Bi-temporal characterization of land surface temperature in relation to impervious surface area, NDVI and NDBI, using a sub-pixel image analysis

YoushuiZhanga, Inakwu O.A. Odehb,,and ChunfengHana

aCollege of Geography, Fujian Normal University, Fuzhou 350007, China

bFaculty of Agriculture, Food and Natural Resources, The University of Sydney, Sydney, NSW 2006, Australia

Received 1 August 2008;

accepted 10 March 2009.

Available online 9 April 2009.

Abstract

As more than 50% of the human population are situated in cities of the world, urbanization has become an important contributor to global warming due to remarkable urban heat island (UHI) effect. UHI effect has been linked to the regional climate, environment, and socio-economic development. In this study, Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) imagery, respectively acquired in 1989 and 2001, were utilized to assess urban area thermal characteristics in Fuzhou, the capital city of Fujian province in south-eastern China. As a key indicator for the assessment of urban environments, sub-pixel impervious surface area (ISA) was mapped to quantitatively determine urban land-use extents and urban surface thermal patterns. In order to accurately estimate urban surface types, high-resolution imagery was utilized to generate the proportion of impervious surface areas. Urban thermal characteristics was further analysed by investigating the relationships between the land surface temperature (LST), percent impervious surface area, and two indices, the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Built-up Index (NDBI). The results show that correlations between NDVI and LST are rather weak, but there is a strong positive correlation between percent ISA, NDBI and LST. This suggests that percent ISA, combined with LST, and NDBI, can quantitatively describe the spatial distribution and temporal variation of urban thermal patterns and associated land-use/land-cover (LULC) conditions.

Keywords:Urban heat island; Land surface temperature; Impervious surface area; NDVI; NDBI

Article Outline

1.Introduction

2.Methods

2.1.Study area and data

2.2.Image pre-processing

2.3.Derivation of LST, NDVI and NDBI from TM and ETM+ imageries

2.3.1.LST

2.3.2.NDVI and NDBI

2.4.The derivation of urban percent ISA

3.Results and discussion

3.1.Land surface imperviousness and LST

3.2.Relationship among imperviousness, NDVI, NDBI and LST

3.3.Relationship between UHI and LULC patterns

4.Conclusions

References

1. Introduction

Changes in land cover (which is the biophysical attributes of the earth's surface) and land use (the utilization of land for a given human purpose) have been reported to be the main driver of environmental change, while climatic change and climate variability may influence land-use preferences differently in different parts of the world (Brunsell, 2006). Therefore accurate and up-to-date information on land cover and the state of the environment is critical to environmental monitoring, management and planning (Assefa, 2004). However, the interaction between land-use/land-cover (LULC) change and the spatial–temporal climatic variability is poorly understood which requires the development of new models linking the climate variability with the changes in land use and land cover, especially at the urban–rural regional scale.

Recent developments in remote sensing and image analysis have identified land surface temperature (LST) as one of the key parameters controlling the physical, chemical and biological processes at the interface between the Earth and the atmosphere. It is an important factor for the study of urban climate ([Voogt and Oke, 2003]and[Small, 2006]). LST has been shown to be an effective means of partitioning latent heat fluxes and thus surface radiant temperature response as a function of varying surface soil water content and vegetation cover (Owen et al., 1998). These findings have encouraged investigations of the relationship between LST and vegetation abundance (e.g.,[Gallo and Owen, 1998a],[Gallo and Owen, 1998b],[Weng, 2001]and[Weng et al., 2004]). Urbanization and industrialization can lead to modification of land surface and near-surface atmospheric conditions, which in turn could cause change in thermal properties of urban areas causing them to be warmer than the surrounding non-urbanized areas.This phenomenon is called urban heat island (UHI), which is mainly caused by replacement of vegetated areas by non-evaporating and impervious materials such as asphalt and concrete ([Dousset and Gourmelon, 2003],[Kim, 1992]and[Ruiliang et al., 2006]). The UHI phenomenon can influence the radiative fluxes in the near-surface flow because in urban areas, the higher level of sensible heat fluxes is caused by LULC changes caused by the removal of the original vegetated areas that were characterized by lower heat fluxes. Furthermore, urbanization generally leads to reduced evapotranspiration and more rapid runoff of rainwater.

