Arab J Geosci DOI 10.1007/s12517-013-1244-3
ORIGINAL PAPER
Assessment of spatial relationship between land surface temperature and landuse/cover retrieval from multi-temporal remote sensing data in South Karkheh Sub-basin, Iran Yasser Ghobadi & Biswajeet Pradhan & Helmi Zulhaidi Mohd Shafri & Keivan Kabiri
Received: 27 September 2013 / Accepted: 11 December 2013 # Saudi Society for Geosciences 2013
Abstract Land surface temperature (LST) is a required input data for modeling in climatological, hydrological, agricultural, and change detection studies. In the current study, remotely sensed thermal infrared data were applied to assess LST in south Karkheh sub-basin, Iran. This research deals with the extraction of LST and land surface emissivity (LSE) and the relationship between vegetation abundance using Landsat 5 TM and Landsat 7 ETM+ for October 1998 and 2002 over the studied area. The landuse/land cover (LULC) map was derived by using maximum likelihood classifier. The variety of LSE was investigated and extracted on the basis of NDVI threshold method, as well as the Plank equation was used to derive LST over the study region. The relationship between LST and different LULC is determined by zonal GIS and regression analysis. The results illustrate that the emissivity lies in the range of 0.860–0.992 and 0.855–0.992 in 1998 and 2002, respectively. The maximum value of LSE equal to ∼0.99 was observed over the dense vegetation, while the minimum LSE value equal to ∼0.880 was found over the sand dune. LST and NDVI analysis exhibits a strong inverse correlation except over water bodies. The strongest coefficient correlation was found over wetland (−0.76 and R2 =0.78). Keywords Land surface temperature (LST) . Normalized difference vegetation index (NDVI) . Land surface emissivity (LSE) . Remote sensing . GIS
Y. Ghobadi : B. Pradhan (*) : H. Z. M. Shafri : K. Kabiri Department of Civil Engineering, Faculty of Engineering, University Putra Malaysia, 43400, Serdang Selangor, Malaysia e-mail:
[email protected] B. Pradhan e-mail:
[email protected]
Introduction Land surface temperature (LST) plays an important role in studies about the global change, heat balance as a factor for control of climate change, and urban heat island (Weng et al. 2004). LST is defined as the surface radiometric temperature corresponding to the instantaneous field of view of the sensor (Norman and Becker 1995; Srivastava et al. 2010). LST and emissivity (ε) are significant factors in energy budget assessment, land cover valuation, and other related studies to earth surface characteristics. This provides better understanding of the overall landuse and land cover (LULC) classes and environmental studies (Mallick et al. 2012; Tehrany et al. 2012, 2013a, 2013b; Jebur et al. 2013; Biro et al. 2013; Yusuf et al. 2013). Moreover, LST can be easily derived from remotely sensed data. It can be used to assess and evaluate the spatial relationship between LST and different LULC in urban areas and environments (Amiri et al. 2009). In urban areas, LULC is associated with LST (Weng et al. 2004). As LST is sensitive to the vegetation and the content of moisture in the soil, it can be used to detect the LULC changes (Dash et al. 2002). Remote sensing offers the high resolution, consistent, and repetitive coverage as well as it enables us to measure a large area of study (Owen et al. 1998). Remotely sensed thermal infrared data are extensively used for retrieval of LST and urban heat islands (Weng et al. 2004). Moreover, remote sensing in combination with geospatial information system (GIS) can be used in large number of studies (Pradhan and Youssef 2010; Youssef et al. 2011; Vijith et al. 2012; Khodaei and Nassery 2013; Wakode et al. 2013; Tehrany et al. 2013a). Several satellite sensors are available to estimate the LST over the urban areas and environmental locations. In satellite imagery technique, various thermal infrared sensors (TIR) exist in order to measure the LST, e.g., the geostationary operational environmental satellite, NOAA-advanced very high-resolution radiometer, and Terra and Aqua moderate resolution imaging
Arab J Geosci
