Urban Ecosystem

The Heat Island Effect Response to the Urban Landscape Pattern of Haikou based on the “Source-Sink” Theory

  • LI Yujie , 1 ,
  • FU Hui , 2, *
  • 1. Faculty of Civil and Environmental Engineering, Hunan University of Science and Engineering, Yongzhou, Hunan 425199, China
  • 2. Forestry College, Hainan University, Haikou, Hainan Province 570228, China
* FU Hui, E-mail:

LI Yujie, E-mail:

Received date: 2020-10-30

  Accepted date: 2021-11-12

  Online published: 2022-03-09

Supported by

The Natural Science Foundation of Hainan Province(421MS015)

The Natural Science Foundation of Hainan Province(421QN200)

The Hainan Province Philosophy and Social Science Planning Project HNSK(ZC)(21-126)


The Landsat images of the 2000, 2005, 2010, 2015, 2018 are selected as the data source to retrieve land cover and surface temperature data. The contribution of Sink-Source landscape pattern to the heat island and its ecological effects on urban and rural gradient were analyzed by using Heat Index (HI), Sink and Source Landscape Contribution (CIsink, CIsource) and Landscape Effect Index (LI) in Haikou. The results show that the heat island is concentrated on the West Coast, and in the central urban and Jiangdong New Area; the HI shows a pattern of decreasing value with the following land types: “Bare land>Artificial surface﹥Source landscape>Shrub grassland>Farmland>Sink landscape>Woodland>Water body”. In the central city section, the CIsink and CIsource are relatively large in these five periods. The LI decreases rapidly along the urban-rural gradient, promoting the Urban Heat Island (UHI) to a large degree. In contrast, the suburban area contributes to a lesser degree. Overall, the LI fluctuates, the proportion of mitigating UHI is large, and there is a second peak outside the city center. The existing Source-Sink Landscape contributes the most to UHI in the central urban area, and this contribution decreases along the urban-rural gradient. With the continuous expansion of city-town areas, the proportion of Sink areas has increased along the gradient, and the proportion of Source areas has subsequently declined, resulting in the spatial transfer and diffusion of UHI. Therefore, a UHI mitigation strategy based on the theory of regional landscape systems is proposed here.

Cite this article

LI Yujie , FU Hui . The Heat Island Effect Response to the Urban Landscape Pattern of Haikou based on the “Source-Sink” Theory[J]. Journal of Resources and Ecology, 2022 , 13(2) : 257 -269 . DOI: 10.5814/j.issn.1674-764x.2022.02.009

1 Introduction

Urbanization and the subsequent population growth have resulted in great change in Land Use and Land Cover (LULC). In particular, the shift from natural landscapes to impervious surface has augmented both the release of man-made heat and the absorption of solar radiation from the underlying surface, which has further resulted in the deterioration of the urban thermal environment. High urban temperatures adversely affect energy consumption, outdoor thermal comfort, air quality, and human health (Qi et al., 2019), which leads to a series of ecological environment and social sustainable development issues. An urban landscape pattern is the manifestation of the characteristics of its surface coverage, and it has tremendous impact on the urban thermal environment. Therefore, clarifying the relationship between the urban landscape pattern and the Urban Heat Island (UHI) will further our understanding on the reciprocal feedback mechanism between landscape patterns and the ecological processes.
At present, Surface Urban Heat Islands (SUHI) have been widely used in studies on the relationship between UHI's temporal and spatial landscape patterns. In particular, this work has become a scientific foundation on which to base urban planning and management efforts (Peng et al., 2016). A study by Sheng et al. (2017) in Hangzhou proved that LULC-driven indicators explain UHI better than UHI-driven indicators. Current research mostly focuses on the impact of LUCC on the distribution of UHI. There are three main aspects: 1) The Land Surface Temperature (LST) is calculated for each discrete landscape type. Different LULC types have different LST characteristics (Chen et al., 2016; Du et al., 2019; Liu et al., 2019). The flaw with this method is that the influence of other factors cannot be ruled out (Chen et al., 2012), which increases the uncertainty of the results. 2) The gradation of continuous surface remote sensing indexes (such as NDBI, NDVI and NDWI, etc.) characterizes landscape characteristics (Peng et al., 2016; Estoque et al., 2017a; Li et al., 2017a). This differs from the landscape pattern analysis method, and LST does not always correlate significantly with a single indicator (Wang et al., 2017), which limits being able to study the LST coupling mechanism. 3) The impact of landscape structure on LST is studied (Song et al., 2014; Qiao et al., 2019), but the simple landscape pattern index analysis cannot truly reflect the ecological effects of the thermal landscape pattern. These above methods clarify the structural relationship between UHI intensity and LULC, but they fail to consider the individual impact of LULC on UHI, and they do not evaluate local UHI intensity holistically. Simply comparing LST in urban areas versus in surrounding rural areas is not enough to quantify UHI intensity on a local or regional scale. Taking the quantitative description of landscape type as the end point fails to explore the strength of the UHI effect within the region or place, which increases the uncertainty of analyzing the relationship between the landscape pattern and UHI. Therefore, a regional-scale method for analysis is needed. The “Source-Sink” landscape theory provided a quantitative approach to more accurately study the thermal effect model of landscape structures (Chen et al., 2006). At present, only Lei et al. (2019) has analyzed the coupling relationship between LULC and LST by using statistics of LST and the Landscape Pattern Index via landscape classification. Therefore, this study aims to fill gaps in theoretical and regional research by exploring the response mechanism of long-term urban landscape patterns and UHI effects based on the “Source-Sink” theory.
The concepts of “Source” and “Sink” are often used in air pollution (Lal and Sheel, 2000) and carbon cycle research (Canadell et al., 2007). The source is the origin of the process, and the sink is where the process disappears. In a thermal environment, the heat “Source” would be the type of LULC that produces UHI, and the heat “Sink” would be the type of LULC that helps reduces UHI (Lal and Sheel, 2000). At present, few studies combine UHI with landscape pattern analysis based on the “Source-Sink” theory. Some examples include the following: Li et al. (2013) introduced the Source-Sink Index to discuss the statistical relationship between LULC and LST; Li et al. (2014) discussed the correlation between the Source-Sink Index and the remote sensing index through the moving pane method; Li et al. (2017b) analyzed the spatial distribution of strong and weak sources, and also strong and weak sinks through the moving pane method; Gao et al. (2019) analyzed the spatial differentiation of the Source-Sink Index on the urban-rural gradient. Current research focuses on the spatial heterogeneity of source and sinks across a landscape, but overlooks the overall pattern of temporal and spatial differentiation. In addition, determining the threshold by which UHI is affected by landscape pattern under different variables is more worthwhile than simply discussing the significance of the impact (Liu et al., 2017).
Since the beginning of the new century, with the continuous expansion of urbanization in Haikou, the thermal environment has become increasingly problematic. Multiple clusters have developed along the coast and along the river, leading to the current landscape pattern and UHI. At present, only Lei et al. (2019) has analyzed and explored the coupling relationship between LULC and LST using the statistics of landscape classification LST and the landscape pattern index, so this study will build on that by integrating the “Source-Sink” theory. The main research purposes here are: 1) To introduce the UHI Source-Sink Index, and to analyze the contribution of source and sink Landscape types to the heat island effect along the urban-rural gradient method. 2) To identify the threshold at which the landscape pattern significantly affects UHI at a local scale, to provide a quantitative control range for the landscape components of urban planning, and to guide landscape planning and design to effectively reduce heat islands.

