According to the goals set out in the 14th Five-Year Plan (2021–2025) for national economic and social development of China, 65% of the population lives in urban areas. In the next two to three decades, China's urbanization population will gradually rise to more than 80%. The continuous improvement of urbanization is also ongoing in Asia and Sub-Saharan Africa1,2. Urban expansion leads to the variation of land use (land cover) types and results in a differential heating process3. For instance, the increasing expansion and intensification of natural surface transferred to impervious surface are considered to enhance urban heat island (UHI) effect, because impervious surface can capture heat and lessening evaporative cooling in contrast to naturally surface4,5. With the advance of global warming, extreme hot weather occurs frequently, which further exacerbates the heat environment of urban6,7. UHI can causes negative environmental and health impacts8, such as urban polluted air9,10, more energy and water consumption11,12, and extreme heat-related morbidity and mortality13.

The land cover changes (urban growth), such as the conversion of natural surfaces to impervious surfaces, often result in land surface temperature (LST) rise, referred as the main cause of the UHI effect14 Data of LST can be effectively obtained based on remote sensing in large scale, real time, and continuous manner in the recent decades15. The development of landscape ecology has facilitated to research urban heat environment. Landscape pattern metrics of land cover can provide a comprehensive description of heat island pattern from landscape composition and spatial arrangements16. Landscape composition is described as the proportion of each land cover type, including green space, water body, construction land and barren land7,17,18, which has been applied to research its relationship with LST based on correlation analysis3,19. In addition, landscapes configuration (such as size, number, shape of patch) also obviously affects the magnitude of LST15,16. Some studies have indicated that landscape configurations are more important factor influencing the cooling effect of water and green space than landscape composition20,21. However, these studies have commonly focused on a discrete area of landscape and the effects of the area's size and shape, rather than the spatial arrangement of different landscape types17. Other spatial landscape metrics such as distance, interspersion, patch density, aggregation, contagion, diversity are rarely used to describe the variation of LST from the overall characteristics of different land cover22. Hence, the role of spatial configuration of different land cover in affecting LST should be further quantified23, and it would be necessary and helpful for urban planning.

This research focused on the dynamics of LST and land use types in the urbanization process during 2001 and 2020 based on Landsat and various geospatial approaches. The purpose were: (1) to investigate the role of landscape composition and configuration in indicating LST; (2) to quantified the coupling relationships between LST and landscape pattern of different land cover. The results can provide a reference for better managing the land use changes and landscape planning within cities in order to mitigate the thermal environment.

Materials and methods

Study sites

Nanchang is an inland urban (Fig. 1), the capital of Jiangxi with population of 6 millions (as of 2021), is considered as the “furnace” city in central China with air temperature exceeds 30–40 °C in summer. The mean annual temperature and precipitation in Nanchang is approximately 23 °C and 1600 mm. The landscapes of the urban are a mosaic of built-up (construction) land, cropland, grassland, and water body including parts of the lakes, rivers and ponds. This study was conducted using data from 2001 to 2020, a period of the typical developing stage of urbanization in China15,22.

Figure 1
figure 1

Location map of urban Nanchang and adjacent provinces in China.

Data processing

Thermal remote sensing (RS) has been used as a powerful tool in the exploration of LST for its broad spatial coverage, frequent revisit cycle and multiple data source23,24,25. The RS images ( applied in this research were selected from 13/5/2001, 16/5/2008, 1/5/2014, 15/4/2020 and 6/9/2020 under the stability of weather conditions (gentle breeze and sunny) when the satellite had to pass the same location26,27.

LST retrieval

The radiative transfer equation (RTE) method is applied to invert the land surface temperature (LST), can reach an accuracy of 0.6°C28. The software ENVI5.3 is applied for radiometric calibration and atmospheric correction of the raw satellite image29, and AicGIS10.2 is used to crop the study area. The normalized vegetation index (NDVI) and vegetation coverage (FVC) are used to calculate the surface emissivity (εi). The formula was as follows:

$$FVC = \frac{{{\text{NDVI}} - {\text{NDVI}}_{{{\text{soil}}}} }}{{{\text{NDVI}}_{{{\text{veg}}}} - {\text{NDVI}}_{{{\text{soil}}}} }}$$
$${\upvarepsilon }_{{\text{i}}} = 0.004{\text{FVC}} + 0.986$$

where NDVIsoil and NDVIveg are the normalized vegetation index of the non-vegetated area and the normalized vegetation index of the total vegetated area, respectively. During atmospheric correction, the atmospheric radiation is simulated based on the principle of the RTE by the model in MODTRAN4.015. Through this simulation, the downward and upward atmospheric radiance Li↓, Li↑ and atmospheric transmissivity τi(θ) can be estimated. With εi, the radiation intensity B(Ts) can be obtained by Eq. (3). Consequently, the LST (Ts) can be calculated by Eq. (4). More detailed process can be checked in the reference15,30.