spectroradiometer (Liu et al. 2006). In addition, there exist highresolution data from Landsat thematic mapper (TM) (Qin et al. 2001), enhance thematic mapper (ETM+) (Rodriguez-Galiano et al. 2011), and advanced spaceborne thermal emission and reflection radiometer (Liu et al. 2006). The Landsat TM 5 is one of the most useful sensors to drive LST and land surface emissivity (LSE) in environmental studies (Dash et al. 2002). Numerous types of methods and algorithms are available for retrieving LST. Radiative transfer equation is the first algorithm (Berk et al. 1989). Mono-window algorithm which was developed by Qin and Karnieli (2001) and generalized single channel was the next method proposed by Jimenez-Munoz and Sobrino (2003). Furthermore, some other approaches are used to determine LST such as split window (Sobrino et al. 2003) and multi-angle algorithm (Dash et al. 2002). Weng et al. (2004) derived LST to study urban heat islands based on Planck’s function. The study is based on investigation of the applicability of vegetation fraction derived from a spectral mixture model as an alternative indictor of vegetation abundance. In addition, the relationship between LST and NDVI (normalized different vegetation index) associated with urban landuse type and landuse pattern was deliberated in Shanghai city using Landsat ETM+ and aerial photographic systems. Results indicated that a positive correlation between LST and Shannon diversity index (SHDI) and a negative correlation between NDVI and SHDI (Yue et al. 2007). Prior of the evaluation of LST from multi-spectral TIR remote-sensed data, a precise measurement of the LSE values is required (Weng 2009). To estimate LSE from satellite data, some different algorithms are available and have been applied in different studies, e.g., classification-based emissivity methods (Sobrino et al. 2012), NDVI-based emissivity methods (Valor and Caselles 1996), temperature-independent spectral indices based method (Becker and Li 1990), temperate, and emissivity separation method (Gillespie et al. 1998). The present study aims to investigate the impacts of variations in LULC on LST values. To this end, two Landsat images of October 1998 (TM) and 2002 (ETM+) were used to assess the spatiotemporal dynamics of LST over the studied area. Two main parameters including NDVI and LULC were utilized in this study. Afterwards, emissivity value (ε) was estimated in order to compute the LST. The study was conducted in two periods: before (1998) and after (2001) construction of Karkheh dam; in order to consider the changes in the LULC caused by dam construction. So it investigates any distribution of LST over the studied area particularly surrounding a wetland in the area. Eventually, the relationship between LST and NDVI was assessed over different LULC for the studied area. Description of the study area The Karkheh River Basin (KRB) is one of the largest watersheds in Iran after Karoon and Dez. It is located on
the western border of the country in a strategic position. The KRB is situated in the southwest of Iran, between 30°58′ to 34°56′N latitude and 46°06′ to 49°10′E longitude. This watershed comprises five sub-basins, namely Kashkan, Qarasou, Gamsiab, Seymareh, and South Karkheh (Fig. 1). The area of Karkheh watershed is ∼50,760 km2 and contains a wide variety of elevation beginning less than 4 m above mean sea level in the downstream (near the Hoor Al Azim wetland) to more than 3,600 m in the upstream of Watershed (in the Karin Mountains). Two-thirds of the KRB approximately lies in the mountains at altitudes of 1,000–3,500 m. It covers ∼9 % of Iran’s irrigated farmland and produces between 10 and 11 % of the wheat production of the country. The climate of the basin is classified as semiarid to arid. The mean annual precipitation in the wetland and the surrounding area is ∼450 mm. South Karkheh receives ∼150 mm annually; whereas in the rest of KRB, annual precipitation is ∼750 mm. Precipitation in the south Karkheh area is regarded as inadequate to cover the water requirement for crop and irrigated farmland depends on water from the Karkheh dam, as well as groundwater resources. The mean annual temperature of KRB is 25 °C, while the mean annual rate of evapotranspiration is ∼1,900 mm. The main stream in the KRB is Karkheh River with ∼900-km lengths. Water in the basin is mainly used for agricultural products, domestic supplies, and fish farming, as well as serving the environment sustainability. Karkheh dam was constructed on Karkheh River and has been operating since 2001. It is a multi-purpose dam that provides irrigation water for ∼350,000 ha of agricultural land in Khuzestan plain, and generates hydropower and controls the flood. In the current study, South Karkheh Sub-basin (SKS) is selected as a study region for analysis where SKS includes Hoor Al Azim wetland (Fig. 1). It is because of high air temperature variations in the hole of the KRB.
Material and methods Satellite data Three scenes of Landsat 5 TM (1998) and three scenes of Landsat 7 ETM+ (2002) were acquired over the KRB and then mosaicked based on the nearest neighbor algorithm (Fig. 2). All TM and ETM+ remotely sensed data are downloaded from USGS website in level 1G product (see http://www.earthexplorer.usgs.gov/). All images were cloudfree and had highly clear atmospheric condition. The spatial resolution for Landsat TM data is 120 m, whereas for ETM+ is 60 m. These data were used to generate LULC, LSE, NDVI, as well as LST. Details of the remotely sensed data applied in this study are shown in Table 1.
