1 Introduction

In the globalizing world, more than half of the world’s population lives in urban areas. The projected world population in 2050 will cause many changes depending on the size and spatial distribution of the global population (United Nations 2014). The need for natural resources is increasing day by day with the rapid population growth in the world, and it seems that this situation puts an increasing pressure on the natural habitats (FAO 1997). Better introducing and management of the urban environment, given population growth and natural resources, is made possible by urban sustainability (Imhoff et al. 2010). In this regard, LULC changes should be considered as the most basic information in land management (Chaudhuri and Mishra 2016). These kinds of changes are due to the rapid dissemination that involving large areas rapidly (Rembold et al. 2000), which contributes to LU policy and decision makers in predicting the direction of the human induced environmental changes (Xiao and Weng 2007; Zhang et al. 2013). To fully understand the LU change caused by human activity, LU change information is needed (Omran 2012; Sajikumar and Remya 2015). As stated by Melesse (2004), the current land-use change information is important in environmental monitoring, management and planning (Zhang et al. 2009). Reported studies show that urban growth, particularly the movement of residential and commercial land to rural areas, occur so rapidly (Hegazy and Kaloop 2015); changes occur rapidly; and updated LULC change maps contribute to the monitoring of urban development (Schneider and Woodcock 2008).

LULC changes is known to be the main focus of sustainable development (Lambin et al. 2000) and are very important concepts in natural resource management and monitoring (Sinha et al. 2015). For this reason, monitoring of LU variations utilizing updated data by means of forward-looking strategies are important for urban planners (Mundia and Aniya 2005). Remote sensing, which is a highly-advanced technique, uses a combination of high resolution images and image processing techniques that reveal the changes in LU in an effective way (Herold et al. 2003; Chaudhuri and Mishra 2016). Remote sensing has been used for LULC mapping and change analysis with the aid of different techniques (Butt et al. 2015; Liu and Yang 2015). It was reported that Landsat satellite images are very convenient for LULC change and also for urban regions analysis (Bagan et al. 2010; Bagan and Yamagata 2012; Mei et al. 2016).

One of the most important indicators of urbanization is the increase of surface temperature with the anthropogenic heat effect and the formation of urban heat island (Kumar et al. 2012). Urban cities are warmer compared to surrounding rural areas (Oke 1973; Chakraborty et al. 2015). Urban heat island (UHI) is known as a problem that occurs as a result of the uncontrolled growth of urban areas (Mirzaei et al. 2012; Yang et al. 2015). The main reason for the formation of urban heat island is warming of the land surface with the effect of material involving the heat due to urban development (Kaya et al. 2012; Hansen 2010). The temperature differences between urban and rural areas are observed due to UHI especially in summer nights (Dhalluin and Bozonnet 2015). Urban heat island affected by the geographical location of a city and the local weather conditions is known as the adverse effect of buildings in urban areas on the climate due to changes in the thermal properties of the urban infrastructure (EPA 2008).

UHI can be expressed generally by air temperature and LST (Li et al. 2014), LST can be easily obtained from remotely-sensed images (Heinl et al. 2015; Feng et al. 2014; Weng 2009). As reported before by Xu et al. (2010), remotely-sensed data are able to demonstrate the relationship between LST and different LULC (Balcik 2014). Landsat TM/ETM+ data have been widely used to perform the analysis of the UHI (Xian and Crane 2006). On the other hand, thermal imagery is very useful to understand UHI taking into account the changes in LULC (Li et al. 2012). The global climate change and increase in the LST arises as a result of rapid urbanization (Streutker 2002; Amanollahi et al. 2016). Therefore, the importance of urbanization on global climate change is greater (Trenberth et al. 2007; Liu et al. 2014; Mathew et al. 2016). The relationship between LULC and LST contributes significantly to determine the effect of LULC change on the LST (Ramachandra and Kumar 2009; Lo and Quattrochi 2003; Deosthali 2000). LULC changes and LST are important components in LU management; and NDVI is a parameter that has a direct influence on UHI analysis (Zhang et al. 2009). Based on the aforementioned relationship between LST and NDVI, LULC change will indirectly affect the LST due to the variation in NDVI among different LULC types (Leong et al. 2015). It can be considered that NDVI and surface temperature depend on land-use changes and on climate change and global warming (Schultz and Halpert 1993). The average surface temperature of our planet is mainly controlled by albedo and atmospheric greenhouse effect (Varotsos 2002; Efstathiou et al. 2003; Varotsos et al. 2014). Albedo reduction due to snow melting causes more sunlight absorption and increases the surface temperature (Pegau and Paulson 2001). The increase in energy absorption emitted by urban surfaces depends on the reductions in Albedo, and the reduction in albedo contributes to increases in air and surface temperature that characterize UHI (Peng et al. 2012; Trlica et al. 2017).