2 Data and methodology

2.1 The study area

Haikou has a land area of 2284.49 km2, and a coastline of 136.23 km. It located in the northern part of Hainan Island at the low-latitude tropical northern edge. Most of it is terraces and plains lie below 100 m. It has a tropical maritime monsoon climate with an average annual rainfall of 2607 mm, more than 2000 sunshine hours per year, and an annual average temperature of 24.3 ℃. The urbanization rate in 2018 was 78.5%. Haikou is the fulcrum city in the national “One Belt One Road Initiative”, which is an effort to construct an international tourism island and international free trade port. This initiative precedes the larger development strategy of “Strengthen Central, Western Control, Eastern Excellence, and Southern Control”, so the pace of urban construction in Haikou has been accelerating. At the same time, the ecological and environmental problems of Haikou have become more serious, adversely affecting the lives of residents and interrupting the sustainable development of the city.

2.2 Data sources

Landsat-5 and Landsat-8 daytime images from USGS (https://landsat.usgs.gov) were acquired by the geospatial data cloud platform. Criteria for image selection includes the following: Low cloud cover in the study area, good weather conditions, similar temperature, and centralized time. The metadata is presented in Table 1.
Table 1 POI data classification statistics
Year Line code Sensor Date Land cloud cover (%)
2000 12446 Landsat_5TM 2000-06-19 13.00
2005 12446 Landsat_5TM 2005-06-17 11.00
2010 12446 Landsat_5TM 2010-07-27 8.00
2015 12446 Landsat_8OLI_TIRS 2015-06-26 0.78
2018 12446 Landsat_8OLI_TIRS 2018-06-21 0.48
Google Earth image with an accuracy of 4.49 m were also used.
Data was collected from Haikou Infrastructure Point of Interest (POI), which is based on Baidu’s online map API interface from November 2018.
The current Land Cover (LC) map of Haikou in 2011 was used here.
The vector maps of the administrative boundary of Haikou were also used.

2.3 Data processing

2.3.1 Land cover

By means of supervised classification, land cover types were classified into 6 categories: Woodland, Shrub land, Farmland, Water body, Artificial surface and Bare land based on ENVI 5.3. The five Kappa coefficients were 0.91, 0.88, 0.87, 0.93 and 0.90, respectively. These images were combined with the corrected Google Earth images and the LC map, and field investigation was carried out to verify the identification blur (Fig. 3e). The overall accuracy reached more than 92%. Finally, five phases of the LC map were generated at a resolution of 30 m (Fig. 1).
Fig. 3 Distribution and density of POI and distribution of field sampling points in research area

Note: The numbers in the legend represent the reclassification results based on kernel density analysis and larger numbers indicate higher kernel density.