$$B_{{\text{i}}} (T_{{\text{s}}} ) = [B_{i} (T_{i} ) - L_{i} \uparrow - \tau_{i} (\theta ) \cdot (1 - \varepsilon_{i} ) \cdot L_{i} \downarrow ]/\tau_{i} (\theta ) \cdot \varepsilon_{i}$$
$$T_{S} = \frac{{{\text{k}}_{2} }}{{\ln \left( {1 + \frac{{k_{1} }}{{B(T_{s} )}}} \right)}}$$

where Li is the radiation intensity (W·m−2·sr−1·µm−1) of the wave band with index number i measured by the satellite sensor. K1 (unit mW·m−2·sr−1·µm−1) and K2 (unit K) are preset constants before the launch of the satellite (Table 1).

Table 1 The value of K1 and K2.

Classification of LST

After retrieval, LST are classified into seven types as Extremely High Temperature Zone (EHTZ), High Temperature Zone (HTZ), Slight High Temperature Zone (SHTZ), Medium Temperature Zone (MTZ), Slight Low Temperature Zone (SLTZ), Low Temperature Zone (LTZ), and Extremely Low Temperature Zone (ELTZ) with the method of standard deviation classification15,31. The calculation of the boundaries is shown in Eq. (5).

$${\text{D}} = {\text{X }} \pm {\text{ a }} \times {\text{ S}}$$

where D is the boundary of each type, a is times of 0.5 from 0 to 2.5, X is the average LST of the image, and the value S is the standard deviation. The seven types are listed in Table 2. The X and s for the four years were calculated separately.

Table 2 The classification of the land surface temperature.

Classification of land use

The land use maps were derived from the open-source datasets ( This dataset with a spatial resolution of 30 m were extracted from Landsat satellite imageries by supervised classification. High resolution images from Google Earth™ were used as the reference layers to assess the classification accuracy23,32. In this study, the land use at 30-m resolution in 2001, 2008, 2014 and 2020 was classified into five major landscape types: construction land (CL), green land (GL), water landscape (WL), agricultural land (AL), and barren land (BL).

Landscape pattern index

The surface coverage data was calculated using 1000 m × 1000 m gridded fishnet on the ArcGIS 10.2 software platform for the surface coverage weighted center identification33. A total of 333 randomly selected polygon grids for study area were used to clip the study area's land cover map in 2020 (Fig. 2a,b). Eleven landscape pattern indices from the perspective of landscape ecology were selected to quantitatively describe the landscape pattern characteristics of land coverage in this study area. These indices are mainly from the two aspects of patch level and landscape level34,35. The patch level indices focus on the analysis of the number, morphology and structure of land use, including number of patch (NP), percent of landscape composition (PLAND), largest patch index (LPI), landscape shape index (LSI), mean closest distance (MNN), interspersion and juxosition index (IJI). The landscape level indices are used to describe the overall characteristics of land use status including patch density (PD), aggregation index (AI), contagion index (CONTAG), shannon’s index (SHDI), shannon’s evenness index (SHEI), LPI, LSI and IJI. These indices were selected and calculated by using Fragstats v4.2.1 software3, and were described in Table s1 (supplementary materials). Because the significant UHI and its obvious adverse impact on human sense of comfort occurs in summer, the subsequent section of the study focused on the urban LST and associated landscape pattern factors was selected on 6 September 2020 in the hottest summertime.

Figure 2
figure 2

(a) Slice of land use type figure (b) Slice of LST grayscale figure.


