Introduction

Wetlands, which are recognized as distinctive ecosystems, link terrestrial ecosystems with aquatic ecosystems (Cong et al. 2019; Salimi et al. 2021). They are also recognized as the “natural gene bank” (Bian et al. 2020; Zhou et al. 2009), and hold immense importance for hydrological regulation, climate moderation, biodiversity conservation, and water purification (Kumari et al. 2020; Liu et al. 2018a; Meng et al. 2017). At the same time, wetlands are recognized as one of the most delicate ecosystems on Earth. It is estimated that over half of wetlands globally have experienced degradation or disappearance as a result of the synergy between anthropogenic disturbances and variability in their ecological surroundings (Davidson 2014; Desta et al. 2012; Kirwan and Megonigal 2013). Wetland areas have been repeatedly occupied and destroyed by China's rapid urbanization and industrialization after its reform and opening up (Zhao et al. 2019; Mao et al. 2018). Concurrently, industrial pollution, domestic wastewater, and aquaculture have hastened the degradation of wetlands (Bian et al. 2020). The growing conflict between urban construction and wetland ecological conservation has been increasingly pronounced in this process. Urban wetlands serve multifaceted functions, such as alleviating the heat island effect and preventing floods and droughts (McLaughlin and Cohen 2013; Xue et al. 2019). They are essential in creating environmentally friendly cities and promoting sustainable development (Peng et al. 2020). Therefore, a comprehensive exploration of urban wetlands' spatial and temporal evolution characteristics, coupled with an in-depth investigation into their driving mechanisms, is crucial to promoting urban wetland management and sustainable development.

The dynamic changes in urban wetlands stem from the synergistic impact of various indicators, categorized into two primary dimensions: ecological surroundings and anthropogenic interventions (van Asselen et al. 2013; Xiong et al. 2023). Ecological surrounding factors include climate, soil properties, and terrain. Climate can impact wetland ecosystems through temperature elevation and alterations in hydrological patterns (Erwin 2009). Precipitation is the primary source of recharge for wetlands, and it has a direct impact on their formation and growth by influencing the distribution of water resources in terms of both time and space (Erwin 2009). At the same time, evapotranspiration and temperature influence the distribution of wetlands by influencing water circulation (Havril et al. 2018). Soil properties affect the ecosystem structure and function of wetlands. Studies have found that the soil's water-holding capacity increases with higher levels of organic carbon content (Masood and Ali 2023; Wang et al. 2020). In contrast, sand has a weaker water-holding capacity than clay, making it easier for water to penetrate and difficult to form a wetland (Herawati et al. 2021; Riaz and Marschner 2020). The soil’s pH level influences vegetation and is a pivotal factor influencing the formation and spatial distribution of wetlands (Liu et al. 2018b). It governs the distribution of relatively depressed landforms within the region, thereby influencing the character of regional water flow (Jin et al. 2017; Job and Sieben 2022). Anthropogenic interventions, such as changes in land use, directly affect wetlands and also indirectly affect them by influencing ecological surrounding factors (Maneas et al. 2019; Newton et al. 2020).

Different levels of urbanization across distinct urban–rural gradients correspond to different levels of human pressure on wetland ecosystems. However, most existing studies on the dynamic change mechanisms of urban wetlands analyze the correlation between wetland dynamics and its drivers from the perspective of the entire region (Long et al. 2022; Wang et al. 2022; Zhang et al. 2021b). This approach may overlook crucial details. Thus, analyzing the response of urban wetland dynamic change to various drivers from the perspective of subdivided geographical space is of great significance. Urban–rural gradient analysis is widely employed in ecological studies (Hou et al. 2020; Inostroza et al. 2019), and has been proven to be an instrumental tool for studying the impact of human intervention on ecosystems (Arnaiz-Schmitz et al. 2018; Kaminski et al. 2021). Dividing urban areas according to urban–rural gradients and exploring the driving mechanisms of urban wetland evolution under these gradients can help formulate more targeted wetland restoration, protection, and management plans. It also contributes to improving the synergy between urban–rural development and ecological conservation.

