Spatiotemporal effects of urban sprawl on habitat quality in the Pearl River Delta from 1990 to 2018

Study area

The Pearl River Delta (112°45′–113°50′ E, 21°31′–23°10′ N) is located in the central and southern parts of Guangdong Province, including the lower reaches of the Pearl River, adjacent to Hong Kong and Macao, and facing Southeast Asia across the sea with convenient land and sea transportation. As shown in Fig. 1, the Pearl River Delta region includes nine prefecture-level cities, namely Guangzhou, Shenzhen, Zhongshan, Zhuhai, Dongguan, Zhaoqing, Foshan, Huizhou, and Jiangmen.

Figure 1

Geographical location of Pearl River Delta drawn in ArcGIS 10.6.

Full size image

Data source

The research framework of this paper is shown in Fig. 2, and the data sources are as follows. Taking the basin as the research unit, the raster data of 30 m and 1 km were analyzed by zoning statistics:

(1)
China’s land-use raster data for 1990, 2000, 2010, and 2018 were obtained from the Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences (http://www.resdc.cn), with a spatial resolution of 30 m. According to land resources and their utilization attributes, the dataset divides land cover types into six first-level categories: cultivated land, woodland, grassland, water area, construction land, unused land, and land reclamation from ocean. The land urbanization rate (LUR) refers to the proportion of construction land in the whole region, which is calculated by dividing the area of construction land by the area of all land use types.
(2)
Raster data of population density (POP) from 1990, 2000, 2010, and 2015 were obtained from the Environment and Resources Data Cloud Platform of the Chinese Academy of Sciences, with a spatial resolution of 1 km. Owing to the stable growth of population density under normal circumstances, the population density data of 2018 were obtained by linear fitting based on POP data from 2010 and 2015.
(3)
Nighttime Light (NTL) raster data from 1992 to 2018 were obtained from the Nature journal data (https://doi.org/10.6084/m9.figshare.9828827.v2) with a spatial resolution of 500 m⁴⁵ Calibration was performed to eliminate the differences in the DMSP (1992–2013) and VIIRS (2012–2018) data, generating a complete and consistent NTL dataset on a global scale.

Figure 2

Research framework.

Full size image

Land-use information TUPU

The land-use information graph is a geospatial analysis model combining attributes, processes, and spaces, which can reflect the spatial differences and temporal changes in land-use types⁴⁶. In its function expression, let the state variables be (pleft( {p_{1} ,p_{2} ,p_{3} , ldots ,p_{n} } right)), and then set p as a function of spatial position r and time t, as follows:

$$ begin{array}{*{20}c} {p = fleft( {r,t} right)} end{array} $$

(1)

where (p) represents land-use characteristics. (1) To realize the spatial description of land attributes, when t is constant, the function relation of (p) changing with (r) is constructed. (2) The process description of land attributes can be realized, and when (r) is constant, the function relation of (p) changing with (t) can be constructed. The combination of these two functions can form a conceptual model of the land-use information graph and realize a composite study of land space, process, and attributes.

Habitat quality

Habitat quality evaluation

We used InVEST-HQ to evaluate the habitat quality in the Pearl River Delta region. Based on land-use types, InVEST-HQ calculated the habitat degradation degree and habitat quality index by using threat factors, the sensitivity of different habitat types to threat factors, and habitat suitability¹⁵. The InVEST-HQ model was co-developed by Stanford University, the Nature Conservancy, and the World Wide Fund for Nature¹⁵. InVEST-HQ has a low demand for data and a better spatial visualization effect, which is widely used in the field of urban ecology^47,48,49. For example, The InVEST-HQ model has been used to assess dynamic changes in habitat quality in Scottish¹¹, China^50,51 and Portugal⁴⁷. Habitat degradation and habitat quality were calculated using the following formulas:

$$ begin{array}{*{20}c} {Q_{{xj}} = ~H_{j} left[ {1 – left( {frac{{D_{{xj}}^{2} }}{{D_{{xj}}^{2} + k^{2} )}}} right)} right]} end{array} $$

(2)

$$ begin{array}{*{20}c} {D_{{xj}} = ~mathop sum limits_{{r = 1}}^{r} mathop sum limits_{{y = 1}}^{y} left( {frac{{w_{r} }}{{mathop sum nolimits_{{r = 1}}^{r} w_{r} }}} right)r_{y} i_{{rxy}} beta _{x} S_{{jr}} } end{array} $$

(3)

where (Q_{{xj}}) is the habitat quality of grid x in land-use type j, (H_{j}) is the habitat suitability of land-use type j, (D_{{xj}}) is the habitat degradation degree of grid x in land-use type j, k is the half-satiety sum constant, r is the number of threat factors, and y is the relative sensitivity of threat sources. (r_{y} ,w_{r}), and (i_{{rxy}}) are, respectively, the interference intensity and weight of the grid where the threat factor r is located, and the interference generated by the habitat. (beta _{x} ,S_{{jr}}) are the anti-disturbance ability of habitat type x and its relative sensitivity to various threat sources, respectively.