Many studies have investigated the relationship between LULC and LST using remote-sensing imagery on regional and global climate (Mohan and Kandya 2015; Chen et al. 2017; Zhang et al. 2016; Odindi et al. 2015; Jiang et al. 2015; Shen et al. 2015). The relationship between LULC and LST is very important in land management and global climate change studies. This study performed based on remote sensing techniques, reveals the relationships between LULC, LST and NDVI. The study carried out by Karakus et al. (2014) emphasized the importance of RS and GIS techniques in determining changes in LU and, in particular, how this information may be utilized in order to understand changes in LULC using aerial photographs for the years between 1973 and 2005 and Landsat TM/ETM+ satellite images for the years between 1987 and 2002 for the city of Sivas and its environment. The present study aimed to determine the effects of LU change on LST by examining LULC changes and LST distributions and UHI intensity for the Sivas city center and its surroundings years from 1989 to 2015 using Landsat 4 TM/Landsat 7 ETM+/Landsat 8 OLI images.

2 Materials and Methods

2.1 Study Area

Sivas is located in the upper parts of the Kızılırmak River within the central Anatolian region. In Turkey, the province of Sivas is ranked second in terms of surface area (28,488 km2) after the province of Konya, and is located between 36° and 39°E longitudes and 38° and 41°N latitudes. This study was carried out in the vicinity of the city center of Sivas at 319000–339000–4,389,000-4,415,000 UTM coordinates (Fig. 1). The study area covers the city center of Sivas and its immediate surroundings and covers an area of approximately 206 km2. A large part of the study area is located in the height class between 1250 and 1300 m where Sivas city center is located. A large part of Sivas city center and its immediate surroundings is located between 0 and 2% and 2–6% slope group interval. The vast majority of the study area consists of conglomerate-sandstone-mudstone, gypsum and alluvial deposits, which are not suitable for settlement areas (GDMRE 2005). Considering the city center of Sivas; the city is located within the alluvial unit which is unsuitable for basic conditions and settlement conditions (GDMRE 1997). Much of the research area is occurs from III., IV. and VII. class lands. The most suitable areas (I. class lands) for agriculture are mostly concentrated along the Kızılırmak river. The unsuitable areas (VIII. class lands) for agriculture are located in the southern part of the study area. The Kızılırmak river passing through the middle of the study area constitutes the most important surface water source of Sivas city and its immediate vicinity (GCM 2005).

Fig. 1
figure 1

Location of study area: a) Sivas province in Turkey b) Study area in Sivas province

Full size image

It is widely known that Sivas is the coldest city of the central Anatolian region; its temperature may reach 40 °C in summer and can drop to −33 °C in winter. The mean annual precipitation is 420 mm, for four seasons, comprising precipitation of 22% in autumn, 36% in spring, 32% in winter, and 10% in summer. When it comes to plant cover, steppe is the main vegetation of Sivas city and its surroundings (Sivas Governorship 2006). Approximately 50% of the city of Sivas is suitable for agriculture. The population of the city of Sivas began to increase significantly in the late 1930s, especially when state investments began, and as a result, the level of employment increased (Mahiroğulları 2003). According to the 1990 census the population of Sivas city center was 221.512, according to the 2000 census the population of Sivas city center was 251.776, according to the 2015 census the population of Sivas city center was 356.884 (TUIK 2015). When the time interval covering the working period is considered, because of the migration due to economic and social problems, the population greatly increased in the Sivas city center.

2.2 Data Description and Pre-Processing

Landsat-4 TM (acquired on August 27, 1989), Landsat-7 ETM+ (acquired on August 07, 1999) and Landsat-8 OLI (acquired on August 11, 2015) were used in order to determine LULC changes, retrieval of LST, and detect the changes in the UHI in the study area. All image bands 1–5 and 7 have a spatial resolution of 30 m, and the thermal infrared band has a spatial resolution of 60 m for Landsat-4 TM and Landsat-7 ETM+ images and 100 m for Landsat-8 OLI image (Table 1). All Landsat images were provided by the U.S. Geological Survey (USGS). Landsat data are frequently used for land-use and land-cover change analyses (Yuan et al. 2005; Bagan et al. 2010) and are useful for deriving the surface temperature (Kumar et al. 2012). All of the images were clear and nearly free of clouds (total cloud cover less than 10%)with high resolution.