Arab J Geosci
Fig. 1 Study area (South Karkheh Sub-basin, SKS) located in the Southwestern part of Iran in South of Karkheh River Basin, KRB
Satellite data image pre-processing At the first step, ENVI 4.8 software was employed to convert raw digital numbers to radiance values. Then, all satellite images were geo-registered and corrected geometrically. To do this, the digital topographic maps (scale=1:25,000, obtained from National Cartographic center, NCC, Iran) were utilized to extract ground control points. Afterwards, the images were resampled to their spatial resolution using the nearest neighbor algorithm with a pixel size of 30×30 m for all scenes. The RMSE between images are less than 0.435 for 2002 and 0.482 for 1998. Finally, to minimize
the destructive effects of aerosols, the FLAASH module in ENVI 4.8 software was applied to implement atmospheric corrections. Generating LULC maps for SKS In this study, to assess the LULC of SKS, false color composite images (as shown in Fig. 2) were selected for analysis. Furthermore, maximum likelihood classifier (MLC) algorithm was applied on satellite images during the classification analysis. Based on MLC, seven classes for this area were considered that is water body (including lakes and rivers), vegetation
Arab J Geosci
Fig. 2 South Karkheh Sub-basin (SKS) in 1998 (TM) and 2002 (ETM+) in false color composite (FCC) 4, 3, 2 bands and the location of Karkheh Dam as well as Hoor Al Azim wetland in the study region
(including agricultural areas), rangeland, wetland, sand dune, forest, and bare soil. Retrieval of LSE and calculate NDVI Regardless of LSE assessment, we will have an error between 0.2 and 1.2 K for LST values in mid-latitude summertime where this error can be between 0.8 and 1.4 K for the wintertime for an emissivity of 0.98 and at the ground heights of 0 km (Dash et al. 2002; Weng 2004). NDVI approach is an Table 1 Landsat TM and ETM+ coverage Karkheh River Basin during 1998 and 2002
easy, accurate, and operative method to estimate LSE. Consequently, an attempt has been made in the present study to retrieve LSE values based on this method. Red band (0.63–0.69 μm) and near-infrared band (0.76– 0.90 μm) are used in satellite images to compute NDVI using the following equation:
NDVI ¼
ðρNIR −ρRed Þ ρNIRþ ρRed
ð1Þ
Path
Row
On-board sensor
Date of acquisition
Spectral band
Sun elevation
Local acquisition time
165 165 166 166 166 166
38 38 37 37 38 38
TM ETM+ TM ETM+ TM ETM+
Oct 16, 1998 Oct 06, 2002 Oct 23, 1998 Oct 26, 2002 Oct 23, 1998 Oct 26, 2002
V-NIR-SWIR-TIR V-NIR-SWIR-TIR V-NIR-SWIR-TIR V-NIR-SWIR-TIR V-NIR-SWIR-TIR V-NIR-SWIR-TIR
43.62 48.51 40.41 40.34 41.51 41.50
06:54 07:03 07:00 07:09 07:00 07:09
Arab J Geosci
Where ρNIR and ρRed are the reflectance of the near-infrared and red bands, respectively. Sobrino et al. (2004) proposed emissivity based NDVI in three different cases: NDVI< 0.2, NDVI>0.5, and 0.2 ≤ NDVI≤0.5. The first case is considered as bare soil and the emissivity values are achieved from reflectivity values in the red region. In the present study, more than 90 % of study area is covered by the NDVI less than 0.2. Based on this method, the Eq. (2) was used to determine emissivity values where NDVI is less than 0.2. ε ¼ 1:0094 þ 0:047lnðNDVIÞ
ð2Þ
It should be noted that the emissivity value for water bodies equals 0.989 (Snyder et al. 1998; Srivastava 2009). Typically, for areas with NDVI values of more than 0.5, emissivity value equals 0.99. NDVI values between 0.2 and 0.5 are composed of both soil and vegetation. Therefore, surface emissivity was obtained by using the following equation: ε ¼ εv Pv þ εs ð1−Pv Þ þ dε
dε = (1− εs)(1− Pv)Fεv. According to Carlson and Ripley (1997), Pv can be estimated as follows:
NDVI−NDV I min Pv ¼ NDV I max −NDV I min
2 ð4Þ
Here, NDVImin is 0.2 and NDVImax is 0.5 and also F is a shape factor whose mean value, assuming different geometrical distribution and, ∼0.55 (Sobrino et al. 1990). In this case and in the absence of in situ observations, εv and εs might be considered equal to 0.985±0.007 and 0.960± 0.010, respectively (Caselles et al. 1996). Subsequently, in the current study, the values equal to 0.992 and 0.965 are considered for εv and εs. Eventually, LSE could be obtained using Eq. 5 as follows: ε¼
εv− εs εs ðiv þ d i Þ−εv ðis þ d i Þ NDVI þ þ dε iv −is iv −is
ð5Þ
Where iv is the pure vegetation NDVI, and is is the pure soil NDVI.