Spatial and temporal patterns of urban land use and LST

Figures 3 and 4a showed the obvious spatial and temporal variations of areas of land use types from 2001, 2008, 2014 to 2020. Amongst them, area of CL (red) was gradually increasing in spatial pattern, contrastively, the spatial pattern of area of AL (yellow) was reductive with the year. These changes reflected the urbanization process of Nanchang in 2001–2020 (Figs. 3, 4a). In 2001, the urban CL was distributed mainly within Xihu and Donghu districts (the old urban center districts) in the east of Ganjiang River, while other districts had higher agriculture coverage rates. In 2008, Nanchang began to expand the urban area by planning of Nanchang Municipal Government, including the construction of Honggutan New District of the west of Ganjiang River, Qingyunpu District, and Qingshanhu District, the CL was scattered distribution with very few connected to each other. In 2014, continuous constructive districts were formed in six districts. Compared to 2014, in 2020, the density of CL had increased significantly and AL area had decreased obviously. GL, WL, and BL relatively little change than CL and AL. According to Figs. 3 and 4a, compared to 2001, the land use in 2020 had changed dramatically within the outer ring road of Nanchang. The urban area of Nanchang had increased from 85 km2 in 2001 with a population of 4,434,200 (the 2001 census) to more than 350 km2 in 2020 (satellite map) with a permanent population of 6,255,000.

Figure 3
figure 3

The districts map of urban Nanchang.

Figure 4
figure 4

The land use maps (a) and LST zones (b) in 2001–2020.

As shown in Fig. 4b, Nanchang city had apparent hot spots in any given years, displaying that the total area of EHTZ and HTZ had amplified obviously while the MTZ and SLTZ had reduced obviously. In 2001, the UHI were assembling only in the old city center of Xihu and Donghu districts like that of CL. In 2008, the UHI downtown extended to the western where Honggutan District is located. This area is a new commercial and economic center with plenty of building beside the Ganjiang River. In 2014, the UHI expanded toward the suburban areas of each districts, and a series of discrete EHTZ and HTZ zones formed due to the development of new city areas. Compared to 2014, in 2020, hot spots had appeared obviously in the almost half of the city where was covered by higher temperature zones, reflecting further expansion of hot spots.

Average LST had a similar spatial pattern with Land use (Fig. 4a). i.e., the distribution of EHTZ and HTZ (red zone) were consistent with that of CL (red); the distribution of SHTZ (yellow) responded to that of AL and BL with higher vegetation coverage rates; the distributions of MTZ and SLTZ (nattier blue) were corresponding with that of GL; and the distribution of LTZ and ELTZ (deep blue) responded to that of WL with large heat capacity and low thermal conductivity21,36. In 2020, obvious hot spots had appeared in the almost half of the city due to large contiguous tracts of CL. These results showed the expansion of urban LST is basically consistent with the expansion pattern of urban land use types.

The variations of areas of land uses were also presented in Fig. 5a and Table s2. Area of CL accounted for 15.93% in 2001 to 38.32% in 2020, rapidly increased by 22.5%, inversely, AL areas accounted for 68.44% to 49.69% from 2001 to 2020 and a sharp decrease by 19%, and WL areas decreased by 2% from 12.06% to 10.15%. The LST features of land use types are achieved attribute to different absorbing and reflection rate of the solar radiation by land coverage14,37. According to Fig. 5b, LST of land uses significantly varied with the years. These changes indicated an adverse effect on the thermal environment through the urbanization process from 2001 to 2020. The LST of CL was highest and increased with the years because of amplified CL area, showing that CL was the most significant factor contributing to LST and thereby heat environment effect9,27. LST of GL and WL showed significantly lower than that of CL and BL. The sequence of average LST rankings from the high to the low were CL, BL, AL, GL and WL. The overall average LST was hotter and hotter with the years when massive AL were replaced by CL (Figs. 4a,b and 5a,b).

Figure 5
figure 5

The area percent (a) and LST (b) of each land use types with the years.

The relation between grade of LST and area percent of land use

The area ratios of land uses in each LST zones were calculated by the spatial statistic function of ArcGIS (Fig. 6). Combined Table s3 (supplementary materials), the results showed the temperature zones were significantly variation with the years. The EHTZ, HTZ and SHTZ increased whereas the MTZ, SLTZ and LTZ decreased with the years. The maximum change in ratios happened in the MTZ and HTZ, EHTZ, showing area percent of CL increased whereas that of AL decreased. And the ratio changes of SLTZ (− 4.77%), MTZ (− 6.95%), SHTZ (6.17%), HTZ (5.28%), and EHTZ (3.86%) were more significant between 2014 and 2020 than those between 2001 and 2008 (2.67%, − 4.84%, − 3.31%, 1.65%, 0.93%, respectively) (Table s3), indicating that heat intensity increased rapidly between 2014 and 2020. The area ratios of land uses in different temperature zones significantly varied, however, the result of each year showed the same features, i.e., CL was the largest portion of EHTZ and HTZ showing CL contributed the most to heat effect38; AL was the largest part in SHTZ, MTZ and SLTZ; and WL was the largest part in LTZ and ELTZ, showing WL contributed the most to cooling effect; GL was small part in LTZ and SLTZ. Area percent of BL was small and can’t be displayed.