Table 1 Landsat images used in this study
Full size table

Ancillary data (below) used in this study served as the reference data to determine the LULC variation, aided by the use of satellite images:

  1. a)

    digitized topographic maps at the scale of 1/25.000 provided by General Command of Mapping of Turkey (GCM)

  2. b)

    aerial photographs at the scale 1/5.000 from the year 2005 provided by General Command of Mapping of Turkey (HGK)

  3. c)

    high spatial resolution remote sensing data such as IKONOS (acquired on 2002) provided by Sivas Provincial Directorate of Agriculture (Turkey)

  4. d)

    LU maps obtained from relevant institutions and organizations for the study area

  5. e)

    point reference data obtained by GPS during field work

All of the images and data were converted in the same coordinate system (UTM/WGS84) to analyze the changes on LULC, LST and NDVI in the study area. All Landsat images were geometrically rectified to a common map reference system (UTM map projection Zone 37 North, WGS84 geodetic datum) based on the IKONOS image and 1/25.000 scale digitized topographic maps. Then all the images (using all bands) were resampled to 30 m using the nearest neighbor method (Xiao and Weng 2007). The RMSE of rectification was less than 0.5. In this study, ERDAS 9.1 was used to determine the LULC variations and for the image processing; and ArcGIS 10.1 was used for the spatial analyses. Furthermore, field type hand-held GPS were also used to identify ground control points.

2.3 Image Processing and Classification

Multi-band satellite images of 1989, 1999 and 2015 were obtained by using all bands of each image and with the help of the ERDAS 9.1 software. A study area covering approximately 206.7 km2 was extracted as a subset from the Landsat OLI images dated 2015, Landsat ETM+ whole-frame images dated 1999, and Landsat TM images dated 1989, covering an area of 185 km × 185 km. Then false color images produced by assembling bands 5, 4, and 3 of the Landsat TM/ETM+/OLI for visual interpretation in order to correct classification. Visual interpretation can give an idea concerning land cover variation over a particular time period (Shalaby and Tateishi 2007). The goal of image enhancement is to simplify the image interpretability (Karakus et al. 2014).

Remote Sensing (RS) has been used to classify and map the land cover and LU changes using Landsat images in the classification of different landscape components (Ozesmi and Bauer 2002). In this study, Anderson Land Cover Classification System was used. It was originally introduced by Anderson et al. (1976). The LU and LC classification system presented by Anderson et al. (1976) includes only the more generalized first and second levels. “Level 1” includes urban or built-up land, agricultural land, rangeland, forest land, water, wetland, barren land, tundra and perennial snow or ice. To determine variations in LULC and determine the effects of LU changes on LST five LULC classes were identified in this study (Table 2); agriculture, vegetation, urban/built up, water and bare land.

Table 2 Description of LULC classes of in this study area
Full size table

Subsequently, supervised classification was carried out using aerial photographs (inclose dates), ancillary reference data (topographic maps, zoning sheets, etc.), and data gathered in the field. A supervised classification process involving five land classes has been applied with the help of the Maximum Likelihood Algorithm, using all the spectral bands except the 6th spectrum (band), which has thermal characteristics parameters (Karakus et al. 2014). For each image, 100 training sites were selected in order to all spectral range have greater accuracy (Li et al. 2012). Because urban areas with bare land (particularly inclined areas covered with limestone and rocky areas) had similar spectral characteristics, it was difficult to separate these classes in the classification step. The classified images (Fig. 2) were exported to the ArcGIS 10.1 for map preparation.

Fig. 2
figure 2

Spatial distribution of land use/land cover (LULC) within study area

Full size image

2.4 Accuracy Assessment

The accuracy evaluation can be done by comparing a reference data with a data obtained with the help of remote sensing (GEOG 2016). In order to perform the accuracy assessment properly, according to Lu and Weng (2007), the number of training samples and their representativeness is important for image classification and the ground data should be collected at a time close to the historical image (Wondrade et al. 2014). To increase classification accuracy of all classified images, ancillary data including digitized topographic maps at the scale of 1/25.000, aerial photographs at the scale 1/5.000 from the year 2005, IKONOS satellite image from the year 2002, point reference data obtained by GPS, Google Earth observations, visual interpretation and our knowledge of the area were used as reference data (Zhang et al. 2013; Li et al. 2012). Reference data were developed for accuracy assessment (Yuan et al. 2005). The reference points formed by a stratified random sampling method was used for accuracy assessment (Xiao and Weng 2007; Wondrade et al. 2014). According to Ji and Niu (2014), confusion matrix computes the accuracy of each LULC maps (Ghebrezgabher et al. 2016). To calculate producer’s accuracy, user’s accuracy, and Kappa statistics using reference points, classification results were compared with reference data (Churches et al. 2014). Subsequently, classification results were associated with the reference data for accuracy assessment analysis.