ð3Þ Measurement of LST
Where εv is the vegetation canopy emissivity and εs is the bare soil emissivity, Pv is the vegetation proportion achieved and dε is the internal reflection emissivity due to cavity effect,
Thermal infrared bands measure top of atmosphere (TOA) radiances. The brightness temperature can be retrieved by
Fig. 3 Spatial distribution of land use/cover by maximum likelihood classifier over SKS during October 1998 (a) and 2002 (b)
Arab J Geosci Table 2 LULC distribution and statistical analysis of MLC over the SKS in October 1998 and 2002 Class
1998 (%) Area
Producer accuracy
2002 (%) User accuracy
Commission Omission Area
Water body 27.98 98.64
98.26
1.74
1.36
Vegetation Rangeland Wetland Sand dune Forest Bare soil
89.30 98.57 92.71 55.37 98.62 94.03
10.70 1.43 7.29 81.96 1.38 5.97
9.08 32.32 5.90 11.32 3.37 0.08
0.87 60.20 0.34 2.01 5.16 3.42
90.92 67.68 94.10 88.68 96.63 99.92
Planck’s equation (Dash et al. 2002). The TOA radiance is the mixed result of three fractions of energy as follows: (1) emitted radiance from Earth’s surface, (2) upwelling radiance from the atmosphere, and (3) downwelling radiance from the sky. In the current study, TIR bands of Landsat TM and ETM+ are used based on reference values, calibration data, and empirical models to derive LST during October 1998 and 2002 over SKS by applying following steps. At the first step, the digital numbers were converted to the spectral radiance using Eq. 6 according to the
Producer accuracy
User accuracy
Commission Omission
29.27 98.50
99.47
0.53
3.50
1.12 59.69 0.14 1.36 3.53 4.85
91.73 94.01 90.57 53.72 99.54 89.03
8.27 5.99 9.43 60.41 0.46 10.97
15.82 36.15 7.75 5.94 0.16 3.47
84.18 63.85 92.25 94.16 96.84 96.03
reference values in the sensor Landsat handbook as follows (Irish 2000):
ðLmaxλ− Lminλ Þ Lλ ¼ ðDN−QCal min Þ þ Lminλ ð6Þ ðQCal max −QCal min Þ Where Lmaxλ and Lminλ are the spectral radiance for each band (TIR band) at digital numbers 255 and 1, respectively (Wm−2 sr−1 μm−1), λ is the wavelength, and QCalmin=1, QCalmax=255
Fig. 4 Spatial distribution of NDVI in year 1998 by Landsat TM (a) and 2002 by Landsat 7 ETM+ (b)
Arab J Geosci
Secondly, spectral radiances values were converted into LST using Planck equation as follows (Eq. 7): TB ¼
K 2 K1 ln þ1 Lλ
ð7Þ
Zonal GIS statistical analysis and ancillary data
Where TB is the effective at-satellite temperature in Kelvin, K1 and K2 are the coefficient determined by wavelength of a satellite sensor which values in ETM+ are 666.09 Wm−2 μm−1 sr−1 and 1,282.71 K, respectively, and 607.76 Wm−2 μm−1 sr−1 and 1,260.56 K in TM data. Using this formula, surface temperature for the study area was obtained. Additional correction for spectral emissivity is required according to the nature of landuse. Therefore, for correction of radiant temperature, thresholding NDVI method are applied. Consequently, the land surface temperature values are computed as follows (Artis and Carnahan 1982): T B Ts ¼ λT B 1þ lnε ρ
10−2 m K. σ is Boltzmann constant (1.38 × 10−23 J/K), h= Planck’s constant (6.626 × 10 −34 J s), c= velocity of light (2.998×108 m/s), Ts is the land surface temperature in Kelvin, and ε is the spectral surface emissivity.
ð8Þ
Where λ is the wavelength of emitted radiance (λ=11.5 μm based on Markham and Barker 1985), ρ=h × c/σ=1.4380×
In some research, it is sometimes necessary to evaluate and know about the surface temperature of each LULC. Zonal statistical tool is a suitable way to assess the above issue. It summarizes the values of a raster within the zones of another dataset (either raster or vector) and reports the results to a table. Therefore, to demonstrate the values of LST and NDVI in different LULC, zonal statistics in ArcGIS Spatial Analyst 10 were applied. According to this, a number of sample points of LST and NDVI for each LULC were randomly selected. Moreover, in the case of the relationship between LST and NDVI, the correlation and regression analysis was performed by the 4,537 randomly selected sample points over the different LULC categories in October 1998 in the study region. On the other hand, for further evaluation, some synoptic stations in and around the SKS were selected and the observed means of annual temperature and precipitation were assessed.
Fig. 5 Spatial distribution of land surface emissivity over the study region in 1998 (a) and 2002 (b) by Landsat TM and ETM+
Arab J Geosci
Results and discussion Results obtained from classification The generated classified maps of October 1998 and 2002 are shown in Fig. 3. However, the related statistical analysis is displayed in Table 2. It is found that the most area of SKS (∼60 % of area) is covered by the rangeland, whereas the wetland and vegetation areas covered a small part of the sub-basin. The results of MLC showed that a part of rangeland (mostly located in northern areas) is classified as sand dune due to similar spectral reflectance characteristics. On the other hand, the user accuracy is lower for sand dune (∼53 and ∼55 %) and the highest user accuracy is observed in water body (∼98 %). Distribution of NDVI and LSE results The NDVI is a measurement of the balance between the energy received and the energy emitted by objects on earth. When it is applied to plant communities, this index establishes a value for how green the area is, the quantity of vegetation present in a given area, and its state of health or vigor of growth. In a practical sense, the values that are below 0.1