Fig. 1 Distribution of source and sink landscapes along the urban-rural gradient in 2000, 2005, 2010, 2015 and 2018.

Note: T1-T21 mean the 21 gradient zones.

2.3.2 LST retrieval

Research by Wang (2018) confirmed that the Single-window Algorithm (SWA) is the most accurate method for inverting surface temperature in Haikou. In addition, because the temperature base station data for Haikou could not be verified, this study relied on the TM/ETM+ SWA and the TIRS SWA to retrieve the surface temperature instead (Wang, 2018). The surface temperature parameters in different periods are derived from the National Meteorological Science Data Sharing Service Platform. Finally, the surface temperatures were cropped based on the administrative boundary of Haikou to generate five phases of surface temperature raster data at a resolution of 30 m (Fig. 2). Now that large areas of clouds and shadows will interfere with subsequent analysis, the abnormally low temperature regions in 2000 and 2005 (the white part within the boundary line in Fig. 2) were cut out.
Fig. 2 LST for 2000, 2005, 2010, 2015 and 2018 in the study area.

2.3.3 Infrastructure POI data processing

POI data is widely used in urban center system identification and feature analysis because POI reflect the urban spatial structure and define the urban center area (Gao et al., 2017; Wang and Xu, 2019). The POI were cleaned to remove duplicates or invalid points, and then they were reclassified (Wang and Xu, 2019) (Table 2) according to the service radius. The resulting data was imported into ArcGIS to create a POI distribution map (Fig. 3a). Next, five types of Kernel density grid maps were generated for each data type according to its corresponding search radius, and the grid calculator was added to obtain a POI Kernel density grid map with a resolution of 30 m. Finally, it was reclassified into 20 levels (Fig. 3b).
Table 2 POI data classification statistics
Service radius (m) POI category Number of effective points Kernel density
search radius (m)
300-500 Catering, Resident services, Education and culture 20074 400
500-1000 Wholesale and retail, Financial insurance, Automobile sales and services 23485 750
1000-1500 Transportation and storage, Public facilities, Commercial facilities and services, Sports and leisure, Accommodation, General hospitals 14884 1250
1500-2000 Health and social security, Agriculture, Forestry, Animal husbandry and fishery, Science and technology services, Scenic spots and golf, Park and squares 1672 1750
2000-3000 Villages, Towns, Areas of interest (university towns and international business districts) 2551 2500

2.4 Methods

2.4.1 Heat island strength calculation

In this study, the Heat Index (HI) was used to define UHI Intensity (UHII) (Li et al., 2018). The temporal and spatial distribution characteristics of SUHI were compared and analyzed longitudinally through standardized processing. The HI calculation formula is:
$HI = {{\left( {Tn - T{\rm{mean}}} \right)} \mathord{\left/ {\vphantom {{\left( {Tn - T{\rm{mean}}} \right)} {T{\rm{mean}}}}} \right.} {T{\rm{mean}}}}$
In the formula, HI represents the relative surface temperature, which is a dimensionless value; Tn represents the surface temperature at the n-th point in the study area; Tmean represents the average surface temperature within the study area.

2.4.2 Source and sink landscape contribution

According to the “Source-Sink” theory, artificial surfaces and bare land that promote UHI are regarded as heat “source” landscapes, and UHI forest land, shrubland, farmland and water bodies are regarded as heat “sink” landscapes. The contribution of these heat source/sinks differs between UHI and UHII. Therefore, the normalized average of all pixels in the source/sink area is taken as its HI. The product of the source/sink area HI and the regional Mean HI (MHI) and percentage determines its contribution to UHI CI (Xu, 2009). The formula for CI is:
$CI = \left( {Ti - T{\rm{mean}}} \right) \times {{Si} \mathord{\left/ {\vphantom {{Si} S}} \right.} S}$
In the formula, Ti and Si respectively represent the source/sink area MHI and total area of a given region; Tmean and S represent the MHI and area of the region, respectively; Heat source areas have CI>0, and heat sink areas have CI<0.

2.4.3 Landscape effect index

Regional UHII depends on the contribution of the regional heat source/sink landscape to UHI, and then the landscape effect index (LI) comes in to compare UHI across different regions. The LI is defined as the absolute value of the CI ratio of the heat source and the heat sink areas. The LI formula is:
$LI = \left| {{{CI{\rm{sink}}} \mathord{\left/ {\vphantom {{CI{\rm{sink}}} {CI{\rm{source }}}}} \right.} {CI{\rm{source }}}}} \right|$
In the formula, CIsink and CIsource respectively represent the sink area in the landscape, and the source area. LI>1 indicates that the Source-Sink Landscape slows the heat island effect; LI=1 indicates that the sources and sinks do not significantly affect the UHI; LI<1 indicates that the Source or Sink Landscape enhances UHI.

2.4.4 Urban-rural gradient analysis

Gradient analysis can explain the ecological response mechanism of UHI to urbanization (Gao et al., 2019), assist in assessing the spatial heterogeneity of UHI along the urban-rural gradient, and affirm the urban-rural gradient differentiation pattern of Source-Sink Landscapes. The geographic center (near the Mingzhu Square) determined by the POI core density serves as the origin for 21 circular buffers with a width of 3 km (Zhang et al., 2016). In each buffer zone, the MHI, CIsink/CIsource and LI were calculated to determine the correlation between MHI and Source-Sink Landscape pattern.