2.5 Retrieval of Brightness Temperature from the Landsat 4 TM/Landsat 7 ETM+/Landsat 8 OLI Images

LST obtained from satellite remotely-sensed thermal infrared (TIR) imagery has an important role in emphasizing the importance of urbanization (Fu and Weng 2016). Thermal band of Landsat satellite images were used in order to reveal the surface temperature spatial distribution (Ding and Xu 2008) and the impact of LULC on the LST within the study area. The procedure was applied about the LST computation described by Malaret et al. (1985) and Guo et al. (2016). Surface temperature was detected to calculate brightness temperature from thermal band of Landsat satellite images taking into account the minimum and maximum numerical values of each image (Yuksel 2008).

The temperature data can be obtained using the digital images in the thermal band (DN 0–255) of Landsat TM / ETM + / OLI images. Digital numbers (DN) on heat bands (band6, band10) is converted to the Kelvin temperature in two steps. First, the digital number (DN) values of the Landsat images thermal infrared band was converted into spectral radiance using the following equation (Landsat Project Science Office 2002, 2016; Sekertekin et al. 2016):

$$ {\mathrm{L}}_{\uplambda}={{\mathrm{M}}_{\mathrm{L}}}^{\ast}\mathrm{Qcal}+{\mathrm{A}}_{\mathrm{L}} $$
(1)
$$ {\mathrm{L}}_{\uplambda}=\frac{{\mathrm{L}\mathrm{max}}_{\uplambda -}{\mathrm{L}\mathrm{min}}_{\uplambda}}{\left({\mathrm{QCal}}_{\mathrm{max}}-{\mathrm{QCal}}_{\mathrm{min}}\right)}\ \mathrm{x}\ \left({\mathrm{QCal}}_{\mathrm{max}}-{\mathrm{QCal}}_{\mathrm{min}}\right)+{\mathrm{L}\mathrm{min}}_{\uplambda} $$
(2)

where;

Lλ is spectral radiance W/(m2 sr μm), ML is radiance multiplicative scaling factor for the thermal bands, cal in Qcal is calibrated standard product, AL is radiance additive scaling factor for the bands, Qcal is pixel value in DN (digital number), Qcalmax = 255 (Qcalmax is the maximum number of each band), Qcalmin = 1 (Qcalmin is the minimum number of each band), and Lmax and Lmin are spectral radiance for thermal bands at DN 255 and 1, respectively (W/m2 ster μm) (Kumar and Shekhar 2015).

Second TIRS data can also be converted from spectral radiance (as described above) to brightness temperature, which is the effective temperature viewed by the satellite under an assumption of unity emissivity. The conversion formula is as follows (Landsat Project Science Office 2016; Kumar and Shekhar 2015).

$$ \mathrm{T}=\frac{{\mathrm{K}}_2}{\ln \left[\left(\frac{{\mathrm{K}}_1}{{\mathrm{L}}_{\uplambda}}\right)+1\right]} $$
(3)

where;

T is brightness temperature in Kelvin, K1 and K2 are thermal conversion constant for the band. For Landsat4 TM, K1 = 607.76 mW/cm2/sr/μm and K2 = 1260.56 Kelvin; for Landsat7 ETM+, K1 = 666.09 mW/cm2/sr/μm and K2 = 1282.71 Kelvin; for Landsat8 OLI, K1 = 774.89 mW/cm2/sr/μm and K2 = 1321.08 Kelvin.

2.6 NDVI (Normalized Difference Vegetation Index)

NDVI is a measure of relative greenness (Raynolds et al. 2008) and the NDVI can be used as a general indicator of vegetation cover (Bakr et al. 2010; Wardlow et al. 2006). NDVI is also useful to determine the production of green vegetation as well as detect vegetation changes (Gandhi et al. 2015). NDVI is described as the difference between the near infrared (NIR) and red bands and NDVI values varying between −1 and + 1 values. Non-vegetated surfaces were shown with negative values (Silleos et al. 2006). Higher NDVI values typically indicate areas with dense vegetation (Yuan and Bauer 2007). NDVI was calculated from the red and near infrared bands (Johansen and Tømmervik 2014):

$$ \mathrm{NDVI}=\left(\mathrm{NIR}-\mathrm{RED}\right)/\left(\mathrm{NIR}+\mathrm{RED}\right) $$
(4)

where; NIR represents the Near Infrared band 4 (0.76–0.90 μm) of Landsat 4 and 7 and RED the corresponding band 3 (0.63–0.69 μm). NIR represents the Near Infrared band 5 (0.85–0.87 μm) of Landsat 8 and RED the corresponding band 4 (0.63–0.67 μm) (Landsat Project Science Office 2016). According to Carlson and Ripley (1997), NDVI values lower than 0.2 represent sparsely vegetated or non-vegetated areas.