correspond to water bodies and bare soil, while higher values are indicators of high photosynthetic activity linked to scrub land, temperate forest, rain forest, and agricultural activity. Figure 4 shows the spatial distribution of NDVI in SKS for the years 1998 and 2002. The extracted NDVI values over the study region for 1998 ranges between −0.47 and 0.71 with the mean value and standard deviation equal to ∼0.123 and ∼0.093, respectively. The low value of NDVI is located in south of Hoor Al Azim wetland which is mostly covered by water body. However, the high value of NDVI (bright area) is located in the north wetland, covering dense vegetation beside the Karkheh River. For 2002, the distribution of NDVI varies from −0.51 to 0.78 with the mean value of ∼0.117 and the standard deviation of ∼0.1. In contrast with 1998, the low value of NDVI was observed in the Karkheh lake dam while the high value of NDVI was almost similar to 1998 which was located in the north of wetland with high dense plant and vegetation area (some crop land areas close to Karkheh River). In case of LSE, the emissivity value lies between of 0.860 and 0.992, with a mean LSE value of 0.899 and a standard deviation of 0.057 in 1998. In 2002, the LSE ranges between 0.855 and 0.992, having a mean value of 0.897, and a standard deviation of 0.061. The maximum value of LSE is observed
Fig. 6 Spatial distribution of LST over South Karkheh Sub-basin during October 1998 (a) and 2002 (b) using Landsat TM and ETM+
Arab J Geosci
Fig. 7 Mean annual temperature (a) and precipitation (b) in some synoptic station located in the SKS
over the dense vegetation which was ∼0.99 and the minimum LSE value found over the sand dune which was ∼0.880. Figure 5 illustrates the spatial pattern of LSE over the south SKS during October 1998 and 2002. LST estimated and GIS zonal statistics Figure 6 provides information for the spatial distribution of LST over the SKS with different LULC in October 1998 and 2002. The LST retrieval in 1998 ranges between 281 and 317 K with a mean temperature of ∼305.55 K and the standard deviation of ∼4.73 K. In addition, the analysis for 2002 shows that the LST ranged from 281 to 327 K, with the mean temperature and standard deviations of 307 and ∼4.71 K, respectively. A comparison between Figs. 3 and 6 revealed that the temperature over the central part of the sub-basin, sand dune, and bare soil classes is higher than other areas, while the temperature over the forest shows the lowest temperature. The distribution of surface temperature is very different and has wide ranges. The thermal signature of each LULC must be studied to gain a better insight into the relationship between LST and LULC.
Therefore, the statistics of LST and LULC are obtained by overlaying LULC images with LST maps. Table 3 confirms the results of zonal GIS analysis and lists the statistical analysis, the relationship between different LULC, and the LST retrieval by satellite data. As it was mentioned earlier, ∼60 % of the watershed is covered by rangeland class and distributes over the whole study area with high difference in elevation. The distribution of temperature over this class varies between 11 K in 1998 to 15 K in 2002. It can be perceived from Fig. 6 and Table 3 that the value of highest temperature observed over sand dune (1998) ranges from 309 to 317 K, mean temperature of 313 K, and taking standard deviation of ∼3.61 K. The bare soil class (followed by sand dune) shows the maximum temperature of 308 to 315 K, mean temperature of 308.5 K, and a standard deviation of ∼3.60 K. Meanwhile, in 2002, the highest temperature was observed over the bare soil (311 to 327 K), with mean value of 323 K and standard deviation of ∼3.60 K. Furthermore, the results of year 2002 illustrate that the temperature for the sand dune is 317–325 K, with mean temperature of 321 K, and standard deviation of ∼2.24 K. In contrast, Table 3 also shows that the lowest temperatures are
Table 3 Zonal statistical descriptive of LST in 1998 and 2002 over different LULC Class
Water Body Vegetation Rangeland Wetland Sand dune Forest Bare soil
1998 (Kelvin)
2002 (Kelvin)
Min. temperature
Max. temperature
Mean. temperature
Std. deviation
Min. temperature
Max. temperature
Mean. temperature
Std. deviation
291 296 302 293 309 281 308
300 307 313 299 317 293 315
293 300 308 295 313 288 310
2.56 1.87 2.13 1.45 3.61 2.91 3.60
291 296 306 295 310 281 312
301 309 321 306 319 297 327
295 303 311 302 314 291 323
2.76 1.57 4.18 1.58 2.24 3.18 3.60
Arab J Geosci Table 4 Descriptive of zonal statistics of NDVI value associated with different LULC in 1998 and 2002
Class
1998
2002
Min
Max
Mean
Std. deviation
Min
Max
Mean
Std. deviation
Water body
−0.47
−0.08
−0.276
0.296
−0.51
−0.09
−0.3
0.303
Vegetation Rangeland Wetland Sand dune Forest Bare soil
0.22 0.1 0.22 0.05 0.22 −0.07
0.65 0.13 0.78 0.11 0.39 0.11
0.41 0.125 0.495 0.07 0.295 0.091
0.301 0.03 0.403 0.042 0.134 0.083
0.22 0.07 0.22 0.05 0.22 −0.07
0.62 0.12 0.79 0.11 0.42 0.11
observed over the forest in both 1998 and 2002 with 281– 293 K, with mean temperature of 288 K and standard deviation of ∼2.91 K in the year 1998 and 281–297 K, mean temperature value of 291 K, and standard deviation of ∼3.18 K for the year 2002, respectively. Green area, vegetation, and wetland show the minimum variety in temperature and exhibit approximately similar mean value temperature throughout October 1998 and 2002, and their mean temperature is 300 and 296 K (1998) and 303 and 302 K (2002). In fact, vegetation and wetland display the moderate temperature