The monitoring of UHI phenomenon and the physical processes associated with it have traditionally been conducted by ground-based observations taken from fixed thermometer networks or by traversing a targeted area with a thermometer mounted on vehicles ([Voogt and Oke, 2003]and[Weng et al., 2004]). However, the advent of remote sensing technology has made it possible to study UHIs using satellite remote sensing data, especially thermal data taken by either satellite-borne or air-borne sensors.Rao (1972)was the first to demonstrate the possibility of identifying urban areas based on the analyses of thermal infrared data acquired by a satellite sensor.Following this,Gallo et al. (1995)reviewed the validity and utility of UHI derived from the satellite-acquired imagery.(Gallo and Owen, 1998a)and(Gallo and Owen, 1998b)and(Streutker, 2002)and(Streutker, 2003)derived land surface temperature and evaluated the UHI phenomenon using National Oceanic and Atmospheric Administration Advanced Very High-Resolution Radiometer (NOAA AVHRR) data for regional-scale urban temperature mapping. Similar studies byChen et al. (2006)andWeng (2001)used Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) thermal infrared (TIR) data to assess the local patterns of UHI. Generally, with increasing availability of thermal remote sensing data from Landsat Thematic Mapper, Advanced Space-borne Thermal Emission and Reflection Radiometer (ASTER), and the NOAA AVHRR, research of LST, using thermal images, has been a topic of great interest in the remote sensing community for the past three decades. The application of low spatial resolution remotely sensed data is requisite for understanding the affects of UHI on climatic patterns. However, understanding the thermal response of individual land covers within a city is equally valuable, as radiative transfer over such intensively developed environment varies significantly over a short space due to the diversity of urban land covers and their respective physical properties (Renee et al., 2006).

In comparison with thermal remote sensing of natural and agricultural surfaces, thermal remote sensing of urban areas has been slow to advance beyond qualitative description of thermal patterns and simple correlation analysis (Ruiliang et al., 2006). Recently,Voogt and Oke (2003)reviewed most of previous research on UHI study and listed three themes of research: examination of the spatial structure of urban thermal patterns and their relation to urban surface characteristics (e.g.,[Balling and Brazel, 1988]and[Dousset and Gourmelon, 2003]); thermal remote sensing for urban surface energy balances (e.g.,[Assefa, 2004]and[Kim, 1992]); and study on the relation between atmospheric heat islands and surface urban heat islands (e.g.,[Ben-Dor and Saaroni, 1997]and[Caselles et al., 1991]). The main objective of this study is to address the first theme.

Many of the previous remote sensing studies of the urban environment used Normalized Difference Vegetation Index (NDVI) as a major indicator of urban climate ([Lo et al., 1997],[Gallo and Owen, 1999]and[Yuan and Bauer, 2007]). However, NDVI is subject to seasonal variations which may influence the results of land surface UHI analysis. Moreover, the relationship between NDVI and LST is well known to be nonlinear, due to the predominantly bare ground surfaces which tend to exhibit larger variation in surface radiant temperature than the densely vegetated LULC types ([Price, 1990],[Gillies and Carlson, 1995],[Owen et al., 1998]and[Chen et al., 2006]). The variability and nonlinearity suggest that NDVI alone may not be sufficient to quantitatively study UHI. The intensity of UHI is related to the spatial extent and composition of vegetation and built-up areas and their temporal changes. Quantitative studies of the relationship between LULC patterns and LST are important for land-use management and planning. Furthermore while NDVI has been used for the estimation of vegetation productivity and rainfall in semi-arid areas ([Chen et al., 2004]and[Wang et al., 2004]), the Normalized Difference Built-up Index (NDBI) has been developed for the identification of urban and built-up areas (Zha et al., 2003). It is therefore possible that the utilization of both NDVI and NDBI as surrogates of LULC can reveal the relationships between different indices such as NDVI, NDBI, and land surface temperature in UHI studies.

