AN ANALYSIS OF HUMAN-INDUCED LAND TRANSFORMATIONS IN THE SAN FRANCISCO BAY/SACRAMENTOAREA

• David Kirtland U.S. Geological Survey 519 National Center Reston, VA 22092 USA703-648-4712
• Leonard Gaydos U.S. Geological Survey Building 242 Ames Research Center 242-2 MoffettField, CA 94035 USA 415-604-6368
• Keith Clarke Department of Geology and Geography Hunter College 695 Park Avenue New York, NY 10021 USA 212-772-5322
• Lee DeCola U.S. Geological Survey 521 National Center Reston, VA 22092 USA 703-648-4178
• William Acevedo U.S. Geological Survey Building 242 Ames Research Center 242-2 MoffettField, CA 94035 USA 415-604-5299
• Cindy Bell, Johnson Controls World Services Building 242 Ames Research Center 242-2Moffett Field, CA 94035 USA 415-604-391

SUMMARY

INTRODUCTION

Background
The importance of land cover in land-atmosphere interactionshas been recognized in global change research and is the subject of intense research andmodeling activities. Of the forces that influence land cover, those that are human induced(resulting from human land use), are less well researched than others. Urban landrepresents a significant human use of land. The impact that urban land has on environmentalsystems at a regional scale is extremely large compared to its spatial extent. TheInternational Geosphere- Biosphere Programme (IGBP) and the Human Dimensions of GlobalEnvironmental Change Programme (HDP) recommend a major worldwide study of land use and itsrelation to land cover change (Turner and others, 1993). With the goal of projecting futurestates of land cover, the IGBP and HDP recommend understanding the determinants of land use(for example, demographics, technology, and levels of affluence) so that future patterns ofland use and land cover can be projected through modeling. Their justification is that"understanding the past and future impacts of changes in land cover is central to the studyof global environmental change and its human driving forces and impacts, includinghydrology, the climate system, biogeochemical cycling, ecological complexity, and landdegradation and its impacts for agriculture and human settlement" (Turner and others, 1993,p. 6).

The U.S. Geological Survey (USGS) has a long tradition of studying land use andland cover and classifying land use potential, particularly lands to be irrigated (Powell,1879). In the 1970's the USGS developed a land use and land cover classification system foruse with remotely sensed data (Anderson and others, 1976). The USGS began a program toproduce a series of land use and land cover maps for the Nation by interpretinghigh-altitude aerial photographs. Complete coverage of the conterminous United States at1:250,000 scale was completed in 1986. During the 1980's the USGS developed techniques formapping land cover by using Landsat data. A project is now underway to produce a data basefor North America consisting of three Landsat multispectral scanner (MSS) scenesrepresentative of land cover in the 1970's, the 1980's, and the 1990's (Kirtland, 1993). In1990 the USGS began producing biweekly composites of advanced very high resolutionradiometer (AVHRR) images for the conterminous United States at 1-km2 (Eidenshink, 1992). These data were used to derive a prototype land cover characteristics data set for theconterminous United States (Loveland and others, 1991). The land cover characteristics dataset is being used as the model for a global land characteristics data base that will serve awide variety of clients and help meet IGBP objectives. The AVHRR data base for the UnitedStates is also being combined with Landsat thematic mapper (TM) data as part of amultiresolution land characteristics monitoring system for research requiring higherresolution data.

Objectives
This project is the initial phase of a study that is usingremotely sensed spatial data integrated with other data in a geographic information system(GIS) environment to investigate, model, and predict global human- induced land coverchange. The initial phase of the project emphasizes urban land transformations as a resultand cause of other human-induced land transformations within a region. Three differentstrategies are employed to track historical and predict future urban growth at a regionalscale and identify data sets useful for focusing attention on areas experiencing discerniblechange: animation, prediction, and description. Animation is being used to visualize andcommunicate the extent of change in urban areas over time with attention focused onsignificant changes within a region. Historical analysis of urban growth in the SanFrancisco Bay/Sacramento area is providing a relatively high-resolution look at thedevelopment of the area during the last 150 years. A simulation model is being developedto help predict future regional landscapes. The data gathered for the historical analysisprovide a means to calibrate the simulation model. To test the use of regional land covercharacteristics and related data, scientists are exploring potential relationships amongvariables representative of the natural and human-induced aspects of the region byintegrating several data sets with multivariate analysis tools. The next phase of thestudy will evaluate the use of the regional-scale integrated data sets to drive theanimation and prediction of land cover change. If this strategy proves successful inproviding insight into human- induced changes in the landscape, then subsequent projectphases will scale up the size of the region studied and generalize the approach for useanywhere in the world. Figure 1 shows the study region.

