Pergamon

Comput., Environ. and Urban Systems,Vol. 21, No. 3/4, pp. 259-276, 1997 © 1998ElsevierScienceLtd. All rights reserved Printed in Great Britain 0198-9715/98$19.00 + 0.00 PII: S0198-9715(97)01002-8

APPLICATION OF LANDSAT TM DATA FOR QUALITY OF LIFE ASSESSMENT IN AN URBAN ENVIRONMENT C. P. Lo 1 Department of Geography, University of Georgia, Athens, Georgia 30602, U.S.A. ABSTRACT. Land.sat TM data were used in conjunction with census data for

urban quality of life assessment, using the Athens-Clarke County, Georgia in the United States as an example. The satellite image data were used to extract environmental variables of land cover, vegetation index, and surface temperature, while socioeconomic variables were obtained from the census data. Using the social space conceptualization put forth by Chombart de Lauwe, quality of life as a collective attribute that adheres to groups of people was assessed with reference to the morphological and sociocultural environment. A principal components analysis and a raster geographic information system overlay of variables were found to be useful approaches that effectively integrated biophysical and socioeconomic data together to give an accurate assessment of quality of life within the study area. It is concluded that high resolution satellite image data are a useful complement to census data by giving the environmental perspective in urban analysis. © 1998 Elsevier Science Ltd. All fights reserved

INTRODUCTION This research is an attempt to expand the use of high-resolution satellite image data (in both spatial and radiometric sense) for urban analysis. In the past, aerial photography was employed in conjunction with socioeconomic data from the census to measure the quality of the urban environment, typified by the work of Green (1957) and subsequently expanded on by Mumbower and Donoghue (1967) and Metivier and McCoy (1971), using a manual approach. Forster (1983) drew attention to the use of spectral reflectance data derived from Landsat MSS images in the city of Sydney, Australia to develop a residential quality index, using house size and vegetative content as a positive indicator of quality and roads and non-residential buildings as a negative indicator. Because house value is strongly related to house size, Forster (1983) has demonstrated that Sydney house values can be predicted from Landsat data over extended areas. Most recently, Weber and Hirsch (1992) made use of the high-resolution SPOT XS (multispec~al) image data in combination with cartographic and census data to measure the urban life quality of Strasbourg, France, ITel: (706)542-2330; Fax: (706)542-2388; E-mail: [email protected] 259

260

C. P. Lo

using a computer-assisted digital approach. In these studies, urban quality was expressed by means of scales or indices which combined the socioeconomic and physical data together for a complete evaluation. Green (1957), for example, made use of the Guttman Scale to combine several qualitative variables to form a "residential desirability scale" based on the physical data extracted from the aerial photographs and a "socioeconomic status scale" based on the census data. These two scales were found to be strongly correlated. Inspired by Forster's 0983) work, Weber and Hirsch 0992) constructed three indices for Strasbourg's urban quality: (a) a housing index which relates to the types of suburban housing, CO) a quality index which measures the degree of housing density and greenness of the environment, and (c) an attractivity index which relates housing to the percentages of land cover in vegetation, industry, commerce, and parking, solely extracted from the SPOT images. The conceptual underpinning of these urban analysis research efforts is derived from the French sociologist, Chombart de Lauwe, who, with an interest in the use of aerial photography, developed the concept of "social space" in 1952. Social space, the total environment in which people live, consists of two components: (a) the morphological environment and Co) the sociocultural environment (Chombart de Lauwe, 1952, cited in De Haas, 1966). The morphological environment, in turn, is made up of biophysical and dcmographical features. Aerial photographs can be used to extract the physical characteristics of the environment, which are basically the land cover and building types. The socio-cultural environment represents people's perception of their living environment, which presumably is affected by the quality of the physical environment. In recent years, advances in satellite remote sensing, computer and geographic information system (GIS) technologies have made possible a new automated approach to urban analysis. For this research, instead of focusing on the residential quality of the urban environment, the quality of life of people will be used. Quality of life is a collective attribute that adheres to groups of people, not to individuals. A great variety of indicators can be used to measure quality of life, which consists of both objective and subjective elements (Liu, 1970; Wallace, 1971; Szalai, 1980; Shelton, Grubber, & Godwin, 1983; Bederman & Hartshorn, 1984; Andrews, 1986). Thus, capita income is an objective measure, but how satisfied people are about their incomes is subjective. This shift of focus from urban environment quality to quality of life in this research will fit in better with Chombart de Lauwe's social space formulation, wherein the sociocultural environment is people's perception of the morphological environment. The interaction between the two types of environment determines the quality of life. A major emphasis of this research is that a green environment is a desirable one for most urban residents and should be an important consideration in quality of life assessment. Greenness (in the form of gardens, parks, and trees in city landscaping) is believed to be a good surrogate of socioeconomic conditions. Greenness is measurable in the form of a vegetation index. In previous studies, the amount of vegetation cover in a city, rather than the vegetation index, has been used in urban quality assessment. The vegetation index, in its many forms, is easily extractable from multispectral satellite images sensing in the red and infrared bands. A major objective of this paper is to demonstrate that through quality of life assessment, biophysical variables from satellite image data and socioeconomic variables from census data can be integrated through a GIS approach.

