Tải bản đầy đủ - 0 (trang)
2 Case Study: Land-Use Zoning and LST Variations

2 Case Study: Land-Use Zoning and LST Variations

Tải bản đầy đủ - 0trang

212



Chapter Seven

with different spatial resolutions, that is, two visible bands and one

near-infrared (VNIR) band with 15-m spatial resolution, six shortwave infrared (SWIR) bands with 30-m spatial resolution, and five

thermal infrared (TIR) bands with 90-m spatial resolution. The level

1B ASTER data were purchased, and they consisted of the image data,

radiance conversion coefficients, and ancillary data (Fujisada, 1998).

Although these ASTER images have been corrected radiometrically

and geometrically, the root-mean-square error (RMSE) was so large

that the images could not meet our research needs. We conducted a

geometric correction for each image using 1:24,000 U.S. Geological

Survey (USGS) digital raster graphs (DRG) as reference maps. At

least 30 ground control points were selected for each georectification,

and the nearest-neighbor resampling method was applied with a

pixel size of 15 m for all VNIR, SWIR, and TIR bands. An RMSE of

less than 0.5 pixel was achieved for all georectifications.



7.2.2



LULC Classification



LULC maps were produced with an unsupervised classification algorithm called iterative self-organizing data analysis (ISODATA). The

images (VNIR and SWIR bands) were classified into six LULC types,

including agricultural land, aquatic systems, barren land, developed

land, forest land, and grassland. Agricultural land is characterized by

herbaceous vegetation that has been planted or is intensively managed for the production of food. This category includes croplands

such as corn, wheat, soybeans, and cotton and also includes fallow

land. Aquatic systems refer to all areas of open water, including lakes,

rivers, streams, ponds, and outdoor swimming pools. Barren land is

characterized by bare rock, gravel, sand, silt, clay, or other earthen

material and no green vegetation present, including quarries, bare

dunes, construction sites, and mines. Developed lands are defined as

the areas that have a high percentage (30 percent or greater) of constructed materials, such as asphalt, concrete, and buildings. This category includes commercial, industrial, transportation, and low and

high residential uses. Forest lands are the areas characterized by tree

cover (natural or seminatural woody vegetation), including natural

deciduous forest, evergreen forest, mixed forest, urban and suburban

forest, and shrubs. Grasslands refer to the areas covered by herbaceous vegetation, including pasture/hay planted for livestock grazing or the production of hay. It also includes the urban/recreational

grasses planted in developed settings for recreation, erosion control,

or aesthetic purposes, such as parks, lawns, golf courses, airport

grasses, and industrial-site grasses.

Remotely sensed data are often highly correlated between the

adjacent spectral wavebands, and redundant bands would slow down

the image processing if all bands were used. Principal components

analysis (PCA) was applied to identify which bands of the nine VNIR



U r b a n L a n d S u r f a c e Te m p e r a t u r e A n a l y s i s

and SWIR bands would contain more information. Based on the

correlation coefficients between the bands and the first component,

five to six bands normally would be selected for classification.

Because the visible (green and red) and near-infrared bands were

found useful in all cases, these three bands were always used in the

image classifications.

Owing to the complexity of LULC types in the study area, each

image scene was first stratified into two subscenes, one for the urban

area and the other for the surrounding rural area. Image classification

was performed separately for each subscene, and the results then were

merged. Fifty spectral clusters were generated with ISODATA. Next,

spectral classes were labeled after referencing to high-resolution aerial

photographs and other geospatial data. After the first unsupervised

classification, if we could not label all spectral classes, we then would

mask out those confused classes and run a second unsupervised classification. The same procedure would be repeated for a third classification if confusion still existed after the second classification.