HISTORICAL ANALYSIS

A time series of land cover data was constructed to visualize theextent of change in a region over time and to calibrate a simulation model. For the SanFrancisco Bay/Sacramento area, a digital data base was assembled from topographic, road, andland use maps and from digital Landsat and elevation data.

Urban extent was inferred from historic maps and from Landsat data. Two time periodswere mapped from USGStopographic maps for the San Francisco- Oakland-San Jose areas. The first was based on1:62,500-scale topographic maps of the area from 1897 to 1906. The Sacramento area,however, was mapped from a 1:125,000-scale map published in 1887 and from a 1:250,000-scalemap representative of the Sacramento Valley from 1903 to 1910. For the second time period,1:62,500- and 1:50,000-scale topographic maps published by the USGS and the Army Map Servicewere used. Aerial photographs used to prepare these maps were taken between 1937 and 1940. Dense street patterns and buildings located on these maps were regarded as built-up areas. Polygons were drawn on mylar overlays around each concentration of these features to mapurban extent. Urban extent around 1925 was obtained from Donley, who used road maps andother sources to map urban extent because topographic maps of that vintage were unavailable(Donley and others, 1979). These maps were published at 1:500,000 scale. The Associationof Bay Area Governments (ABAG) prepared maps of land use for the years 1954 and 1962. Aggregated urban land uses indicated the extent of urbanization. Comparisons with the othersources are difficult, however, because some low density residential areas would not havebeen considered urbanized when mapped from topographic maps. The USGS mapped the urbanizedarea for the ABAG counties in 1970 by using high-altitude aerial photographs and land usecriteria (USGS, 1972).

Landsat, first launched in 1972, was used to create two maps. Two scenes acquired in 1974 using the 80-m resolution multispectral scanner and two scenesacquired in 1990 using the 30-m thematic mapper were interpreted using a digital display. Amanual photo interpretation process was used to delineate the color and pattern indicatingurbanized land.

Delineations of urban extent from maps, satellite imagery, and theABAG land use maps were digitized by scanning. The digitized polygons of historic landcovers and the interpreted Landsat data were then registered to a 30-m Universal TransverseMercator (UTM) grid developed for the area. Even though component data sets ranged inresolution from 30m to 100m, all data were registered to 30m to preserve the finest detailpossible. Donley's maps of highway development from 1920 to 1978 were scanned; Donley hadderived them from source maps prepared by the California Department of Transportation(Donley and others, 1979). Recent highway data were obtained from 1:2,000,000-scaleNational Atlas digital line graphs (DLG). Elevation data were obtained from seven mosaicked3-arc second digital elevation models (DEM). All of these digital data sets were registeredto the 30-m UTM grid to provide a consistent data base for calibrating visualization andsimulation models.

The data base has a relatively high spatial and temporalresolution. To visualize the extent of change for this area over time, it was necessary touse the available historic data to interpolate. Linear interpolation, with the urbanboundary maps as reference, was used to create intermediate data at 1-year intervals. Theinterpolation algorithm calculates a linear distance from the starting urban boundary to theending urban boundary. That distance is then used to assign an urban extent status to thecorresponding pixel when estimating urban extent for each year.

Simple, single-frameanimation techniques were used to visualize these data. The urban data sets weretransmitted sequentially to a computer display in translucent color to show urban growthover time; a recent Landsat image was used as a reference base. Speed and zoom factors ofthe animation can be controlled to show that particular areas can be seen in more detail. The animation provides a strong visual portrayal of extensive land cover transformations inthe area. It easily communicates the pace of urbanization from slow to moderate toexplosive. When the topographic data are added as an alternative base to the animation, thevisualization illustrates how urbanization has been influenced by the physiography of theregion. An example of the changing urban morphology of Sacramento from 1950 to 1990 ispresented in figure 2.