Figure 6
figure 6

The area percent of each land use types in different LST zones in 2001–2020.

Quantitative analysis of LST and landscape composition

According to Fig. 7, there were a good linear fitting relationship (p < 0.01) between LST and the area ratios of CL, WL, GL and AL, indicating the four land use types can better explain the changes of LST. The linear fitting of LST and area ratio of CL was the strongest and had a significant positive correlation (R = 0.78), indicating the larger proportion of CL and the higher LST. There were a significant negative correlation between LST and AL, GL and WL, indicating the higher the proportion of AL, GL and WL can better mitigate the region's high temperature, especially WL had a best cooling effect (R = − 0.76) for this city.

Figure 7
figure 7

The quantitative relationship between LST and area ratios of land uses.

In order to obtain a more accurate model of the relationship between LST and the overall land use, multiple regression analysis was carried out between LST and area percent of land use15, and the contribution value of different land uses to LST was obtained. The prediction model was shown in Eq. (6).

$${\text{T }} = { 39}.{95}0 \, + \, 0.0{\text{16CL }} - \, 0.{19}0{\text{GL }} - \, 0.0{\text{51AL }} - \, 0.{1}0{\text{7WL }} + \, 0.00{\text{1BL}}$$

where T stands for LST. The coefficient of the model was the coefficient of the variable in Table 3. The model fitted well (R = 0.87), can be used to predict the LST and adjust the area of various land use types to achieve the optimal land use planning of urban15.

Table 3 The multivariate linear regression of LST and area ratios of land uses.

Quantitative relation of LST and patch type indices of land use

There were significant correlation between LST and six landscape pattern indices of CL (p < 0.01) (Table 4 and Fig. s1), with significant positive correlations with PLAND (y = 8.20x − 240.98, R = 0.78), LPI (y = 8.82x − 268.20, R = 0.73) and MNN (y = 0.34x − 8.57, R = 0.44), and negative correlations with NP (y = − 0.64x + 28.22, R = − 0.45), LSI (y = − 0.16x + 9.30, R = − 0.35) and IJI (y = − 4.33x + 194.35, R = − 0.34). These results showed that large PLAND and LPI values indicated the larger the area ratio and the more concentrated distribution of CL generated the higher LST, due to CL generally made of concrete and other heat-absorbing materials16,23. Large NP, IJI and LSI intended the more fragment, dispersed and complex shape distribution of CL help to alleviate the thermal environment due to massive surface contact with the surroundings.

Table 4 The correlation between LST and patch type metrics of land use.

Although all the metrics of WL had a significant correlation with LST (p < 0.05) (Table 4 and Fig. s2), PLAND (y = − 7.92x + 301.38, R = − 0.78) and LPI (y = − 7.05x + 268.45, R = − 0.75) had the most strong negative correlation with LST (p < 0.01), meaning the larger the area ratio and the more concentrated distribution of water body, the lower LST. NP (y = − 0.09x2 + 6.37x − 109.27) and LSI (y = − 0.03x2 + 2.05x − 33.84) had significant fitting with LST, showing the more number and complex shape of WL, the more effectively reduce LST. In addition, high temperature vs a very low area proportion (even 0%) of WL in Fig. s2 indicated that temperature surrounding water body was relatively high and water body had a cooling effect. IJI (y = 0.32x2 − 19.56x + 354.85, R = 0.32) had positive correlations with LST, indicating the dispersed distribution of WL can weaken its cooling effect to some extent.

LST had negative significant correlation with five metrics of AL (p < 0.05) except MNN (Table 4 and Fig. s3), showing that AL can reduce heat effect. However, the correlation coefficients were smaller between PLAND (y = − 5.49x + 235.74), LPI (y = − 5.24x + 232.99) of AL and LST than those of WL and LST, indicating AL was weaker than WL to relieve heat effect. This can also be confirmed by the majority of AL in SHTZ, MTZ and SLTZ (Fig. 6).