Accuracy assessment of classification images was conducted by

using an error matrix. Some important measures, such as overall accuracy, producer’s accuracy, and user’s accuracy, can be calculated from

the error matrix (Foody, 2002). In this study, a total of 350 points

were checked on each classified image using a stratified randomsampling method. Digital orthophotographs from 2003 were used

as the reference data. The color orthophotographs were provided by

the Indianapolis Mapping and Geographic Infrastructure System,

which was acquired in April 2003 for the entire county. The orthophotographs had a spatial resolution of 0.14 m. The coordinate system belonged to Indiana State Plane East, Zone 1301, with North

American Datum of 1983. The orthophotographs were reprojected

and resampled to 1-m pixel size for the sake of quicker display and

shorter computing time. Figure 7.1 shows the resulting classified

LULC maps. Overall classification accuracy of 87 percent (October 3,

2002, image), 88.33 percent (June 16, 2001, image), 92 percent (April 5,

2004, image), and 87.33 percent (February 6, 2006, image) was

achieved, respectively. We made a further comparison among the

four classified maps to ensure consistency within the classes and

found that the magnitude and spatial pattern of each class corresponded well with each other but also reflected the seasonal and temporal differences.



7.2.3



Spectral Mixture Analysis



Spectral mixture analysis (SMA) is regarded as a physically based

image processing technique that supports repeatable and accurate

extraction of quantitative subpixel information (Mustard and Sunshine,

1999; Roberts et al., 1998; Smith et al., 1990). It assumes that the spectrum measured by a sensor is a linear combination of the spectra of all



213



214



Chapter Seven



(a)



N

0



2.5



5



10

Kilometers



15



20



Legend

Urban

Forest



Agriculture

Water



Grasslands



Barren lands



FIGURE 7.1 LULC maps of four seasons in Marion County, Indianapolis,

Indiana, derived from ASTER images: (a) October 3, 2000; (b) June 16, 2001;

(c) April 5, 2004; and (d ) February 6, 2006. (Adapted from Weng et al., 2008.)

See also color insert.



components within the pixel (Adams et al., 1995; Roberts et al., 1998).

Because of its effectiveness in handling spectral mixture problems,

SMA has been used widely in estimation of vegetation cover (Asner

and Lobell, 2000; McGwire et al., 2000; Small, 2001), in vegetation or

land-cover classification and change detection (Aguiar et al., 1999;

Cochrane and Souza, 1998), and in urban studies (Lu and Weng, 2004,

2006a, 2000b; Phinn et al., 2002; Rashed et al., 2001; Wu and Murray,

2003). In this study, SMA was used to develop green vegetation, soil,

and impervious surface fraction images. Endmembers were identified initially from high-resolution aerial photographs. An improved

image-based dark-object subtraction model has proved effective and



U r b a n L a n d S u r f a c e Te m p e r a t u r e A n a l y s i s



(b)



N

0



2.5



5



10

Kilometers



15



20



Legend



FIGURE 7.1



Urban



Agriculture



Forest



Water



Grasslands



Barren lands



(Continued)



was applied to reduce the atmospheric effects (Chavez, 1996; Lu

et al., 2002). After implementation of the atmospheric correction and

geometric rectification of the ASTER images, a constrained leastsquares solution was applied to unmix the nine VNIR and SWIR

bands of the ASTER imagery into fraction images, including highalbedo, low-albedo, vegetation, and soil fractions. An impervious

surface then was estimated based on the relationship between highand low-albedo fractions and impervious surfaces. For more details

about the derivation of fraction images, please refer to the article by

Lu and Weng (2006b).



7.2.4



Estimation of LSTs



Various algorithms have been developed for converting ASTER TIR

measurements into surface kinetic temperatures (i.e., LSTs), as



215



216



Chapter Seven



(c)



N

0



2.5



Legend

Urban



FIGURE 7.1



5



10

Kilometers



15



20



Agriculture



Forest



Water



Grasslands



Barren lands



(Continued)



reported by the ASTER Temperature/Emissivity Working Group

(1999) and Gillespie and colleagues (1998). However, a universally

accepted method is not available currently for computing LSTs from

multiple bands of TIR data such as those found on ASTER. In this

study, we selected ASTER band 13 (10.25 to 10.95 μm) to compute

LSTs because the spectral width of this band is close to the peak radiation of the blackbody spectrum given off by the urban surface of the

study area. Two steps were taken to compute LSTs: (1) converting

spectral radiance to at-sensor brightness temperature (i.e., blackbody

temperature) and (2) correcting for spectral emissivity. We adopted

the most straightforward approximation to replace the sensorresponse function with a delta function at the sensor’s central wavelength to invert LSTs with the assumption of uniform emissivity