3 Results

3.1 Temporal and spatial differentiation characteristics of heat island source landscape areas

According to the temporal and spatial changes of LC and UHI in Fig. 1 and Fig. 2, combining with the author’s previous research (Li et al., 2020), we can see that with the expansion of towns, the artificial surface of Haikou increased by 2.98 times from 2000 to 2018, which occupied other ecological land. And UHI are also in a trend of continuous spreading and contiguous. The “spreading” expansion in Haikou’s central city has resulted in a spatiotemporal evolution of the heat source/sink area across the urban-rural gradient zone (Fig. 1). It can be observed in Fig. 4 that the area of heat sources has decreased from bands 1 to 7 along the gradient (hereinafter referred to as T) over the five years of measurement, while the area of heat sinks has increased. The heat source and sink areas have greatly changed in T9 and T21, though not in 2005. Over 18 years in T1 to T7, the increase in proportion of heat source area (or the decrease in the proportion of heat sink area) was 5.17%, 25.46%, 39.94%, 22.70%, 20.20%, 19.21% and 14.32%. Then from T8 to T21, the heat source area increased in proportion by 5.47% The fluctuation of about 0.90% indicates that the heat source area spreads and transfers in the following pattern: “central city→urban edge→urban-rural interlaced zone→suburban”. The heat source and sink area changed in proportion the most in T2 and T3: From 2000 to 2005 the heat source area increased by 10.46% and 10.58%, respectively; from 2005 to 2010 the heat source area increased by 12.85% and 16.44%, respectively. Therefore, the increase in the heat source area and the decrease in the heat sink area are manifestations of the continuous expansion of the artificial surface from the city center to the periphery.
Fig. 4 Percentage landscape area change of sources and sinks

3.2 Analysis on the spatio-temporal pattern in the thermal field variation index

Through analyzing the evolution of HI amidst the urban-rural gradient of Haikou, we can understand the temporal and spatial characteristics of its UHI (Fig. 5, Fig. 6). On the whole, as the distance from the city center increased, the MHI also decreased rapidly between T1 and T7 along the gradient in all years. The MHI generally varies less from T8 to T21, except in 2018. After the initial peak, MHI peaked a second time between T19 and T21. The MHI ranges of the gradient bands, from 2000 to 2018, were 0.4858, 0.2842, 0.1903, 0.5527 and 0.6459, respectively. The small change in 2010 may be explained by the large area of exposed farmland that was present at the time the thermal remote sensing image was taken. This reduced the HI value gap. The biggest change in 2018 can be attributed to the considerable area of water between T15 and T18, which had a minimum HI value. The heat island is concentrated in the West Coast, built-up areas, and the Jiangdong New Area, with MHI values decreasing from T1 to T7, and smaller changes from T8 to T21.
Fig. 5 Distribution of HI along urban-rural gradients in 2000, 2005, 2010, 2015 and 2018.
Fig. 6 Changes in MHI along the urban-rural gradient