2.7 Urban Heat Island Intensity

Urban heat island is the difference in temperature between cities and their surrounding rural zones (Alonso et al. 2007). The meteorological and urbanization factors cause the formation of UHI effects with the increase in urban temperatures due to electricity demand in an urban area (Shahmohamadi et al. 2010). The intensity of urban heat island was defined as the difference between average temperature of UHI area and that of rural area (Ma et al. 2010; Effat et al. 2014; Lee et al. 2012). The TM data was used to perform the UHI detection especially in thermal infrared band (Du et al. 2009).

$$ {\displaystyle \begin{array}{c}\mathrm{UHI}\ \mathrm{intensity}\ \left({}^oC\right)={T}_u- Tr\\ {} LST>\mu +0.5\times \delta\ \mathrm{referred}\ \mathrm{to}\ \mathrm{UHI}\ \mathrm{area}\\ {}0< LST\le \mu +0.5\times \delta\ \mathrm{denoted}\ \mathrm{non}-\mathrm{UHI}\ \mathrm{or}\ \mathrm{rural}\ \mathrm{area}\end{array}} $$
(5)

where; Tu is the urban temperature, and Tr, the rural (surrounding) temperature, μ and δ are the mean and standard deviation of temperatures in study area, respectively.

A positive value indicates the existence of an urban heat island, i.e. the urban temperature is higher than rural. A negative UHI intensity means that the urban temperature is lower than rural temperature and so indicates an urban cool island. The temperatures used in Eq. (5) can be air temperatures, radiant temperatures or surface temperatures. The UHI effect is commonly quoted based on air temperature with the radiant and surface temperature UHI effects being specifically identified, e.g. UHI (radiant) effect or UHI (surface temperature) effect (Cheung 2011). In hot season, for example in summer, water bodies in urban areas such as ponds, canals and rivers may also have a cooling effect on the surrounding urban area. The cooling effect is caused by evaporation and by heat absorption (Hove et al. 2011).

3 Results and Discussion

3.1 LULC Changes between 1989 and 2015

LULC variations have been determined as five LU categories with corresponding definitions (Table 2). Fig. 2 show the changes in LU/LC during nearly 25-year period. When areal distribution values (Table 3) for LULC for the year 1989, 1999 and 2015 are examined, it is observed that urban built-up area increased rapidly. The urban built-up area increased from 13.37% of the study area in 1989 to 18.08% in 1999 and to 25.72% in 2015. However, bare land decreased from 50.13% of the study area in 1989 to 45.98% in 1999 and to 34.48% in 2015. The area of agricultural land increased rapidly from 26.10% in 1999 to 28.95% in 2015 while the area of agricultural land decreased from 26.47% in 1989 to 26.10% in 1999. During the specified time period, the development of the agricultural areas was observed in the areas outside the Kızılırmak river, and these areas are outside the first class agricultural areas. Population has increased due to the increase in migration from villages to villages and accordingly, the development of agricultural land outside of the first class sites has been determined in these areas. The vegetated area increased from 8.86% in 1999 to 10.03% in 2015 while vegetated area decreased from 9.03% in 1989 to 8.86% in 1999. The main reason of this increase is the improvements made in afforestation around Sivas city center by relevant institutions and organizations. The area of water was 1.00% in 1989, 0.97% in 1999 and 0.82% in 2015. When a general assessment is made for land-use changes over the period 1989–2015, it is observed that urban built-up area (mainly organized industrial zones) was constructed on agricultural land in the north-eastern part of the study area. The development of the residential area was monitored toward the Kızılırmak River which is located on the south of the selected study area. It appears that the most dominant terrain types are agricultural land and bare land in the study area. The increase in urban built-up area and agricultural land is 12.35% and 2.48%, respectively; while a decrease of 15.65% was observed in the bare land between 1989 and 2015. These results have also shown that a majority of bare land (approximately 12.35%) have been converted into urban built-up area. All of the results clearly show that urban built-up have had many changes. The transition to city residential settlements from village, which can be seen for many years in Sivas, was supported in agricultural and residential areas.