0.406 0.095 0.467 0.07 0.31 0.091
0.294 0.035 0.405 0.042 0.155 0.083
in the study region among the LULC. When considering the results from Table 3, the mean value of LST over the wetland is increases from 295 to 302 K. The reason is that generally, wetlands are lands that saturate with water and are composed of some types of vegetation. Therefore, the LST over this LULC is low in 1998. But after 1998 and in 2002 (after construction of Karkheh dam in 2001, the LULC changes and a percent of wetland (Hoor Al Azim) shifts to the bare soil and the area of wetland was declined (Ghobadi et al. 2012). As a result, the temperature of bare soil as compared
Fig. 8 Geographical representation of mean land surface temperature (a) and mean NDVI (c) in year 1998 and also mean land surface temperature (b) and mean NDVI (d) in year 2002 attendant with different LULC
Arab J Geosci Table 5 Correlation coefficient and linear regression analysis between LST and NDVI 1998 over different LULC
Relationship between LST and NDVI
Land use/cover Correlation Linear regression coefficient
Coefficient of determination NDVI (R2)
Water body Vegetation Rangeland Wetland Sand dune Forest Bare soil
0.63 0.65 0.51 0.78 0.74 0.58 0.61
For further analysis, the relationship between LST and NDVI is investigated over the different LULC through correlation analysis and regression analysis. Table 5 shows the correlation coefficient and regression analysis between two variables (LST and NDVI) in association with seven classes during the analysis. A negative correlation is seen between NDVI and LST value among the LULC except over the water bodies (Table 5). All LULC have an inverse correlation between LST and NDVI but water bodies showed the different action (Fig. 8). The highest negative correlation coefficient is observed in wetland (−0.7673) followed by vegetation (−0.6606) and sand dune (−0.6440). However, forest, bare soil, and rangeland demonstrate a moderate correlation coefficient (−0.5540, −0.4794, and −0.4147, respectively). The strongest regression coefficient of LST and NDVI is observed over wetland (R2 =0.78) followed by sand dune (R2 =0.74). Previous studies show a negative correlation between LST and NDVI associated with water bodies, i.e., lake, river, and reservoir (Yue et al. 2007). The present study found a positive correlation between LST and NDVI (0.4597). Figure 9 shows the results of regression analysis between LST and NDVI variables (except water bodies). As a result, the regression analysis revealed that there is a significant inverse correlation between mean LST and NDVI values for the LULC (R2 = 0.57). It can be seen in Fig. 8 that the high value of NDVI (green area) exhibit the low LST value. In contrast, the sand dune and bare soil that show the low mean value of NDVI have a high LST values. Since in green land (covered by vegetation, wetland, and forest) the rate of evapotranspiration is high, land surface temperature is high in contrast to other types of LULC. Eventually, according to the aforementioned analysis and results, any changes in the LULC will affect the LST. In this study, while the focus was on the Hoor Al Azim wetland, we found that the LULC was changed due to some engineering project and anthropogenic phenomena, e.g., dam. Karkheh
0.4597 −0.6606 −0.4147 −0.7673 −0.6460 −0.5540 −0.4794
y=12.763 ×NDVI +289.52 y=−27.913 ×NDVI +294.76 y=−20.724 ×NDVI +311.79 y=−10.67 ×NDVI +293.54 y=−23.409 ×NDVI +313.39 y=−27.31 ×NDVI +297.37 y=−9.374 ×NDVI +306.24
to wetland is higher. This factor affects the temperature over the wetland and leads to an increase in the mean value of LST during 2002. However, the weather data of synoptic stations which covered SKS did not show a meaningful difference in the temperature and precipitation duration 1998–2002 (Fig. 7). Table 4 illustrates the results of GIS zonal analysis to relate the mean NDVI values and different LULC. Moreover, Fig. 8 illustrates the results of GIS zonal analysis of dispersion of LST and NDVI over different LULC in 1998 and 2002. Generally speaking, LST and NDVI have a clear inverse relationship. Sand dune class shows the highest mean LST and lowest NDVI values among other classes, followed by sand dune, bare soil, and rangeland showed the low NDVI values. These results are similar for two periods, 1998 and 2002. Meanwhile, water bodies represent neither high NDVI nor high LST. In contrast, at the same time, dense forests as covered fully by vegetation exhibit the highest NDVI values and as a result show the lowest mean LST as compared to other classes. Furthermore, wetland and vegetation represent a low mean LST to consider that these two classes show the high mean NDVI. Fig. 9 Scatter plot of linear regression analysis between LST and NDVI value associate with six LULC during October 1998
Arab J Geosci
dam was launched on 2001 and directly affected the downstream of SKS, particularly Hoor Al Azim wetland. Since it is designed to provide the water for agricultural land in downstream and upstream as well as for hydropower generation, the water discharge to the wetland is decreased regularly. This led to dryness of the most part of the wetland and changed into the bare soil and its LST value also increased within this area.