3.3 Source and sink landscape type and response of thermal field variation index

The contribution of LST to UHI differed greatly among different Source and Sink Landscape types (Zhang et al., 2012). Therefore, the HI characteristics of the different source/sink landscape types were based on Haikou’s UHII and LULC data. From Table 3, it can be seen that MHI follows the pattern of “Bare land > Artificial surface > Source landscape > Grassland > Farmland > Sink landscape > Woodland > Water body”. In light of the low proportion of unoccupied land, the Artificial Surface contributes greatly to UHI. On the opposite side of the spectrum, water bodies have the greatest cooling effect, but the sink strength decreases sequentially in the Woodland, Farmland, and Shrubland.
Table 3 HI characteristics of different Source-Sink landscape types
Year Study area Sink
Woodland Shrubland Water Farmland Source
Artificial surface Bare land
2000 0.0264 0.0092 -0.0076 0.2938 -0.0819 0.0355 0.3763 0.3687 0.4562
2005 0.0158 0.0049 -0.0013 0.1817 -0.0928 0.0226 0.1749 0.1733 0.1929
2010 0.3160 0.3008 0.2854 0.4022 0.2532 0.3219 0.4632 0.459 0.4905
2015 0.1635 0.1198 0.0839 0.3515 0.0112 0.1402 0.4931 0.4846 0.5613
2018 -0.0104 -0.0532 -0.0963 0.1538 -0.0983 -0.0438 0.2542 0.2544 0.2525
In order to further understand the detailed characteristics of HI in different Source and sink landscape types, it was necessary to analyze the temporal and spatial differentiation pattern across this urban-rural gradient. In Fig. 7 it can be observed that the MHI changes consistently in Forest lands and Farmlands. That is, as the distance from the city center increases, the MHI decreases continuously at first, and then fluctuates slowly. The Artificial surface is essentially consistent with the MHI of the overall Source landscape. Save for a few individual values found across the five-stage gradient, the Water body has the lowest HI value of all landscape types, indicating that its cooling effect is the most significant. However, HI fluctuates greatly from year to year. The minimum water body MHI was found in T10, T16 and T19 in 2000, in T15-T17 in 2005, in T14-T16 in 2010, in T9, T10, T14-T16 and T19 in 2015 and in T14-T17 in 2018. The variation is a product of different human factors (water diversion and storage projects, farmland irrigation, aquaculture), climatic factors (rainfall, evaporation, wetland conservation) and the relative proportions of other landscape types that year.
Fig. 7 Changes of mean HI of urban and rural gradients in different source and sink landscapes in 2000-2018
As the distance from the city center increases, the MHI of Source-Sink Landscapes tends to fluctuate and decrease, but the MHI of the sink area is always lower than the MHI of the source area. This is consistent with the changes observed in the MHI across the study gradient zone. Combining Fig. 1 and Fig. 7, it can be seen that urban development has been concentrated in T1-T6 for 18 years. The sink landscape area is small, the mitigation effect on UHI is small, and the MHI value plateaus. The bands from T7-T21 are mainly a heat sink landscape. The overall UHI is relatively weak, but a second peak will appear on some artificial surfaces or where bare land is concentrated. In 2000, T6-T11 had a large proportion of wetlands, reservoirs, ponds, and woodland and a small proportion of artificial surfaces, resulting in low MHI values. In 2005, the MHI minimum was found from T13 to T17, which was further away than in 2000. This shift could be attributed to the urban expansion in the area from T6 to T11 and the occupation of the original ecological space by agricultural reclamation. From 2010 to 2015, the MHI has no obvious minimum, which is consistent with the continuous encroachment of ecological space by agricultural reclamation. In 2018, the minimum was located in T14-T17, which are between Jiazi Town, Sanmenpo Town and Hongming Farm. There are many reservoirs, ponds, irrigated paddy fields and woodlands here. Coupled with factors such as elevation, topography and precipitation, “low valleys” in MHI values appear, which was not expected. In the past 18 years, the “trough” of MHI moved backward at first, and then it gradually disappeared. This phenomenon is inseparable from the increasing intensity of development and the shrinking ecological space. It should be noted that closer to the urban center, the proportion of sink landscape area is smaller and the patches are more scattered, which weakens its cooling effect. Therefore, the sink and source MHI are very similar in T1 at Phase 5. However, as the distance from the city center increased, the difference in MHI between the sink and source area gradually increased as well. Until the convergence landscape dominates, the two MHI gradually converge.

3.4 Analysis of contribution degree of source and sink landscape based on gradient analysis

The LST and landscape index of 21 buffer zones were extracted each year to analyze the contribution of different Source and Sink Landscape types and overall composition to UHI, and to track their temporal and spatial differentiation, as shown in Fig. 8. The CI of the UHI source area (CIsource) and the CI of the UHI sink area (CIsink) have similar spatial characteristics. In 2000, CIsource was positive in all regions and decreased rapidly along the urban-rural gradient. CIsink was mostly negative, and the absolute value of the urban-rural gradient showed a decreasing trend, with a positive value only in T21. The absolute values of both source and sink were extremely large at T2. Then in 2005, CIsource was mostly positive, with the only positive values in T16. At first, it rapidly decreased along the urban-rural gradient (T1-T8). Then it increased (T8-T16), decreased (T16-T18) and increased again (T18-T21). CIsink was mostly negative, with the only positive values appearing in T16 and T21, and with the absolute value generally decreasing along the urban-rural gradient. In 2010, 2015, and 2018, CIsource was positive, and CIsink was negative. The absolute values of both fluctuated and decreased along the urban-rural gradient. The negative value of CIsource in the five-phase gradient occurred because the source area LST was lower than the average LST, and the sink area LST was higher.
Fig. 8 Source and sink landscapes contribution index (CI) and landscape effect index (LI)

Note: The x-axis of all analysis graphs indicates the gradient band number.

The overall CI absolute value (|CI|) changed in the same pattern as the CIsource and CIsink: |CI| showed a decreasing trend along the urban-rural gradient. |CI| was larger in T1-T7, indicating that the heat source dominates in the UHI concentration area. Therefore, the source landscape contributes more to UHI when it’s closer to the urban core area. |CI| was relatively low from T8 to T21 because further from the geographic center, the proportion of Source Landscapes shrinks, while Sink Landscapes gradually dominate. Further, UHI gradually weakens further from the city center, and the contribution of Source Landscapes to UHI likewise continues to decrease. The second and third peaks of |CI| appeared with the concentrated distribution of large residential areas or bare land between T8 and T21. A separate analysis of T1-T7 shows that CIsource and CIsink were almost symmetrically distributed along the urban-rural gradient of the five phases. From 2000 to 2018, the distribution of CI values became more and more balanced in the area from T1 to T7, indicating that urban expansion leads to a dispersion of UHI. The maximum value was at T2 in 2000-2005, but it moved to T6 in 2018, which mirrors the transfer of the UHI concentration to the Jiangdong New Area. As distance from the geographic center increased, the proportion of artificial surface shrank, and the proportion of Woodland, Farmland and Water Body areas grew until it became the foremost landscape type along the gradient after T8.