Landsat TM Data for Quality of Life Assessment

261

THE STUDY AREA

Athens--Clarke County, Georgia in the United States, is selected for this study. Situated about 100 km northeast of Atlanta, it has a total area of 31,376 ha and a total population of 87,594 living in 33,113 households in 1990. Athens, the major city with a population of 45,734, is the site of the University of Georgia, which has a student population of about 28,000. Despite its small size, there is a high degree of contrast between high and low income groups in this county. The 1990 per capita income was $15,715, which is 82% of the national average and 90% of the state average. The majority of the county population is urban (82.4% in 1990) with only 0.19% living in rural farms and the rest rural residential. Because of the location of the University of Georgia, Athens-Clarke county has the second highest educational level in the State of Georgia (Hodler, Lawson, Schretter, & Torguson, 1994). The study area is therefore not a large urban metropolis such as Atlanta, but it is typical of a medium size community in the Sun Belt which attracts people to move in. It has been consistently ranked by Money Magazine as one of the top places in the United States to live. Despite its small size, the biophysical and socioeconomic characteristics are varied enough to affect the quality of life from one part of the county to another.

DATA SOURCES AND DATA VARIABLE EXTRACTION

A Landsat Thematic Mapper (TM) image of the study area in digital form acquired on July 16, 1990, a date not too far away from the 1990 mid-year Census, was used for this research. This early summer date also insures that vegetation was still growing so that the greenness of the environment could be accurately measured. The Landsat TM data had been radiometrically corrected and georeferenced to the Universal Transverse Mercator (UTM) with a resampled spatial resolution of 25 m. In order to provide ground truth data on computer-assisted land cover classification, black-and-white panchromatic aerial photographs of the study area acquired with a Zeiss (Jena) LMK camera (f= 152.27 ram) on January 10, 1990 by Atlantic Aerial Survey in Alabama at a nominal scale of 1:18,000 were employed. Socioeconomic data were extracted from the STF3A CD-ROM, based on a 16-17 percent sample survey of the 1990 Census. The following four variables commonly employed in developing a quality of life index were used: population density, per capita income, median home value, and education level (expressed as the percent of college graduates; Bederman & Hartshorn, 1984). These variables were aggregated at the block group' level (Figure 1). As the name implies, the block group is made up of several street blocks of the census, and is intermediate in size between the very large "census tracts" and the very fine "census street blocks". The census block groups were supplied by the Athens-Clarke County Planning Commission in the form of an ARC/INFO file in State Plane Coordinates, which were then converted to the Universal Transverse Mercator projection using the PROJECT command of the ARC/INFO program so that the block group boundaries could be registered to the Landsat TM data. A test of the rectification accuracy using nine ground control points (UTM coordinates read from the 1:24,000 scale United States Geological Survey topographic maps) distributed throughout the image of

10

31

44

6

0

54

8

12

50

39

Y

FIGURE 1. Ceasus block group boundaries for Athens-Clarke County, 1990.