Table 3 Changes in land use/cover obtained Landsat TM/ETM+/OLI images
Full size table

Accuracy assessment is an important parameter for urban growth and LST (Wang et al. 2016). The map accuracy procedure defined originally by Congalton (1991) was applied for making the classification accuracy assessment (Saadat et al. 2011). Accuracy assessment analysis of the classification results was performed using Erdas 9.1 at specified time intervals. For this purpose, confusion matrices were obtained to confirm accuracy of the LULC maps (Table 4). Field reference points collected by using a hand-held GPS and ancillary data (aerial photo, IKONOS image) for use in accuracy assessment. A set of 551, 512, and 526 are field reference points for 1989, 1999, and 2015, respectively. It was determined that overall accuracy for the 1989, 1999, and 2015 LULC maps were 84.39%, 90.43%, and 94.11%, respectively. The Kappa coefficient of LULC maps were 0.79, 0.87, and 0.92, respectively. Some LU types (vegetation and water in 1999 and water in 2015) showed 100% producer’s accuracy (Table 4). If kappa values are greater than 0.75 and/or 0.80, the classification and reference data are compatible (Wondrade et al. 2014). According to Anderson et al. (1976), minimum requirement value is 85% for accuracy level by using Landsat satellite images in LULC mapping as suggested USGS (Bakr et al. 2010). It is clearly observed that the accuracy evaluation results obtained from our study is compatible with the values recommended in the literature.

Table 4 Confusion matrices for the classifications results
Full size table

3.2 Relationship between LULC and NDVI

NDVI analysis was performed to confirm LU changes as determined from satellite images and to determine the green areas in the study area using Erdas 9.1. Figure 3 shows spatial distribution of NDVI values within study area. NDVI values are in the range of −0.55 to 0.71, −0.56 to 0.68 and − 0.31 to 0.62 in 1989, 1999 and, 2015, respectively. The average NDVI values were 0.06, 0.08 and 0.15 for 1989, 1999 and 2015, respectively and the average NDVI value increased within the study period. The most important reason of this increase is the enrichment of the green areas especially in and around the Sivas city center according to our own observations. Forest land located in the north of the study area had the highest NDVI values (Xiao and Weng 2007), which were 0.51, 0.44 and 0.60 in 1989, 1999 and 2015, respectively and had the average NDVI as 0.34, 0.29 and 0.41 in 1989, 1999 and 2015, respectively. Fig. 5a show NDVI values associated with different LULC. The higher NDVI values were found over the dense vegetation areas (weighted forest). The lowest NDVI values were observed in urban city (Effat and Hassan 2014), agriculture land and water body for the study area (Fig. 5a). Bare land had the second highest average NDVI values over the study period.

Fig. 3
figure 3

Spatial distribution of NDVI within study area

Full size image
Table 5 LST statistic values for Landsat images used in this study
Full size table
Fig. 4
figure 4

Spatial distribution of LST (in Celsius) within study area

Full size image
Fig. 5
figure 5

Graphical representation of mean NDVI (a) and mean LST (b) with different LULC during the study period (Gaylan 2017)

Full size image
Fig. 6
figure 6

Linear relationship between NDVI and LST

Full size image

NDVI mean values obtained for 1989, 1999 and 2015, respectively in this study are compatible with NDVI indices value (NDVIBuilt-up < 0.2, NDVIBare land < 0.2, NDVIVegetation > 0.2, NDVIWater < 0) recommended by Chen et al. (2006) for different LULC types.

3.3 Relationship between LULC and LST

LULC significantly affects LST. Transformations between different LULC types (especially urban expansion) increase the effect of UHI by increasing and strong effect on the number and distribution of hot spots (Tran et al. 2017). The LST distribution map was prepared based on the classification scheme of the standard deviation (Weng et al. 2004) using digital images in the thermal band (DN 0–255) of Landsat TM/ETM+ /OLI images for the study area (Fig. 4). The mean LST values associated with LULC categories were summarized in Table 5. It was observed that during 1989, 1999 and 2015, the temperature changed in the 20 °C–50 °C, 21 °C–45 °C and 22 °C–50 °C range, respectively. The urban built-up had the highest average temperature, followed by bare land, agricultural land, vegetation and water in 1989, 1999 and 2015, respectively and highest average temperatures were observed for all the LULC categories in 1989 (Fig. 5a). The average LST values for urban built-up were 44.10 °C, 37.99 °C, and 42.30 °C in 1989, 1999, and 2015, respectively. The LST of urban built-up had a 6.11 °C decrease from 1989 to 1999 but increased by the maximum LST difference of 4.31 °C between 1999 and 2015. The bare land had the second highest LST followed by agricultural land over the study period for all the LULC categories. The LST of barren land had a 4.53 °C decrease from 1989 to 1999 but did not change significantly between 1989 and 2015. The LST for agricultural land was 37.33 °C in 1989, and changed to 29.38 and 33.28 °C in 1999 and 2015, respectively. The LST of agricultural land had a 4.05 °C decrease while the increase in urban built-up area was 12.35% from 1989 to 2015. The LST of vegetated areas had a 6.65 °C decrease from 1989 to 1999, but increased by 3.34 °C between 1999 and 2015. The lowest LST was observed for all the LULC categories in 1999 and water body and an area covered by green vegetation contains had the lowest average temperature for all the years. There is a maximum LST difference of 16.82 °C between urban built-up and water body in 1999. The average temperature increased from 1999 to 2015 and the average temperature increased from 1989 to 1999 for all the LULC categories (Fig. 5b). Generally, there was no great change in LST between 1989 and 2015.