In the current research on wetland dynamic changes and driving mechanisms, various statistical approaches are frequently utilized for quantitative evaluation, including but not limited to correlation analysis (Lin et al. 2018; Yi and Wang 2021), cluster analysis (Hu et al. 2021), regression analysis (Wang et al. 2022), principal component analysis (PCA) (Zhang et al. 2021a, b), boosted regression tree (BRT) (Li et al. 2020c), and GeoDetector (Li et al. 2022; Zhang et al. 2021c). In particular, GeoDetector is a collection of statistically based methods constructed to detect spatially stratified heterogeneity and identify its drivers (Wang et al. 2010). It is widely employed to detect ecosystems and their driving mechanisms (Wu et al. 2022). In contrast to other methods, GeoDetector can overcome the limitations of conventional statistical analysis methods and does not need linear assumptions. GeoDetector has proven to be an effective tool for discerning the roles of individual factors and their interactions (Wang and Xu 2017). This approach aligns well with the paper's requirements for the quantitative evaluation of the contributions of individual indicators, encompassing both ecological surroundings and anthropogenic interventions, as well as their interactions. However, since GeoDetector is designed to aptly analyze categorical data, continuous data in driving factors need to be discretized (Wang and Xu 2017). Prior studies typically consult the pertinent literature and existing knowledge to determine the appropriate discretization method and the number of breaks, which influence the precise determination of the outcomes (Wang et al. 2023). In contrast, an optimal parameters-based geographical detector (OPGD) model optimizes the spatial data discretization process and the spatial scales of spatial analysis. It can identify the optimal combination of parameters for the GeoDetector model within a specified range and enhance the precision and efficiency of spatial analysis (Song et al. 2020). Hence, to examine the driving mechanism of wetland dynamic changes, OPGD was utilized in this study.

With its rich wetland resources and excellent protection work, Hefei received the certification of “International Wetland City” (Li et al. 2023) at the 14th Conference of the Parties to the Convention on Ramsar. The certification represents the highest global recognition for the ecological conservation of urban wetlands. Over the past few decades, Hefei has seen substantial and rapid development, increasing the risk of the degradation of wetland areas (Li et al. 2020a). However, existing studies on Hefei's wetlands are primarily centered around the Chaohu Lake area (Li and Gao 2016; Ni et al. 2018; Qiu-yu et al. 2022), lacking a comprehensive assessment of wetland dynamic changes and their driving mechanisms across the Hefei region. As a major city in central China, Hefei is poised to play a crucial role in promoting international wetland protection in the future. A scientific and comprehensive understanding of the dynamic change of Hefei's wetlands and their driving mechanisms is crucial for promoting the protection and development of Hefei's wetlands. Therefore, Hefei was chosen as the study area to investigate the dynamic change characteristics of its wetlands. Additionally, this study investigated the impacts of ecological surroundings and anthropogenic interventions, along with their interactions, on wetland dynamics across different urban–rural gradient regions. This study's primary focus was on (1) analyzing the dynamic characteristics of Hefei's wetlands; (2) constructing landscape indicators that divide urban and rural gradients to quantify the characteristics of Hefei's urban and rural gradients at the landscape scale; and (3) investigating the effects of ecological surrounding and anthropogenic intervention factors, as well as their interactions, on wetland dynamics in regions with different urban–rural gradients in Hefei.