MODELING URBAN TRANSFORMATIONS

The following factors control the behavior of the system:

1. A diffusion factor, which determines the overall dispersiveness of the distribution ofsingle (hamlet) grid cells and in the movement of new settlements outward through the roadsystem.
2. A neighbors coefficient, which determines how connected a new cell needs to be tocontinue outward growth or to fill in cells surrounded by growth.
3. A breeding coefficient,which determines the likelihood of a newly generated detached settlement beginning its owngrowth cycle.
4. A spread coefficient, which controls how much normal outward "organic"expansion takes place within the system.
5. A slope resistance factor, which decreases thelikelihood of settlement extending up steeper slopes.
6. A road gravity factor, which hasthe effect of attracting new settlements along the existing road system.
7. A road weight,which makes grid cells more likely to "hit" roads or locate near roads rather than at adistance from them, based on a set of weights for distance.

Some of the factors aremore system-sensitive than others, and in such a system a complex set of interfactordependencies exists. A full set of outcomes can be generated by varying the parameters toextremes; for example, outcomes that result in both zero and extensive growth can besimulated. Extensive growth patterns (that is, those that completely fill an area), whichare linear, exponential, and S-curve type (that is, reaching and stabilizing at some"optimum"), can be simulated over time. Both interactive and batch versions of the modelhave been written, which allow for calibration, scenario construction, model replication,sensitivity analysis, and browsing of outcomes.

After the model was calibrated toensure that it was operating as intended, the control parameters were changed to allowself-modification. This coupling of parameters required another set of rules:

1. When the system growth rate exceeds a critical value, the diffusiveness and spreadfactors are increased.
2. When the system growth rate falls below a second critical value,the diffusiveness and spread factors are decreased.
3. When growth rates exceed a criticalvalue, the breeding probability is increased.
4. The road gravity factor is increased as thesystem builds more and more roads (that is, road layers from several historical periods areread in and used as the number of time cycles increases).
5. As the percentage of landavailable for development decreases, the resistance to expansion onto steeper slopesdecreases.
6. As new growth in a time cycle takes place on steeper slopes, the spread factorincreases (that is, movement outward on flat land becomes faster).

A large number of ways of quantitatively comparing sets of spatial urban forms over time weredeveloped during the testing, including centrographic statistics, edge and areameasurements, and internal age and rule-origin structures. These aids have greatlyincreased the ability of the model to provide meaningful and rapid predictions of futurepatterns of urbanization from the current and historical data.

Model Use
Advantages ofthe model are many. The step rules are simple to explain, understand, and modify as insightinto system processes improves. The model is not dependent on generalized probabilitydistributions derived from observed or hypothetical data but allows each cell to actindependently according to the rules (that is, every single part acts as part of anensemble). This is similar to the way that a city expands, as the result of hundreds ofindividual decisions, made one at a time, but susceptible to the physical, social, economic, cultural, and political landscape (forexample, the overall trends of market, mortgage rates, economic climate, transportationtechnology, and so on). Another important characteristic is the model's use of interactiveand animated computer graphics, allowing point-and-click access to the parameters andimmediate visualizations of the outcomes. Furthermore, multiple applications of the modelfrom various starting conditions allow the computation of Monte Carlo-style averageaggregate output probabilities of any given cell being urbanized. The resulting maps offuture urbanization, while susceptible to the rules and properties of the model, areextremely useful for investigating urban land transformations in a regional context as partof global change research. Different scenarios can be linked to simple environmental modelsand the environmental effects (for example, urban heat islands, loss of other land uses,increased particulate and gas emissions, etc.), can then be more effectively explored. Figure 3 presents an example of model output.

DESCRIPTIVE ANALYSIS

The area shown in figure 1 is extremelycomplex, with elevations from sea level to 2500 meters, ecosystems from arid to humid, andhuman settlement densities from wilderness to metropolitan (Knox, 1991). Separate data setswere integrated into a single, multiscale data base for the analysis. Grid cells of 1-km2(or some multiple up to 256-km2) were selected for this analysis. Figure 4 shows differentways to visualize the landscape.

Land cover in the region is measured in two ways: byusing the NDVI, an index that represents the amount of vegetative biomass within each cell,and by using a land cover class to categorize either the dominant natural or human activityin each cell (EROS Data Center, 1993). DEM data were obtained to represent the topographyof the area. These data were used to generate slope and aspect measurements of the area.

