQM removes the computational load of traditional air-quality modeling while remaining flexible enough for the user to test various future scenarios. The RSim AQM estimates the relative change in the concentration of ground-level ozone in the Columbus area caused by changes in transportation, business and industry, construction, military operations, and other human activities. In addition, the AQM simulates effects on vegetation.

RSim draws on the extensive, state-of-the-art, and thoroughly reviewed ozone air quality model simulations of the Fall Line Air Quality Study (FAQS). Therein, an air-quality model was created that accurately represents a historical ozone episode for the Columbus/Fort Benning area in the year 2000. In RSim, future-year changes in human activities (sources) are used together with the FAQS base case to estimate future-year changes in ozone air quality:

ozonet = ozone2000 + (∂ozone/∂source) ΔSourcet – 2000

In the above equation, sources may change relative to how they were in the year 2000 (∆Sourcet – 2000), for example, from economic growth in the region or changes in transportation patterns, and these can be controlled by the RSim user. The term ∂ozone/∂source is a sensitivity coefficient that is unique to the source and quantifies how a change in the source, ∂source, affects changes in the concentrations of ozone, ∂ozone. These sensitivity coefficients were calculated outside of RSim and cannot be modified by the user. The description above assumes that only one source changes during any given period. As implemented in RSim, the AQM really accounts for multiple changes in many sources throughout the emissions inventory, some of which may exasperate poor air quality and some of which may mitigate poor air quality. Selection of the Default RSim scenario creates a future in which relative changes in emissions sources (∆Sourcet – 2000) are estimated with growth factors from the U.S. Environmental Protection Agency’s Emissions Growth Analysis System (EGAS; U.S. Environmental Protection Agency 2004).

Ozone can cause foliar damage in trees, crops, and other vegetation, as well as other effects. RSim simulates the effects of ozone on vegetation by using the secondary standard for ozone to simulate the relative likelihood of effects of ozone on vegetation. This standard is meant to protect crops and vegetation, as well as other aspects of public welfare. The secondary standard for ozone is equivalent to the primary standard, which states that the fourth highest 8-h ozone concentration cannot exceed 0.08 ppm (parts per million).

B.2 Water quality and nitrogen and phosphorus export modules

The water quality module predicts changes in annual nitrogen (N) and phosphorus (P) exports from watersheds within the five-county (Harris, Muscogee, Marion, Chattahoochee, and Talbot) RSim region surrounding Fort Benning. It is widely established that land use and land cover are principal determinants of nutrient export from terrestrial ecosystems to surface receiving waters (Beaulac and Reckhow 1982). The water quality submodel predicts total (kg yr–1) and normalized (kg ha–1 yr–1) losses of N and P from 48 watersheds within the region over the time frame of RSim scenarios by using export coefficients (Johnes 1996, Johnes et al. 1996, Mattikalli and Richards 1996).

Calculations of annual N and P export are performed for the 48 12-digit hydrological units (HUC) that are included within the RSim region. The method is based on land-cover area (ha) within each watershed and annual nutrient export coefficients (kg element ha–1) specific to each of the eight land cover types. The area (ha) of each land cover category is multiplied by its respective export coefficient, and the products are summed for all land covers to estimate the annual flux (kg element yr–1) of N or P from each watershed. The exports (kg yr–1) are also normalized for the size (ha) of the watershed to yield an area-normalized N or P export (kg element ha–1 yr–1). The 48 12-digit HUCs range in size from approximately 3200 to 12,000 ha.

RSim predictions of N and P exports (kg element yr–1) over time vary depending on the changing patterns of land cover within each watershed. Trial runs with the water quality submodel indicate that the annual fluxes of both N and P exhibit a significant positive correlation with size of the hydrological unit (r = 0.80 and r = 0.48, respectively, P ≤ 0.001). However, the size of a watershed, the types of land cover within a watershed, and the export coefficients selected for different land covers all influence the predicted N and P exports.

B.3 Species of special concern module

RSim considers effects on the two rare species in the vicinity of Fort Benning: Red-cockaded Woodpecker (Picoides borealis) and gopher tortoise (Gopherus polyphemus). RSim simulates changes in Red-cockaded Woodpecker (RCW) clusters based on changes in land cover. These clusters primarily occur in mature longleaf pine (Pinus palustris) forest, so as land changes from evergreen forest it becomes unsuitable for RCW. In the five-county region, most of the clusters are found within Fort Benning. In December 2005, there were 212 known active and 96 inactive RCW clusters at Fort Benning. According to the FWS biological opinion and the installation’s RCW management plan, Fort Benning’s goal is 361 active RCW breeding clusters. RSim reports how this goal is affected by changes in land cover for every year of the projection.

The gopher tortoise habitat module in RSim computes the probability of suitable gopher tortoise habitat in a region according to a logistic regression model described by Baskaran et al. (2006a). The gopher tortoise habitat module of RSim uses land cover variables, distance to stream and road variables, and clay variables as inputs to derive the probability of finding a gopher tortoise. RSim gives the user the option to further define habitat suitability based on habitat patch size, identified within RSim using a modification of the Hoshen-Kopelman algorithm (Berry et al. 1994, Constantin et al. 1997). The outputs from this module are:
  • a map of the probability of occurrence of gopher tortoise habitat;
  • a map of predicted burrow presence and absence;
  • a table of the area of predicted burrows per year.

B.4 Noise module

Noise from military installations may cause human annoyance outside of installation boundaries. Noise can also affect wildlife. RSim uses estimates of exposure to noise from aspects of military training, namely aircraft overflights, large munitions, and small arms. Noise contour maps are developed from three noise simulation models external to RSim (Operational Noise Program 2007):
  • NOISEMAP calculates contours resulting from aircraft operations using such variables as power settings, aircraft model and type, maximum sound levels and durations, and flight profiles for a given airfield;
  • BNOISE projects noise effects around military ranges where 20-mm or larger weapons are fired and takes into account both the annoyances caused by both impulsive noise and vibration caused by the low-frequency sound of large explosions;
  • The Small Arms Range Noise Assessment Model (SARNAM) projects noise effects around small-arms ranges and accounts for noise attenuated by different combinations of berms, baffles, and range structures.
In the implementation of RSim in the region of Fort Benning, noise contour maps represent blast noise simulated by BNOISE, as well as the negligible noise from small arms, but not aircraft noise. RSim uses these contours to estimate human annoyance and to recommend compatible land uses. Residential development and other land uses associated with low-intensity urban land cover are not compatible with blast noise > 115 dBP (peak decibels).