33

21

11

I

16 Kilon'Bters

S

N

O

.O

Landsat TM Data for Quality of Life Assessment

263

the study area gave a root mean square error (RMSE) of 0.33 pixel or 8.19 m, which is good, considering that the spatial resolution of the resampled Landsat TM data is 25 m. The Landsat TM data are an excellent source of biophysical data useful to quality of life assessment (Jensen, 1983). In order to understand the character of the urban environment, the land use/cover will have to be extracted. Image classification using multispectral data is a common technique in remote sensing digital image analysis (Richards, 1986). Initially, a land use classification scheme was developed. For this study area, the following eight land use/cover classes can be identified: (a) water, (b) forest, (c) commercial/industrial, (d) transportation, (e) agriculture in pasture, (f) agriculture in crops, (g) low-density residential, and (h) high-density residential. Using the Desktop Mapping Software (DMS) and a supervised image classification approach with the maximum likelihood classifier, a land use/cover map with an overall accuracy of 74% (determined with 460 random check points) was obtained. The black-and-white aerial photographs provided the ground truth land use data for training sets and accuracy evaluation. The land use/cover map produced was edited by using the DMS software to change the name of the mis-classified polygons of a particular land use/cover class to that of the correct land use/cover class. In other words, this procedure simply changed the name of one land use class to that of another. In this way, a land use/cover map of the study area with an overall accuracy of 99.1% was achieved. From this final land use/cover map, commercial/industrial and transportation uses were grouped into one class called "urban use" which is a negative indicator of environmental quality. The high overall accuracy of the final land use/cover map insured that the delineation of "urban use" was accurate. From the Landsat TM data, a vegetation index was extracted to measure the degree of greenness of the environment. The normalized difference vegetation index (NDVI) was extracted using the following formula for Landsat TM data (Jensen, 1996): NDVI = (TM4 - TM3)/(TM4 + TM3) where TM3 is the red band (0.63-0.69/.,m) and TM4 is the reflective infrared band (0.76-0.90 gm). This index is a ratio transform which contrasts the energy reflectance from dead or senescent vegetation (of the red band) with that from healthy green vegetation (of the infrared band). This ratio was computed using the IDRISI GIS software. The value ranges from - 1 to + 1 as greenness (and biomass) increases. The vegetation index is a positive indicator of environmental quality. The Landsat TM data's band 6 (10.3-12.5/an) is a thermal infrared band which records thermal infrared emission from the land surface. The building layout, building heights, and land cover type will affect the surface temperature in a city, which will have an impact on the development of urban heat island. Therefore, a knowledge of surface temperatures will indicate to what extent the layout of vegetation has alleviated the heating effect of roof and road surfacing materials in the city. This is a particularly important consideration for a city in Southern United States where summer temperatures can reach over 32°C. Sustainable landscaping will help cool down urban heat island, thus helping to reduce the amount of energy used for indoor cooling and hence the heat generated from air conditioners (McPherson, 1994). Therefore, high surface temperature is an indicator of ineffective landscaping, and is not regarded desirable by most people as far as environmental quality is concerned. The following formula, developed by Malaret, Bartolucci, Lozano, Anuta, and McGillem (1985), was used to extract surface temperatures from TM band 6 data:

264

C.P. Lo

T(K) = 209.831 + 0.834DC - 0.00133DC 2 where DC is the digital count between 0 and 255, and T(K) is absolute temperature in Kelvin. Using the IDRISI GIS overlay function, the absolute temperature for each pixel of the Landsat TM band 6 image was computed, which was then converted to Celcius (°C) by subtracting the temperature of the ice point (273.15 K) from it. The surface temperature obtained for each pixel had not taken into account emissivity variations Caused by the different types of land cover materials. In general, emissivities vary from 0.95 for vegetated cover to 0.92 for non-vegetated cover (Nichol, 1994). The resulting image of surface temperature exhibits a range of temperature from 21°C to 38°C. As expected, high surface temperatures were observed along transportation lines and built-up areas. These three environmental variables of urban use, NDVI, and surface temperature are per-pixel data while the socioeconomic variables of population density (Figure 2), per capita income (Figure 3), median home value, and education level are aggregated by block groups. Because of the inaccuracy of interpolating the socioeconomic variables from block groups to individual pixels (which would have been ideal for integrating with the raster image data) (see an attempt to interpolate population by Langford, Maguire and Unwin (1991)), the three environmental variables were aggregated into the block group areal unit instead (Figures 4-6), using a simple EXTRACT command and a geographic definition file of block groups (already accurately registered over the Landsat TM image) in IDRISI GIS software. Such a conversion has the disadvantage of averaging out the values of the environmental variables over the whole areal unit (block group) which unfortunately varies in size according to land use density, with the downtown area having smaller areal units than the periphery (Figure 1). The implication is that the basic element of the mappable spatial pattern of quality of life will have to be the census block group.