Conclusions In this research, an attempt has been made to derive LST by applying an algorithm on Landsat TM 5 and Landsat 7 ETM+ TIR band on October 1998 and 2002 over the south Karkheh sub-basin. A maximum likelihood classifier is applied on the satellite images to derive LULC and seven classes were determined. The classes are water bodies, vegetation, rangeland, wetland, sand dune, forest, and bare soil. To derive LST, atmospheric correction was required. Therefore, NDVI map for two Landsat images is extracted. Then Valor’s method is used to retrieve LSE. Regarding corrected emissivity, land surface temperature is derived using GIS analysis; each LULC is investigated to estimate LST. The relationship between land surface temperature and NDVI is considered with respect of the difference in LULC. The results showed that the temperature over the sand dune exhibit the highest value while this value was minimum over the water bodies. In 2002, the construction of Karkheh dam and change in value of water discharge to the Hoor Al Azim wetland lead to dramatic decline in the area of wetland so its LULC is changed and shifted to the bare soil. As a result, the mean temperature increased over the wetland area. On the other hand, the GIS-based and statistical analysis revealed that LST and NDVI exhibited a strong negative correlation. Subsequently, after the change in NDVI value and converting water and wetland areas to the bare soil, the value of NDVI declined and the value of LST increased. The strongest negative correlation between LST and NDVI is found in wetland with a correlation coefficient of −0.76 and R2 =0.78. This study ascertains that the LST value has a negative correlation with NDVI value and vegetation abundance influences the value of LST and has positive correlation for water bodies LULC. The analysis of regression showed a significant inverse correlation of 0.57. Acknowledgments Thanks to two anonymous reviewers for their valuable comments which helped us to improve the quality of the manuscript.
References Amiri R, Weng Q, Alimohammadi A, Alavipanah SK (2009) Spatial– temporal dynamics of land surface temperature in relation to fractional vegetation cover and land use/cover in the Tabriz urban area, Iran. Remote Sens Environ 113(12):2606–2617. doi:10.1016/j.rse. 2009.07.021
Artis DA, Carnahan WH (1982) Survey of emissivity variability in thermography of urban areas. Remote Sens Environ 12(4):313–329 Becker F, Li ZL (1990) Temperature-independent spectral indices in TIR bands. Remote Sens Environ 32(1):17–33 Berk A, Bernstein LS, Robertson DC (1989) MODTRAN: a moderate resolution model for LOWTRAN 7, technical report GL-TR-89– 0122. Geophysics, Lab, Bedford, MA Biro K, Pradhan B, Buchroithner MF, Makeschin F (2013) An assessment of land use/land-cover change impacts on soil properties in the northern part of Gadarif region, Sudan. Land Degrad Dev 24(1):90– 102. doi:10.1002/ldr.1116 Carlson TN, Ripley DA (1997) On the relationship between NDVI, fractional vegetation cover, and leaf area index. Remote Sens Environ 62:241–252 Caselles V, Coll C, Valor E, Rubio E (1996) Mapping land surface emissivity using AVHRR data: application to La Mancha, Spain. Remote Sens Rev 12(3–4):311–333 Dash P, Gottsche FM, Olesen FS, Fischer H (2002) Land surface temperature and emissivity estimation from passive sensor data: theory and practice–current trends. Int J Remote Sens 23:2563–2594 Ghobadi Y, Pradhan, B, Kabiri K, Pirasteh S, Shafri HZM, Sayyad GH (2012) Use of Multi-Temporal Remote Sensing Data and GIS for Wetland Change Monitoring and Degradation. IEEE Colloquium on Humanities, Science and Engineering. 103-108, doi:10.1109/ CHUSER.2012.6504290 Gillespie A, Rokugawa S, Matsunaga T, Cothern JS, Hook S, Kahle AB (1998) A temperature and emissivity separation algorithm for advanced spaceborne thermal emission and reflection radiometer (ASTER) images. IEEE Trans Geosci Remote Sens 36(4):1113–1126 Irish RR (2000) Landsat 7 Science Data User’s Handbook. National Aeronautics and Space Administration, Report 430-15 Jebur MN, Shafri HZM, Pradhan B, Tehrany MS (2013) Per-pixel and object-oriented classification methods for mapping urban land cover extraction using SPOT 5 imagery. Geocarto Int (Article online first available). http://dx.doi.org/10.1080/10106049.2013.848944 Jimenez-Munoz JC, Sobrino JA (2003) A generalized single channel method for retrieving land surface temperature from remote sensing data. J Geophys Res 108(22). doi: 10.1029/ 2003JD003480 Khodaei K, Nassery HR (2013) Groundwater exploration using remote sensing and geographic information systems in a semiarid area (Southwest of Urmieh, Northwest of Iran). Arab J Geosci 6:1229– 1240. doi:10.1007/s12517-011-0414-4 Liu Y, Hiyama T, Yamaguchi Y (2006) Scaling of land surface temperature using satellite data: a case examination on ASTER and MODIS products over a heterogeneous terrain