Materials and Methods

Study Area

Hefei (116°41′0″–117°53′0″E, 31°30′0″–32°28′0″N) is situated at the core of Anhui Province (Fig. 1) and spans an area of 11,445 km2. Hefei consists of four urban districts, one county-level city, and four counties. Hefei is home to Chaohu Lake, which ranks among China's top five largest freshwater lakes. There are numerous reservoirs, rivers, ponds, and artificial ditches scattered throughout Hefei, making it rich in wetland resources. Hefei currently has 118,200 hectares of wetlands, with a wetland rate of 10.33%. The city experiences a subtropical humid monsoon climate, marked by an average annual temperature of 15.7 °C and an average annual precipitation of around 1,000 mm. In Hefei, climatic conditions manifest substantial heat and precipitation during the summer, contrasting with a temperate winter characterized by significant rainfall. In 2020, the resident population of Hefei grew to 9,369,900 people, its urbanization rate reached 82.3%, and its GDP amounted to 1,004.572 billion yuan. Rapid urbanization poses a severe threat to wetlands. Hence, achieving harmony between economic development and wetland conservation requires a thorough examination of wetland dynamics and the underlying driving forces in Hefei. This is crucial for laying the groundwork for urban planning and safeguarding wetlands.

Fig. 1
figure 1

Location of Hefei

Full size image

Data Sources

The data encompass land use data, socioeconomic data, terrain data, soil data, and meteorological data (Table 1). The land use data utilizes the annual China Land Cover Dataset (CLCD). Compared with existing thematic products, CLCD demonstrated commendable consistency (Yang and Huang 2021). The categories of land use were cropland, forest, shrub, grassland, water, snow or ice, wetland, impervious surface, and barren. Considering that paddy fields have lost numerous many of their ecological functions as artificial wetlands, this study does not categorize them as wetlands. This study amalgamates wetlands and water into a unified category termed wetlands.

Table 1 Data sources used in this study
Full size table

Socioeconomic data consisted of the annual average population density, Gross Domestic Product (GDP), cropland, and impervious surface proportions. Based on land use data, the proportions of cropland and impervious surface were calculated in ArcGIS Pro (version 3.0.1). Terrain data comprises Digital Elevation Model (DEM), slope, and aspect. Slope and aspect were computed using the Spatial Analysis Tools in ArcGIS Pro (version 3.0.1) based on the DEM data. The soil data utilized consisted of clay, sand, silt, organic carbon (OC), and soil pH. Meteorological data included monthly average temperature, precipitation, and potential evaporation datasets. The average temperatures for each decade during the study period were calculated using the raster calculator in ArcGIS Pro (version 3.0.1).

In this study, to ensure consistency across data with different resolutions during analysis, we preprocessed the datasets. Using ArcGIS Pro (version 3.0.1), we used the Resample tool with the NEAREST method to resample all driving factor data to a uniform 30-m resolution, matching the accuracy of the land use data. All datasets were unified into the same geographic coordinates and projection system (geographic coordinates: WGS 1984, projection coordinates: WGS 1984 UTM Zone 50N).

This study is organized into three major sections, as shown in Fig. 2. Step 1: Analysis of dynamic changes in wetlands. Step 2: Identification of the urban–rural gradient. Step 3: Analysis of the driving mechanism.

Fig. 2
figure 2

Overall research framework

Full size image

Dynamic Change Characteristics of Wetlands

Land Use Transition Matrix

The land use transition matrix involves applying the Markov model to analyze changes in land use, a method commonly utilized to describe modifications in the land use patterns of a specific area. This model can quantitatively characterize the dynamics of land use types across different periods (Shi et al. 2019). The following is its calculation formula:

$${S}_{ij}=\left[\begin{array}{cc}\begin{array}{cc}{S}_{11}& {S}_{12}\\ {S}_{21}& {S}_{22}\end{array}& \begin{array}{cc}\cdots & {S}_{1n}\\ \cdots & {S}_{2n}\end{array}\\ \begin{array}{cc}\cdots & \cdots \\ {S}_{n1}& {S}_{n2}\end{array}& \begin{array}{cc}\cdots & \cdots \\ \cdots & {S}_{nn}\end{array}\end{array}\right]$$
(1)

where \(\begin{array}{c}n\end{array}\) stands for the overall count of land use categories, where \(\begin{array}{c}i,j=\text{1,2},...,n\end{array}\). \({S}_{ij}\) represents the area converted from type \(\begin{array}{c}i\end{array}\) to type \(\begin{array}{c}j\end{array}\) during the study period.