DATA INTEGRATION USING PRINCIPAL COMPONENTS ANALYSIS

Principal components analysis (PCA) is a data transformation technique which can convert a large number of correlated variables into a smaller number of uncorrelated components made up of these variables, weighted according to the amount of the total variance that they describe, for a series of sites, or objects, or persons (Daultrey, 1976). Each variable measured is an axis, or dimension, of variability. PCA transforms the data to describe the same amount of variability, the total variance, with the same number of axes, the number of variables, in such a way that the first axis (component) always accounts for as much of the total variance as possible, the second axis accounts for as much of the remaining variance as possible while being uncorrelated with the first axis, and so on for other subsequent axes. The main focus will be on how each principal component can be interpreted in terms of the original variables. For each site, object, or person so analyzed, it will have a value or score measured in the units of the new axes according to the value of each of the original variables (i.e., a new value after the data transformation). PCA is therefore an appropriate approach to integrate the environmental and socioeconomic variables together at the block group level and to develop a quality of life score for each block group. Using PC SAS software, a correlation analysis was first run among the seven variables: (a) population density, Co) per capita income, (c) median home value, (d) percentage of

Landsat TM Data for Quality of Life Assessment

265

UJ v

~|lgll

t i

i 0

¢0

¢

7

7

FIGURE 3. Per capita income by block groups.

0

i

14 Kilomet~

18211 - 2 6 8 2 6

3085 - 6187 6188 - 8726

~ r capita I ~

$

o

O~ O~

8

8

F I G U R E 4. Percentage of urban use by block groups.

0

16 Kilometers r

S

N

8.5-15.59 15.59 - 2 9 . 0 8 29.08 - 51.33

Urban Land Use % 0.11 - 2.75 2.75 - 8.5

~,l

o~

~D

o I'-

r'-

8

8

FIGURE 5. NDVI of Athens-Clarke County by block groups.

0

0.182 - 0.314 0.314 - 0.393

S

N

0.436 - 0.49 0.49 - 0.545

16 Kilometers K

II

~

B B 0.3,93-0.43o

NDVI

I". o

O~ QO

8

8

16 ~ e r s i

FIGURE 6. Apparent surface temperatures (in °C) of Athens-Clarke County by block groups,

0

I I I

S

25.69 ~.9728-38 30-08"

26.97 28.38 30.08 31.9

25.06 - 25.c~

Surface Temperature (°C)

m co

o,,

Jo

~"

=~

270

C. P. Lo

college graduates, (e) percentage of urban use, (f) NDVI, and (g) surface temperature, all aggregated at the block group level. It is noteworthy that NDVI is negatively correlated with the percentage of urban use (r= -0.8498 at the 0.0001 level of confidence), surface temperature (r = -0.8838), and population density (r = -0.4002), but positively correlated with per capita income (r= 0.5637), median home value (r=0.4284), and percentage of college graduates (r = 0.3065) (Table 1). Clearly, NDVI, which measures the greenness of the environment, is directly related to such socioeconomic variables as income, house value, and education. The PCA based on the correlation matrix identified two principal components which have eigenvalues greater than 1 (Table 2). The choice of 1 as the cutoff point is quite reasonable because only seven variables were used in PCA. The communality, which is the sum of the square of the component loadings for each variable, indicates the proportion of the variance for each variable accounted for by the two principal components. Thus, in this study, 90% or more of the variance for the variables of NDVI by block groups and surface temperature were accounted for by the two principal components. With the exception of the population density variable (44%), the two principal components together accounted for 70 to 78 % variance of the land cover and socioeconomic variables. In other words, the two principal components alone reflected very strongly the environmental characteristics and moderately the socioeconomic characteristics of the study area. The eigenvalue (or the sum of squared component loadings for a particular principal component) divided by the total number of variables (seven) employed in the analysis indicated the percent of variance accounted for by that principal component. The first principal component, which accounted for 54% of the total variance, showed high positive Table I. Corrvlation Matrix of Variables