area. Remote Sens Environ 105(2):115–128 Mallick J, Singh CK, Shashtri S, Rahman A, Mukherjee S (2012) Land surface emissivity retrieval based on moisture index from LANDSAT TM satellite data over heterogeneous surfaces of Delhi city. Int J Appl Earth Obstet 19:348–358 Markham BL, Barker JL (1985) Spectral characterization of the Landsat Thematic Mapper sensors. Int J Remote Sens 6(5):697–716 Norman JM, Becker F (1995) Terminology in thermal infrared remote sensing of natural surfaces. Agric For Meteorol 77(3):153–166 Owen TW, Carlson TN, Gillies RR (1998) An assessment of satellite remotely-sensed land cover parameters in quantitatively describing the climatic effect of urbanization. Int J Remote Sens 19(9):1663– 1681 Pradhan B, Youssef AM (2010) Manifestation of remote sensing data and GIS on landslide hazard analysis using spatial-based statistical models. Arab J Geosci 3:319–326. doi:10.1007/s12517-009-0089-2 Qin Z, Karnieli A (2001) A mono-window algorithm for retrieving land surface temperature from Landsat TM data and its application to the Israel-Egypt border region. Int J Remote Sens 22(18):3719–3746 Rodriguez-Galiano V, Pardo-Iguzquiza E, Sanchez-Castillo M, ChicaOlmo M, Chica-Rivas M (2011) Downscaling Landsat 7 ETM+
Arab J Geosci thermal imagery using land surface temperature and NDVI images. Int J Appl Earth Obstet 18:515–527. doi:10.1016/j.jag.2011.10.002 Snyder WC, Wan Z, Zhang Y, Feng YZ (1998) Classification-based emissivity for land surface temperature measurement from space. Int J Remote Sens 19(14):2753–2774 Sobrino JA, Caselles V, Becker F (1990) Significance of the remotely sensed thermal infrared measurements obtained over a citrus orchard. ISPRS J Photogramm Remote Sens 44(6):343–354 Sobrino JA, Kharraz JE, Li ZL (2003) Surface temperature and water vapor retrieval from MODIS data. Int J Remote Sens 24(24):5161– 5182 Sobrino JA, Jiménez-Muñoz C, Paolini L (2004) Land surface temperature retrieval from LANDSAT TM 5. Remote Sens Environ 90(4): 434–440 Sobrino JA, Oltra-Carrió R, Jiménez-Muñoz JC, Julien Y, Sòria G, Franch B, Mattar C (2012) Emissivity mapping over urban areas using a classification-based approach: application to the dual-use European Security IR Experiment (DESIREX). Int J Appl Earth Obstet 18:141–147 Srivastava PK, Majumdar TJ, Bhattacharya Amit K (2009) Surface temperature estimation in Singhbhum Shear Zone of India using Landsat-7 ETM+thermal infrared data. Adv Space Res 43:1563– 1574 Srivastava PK, Majumdar TJ, Bhattacharya AK (2010) Study of land surface temperature and spectral emissivity using multi-sensor satellite data. J Earth Syst Sci 119:67–74 Tehrany MS, Pradhan B, Jebur MN (2012) A comparative assessment between object and pixel-based classification approaches for land use/land cover mapping using Spot 5 imagery. Geocarto International, (Article online first available). http://dx.doi.org/10. 1080/10106049.2013.768300 Tehrany MS, Pradhan B, Jebur MN (2013a) Spatial prediction of flood susceptible areas using rule-based decision tree (DT) and a novel
ensemble bivariate and multivariate statistical models in GIS. J Hydrol 504:69–79 Tehrany MS, Pradhan B, Jebur MN (2013b) Remote sensing data reveals eco-environmental changes in urban areas of Klang Valley, Malaysia: contribution from object based analysis. J Indian Soc Remote Sens 41(4):981–991. doi:10.1007/s12524-013-0289-9 Valor E, Caselles V (1996) Mapping land surface emissivity from NDVI: application to European, African, and South American areas. Remote Sens Environ 57(3):167–184 Vijith H, Suma M, Rekha VB, Shiju C, Rejith PG (2012) An assessment of soil erosion probability and erosion rate in a tropical mountainous watershed using remote sensing and GIS. Arab J Geosci 5(4):797– 805 Wakode HB, Baier K, Jha R, Azzam R (2013) Analysis of urban growth using Landsat TM/ETM data and GIS—a case study of Hyderabad, India. Arab J Geosci, 1-13. DOI 10.1007/s12517-013-0843-3 Weng Q (2009) Thermal infrared remote sensing for urban climate and environment studies: methods, application, and trends. ISPRS J Photogramm 64:335–344 Weng Q, Lu D, Schubring J (2004) Estimation of land surface temperature–vegetation abundance relationship for urban heat island studies. Remote Sens Environ 89(4):467–483 Youssef AM, Pradhan B, Tarabees E (2011) Integrated evaluation of urban development suitability based on remote sensing and GIS techniques: contribution from the analytic hierarchy process. Arab J Geosci 4(3–4):463–473 Yue W, Xu J, Tan W, Xu L (2007) The relationship between land surface temperature and NDVI with remote sensing: application to Shanghai Landsat 7 ETM+data. Int J Remote Sens 28(15):3205–3226 Yusuf YA, Pradhan B, Idrees MO (2013) Spatiotemporal assessment of urban heat island effects in Kuala Lumpur Metropolitan City using Landsat images. J Indian Soc Remote Sens, (Accepted: in press). http://dx.doi.org/10.1007/s12524-013-0342-8