Based on August, the areas determined as agricultural areas within the study area were considered as harvested areas. These areas are also considered as bare land. Studies done by Pu et al. (2006) and Zareie et al. (2016) have shown that the LST values of bare land are higher than the LST values of urban areas. Similar results were obtained in our study. The high LST values in these areas can vary depending on the content of the soil (sand, clay, etc.). The LST values on the dates of satellite images are not dependent on the spatial values of the land use classes, but may depend on the average daily air temperature on the satellite images taken. Temperature measurements should be performed on different LULC properties based on field observations to show how LST results calculated from Landsat TM/ETM thermal band have been verified with actual field surface temperatures (Mallick et al. 2008). As a matter of fact, the daily mean air temperatures of the dates indicated by the LST values stated on the date (the average daily air temperature at 27.08.1989 is 24,1 °C, the daily average air temperature at 07.08.1999 is 19,0 °C, the daily mean air temperature at 11.08.2015 is 23,7 °C) parallel to each other.

Distorted construction resulting from urban growth decreases the amount of leakage of rain water into the soil, whereas surface flows increase. Therefore, underground water levels are decreasing and the amount of groundwater is gradually decreasing. Depending on these two factors, the evapotranspiration (evaporation + sweating) event required for the hydrological cycle (water cycle) is not sufficiently realized. As a result, the water balance deteriorates and this leads to climate change (Yüksel et al. 2011). And so the changes in land use also affect the climate elements such as daily max and min temperature. Land use change causes aerosol emissions and surface albedo changes. For this reason, land use change is one of the basic human activities that may disrupt the radiation balance of the Earth (Myhre and Myhre 2003). For example; air temperature has declined due to the transformation of water-marsh land uses into agricultural lands with high albedo values (Gophen 2014).

3.4 Relationship between LST and NDVI

NDVI and LST had a strong correlation (Kaufmann et al. 2003; Chen et al. 2006) and it was clearly seen that there was an increase in NDVI values (greater proportion of vegetation) during the summer (Julien et al. 2006). The increase in LST values was observed together with a reduction in NDVI values (Bokaie et al. 2016). There is a significant and negative relationship between vegetated land (having very low temperatures) and LST (Li et al. 2014). The surface temperature can be expressed using linear regression between surface temperature and NDVI (Fig. 6) according to certain acceptable range accuracy (Zareie et al. 2016a, 2016b). The coefficients (R2) were 0.95, 0.99 and, 0.96, respectively and this R2 value was an indicative of the strong relationship between LST and NDVI when the correlation made between NDVI and LST was considered. A negative correlation was determined between both parameters when a general assessment was made between NDVI values and LST values (Lo et al. 1997; Kaplan et al. 2018) based on the information in the literature during the study period (Fig. 6).

Areas covered by the highest NDVI values are located in south, southwest and north of the Sivas city center. The lowest NDVI values concentrated in an area covering the central city of Sivas and its surroundings. Areas covered by the highest NDVI values are usually forested areas (Fig. 3). NDVI values increased with the increase of vegetated areas and LST values decreased. On the other hand, an increase in LST was determined by increasing agricultural land and a decrease in LST was determined by decreasing bare land whereas without much change in NDVI values between 1989 and 2015. When an assessment was made for urban areas and bare land for 1989 and 2015, it was observed that LST was decreased by increasing NDVI. According to the research results obtained by Yuan and Bauer (2007), NDVI may be used for analysis of surface UHI effects during summer. Vegetation plays major role to reduce the amount of thermal radiation (Chakraborty et al. 2014).