Popden Percap Medhome PercoU NDVIblk Perurb Surtemp

Popden

Percap

Medhome

Percoll

NDVIblk

Perurb

Surtemp

1.0000 -0.3008 0.0398 -0.0885 -0.4002 0.1964 0.4885

1.0000 0.5907 0.7176 0.5637 -0.5302 -0.4537

1.0000 0.5501 0.4284 -0.4605 -0.2899

1.0000 0.3065 -0.2996 -0.1227

1.0000 -0.8498 -0.8838

1.0000 0.7747

1.0000

Popden, population density; Percap, per capita income; Medhome, median home value; Percoll, percent of college graduates; NDVIblk, NDVI by block groups; Perurb, percent of urban use; Surtemp, surface tempcraturvs. Table 2. Principal Component Analysis Loadings Component loadings Variables Popden Percap Medhome PercoU NDVlblk Perurb Surtemp Eigenvalue % variance

C1

C2

Commuuality

- 0.42881 0.80670 0.64172 0.57380 0.90438 -0.85299 - 0.82029 3.7941 54.20

0.50695 0.36255 0.54280 0.66469 -0.28840 0.16968 0.48673 1.4738 21.05

0.44 0.78 0.71 0.77 0.90 0.76 0.91

Landsat TM Data for Qual#y of Life Assessment

271

loadings from NDVI, per capita income, median home value, and percentage of college graduates, and high negative loadings from percentage of urban use, surface temperature, and population density. Thus, the first component which integrated NDVI with per capita income can be labeled as "the greenness of the environment". The second principal component, which accounted for 21% of the total variance, exhibited high positive loadings from percentage of college graduates, median home value, population density, and surface temperature. Thus, the second principal component is more socioeconomic in nature than environmental, and can be labeled as "personal traits". This matches up very well with the dual sub-division of social space into "morphological" and "sociocultural" environments expounded by Chombart de Lauwe (1952). Figure 7 illustrates this dichotomous relationship between the cluster of "environmental" variables and the cluster of "socioeconomic" variables. Interestingly, the NDVI variable is much closer to the "socioeconomic" cluster than the "environmental" one. Thus, the first principal component which integrated both environmental and socioeconomic characteristics of the study area is a good quality of life indicator. The first principal component score for each block group is a useful measure of quality of life. A map of the first principal component scores by census block groups was produced using standardized principal components of the PCA module of the IDRISI GIS program. Because IDRISI is a raster GIS, the first principal component scores have been scaled into 0-255 range for color display. As a black-and-white illustration for this paper, the raster display has been converted into a choropleth map in vector form using ARC/INFO and ARCVIEW programs (Figure 8). A higher level of quality of life is associated with a higher principal component score. The map reveals high quality of life in the southeast tip [block groups 1502.2 (#54 on Figure 8) and 1502.1 (#50)] and southwest edge [block groups 1502.4 (#40) and 12.1 (#47)] of the county, while low quality of life is found in the downtown area [block group 1.1 (#22) and the surrounding block groups]. A sectoral pattern of spatial variation in quality of life is revealed, indicating the control of highways, urban use, and vegetation cover. The low quality of life block groups extend eastward and westward from the downtown block group 1.1. Retail and commercial development along the Atlanta Highway has been responsible for the low quality of life of the block groups that the highway passes. On the other hand, the high quality of life block groups have low density residential development with excellent tree cover.

DEVELOPMENT OF QUALITY OF LIFE SCORES USING GEOGRAPHIC INFORMATION SYSTEM OVERLAY The principal components analysis represents an objective approach to the determination of the quality of life of the study area by block groups. To complement it, an alternative subjective approach was also adopted. Each of the environmental and socioeconomic variables was ranked according to a person's perception of high and low quality of life. The value of each variable was assigned to one of 10 rank scores, with 1 being the lowest and 10 being the highest. Thus, in the case of "percentage of urban use", "surface temperature", and "population density", the higher the value the less desirable it becomes, so that higher values were given low rank scores. On the other hand, the variables of "NDVI", "per capita income", "median home value", and "percentage of college graduates" were assigned high rank scores with higher values which are more desirable. In each case, the ranking of the variable was determined by finding the difference

1.0

t

-0.4

I

-1.0

-0.8

-0.6

-0.4

-0.2

0.2

I

0.4

I

I

0.6

04

FIGI.,T1LE 7. Relationship between the fix,s t and second principal components.