U r b a n L a n d S u r f a c e Te m p e r a t u r e A n a l y s i s



(d)



N

0



2.5



5



10

Kilometers



15



20



Legend



FIGURE 7.1



Urban



Agriculture



Forest



Water



Grasslands



Barren lands



(Continued)



(Dash et al., 2002; Li et al., 2004; Schmugge et al., 2002). The conversion formula is

Tc =



C2

⎛ C



λ c ln ⎜ 5 1 + 1⎟

⎝ λ c π Lλ ⎠



(7.1)



where Tc = brightness temperature in kelvins (K) from a central

wavelength

Lλ = spectral radiance in W m–3 sr–1 μm–1, λc is the sensor’s central wavelength

C1 = the first radiation constant (3.74151 × 10–16 W m–2 sr–1 μm–1)

C2 = the second radiation constant (0.0143879 m · K)

The temperature values thus obtained are referenced to a blackbody. Therefore, corrections for spectral emissivity ε became necessary



217



218



Chapter Seven

according to the nature of the land cover. Each of the land-cover categories was assigned an emissivity value according to the emissivity

classification scheme by Snyder and colleagues (1998). The emissivitycorrected LST was computed as follows (Artis and Carnhan, 1982):

LST =



Tc

1 + (λ ∗ Tc /ρ)ln ε



(7.2)



where λ = the wavelength of emitted radiance (for which the peak

response and the average of the limiting wavelengths λ =

10.6 μm) (Markham and Barker, 1985)

ρ = h × c/σ (1.438 × 10–2 mK)

σ = Boltzmann’s constant (1.38 × 10–23 J/K)

h = Planck’s constant (6.626 × 10–34 J · s)

c = the velocity of light (2.998 × 108 m/s)

Relative LST is sufficient for mapping of the spatial variations of

urban land surface temperatures. Therefore, the effects of atmosphere

and surface roughness on LST were not taken into account in this

study. Lack of atmospheric correction may introduce a temperature

error of 4 to 7°C for the midlatitude summer atmosphere (Voogt and

Oke, 1998). The magnitude of atmospheric correction depends on the

image bands used, as well as atmospheric conditions and the height

of observation. Errors owing to urban effective anisotropy depend on

surface structure and relative sensor position and can yield a temperature difference of up to 6 K or higher in downtown areas (Voogt

and Oke, 1998).



7.2.5



Statistical Analysis



Based on previous research results (Oke, 1982, 1988), we hypothesized

that the spatial variations of LST were related to four groups of factors

that described LULC composition (six variables), biophysical conditions

(four variables), intensity of human activity (four variables), and landscape pattern (four variables) (Table 7.1). We computed the mean (and

standard deviation) values of LST and each potential factor per general zoning polygon and per residential zoning polygon. Tables 7.2

and 7.3 show the definitions for each zoning category and relevant

attributes. Multiple stepwise regressions then were applied to obtain

independent variables with statistical significance (P < 0.001). Variables that were removed from the stepwise regressions were not considered to be explanatory factors of LST variation and therefore were

excluded in the subsequent statistical analyses. Next, factor analysis

(specifically PCA) was conducted to transform the identified independent variables into a set of uncorrelated principal components. Factors

with eigenvalues greater than 1 were extracted (Kaiser, 1960), and the

factor loadings of each original variable were examined.