3.5 Landscape effect evaluation

The LI can effectively express the contribution of the landscape to the UHI within a certain area (Xu, 2009; Li et al., 2013). Depending on the stage of urban development, the spatial change pattern of LI also differs. In 2000, LI was mainly between 0.5 and 1.0, though it was greater than 1 at T1, T5, T13, T15, T17, and T19, which had a mitigating effect on UHI. T19 had the most significant effect. In 2005, LI was mostly between 0.4 and 1.0, with an LI greater than 1 at T3, T5, T9, T10, T11, and T16. In 2010, LI was mostly between 0.8 and 1.2, in T1, T6, T10, T12, T13, with LI>1 at T15 and T20. In 2015, LI was mostly between 0.5 and 1.0, and LI>1 at T3, T5 and T21. However, LI increased sharply at T21, which was linked to the existence of a strong UHI. In 2018, LI mostly fluctuated between 0.9 and 1.1, steadily along the direction of the gradient, with 11 of the gradient bands having LI>1. The gradient zone in the central urban bands (T1-T6) can slow UHI due to the distribution of large areas of green space and water bodies (such as Hongcheng Lake, Jinniuling Park, old airport, Haidian River, West Lake, East Lake and a large number of wetland ponds). In the suburbs (T7-T21), the gradient zone ratio of LI>1 did not increase significantly, though there was a slight continuous increase. This indicates that the Sink and Source landscape area ratio is not the only factor affecting the landscape effect.

4 Discussion

4.1 Variation of thermal field along the urban-rural gradient

Haikou’s coastal strip-shaped urban structure diminishes the overall HI distribution along the urban-rural gradient from one end to the other: The MHI decreases rapidly in the central urban area (T1-T7), then from T7-T21 it fluctuates and has two peaks. This differs from Wuhan (Gao et al., 2019), Bangkok, Jakarta, and Manila (Estoque et al., 2017), which have peaks 8 km, 9 km, 3 km, and 10 km away from the city center, respectively. This shows that urban structure and planning have a profound influence on the urban thermal environment. Therefore, the spatial relationship between sources and sinks should be considered in any subsequent urban planning effort so as to avoid furthering the UHI via the city’s sprawl expansion.

4.2 The driving forces of source and sink landscapes on the urban heat island

In the five-phase data, source landscapes dominate from T1 to T7. Then as the proportion of sources decreases, UHII fluctuates gradually. From T8 to T21 the proportion of sinks increases, and the UHII fluctuates at low values. The spread of sources characteristic of urban expansion has invaded former sink areas, leading to the continuous spread of the contiguous UHI from T1 to T7, and the formation of a new continuous UHI center in the Jiangdong New District. This shows that heat sources and heat sinks are closely related to LC. Source landscapes are mainly composed of construction land with flat terrain, concentrated buildings, dense population, and large overall energy consumption. Sink landscapes are mainly composed of farmland, water bodies or forest, which also serve a variety of ecological service functions. Therefore, the identification of heat sources and heat sinks can assist in classifying urban and rural surfaces and ecological space, and can provide a theoretical foundation for methods of LC management.
We used SPSS 22.0 to analyze the correlation between LI and Sink-source landscape area ratio under urban-rural gradient (Table 4). The analysis shows that: This correlation occurs only in individual years. On the whole, there is no significant relationship between LI and the sink-source landscape area ratio. Further analysis of correlation between MHI and Sink-source landscape area ratio (Table 5) shows that: The Spearman correlation coefficient of MHI and Sink-source landscape area ratio is the most significant, showing a certain level of correlation. This suggests that the landscape pattern of Sink-Source landscape is an important factor affecting the thermal environment effect except the area. This contradicts the conclusion of Gao et al. (2019), which posited that areas with large sink to source landscape ratios have relatively large landscape effect indexes in Wuhan City. More case studies and comparisons are needed to verify whether this finding has universal applicability. If it not, the scope or conditions of its application must be determined. In the future, the research should focus on the response mechanism of Sink-Source landscape and thermal landscape pattern at local scale.
Table 4 Correlation between LI and Sink-Source landscape area ratio
Year 2000 2005 2010 2015 2018 2000-2018
Pearson correlation coefficient 0.24 0.348 -0.352 0.345 -0.539* 0.241*
Kendall correlation coefficient -0.2 0.063 -0.248 -0.524** -0.19 -0.195**
Spearman correlation coefficient -0.219 0.155 -0.355 -0.631** -0.295 -0.245*

Note: ** and * indicate significant correlation at 0.01 and 0.05 levels (two-sided), respectively.

Table 5 Correlation between MHI and Sink-Source landscape area ratio
Year 2000 2005 2010 2015 2018 2000-2018
Pearson correlation coefficient -0.342 -0.460* -0.426 -0.588** -0.449* -0.174
Kendall correlation coefficient -0.390* -0.495** -0.495** -0.514** -0.448** -0.264**
Spearman correlation coefficient -0.517* -0.655** -0.626** -0.631** -0.544* -0.377**

Note: ** and * indicate significant correlation at 0.01 and 0.05 levels (two-sided), respectively.