-0.2

0.0

J

-0.6

I

-0.8

0.2

0.4

6

•1

1. POPULATION DENSITY 2. PER CAPITA INCOME 3. MEDIAN HONE VALUE 4. PERCENT COLLEGE GRADUATES 5. NDVI 6. PERCENT URBAN USE 7. SURFACE TEMPERATURE

KEY

•

07

0.6

0.8

COMPONENT 2 1.0

03

0.8

I

•

2

o5

1.0

I

COMPONENT 1 o

I,o

8

FIGURE 8. First principal component score by block groups.

0

16 Kilometers r

S

N

1st P.C. Score IITrlTH] -2.37--2.01 ~-~ -2.01 --0.7 -0.7- 0.28 0.28- 1.04 1.04- 1.89

i

o

I"-

274

C. P. Lo

between the highest and the lowest values and then dividing the differenceinto I0 equal intervals. N o weighting on the rank scores of each variable was made, because without conducting any opinion survey, it would be too arbitraryto decide what weight to use for each variable. Using the O V E R L A Y A D D function in the IDRISI (}IS program, the aggregated rank score for all seven variables for each block group was obtained (Figure 9). The highest possible aggregate rank score is 70 while the lowest is 7. The resultant map confirmed the observations made earlierin connection with the principal component scores that the southeast tip and southwest edge of the county have the highest quality of life while the downtown area and areas around the Atlanta highway have much lower quality of life.This ranking score approach gave rise to more subtle differentiationof quality of lifeamong block groups than in the case where the firstprincipal component scores were used. The northern part of the county exhibitslow quality of lifetoo, largelyweighed down by lower per capita income.

CONCLUSIONS This research has demonstrated the usefulness of high-resolution satellite images, such as Landsat TM data, for quality of life assessment in an urban environment. Quality of life is a collective attribute that adheres to groups of people. The satellite images provide environmental information on land cover, which, with the aid of a vegetation index, can be linked to socioeconomic data obtainable from the census. A hybrid raster and vector GIS approach has to be used. The socioeconomic data from the census which are aggregated by block groups have to be converted to raster format to match up with the environmental data extracted from the satellite images. In turn, the per pixel environmental data have to be aggregated to the block group level. In applying to the study area of Athens-Clarke County in Georgia, this new approach in urban analysis has succeeded in integrating the biophysical variables with the socioeconomic variables in characterizing the quality of life of a block group with the aid of principal components analysis (PCA) and GIS overlay. The PCA produced the greenness component and a personal trait component with NDVI providing the key link, displaying very dearly the two main components of the social space: the morphological and the sociocultural environment. The first principal component scores by block groups provide a simplified but accurate picture of quality of life variations in space in the study area. The GIS overlay approach is an alternative approach which ailows an accumulated rank score for each block group within the county to be computed, thus providing a more subtle differentiation of block groups based on the ranking of environmental and socioeconomic variables that affect people's quality of life. Although the study area is a medium-size campus town with a lot of vegetation, the approach reported in this paper should be applicable to larger cities in the United States, and cities of varying sizes in Canada and Australia, where greenness and low density are important considerations for residential quality. Integrating satellite images with census data is the best approach to expand the use of satellite images in urban analysis, particularly in extracting the socioeconomic characteristics of the urban environment. However, further refinement of this approach is still required. Research on disaggregating census data to individual pixels will be required so that a more seamless integration between the raster image data and the vector

8

8

FIGURE 9. Aggregate rank scores obtained by GIS overlay by block groups.

0

51.9 - 60.9

I

16 I~lorneters

S

N

VV E

I

44.9- 51.9

35.9 - 44.9

Rank Score I]]]mt 18 - 23.1 23.1 - 35.9

=.e,.

o

5" "",l

=_,.

Q}

276

C. P. Lo

census data can be achieved. Another research direction should be to study the universal applicability of the greenness concept and its linkages to the socioeconomic characteristics of urban environments in different cultural settings.