Impervious surfaces, defined as land-cover types that impede water infiltration are primarily associated with transportation (streets, highways, parking lots and sidewalks) and building rooftops. In remote sensing, classified impervious surface area (ISA) has been used to quantify and map the degree of urbanization and extent of urban land use ([Yuan and Bauer, 2007],[Xian and Crane, 2006]and[Civco et al., 2002]). With increased concern regarding global climate change, it is important to analyse the relationship between the LST and percent ISA in an urbanized environment as an alternative approach to the study of urban expansion. Compared to the NDVI, the percent ISA is more stable and less affected by seasonal changes in landscape conditions, which means that percent ISA may provide an additional metric for the analysis of LST and urban thermal patterns.

The aims of this study are to investigate the relationships of LST with NDVI, and those of NDBI with percent ISA using Landsat TM and ETM+ data obtained for the city of Fuzhou in south-eastern China; and to quantitatively compare the patterns and intensity of UHI with LULC types. In order to generate verifiable estimates of change in urban extent, the LULC types were quantitatively determined by implementing sub-pixel percent imperviousness estimation and selecting certain threshold values of percent ISA. In developing this method of quantifying the urban LULC types and associated surface thermal distribution using remote sensing data, it is envisaged that the method could be used for other similar geographical regions in China and indeed elsewhere.

2. Methods

2.1. Study area and data

The study area is Fuzhou City, located in the southeast coast of China (Fig. 1). With a population of over 5.75 million, Fuzhou is the major coastal city located between Hong Kong and Shanghai. The city is on a subtropical plain sandwiched between the Gu and Qi mountains with potential for expansion in all directions. Like many other Chinese cities, the population of Fuzhou is rapidly increasing leading to increased urban expansion. This urban growth is encroaching into the adjacent agricultural and other non-urban land. With sweltering summer and mild winter, the city has several advantages that make it suitable for our study. It is characterized by a diversity of land-cover types transversed by the Min River. The city is also characterized by high-, medium- and low-density urban developments in the central portion and several rural land-cover types – predominantly agricultural fields, forests, water and bare land in the surrounding landscapes. The built environment consists of buildings and roofs made up of concrete, brick tiles and metal plates, and majority of the roads are covered by asphalt and concrete. The city is therefore ideally suitable for the analysis of UHI phenomenon due to its diversity of land-cover types and the rapid urbanization.

/ Full-size image(57K)
High-quality image(545K)

Fig. 1.Location map of study area showing the aerial photograph image.

To quantitatively derive LST and compute UHI intensity, Landsat 5 TM image (acquired on June 15, 1989) and Landsat 7 ETM+ image (acquired on March 4, 2001) were used. While bands 1–5 and 7 images have nominal spatial resolution of 30m, the thermal infrared band (band 6) has 120m spatial resolution for TM image and 60m for ETM+ image. In addition, an IKONOS image (acquired on October 29, 2000) with 1m spatial resolution and aerial photographs (acquired on May 20, 1988) with 2m spatial resolution, rectified to the Universal Transverse Mercator (UTM) coordinate system, were used to calculate the percent ISA. Land-cover classification of the remote sensing imageries was carried out using ancillary data obtained from 1:10,000 scale digital topographic maps.

2.2. Image pre-processing

To analyse the changes in temperature in relation to LULC types, the bi-temporal TM/ETM+ images were geo-referenced to a common UTM coordinate system based on the rectified high-resolution IKONOS image, aerial photograph and the 1:10,000 scale topographic maps. The RMSE of rectification is less than 0.3pixels (≈9m). Using the radiometric correction method ofSchroeder et al. (2006), the original digital numbers of bands 1–5 and 7 images were converted to at-satellite radiance, at-satellite reflectance, and further converted to surface reflectance. While bands 1 through 5 and band 7 are at a spatial resolution of 30m, the thermal band (band 6) comes at an original spatial resolution of 120m for TM and 60m for ETM+. In order to carry out further analysis on a common spatial resolution, bands 1–5 and band 7 of both Landsat imageries were resampled onto 120m using the cubical convolution algorithm.