4.3 Landscape management strategies to alleviate urban heat islands

According to many studies, UHI mitigation measures must be emphasized and implemented systematically as part of the urban planning and policy making processes. Therefore, based on the author’s previous research (Li et al., 2019), the following regional-scale strategies are proposed:
(1) In the central city in bands T1 to T3, wedge-shaped green space and small water bodies are the first choices for construction that will alleviate UHI. A dotted green layout is flexible and can effectively increase the microclimate variety in densely built areas. Alternatively, lines of blue and green spaces can enhance heat exchange and increase urban permeability by forming ventilation corridors. A study in Beijing also showed that the higher the fragmentation of ecological land, the higher the regional temperature (Peng et al., 2016). Accordingly, green spaces and water bodies should be integrated into the existing urban structures of Haikou. For example, riparian zones should have more restrictions protecting against development, artificial water bodies should be introduced in parks, and dot-shaped green spaces should be inserted into old areas. Green belts should be set up downwind of roads, and cold belts should be introduced along the main roads and loops to increase permeability. Cooling points can even be inserted into the central area of the continuous heat island to protect and bolster the available cooling surface at the city’s edge.
(2) Within the main urban area, T4-T9 (Jiangdong Group) and T4-T7 (Changliu Group) offer opportunities for new construction of urban parks or country parks in appropriate areas to avoid magnifying the central urban heat islands. At the same time, a large wedge-shaped green space can be embedded in the main urban area to facilitate heat exchange with the suburbs. For example, the planned Wuyuan River Forest Park, Yongzhuang Reservoir Forest Park, Shapo Reservoir Forest Park, Yulongquan National Forest Park, and Dongzhaigang National Nature Reserve can all serve as wedge-shaped green spaces in future urban green space construction.
(3) The area in the suburbs to the west of the Nandu River, from T5-T13, is not only the development site of the new Satellite City (Mission Hills New Town), but is also home to Shishan Volcanic Group National Geological Park and the vast Yangshan Wetland area. These are important urban ecological “cold sources” which need to be protected.
(4) The area in the suburbs to the east of the Nandu River, from T5 to T21, is the main farming and breeding area. Here, it is necessary to protect the arable land, wetland, water bodies and the shelter forest system to avoid forming future extreme suburban heat islands. At the same time, in the construction of small towns, it is necessary to avoid contiguous development by appropriately retaining ecological spaces and building rural parks.

5 Conclusions

Our research shows that, on the urban-rural gradient, MHI is correlated with the type of source/sink landscape, patch size, temporal and spatial distribution, and mix ratio. As the distance from the city center increases, the effect of the source landscape weakens, and the effect of the sink landscape strengthens. With the continuous expansion of towns, the proportion of source areas is expected to increase along the urban-rural gradient. Source and sink landscapes contribute the most to the UHI in the central urban area. Closer to the city center, the landscape can better alleviate the UHI, but the sources are also stronger. With the continuous expansion of cities and towns, the proportion of sink areas continues to decrease along the urban-rural gradient, thus leading to the spatial transfer and spread of UHI. In this study, the LI was used to evaluate the contribution of source and sink landscapes to UHI in Haikou, and it was found that the landscape in T1-T7 promoted a large proportion of the UHI, while the landscape in T8-T21 alleviated a large proportion of the UHI. Along the urban-rural gradient, there is great heterogeneity between source and sink landscape types and the thermal environment. Further away from the geographic center, the absolute values of the five-phase CIsource and CIsink fluctuated gradually along the urban-rural gradient. The artificial surface contributed the most to UHI. The degree of UHI also decreased along the gradient; closer to the core area, the landscape contributed more to alleviating UHI. Therefore, it is possible and necessary to plan and design an urban landscape pattern to transform and alleviate the heat island effect.

We would like to thank Elizabeth Tokarz at Yale University for her assistance with English language and grammatical editing.

Canadell J G, Kirschbaum M U F, Kurz W A, et al. 2007. Factoring out natural and indirect human effects on terrestrial carbon sources and sinks. Environmental Science & Policy, 10(4): 370-384.

Chen A L, Sun R H, Chen L D. 2012. Studies on urban heat island from a landscape pattern view: A review. Acta Ecologica Sinica, 32(14): 4553-4565. (in Chinese)


Chen K L, Gong J Z, Chen X Y, et al. 2016. The pattern relationship research of green space and surface temperature in Guangzhou City. Ecology and Environmental Sciences, 25(5): 842-849. (in Chinese)

Chen L D, Fu B J, Zhao W W. 2006. Source-sink landscape theory and its ecological significance. Acta Ecologica Sinica, 26(5): 1444-1449. (in Chinese)

Du H Y, Ai J Q, Cai Y L, et al. 2019. Combined effects of the surface urban heat island with landscape composition and configuration based on remote sensing: A case study of Shanghai, China. Sustainability, 11(10): 2890. DOI: 10.3390/su11102890.


Estoque R C, Murayama Y, Myint S W. 2017. Effects of landscape composition and pattern on land surface temperature: An urban heat island study in the megacities of Southeast Asia. Science of the Total Environment, 577: 349-359.