3.5 Intensity of Urban Heat Island

A procedure is applied based on the Eq. (5) in order to calculate the intensity of UHI. UHI intensity calculated using the mean temperature (μ) and standard deviation (δ) values of LST statistics for the specified year. The mean temperatures were 37.20 °C, 37.85 °C, and 38.55 °C in 1989, 1999, and 2015, respectively (Table 5).

The UHI intensity were 6.77 °C, 3.61 °C, and 9.02 °C in 1989, 1999, and 2015, respectively. The UHI intensity had a 3.16 °C decrease from 1989 to 1999 but increased by the UHI intensity difference of 5.41 °C between 1999 and 2015. It has been noticed that UHI intensity had a 2.25 °C increase when the general assessment made between 1989 and 2015 and it has showed that a fluctuating trendof UHI intensity. The mean temperatures revealed a positive UHI intensity due to the lowest mean temperatures that were recorded in the urban area when compared to the mean temperatures in urban and rural areas (Table 6).

Table 6 UHI intensity values which calculated during the study period (adapted from Hala Adel Effat, 2014)
Full size table

4 Conclusion

This study examined the effects of LU change on LST and UHI intensity, which were determined for the Sivas city center and its surroundings from 1989 to 2015, using Landsat satellite images. Accuracy analysis results revealed obviously that the accuracy of the classification according to our study also supported to previous studies. When an assessment was made on obtained results, it was seen that the urban built-up areas and agricultural lands increased while barren land decreased over the study period. The main reason for the increase in agricultural and urban areas is the transition to city residential settlements from villages. The vegetated areas decreased from 1989 to 1999, whereas they increased from 1999 to 2015. Agricultural and bare land are the dominant LU classes in the study area. The importance of agricultural activity is great on the interpretation of LULC categories (Chen et al. 2006). When examining changes in LU, a majority of bare land (approximately 12.35%) have been converted to urban built-up area. The city of Sivas has grown mainly towards the north-east, south and south-west. Part of the agricultural land in the study area has also turned into residential area. When increasing development of settlement areas towards the Kızılırmak River is taken into consideration, it has been deemed that this situation might present a threat to the river. Such drawbacks can be observed as a result of population pressure on the land and inadequate land use planning.

The liberal economic policies adopted in Turkey since 1950 and the mechanization of agriculture have accelerated the migration from rural to urban areas. As a result, the urban population has increased steadily to daylight. Especially in the years 1980 and after 1980, the rapid growth of the scale in the cities gave birth to the result. With the 1980s, approaches have been adopted aiming to bring the areas that lost their function in the city center to the urban economy (Yedekci 2015). In line with these explanations, the period of time when the urban growth is the most in the city center of Sivas is striking as 1999–2015 compared to the interval between 1989 and 1999. In parallel with this increase seen in the overall Turkish cities, a similar increase was determined in Sivas city center. Population growth between 1999 and 2015 also supported urban growth.

With the investigated relationship between LULC, NDVI and LST, the spatial distribution maps of NDVI and LST were created carefully. When changes in LULC were accompanied by changes in NDVI and LST, the higher NDVI values were found surprisingly to be over the dense vegetation areas and the urban built-up land had the highest average temperature over the study period. A negative correlation was determined between NDVI and LST, and a lower average LST was observed in vegetated areas. The differences in air temperature on the date of receipt of satellite images had led to the different surface temperature values. Urban built-up land, particularly with an average temperature of 44.10 °C (1989), are the hottest places while water bodies with an average temperature of 27.68 °C (2015) are the coolest places. Urban built-up and bare lands have the highest LST in the study area. The urban built-up areas have a higher surface temperature than the agricultural land due to the city’s unique geographical location, physical and morphological structure and mainly depending on the construction material.

It is generally known that industrial and residential areas are factories which contribute to the development of UHI. The urban built-up areas increased but on the other hand a positive UHI intensity was occurred. The obtained results have clearly shown there was increase in urban areas, in parallel with the intensity of UHI in the study area was an increase tendency between 1989 and 2015. Because of the urban temperature is higher than rural temperature, an urban heat island formation was determined for the Sivas city center and its surroundings.

Urban heat islands have led to negative effects such as health problems, increase in energy consumption, air pollution and water shortage (Agarwal and Tandon 2010; Amanollahi et al. 2013). According to the results obtained from our research, Sivas is a city that poses a problem in terms of urban heat island and similar problems. As a result, planning studies should be carried out carefully taking into consideration the natural capacities of the land in order to prevent its misuse, and the government should take the necessary measures immediately in the region with the urban heat island.