The value range of habitat degradation degree is [0, 1], and the larger the value, the more serious the habitat degradation. The value of habitat quality is between 0 and 1, and the higher the value, the better the habitat quality.

$$ begin{array}{*{20}c} {Linear,attenuation:~i_{{rxy}} = 1 – left( {d_{{xy}} /d_{{r,max}} } right)} end{array} $$

(4)

$$ begin{array}{*{20}c} {Exponential,decay:~i_{{rxy}} = expleft[ { – 2.99d_{{xy}} /d_{{r{text{~}}max}} } right]} end{array} $$

(5)

where (d_{{xy}}) is the straight-line distance between grids x and y, and (d_{{r,max}}) is the maximum threat distance of threat factor r.

Five categories of documentation are prepared before using InVEST-HQ: LULC maps, threat factor data, threat sources, accessibility of degradation sources, habitat types and their sensitivity to each threat. Threat sources were divided into Cropland, City/town, Rural settlements, Other construction land, Unused land, and land applications. The maps of threat sources are generated in ArcGIS. For example, in the map of threat sources of cultivated land, the raster value of cultivated land is set to 1, and the raster value of other land types is set to 0. Distance between habitats and threat sources, weight of threat factors, decay type of threats factors, habitat suitability and the sensitivity of different habitat types to threat factors were derived from previous studies in similar regions^{2,25,38,39,50} and user guide manual of InVEST model¹⁵, as shown in Tables 1 and 2.

Table 1 Threat factors and related coefficients.

Full size table

Table 2 Sensitivity of habitat types to each threat factor.

Full size table

Habitat quality change index and contribution index

The CI was used to analyze the causes of the changes in habitat quality, and the following formula was used to qu^{2,25,38,39,50}antitatively represent the contribution of land-use conversion to habitat quality change. In this study, the total value of habitat quality loss caused by land transfer in areas related to construction land expansion from 1990 to 2018 can be expressed as follows:

$$ begin{array}{*{20}c} {CI~ = ~frac{{mathop sum nolimits_{1}^{n} left( {Q_{{ij2018}} – Q_{{xj1990}} } right)}}{n}} end{array} $$

(6)

where n is the grid number of cultivated land transferred to construction land.

To analyze the relationship between land-use change and habitat quality, the HQCI was constructed to describe the mean value of habitat quality reduction caused by land transfer in the areas related to construction land expansion during the study period. The formula is as follows:

$$ begin{array}{*{20}c} {HQCI~ = CI_{{ij}} /S_{{ij}} } end{array} $$

(7)

where (CI_{{ij}}) represents the total value of habitat quality change when land-use type (i) is converted into land-use type (j), and (S_{{ij}}) represents the area converted from land-use type (i) into land-use type (j). The positive and negative values of HQCI, respectively, represent the positive and negative impacts of land-use change on the habitat, and the higher the absolute value of HQCI, the greater the impact.

Correlation analysis

Geographically weighted regression

Based on traditional OLS, GWR establishes local spatial regression and considers spatial location factors, which can effectively analyze the spatial heterogeneity of various elements at different locations⁵². The calculation formula is as follows:

$$ Y_{i} = ~beta _{0} left( {mu _{i} ,v_{i} } right) + sum kbeta _{k} left( {mu _{i} ,v_{i} } right)X_{{ik}} + varepsilon _{i} $$

where (Y_{i}) is the coupling coordination degree of the ith sample point, (left( {mu _{i} ,v_{i} } right)) is the spatial position coordinate of the ith sample point, (beta _{k} left( {mu _{i} ,v_{i} } right)) is the value of the continuous function (beta _{k} left( {mu ,v} right)) at (left( {mu _{i} ,v_{i} } right)), (X_{{ik}}) is the independent variable, (varepsilon _{i}) is the random error term, and k is the number of spatial units.

To simplify the complicated urbanization process, it was divided into three aspects: economic urbanization, population urbanization, and land urbanization according to the existing research³⁸. The NTL, POP, and LUR were used to represent the economic development, population scale, and land urbanization level of the city.

The research unit is a river basin, which has both natural and social attributes. It is a relatively independent and complete system, which can connect and explain the coupling phenomenon of society, economy, and nature⁵³. The hydrological analysis module in ArcGIS was used to divide the research area into 374 small basins. When calculating the cumulative flow of the grid, 100,000 was used as the threshold value, and basins less than 5 km² were combined with the adjacent basins.

Zone classification using the Self-organizing feature mapping neural network

The SOFM neural network was proposed by Kohonen, a Finnish scholar, and constructed by simulating a “lateral inhibition” phenomenon in the human cerebral cortex. It has been widely applied in classification research in geographic and land system science^42,43. The advantages of the SOFM neural network in classifying the coupling relationship between urbanization and habitat quality are as follows : (1) it simulates human brain neurons through unsupervised learning, which is objective and reliable. (2) It maintains the data topology during self-learning, training, and simulation to obtain reasonable partition results and identify the differences between different basins. (3) For massive data, the SOFM network has a good clustering function while maintaining its characteristics and uses the weight vector of the output node to represent the original input. The SOFM neural network can compress the data while maintaining a high similarity between the compression results and the original input data⁵⁴. We exported the data from ArcGIS, and conducted cluster analysis on the four factors of NTL, POP, LUR and habitat quality using SOFM. Finally, the analysis results are imported into ArcGIS for display.

Source: Ecology - nature.com