Gao J, Gong J, Li J Y, et al. 2019. Effects of source and sink landscape pattern on land surface temperature: An urban heat island study in Wuhan City. Progress in Geography, 38(11): 1770-1782. (in Chinese)

Gao S, Janowicz K, Couclelis H. 2017. Extracting urban functional regions from points of interest and human activities on location-based social networks. Transactions in GIS, 21: 446-467. (in Chinese)


Lal S, Sheel V. 2000. A study of the atmospheric photochemical loss of N2O based on trace gas measurements. Chemosphere-Global Change Science, 2(3-4): 455-463.

Lei J R, Chen Z Z, Wu T T, et al. 2019. Spatio-temporal evolution and interrelationship between thermal environment and landscape patterns of Haikou City, 1989-2015. China Environmental Science, 39(4): 1734-1743. (in Chinese)

Li B, Wang H M, Qin M Z, et al. 2017a. Comparative study on the correlations between NDVI, NDMI and LST. Progress in Geography, 36(5): 585-596. (in Chinese)

Li J, Zhao T, Zhu W, et al. 2018. Urban heat island effect based on Landsat8 image in urban districts of Chongqing, China. Mountain Research, 36(3): 452-461. (in Chinese)

Li L G, Wang H B, Zhao Z Q, et al. 2014. Characteristics of urban heat island (UHI) source and sink areas in urban region of Shenyang. 2014 Third International Workshop on Earth Observation and Remote Sensing Applications (EORSA). June 11-14, 2014, Changsha, China. IEEE: 62-66.

Li L G, Xu S L, Wang H B, et al. 2013. Urban heat island effect based on urban heat island source and sink indices in Shenyang, Northeast China. Chinese Journal of Applied Ecology, 24(12): 3446-3452. (in Chinese)


Li W F, Cao Q W, Lang K, et al. 2017b. Linking potential heat source and sink to urban heat island: Heterogeneous effects of landscape pattern on land surface temperature. Science of the Total Environment, 586: 457-465.


Li Y J, Fu H, Xu X S. 2019. Construction of regional landscape system in Huaihua from the perspective of Global Tourism. Guangdong Landscape Architecture, 41(5): 52-56. (in Chinese)

Li Y J, Fu H, Zhang Y, et al. 2020. Landscape pattern gradient research on spatio-temporal differentiation characteristics and response mechanism in Haikou. Journal of Northwest Forestry University, 2020, 35(6): 258-268. (in Chinese)

Liu Y X, Pemg J, Wang Y L. 2017. Relationship between urban heat island and landscape patterns: From city size and landscape composition to spatial configuration. Acta Ecologica Sinica, 37(23): 7769-7780. (in Chinese)

Liu Y, Gu J R, Pan H J. 2019. Study on the effect of the landscape pattern on the thermal environment in Chengdu. Hubei Agricultural Sciences, 58(6): 58-69. (in Chinese)

Peng J, Xie P, Liu Y X, et al. 2016. Urban thermal environment dynamics and associated landscape pattern factors: A case study in the Beijing metropolitan region. Remote Sensing of Environment, 173: 145-155.


Qi J, Ding L, Lim S. 2019. Ontology-based knowledge representation of urban heat island mitigation strategies. Sustainable Cities and Society, 52: 101875. DOI: 10.1016/j.scs.2019.101875.


Qiao Z, Huang N Y, Xu X L, et al. 2019. Spatio-temporal pattern and evolution of the urban thermal landscape in metropolitan Beijing between 2003 and 2017. Acta Geographica Sinica, 74(3): 475-489. (in Chinese)

Sheng L, Tang X L, You H Y, et al. 2017. Comparison of the urban heat island intensity quantified by using air temperature and Landsat land surface temperature in Hangzhou, China. Ecological Indicators, 72: 738-746.


Song J E, Du S H, Feng X, et al. 2014. The relationships between landscape compositions and land surface temperature: Quantifying their resolution sensitivity with spatial regression models. Landscape and Urban Planning, 123: 145-157.


Wang H C, Xu H. 2019. Shanghai urban-rural ecotone distribution research based on the POI and kernel density analysis. Journal of Shanghai Jiaotong University (Agricultural Science), 37(1): 1-5, 18. (in Chinese)

Wang L, Zhu Y P, Wei B L. 2017. Research progress on correlation between urban heat island effect and landscape pattern. Environmental Science and Management, 42(11): 156-160. (in Chinese)

Wang Z Z. 2018. Urban heat island effect research and system design based on remote sensing image. Diss., Haikou, China: Hainan Unitersity. (in Chinese)

Xu S L. 2009. An approach to analyzing the intensity of the daytime surface urban heat island effect at a local scale. Environmental Monitoring and Assessment, 151(1-4): 289-300.


Zhang J M, Wang P L, Ma N. 2012. Spatial-temporal evolution of urban heat island effect in basin valley-A case study of Lanzhou City. Scientia Geographica Sinica, 32(12): 1530-1537. (in Chinese)

Zhang Z H, Tu Y J, Li X. 2016. Quantifying the spatiotemporal patterns of urbanization along urban-rural gradient with a roadscape transect approach: A case study in Shanghai, China. Sustainability, 8(9): 862. DOI: 10.3390/su8090862.