2.3. Derivation of LST, NDVI and NDBI from TM and ETM+ imageries

2.3.1. LST

LST is the radiative skin temperature of the land surface which plays an important role in the physics of the land surface through the process of energy and water exchanges with the atmosphere. The derivation of LST from satellite thermal data requires several procedures: sensor radiometric calibrations, atmospheric and surface emissivity corrections, characterization of spatial variability in land-cover, etc. As the near-surface atmospheric water vapour content varies over time due to seasonality and inter-annual variability of the atmospheric conditions, it is inappropriate to directly compare temperature values represented by the LST between multiple periods. Therefore the focus here is on the UHI intensity and its spatial patterns across the study region. UHI intensity is estimated as the difference between the peak temperatures (LST) of the urban area and the background non-urban temperatures (Chen et al., 2006). This UHI effect can be determined for the individual thermal images and then compared between two or more periods. However, before we compute UHI effect, we must first derive the LST based on different methods for TM and ETM+ images.

As described above the TM and ETM+ thermal infrared band (10.4–12.5μm) data were used to derive the LST.Yuan and Bauer (2007)proposed a method of deriving LST in three steps: Firstly, the digital numbers (DNs) of band 6 are converted to radiation luminance or top-of-atmospheric (TOA) radiance (Lλ, mW/cm2sr) using:

(1)

whereb6is the pixel digital number for band 6,Lmax=1.896 (mW/cm2sr), andLmin=0.1534 (mW/cm2sr). In the case of Landsat 7 ETM+ image, TOA is derived by:

(2)

whereQCALmin=1, QCALmax=255, andLmax=17.04W/(m2srμm), andLmin=0.

Secondly, the TOA radiance is converted to surface-leaving radiance by removing the effects of the atmosphere in the thermal region. An atmospheric correction tool – MODTRAN 4.0 for the thermal band of Landsat sensors was applied. This tool uses the MODTRAN radiative transfer code and a suite of integration algorithms to estimate three parameters – atmospheric transmission, and upwelling and downwelling radiance – which enable the calculation of the surface-leaving radiance – LT or the radiance of a blackbody target of kinetic temperatureT, in the form of (Eq.(3)):

(3)

where TOA is the radiance derived for the instrument,Lμis the upwelling or atmospheric path radiance,Ldis the downwelling or sky radiance,τis the atmospheric transmission, andis the emissivity of the surface specific to the target type. Radiance values are in units of W/(m2srμm) and the transmission and emissivity is unitless. The emissivity could be based on the land-cover classification (Yuan et al., 2005) or the emissivity values as derived bySnyder et al. (1998).

Lastly, the radiance (LT) is converted to surface temperature (LST) using the Landsat specific estimate of the Planck curve (Eq.(4)) (Chander and Markham, 2003):

(4)

where LST is the temperature in Kelvin (K),K1is the pre-launch calibration constant in W/(m2srμm) andK2is another pre-launch calibration constant in Kelvin. For Landsat 5 TM,K1=607.76W/(m2srμm) andK2=1260.56K; for Landsat 7 ETM+,K1=666.09W/(m2srμm) andK2=1282.71K.

The LST image from the thermal band of ETM+ image (band 6) with original spatial resolution of 60m was resampled to 120m using the nearest neighbour algorithm to match the pixel size of the LST image from TM image.

2.3.2. NDVI and NDBI

The NDVI and NDBI indices were required to characterize the LULC types and to explore the quantitative relationships between LULC types and UHI. NDVI (Eq.(5), which has generally been used to express the density of vegetation (Purevdorj et al., 1998) is of the form

(5)

whereρ3is the reflectance value of red band (band 3) andρ4is the reflectance value of near-infrared band (band 4), both of the Landsat images. The NDVI values range from −1 to 1, with positive values indicating vegetated areas and negative values signifying non-vegetated surface features.

Another index used in this study that is sensitive to the built-up area is NDBI (Zha et al., 2003), derived as

(6)