The relationship between LST and patch type indices of GL was similar to that between LST and patch type indices of AL (Table 4 and Fig. s4). In other words, the cooling effect of a large area of GL was stronger than that of multiple small areas of equal area, and the more concentrated distribution and more complex shapes of GL patches were better the cooling effect3,16.

Quantitative correlation between LST and landscape level indices

Table 5 showed the correlation between LST and landscape pattern metrics at landscape level. LST was significantly correlated with seven landscape level indices (p < 0.01), amongst them, LST was significantly negatively correlated with PD, LSI, IJI, SHEI and SHDI, in contrast to LPI, CONTAG and AI. In other words, large PD, SHEI and SHDI values indicated the richer diversity of landscape corresponded to the lower LST; The large the LSI values showed the more complex shape characteristic, the lower LST; The high LPI, CONTAG and AI values showed the lower fragmentation degree of landscapes were consistent with the higher LST. When considering the effect of mixed landscape configurations, the fragmentation degree, shape characteristic and diversity were more important to alleviate LST than other spatial arrangement features. It was noticed some studies also showed a significant effect of landscape metrics in explaining the thermal environment intensity based on fragmentation, shape, connectivity and patch size39,40, for instance, relatively large water body presented a higher cooling effect than equally distributed small waterbodies41,42, while the fragmentations of construction land were conducive to reduce the thermal environment23, these results were obtained from individual patch characteristic of land cover (patch level), not from the overall characteristics of different land cover (landscape level).

Table 5 The correlation between LST and landscape pattern indices at landscape level.


There were strong correlations between LST and urban land uses43. With the rapid urbanization development of Nanchang in the past 20 years in this study, urban heat spots had covered most areas of the city due to increasing in the coverage of impervious surfaces and to decreasing in natural surface, consequently, high heat capacity and less evapotranspiration from surface had existed27,44. Similarly, construction land corresponded to extremely high temperature zone (EHTZ) and high temperature zone (HTZ) in this study. On the other hand, water body was consistent with extremely low temperature zone (ELTZ) and low temperature zone (LTZ). Hence, in landscape planning and design, the landscape of CL should be dispersed as far as possible by the landscape of GL or WL, which can effectively alleviate urban thermal environment. The conclusion was also consistent with the overall characteristics of different land cover (landscape level) in affecting LST.

Some studies had shown a significant effect of landscape pattern metrics on LST39,40. The area proportion of land use was the most important indicator of the contribution on LST. The larger the area ratio and the more concentrated distribution of CL generated the higher LST, whereas the more fragment, dispersed and complex shape distribution of CL help to alleviate the thermal environment. WL, AL and GL Landscapes can be regarded as ecological landscapes, indicating their cooling effects for the entire city44,45. The area percentage of ecological landscapes had the most obvious negative correlation with LST, followed by the fragmentation and the shape indices, showing that compositions and configuration features of landscapes were all important in affecting LST30,46. The multiple regression analysis was carried out between LST and area percents of land uses15, and the contribution value (0.87) of different land use areas to LST can be obtained. However, the contributions of many landscape metrics to the heating/cooling effect are still unclear14,40. Therefore, a careful planning of land use is important based on landscape configurations in order to maximise the benefit of the policy strategy.


  1. 1.

    Heat environment of Nanchang had become more worse with continuous expansion and enhancement of urbanization area in the past 20 years. In 2001, hot spots were mainly concentrated on the old downtown. In 2020, with comprehensive expansion of construction land, hot spots had gradually occupied a dominant position all around the city.

  2. 2.

    LST of different land uses were significantly different, and the order was: CL > BL > AL > GL > WL. CL contributed the main heating effect. GL has a certain transpiration and shading effect, but GL occupied a relatively small area in this study. WL formed sheet surfaces and continuous bands and had a certain land area, therefore, WL had a better cooling effect.

  3. 3.

    The relations of LST and patch indices of land uses showed that the larger area of GL, AL, WL, the smaller landscape fragmentation degree, the more concentrated the distribution, and the more complex shape responded to the lower LST, in contrast to those of CL.

  4. 4.

    The relation between LST and landscape level indices showed that LST was negatively correlated with PD, LSI, IJI, SHEI and SHDI, and positively correlated with CONTAG, LPI and AI. Hence, in order to reduce the surface temperature in the region, a variety of different types of landscapes should be set as far as possible, and the degree of aggregation between different landscapes should be low.