U r b a n L a n d S u r f a c e Te m p e r a t u r e A n a l y s i s



Categories of

Variables

LULC

composition



Biophysical

conditions



Intensity of

human activity



Landscape

patterns



TABLE 7.1



Variable



Meaning of Variable



Per_Ur



Percentage of built-up land



Per_Ba



Percentage of barren land



Per_Gr



Percentage of grassland



Per_Ag



Percentage of agricultural land



Per_Fo



Percentage of forest land



Per_Wa



Percentage of water bodies



NDVI



Mean value of normalized difference

vegetation index



GV



Mean value of green-vegetation

fraction derived from SMA



IMP



Mean value of impervious surface

fraction derived from SMA



SOIL



Mean value of soil fraction derived

from SMA



PAVEMENT



Percentage of pavement area per

zoning polygon



BLDG_ARE



Percentage of building area per zoning

polygon



MEAN_POP



Mean population in a zoning polygon



POP_DEN



Population density in a zoning polygon



SHAPE_IN



Shape index of a zoning polygon



FRACTAL



Fractal dimension of a zoning polygon



LANDSIM



Landscape similarity index ( percent)



DIVERS



Shannon’s diversity index of LULC

composition within a zoning polygon



Variables Applied to Examine LST Variations



7.2.6



Landscape Metrics Computation



Landscape-pattern metrics have been employed frequently in landscape ecology to characterize the arrangement of species, communities, and habitat patches within landscapes (Read and Lam, 2002).

Their potential for monitoring ecosystem changes and linking with

ecologic and environmental processes has been recognized (Li and

Reynolds, 1994). These metrics can be applied to create quantitative

measures of spatial patterns found on a map or remotely sensed

imagery. When applying landscape metrics to remotely sensed data,



219



220

Number of

Polygons



Percent of

Landscape



Mean

Polygon

Size (ha)



Mean

Shape

Index



Mean Fractal

Dimension



Zoning Category



Code



Description



Historical

preservation



HP



Historic preservation

district including a variety

of land uses

(mostly residential)



1



0.01



8.64



1.44



1.29



Special uses



SU



Wide variety of uses,

such as schools,

utility infrastructure,

cemeteries, libraries,

community centers,

charitable organizations,

golf courses, and penal

institutions



1232



17.84



14.09



1.38



1.32



University



UQ



Variety of land uses

typical of highereducation institutions,

including classroom,

office, dormitory, facility

maintenance, and parking

structures



13



0.19



19.24



1.28



1.35



Agriculture



DA



Agriculture and single

family, very low density



781



11.31



23.76



1.47



1.32



D



Variety of residential

categories summarized in

Table 7.3



1926



27.88



18.72



1.41



1.30



CBD



CBD



Central business district:

core activities of all types

with a variety of related

land uses



44



0.64



24.60



1.38



1.29



Commercial



C



Includes office buffer,

high-intensity office/

apartment, neighborhood

commercial, thoroughfare

service, and corridor

commercial districts



2185



31.63



4.58



1.38



1.34



Hospital



HD



Major hospital complexes

and campuses



33



0.48



17.45



1.44



1.30



Industrial



I



Variety of industrial uses,

including urban and

suburban and light,

medium, and heavy

industry



498



7.21



22.12



1.51



1.32



Park



PK



Permits all sizes and

ranges of public park land

and facilities, including

park peripheral areas

ensuring compatibility of

adjacent land use



166



2.40



42.92



1.53



1.31



221



Residential



TABLE 7.2



General Zoning (Polygon) Attributes



222

Zoning Category



Code



Description



Airport



A



Public airports,

municipally owned or

operated, including all

necessary

navigation and flightoperation facilities and

accessory uses



Number of

Polygons



Percent of

Landscape



Mean

Polygon

Size (ha)



Mean

Shape

Index



Mean Fractal

Dimension



28



0.41



97.67



1.53



1.33



Note: A zoning geographic information system (GIS) data layer was provided by the Indianapolis Mapping and Geographic Infrastructure System, City

of Indianapolis. Information on zoning category, code, and description was provided by the Metropolitan Planning Department, City of Indianapolis.

Other attributes in the table are computed and compiled by the authors. We also referred to Wilson and colleagues (2003) in compiling this table.



TABLE 7.2



General Zoning (Polygon) Attributes (Continued)



Tài liệu bạn tìm kiếm đã sẵn sàng tải về

2 Case Study: Land-Use Zoning and LST Variations

Tải bản đầy đủ ngay(0 tr)

×