ACKNOWLEDGEMENTS~The author gratefully acknowledges the financial support of EOSAT for this research. He also wishes to thank Ben Faber for research assistance. The author appreciates the comments of three anonymous reviewers, which helped improve this paper.

REFERENCES Andrews, F. M. (1986). Research on the quality of life. Ann Arbor, MI: Survey Research Center. Bederman, S. H., & Hartshorn, T. A. (1984). Quality of life in Georgia: The 1980 experience. The Southeastern Geographer, 24, 78-98. Chombart de Lanwe, P. H. (1952). Paris et l'agglomeration parisienne; L L'espace social darts ane grande citd. Paris: Presses Universitaires de France. Daultrey, S. (1976). Principal Components Analysis. Concepts and Techniques in Modern Geography, No. 8. Norwich: Geo Abstracts Ltd., University of East Anglia. De Haas, W. G. L. (1966). Integrated surveys and the social sciences. Delft, The Netherlands: Publications of the ITC-UNESCO Centre for Integrated Surveys. Forster, B. (1983). Some urban measurements from Landsat data. Photogrwnmetric Engineering, 49, 1693-1707. Green, N. E. (1957). Aerial photographic interpretation and the social structure of the city. Photogrammetric Engineering, 23, 89-96. Hodler, T., Lawson, N., Schretter, H., & Torguson, J. (1994). The interactive atlas of Georgia. Athens, GA: Institute of Community and Area Development. Jensen, J. R. (1983). Biophysical remote sen=rig. Annals of the Association of American Geographers, 73, 111-132. Jensen, J. R. (1996). Introductory digital image processing: A remote sensing perspective (2nd ed.). Upper Saddle River, NJ: Prentice Hall. Langford, M., Maguire, D. J., & Unwin, D. J. (1991). The areal interpolation problem: Estimating population using remote sensing in a GIS framework. In I. Masser & M. Blakemore (Eds.), Handling geographical information: Methodology and potential applications (pp. 55-77). New York: Longman Scientific and Technical. Liu, B. C. (1970). Quality of life indicators in U.S. metropolitan areas, 1970. U.S. Environmental Protection Agency. McPhersen, E. G. (1994). Cooling urban heat islands with sustainable landscapes. In R. H. Platt, R. A. Rowntrec, & P. C. Muick (F_xls.), The ecological city." Preserving and restoring urban biodiversity (pp. 151-171). Amherst, MA: The University of Massachusetts Press. Malaret, E., Bartolucci, L. A., Lozano, D. F., Anuta, P. E., & MeGillem, C. D. (1985). Landsat-4 and Landsat-5 Thematic Mapper data quality analysis. Photogrammetric Engineering and Remote Sensing, 51, 1407-1416. Metivier, E. D., & McCoy, R. M. (1971). Mapping urban poverty housing from aerial photographs. In Proceedings of the Seventh International Symposium on Remote Sensing of Environment (pp. 1563-1569). Ann Arbor, MI: University of Michigan. Mumbower, L. E., & Donoghue, J. (1967). Urban poverty study. Photogrammetrie Engineering, 33, 610-618. Nichol, J. E. (1994). A GIS-based approach to microclimate monitoring in Singapore's high-rise housing estates. Photogrammetric Engineering and Remote Sensing, 60, 1225-1232. Richards, J. A. (1986). Remote sensing digital image analysis: An introduction. Berlin: Springer-Veriag. Shelton, G. G., Grubber, K. J., & Godwin, D. D. (1983). The effect ofhoasing type on the quality of living: A comparison of residents of conventional homes, mobile homes, and apartments in rural North Carolina. Greensboro, NC: North Carolina A&T State University and Cooperative State Research Service, USDA. Szalai, A. (1980). The meaning of comparative research on the quality of life. In A. Szalai & F. M. Andrews (Eds.), The quality of life: Comparative studies (pp. 7-21). London: Sage Publications. Wallace, S. (1971). Quality of life. Journal of Home Economics, 66, 7-8. Weber, C., & Hirsch, J. (1992). Some urban measurements from SPOT data: Urban life quality indices. International Journal of Remote Sensing, 13, 3251-3261.