Total annual precipitation and average annual temperature for the past 30 years werecompiled to describe the region's climate. These data represent the records from 82stations in the region, 55 of which report only precipitation. A spline technique was usedto integrate these measurements into the data base (see Legates and Willmott (1990) for amore sophisticated approach to this problem).

Human population data for the regionwere compiled from more than 18,000 block group counts from the 1990 census of population. These data were originally TIGER Line Files (United States Bureau of the Census, 1993) thatwere transformed into ARC/INFO polygons, then into point locations, and finally into adistance-weighted surface (Martin and Bracken, 1991).

The use of this integrated database to analyze environmental interactions is just beginning. For example, what effects dopopulation, elevation, temperature, and precipitation have on vegetation density? Anordinary least-squares regression model will be used to estimate the relationships inherentin the data. This model can be used to investigate numerous other statistical scenarios. The integrated data base can also provide direct and indirect evidence of human influence onthe land. For example, the graphic portraying population in figure 4 provides directevidence of human settlement, and the agricultural land use in the lower elevations with lowannual precipitation offers indirect evidence of irrigation.

CONCLUSION

While thisinitial phase of the project focuses on one region, the underlying objective is to determinethe usefulness of integrating remotely sensed data about the Earth's surface, at a scale of1 km, with other data on population and climate, in order to drive historical, predictive,and descriptive analyses for any region, worldwide. Whether or not these 1-km data sets(several of which are being developed as part of the USGS Global Change Research Program)can be used as source for global-scale analyses of human-induced land transformations, isthe focus of the next phase of this project.

REFERENCES

Batty, M., P. Longley, and A.S. Fotheringham, Urban growth and form: scaling, fractalgeometry, and diffusion-limited aggregation, Environment and Planning A, 21, 1447-1472(1989).

Coucleilis, H., Cellular Worlds: A framework for modeling micro-macro dynamics, Environmentand Planning A, 17, 585-596 (1985).

Donley, M., S. Allan, P. Caro, and C. Patton, Atlas of California, Academic Book Center,Culver City, California (1979).

Eidenshink, J.C., The 1990 conterminous U.S. AVHRR data set, Photogrammetric Engineering andRemote Sensing, 58, 809-813 (1992).

EROS Data Center, The 1990 conterminous U.S. land cover characteristics data set, CD-ROM,U.S. Geological Survey, EROS Data Center, Sioux Falls, South Dakota (1993).

Kirtland, D.A., Progress in global land data sets and land characterization: USGS researchprogram, World Resource Review, 5, 286-302 (1993).

Knox, J.B. and A. Scheuring, Global climate change and California: potential impacts andresponses, University of California, Berkeley (1991).

Legates, D.R. and C.J. Willmott, Mean seasonal and spatial variability in global surface airtemperature, Theoretical Applied Climatology, 41, 11-21 (1990).

Loveland, T.R., J. Merchant, D.O. Ohlen, and J. Brown, Development of a land covercharacteristics data base for the conterminous U.S., Photogrammetric Engineering and RemoteSensing, 57, 1453-1463 (1991).

Loveland, T. R., D.O. Ohlen, J.F. Brown, B.C. Reed, J.W. Merchant, and L.T. Steyaert,Prototype 1990 conterminous United States land cover characteristics data set CD-ROM, EROSData Center, U.S. Geological Survey CD-ROM set, (1993).

Martin, D. and I. Bracken, Techniques for modelling population-related raster databases,Environmental and Planning A, 23, 1069-1075 (1991).

Powell, J.W., Report on the lands of the arid regions of the United States, with a moredetailed account of the lands of Utah, U.S. Geographical and Geological Survey of the RockyMountain Region, Washington, D.C. (1879).

Turner, B.L., R.H. Moss, and D.L. Skole, (eds.) Relating land use and global land-coverchange: a proposal for an IGBP-HDP core project, International Geosphere-BiosphereProgramme, IGBP Report No. 24, HDP Report No. 5 (1993).

United States Bureau of the Census, 1993 TIGER/line files technical documentation, U.S.Government Printing Office, Washington DC (1993).

United States Geological Survey, San Francisco atlas of urban and regional change, U.S.Geological Survey, Reston, Virginia (1972).

White, R. and G. Engelen, Cellular automata and fractal urban form: a cellular modellingapproach to the evolution of urban land use patterns, Research Institute voorKennis-Systemen, Publication Number 915000/9264 (July 1992).