Single Land Use Dynamic Degree

We chose the model to investigate quantitative alterations in wetlands (Hu et al. 2021). This model examines numerical shifts in a particular kind of land use at a given location over a specified period. The following is the calculating formula:

$$\text{K}=\frac{{\text{U}}_{\text{b}}-{\text{U}}_{\text{a}}}{{\text{U}}_{\text{a}}}\times \frac{1}{\text{T}}\times 100\text{\%}$$
(2)

where \(\begin{array}{c}\text{K}\end{array}\) stand for the dynamic degree; T is the study period in units of years; and \({\text{U}}_{a}\) and \({\text{U}}_{b}\) represent the acreages of a particular land use type at the start and end of the research period, respectively.

Urban–Rural Gradient

This study used a clustering indicator based on land cover composition and configuration to define the urban–rural gradient (Kaminski et al. 2021). This method integrates landscape pattern indicators, performs clustering to identify gradients, and uses the resulting gradient types to describe administrative units in the study area. More landscape complexity can be captured using this method.

Morphological Spatial Pattern Analysis (MSPA)

MSPA can recognize the spatial topological relationships between target pixel sets and structural elements (Lin et al. 2021). Additionally, there are seven different categories for the identified target pixel sets: core, islet, loop, bridge, perforation, edge, and branch. The literature provides information on the specific meanings and ecological significance of each category (Vogt et al. 2007). MSPA is implemented using the open-source software Guidos Toolbox (version 3.1), which is publicly available at (https://forest.jrc.ec.europa.eu).

This study utilized land use data for this analysis. First, the analysis focused on ecological benefit regions (wetlands and green lands) as foreground, considering other remaining regions as background. Next, the analysis was conducted with developed regions (impervious surfaces) as foreground, with other remaining regions considered as background. Drawing on previous experiences and achievements and taking into account the study area's size parameters, as well as the seven categories regions' numerical patterns (Kaminski et al. 2021), the core type and island type of ecological and built-up areas with a size parameter of 30 m in the MSPA analysis were chosen to represent the key spatial patterns of the city in this study. The towns and streets are selected as the smallest analysis units. Three land use types and four landscape pattern indicators are chosen, and their respective proportions in the analysis unit are calculated to construct a landscape indicator system for urban–rural gradient division, as illustrated in Table 2.

Table 2 Landscape indicators and calculation formulas for dividing urban and rural gradients
Full size table

Principal Component Analysis (PCA)

PCA is a commonly used statistical method for dimensionality reduction. It has the capability to distill numerous indicators into a handful of principal components that reflect the majority of information present in the original data (Bro and Smilde 2014). In this study, considering multiple dimensions, a landscape indicator system for urban–rural gradient division is constructed based on land cover composition and configuration. For example, some indicators exhibit a link between ecological land and the ecological land's core region, which makes statistical analysis difficult (Samuelson and Leadbeater 2018). To address this, PCA is applied to the indicator system to reduce dimensionality and extract the main components that characterize urban–rural gradient features. The transformation formula for extracting principal components is as follows:

$$\left\{\begin{array}{c}{y}_{1}\\ {y}_{2}\\ \\ {y}_{k}\end{array}\right.\begin{array}{c}=\\ =\\ \\ =\end{array}\begin{array}{c}{\mu }_{11}\\ {\mu }_{21}\\ \\ {\mu }_{k1}\end{array}\begin{array}{c}{x}_{1}\\ {x}_{1}\\ \\ {x}_{1}\end{array}\begin{array}{c}+\\ +\\ \\ +\end{array}\begin{array}{c}{\mu }_{12}\\ {\mu }_{22}\\ ...\\ {\mu }_{k2}\end{array}\begin{array}{c}{x}_{2}\\ {x}_{2}\\ \\ {x}_{2}\end{array}\begin{array}{c}+\\ +\\ \\ +\end{array}\begin{array}{c}...\\ ...\\ \\ ...\end{array}\begin{array}{c}{\mu }_{1n}\\ {\mu }_{2n}\\ \\ {\mu }_{kn}\end{array}\begin{array}{c}{x}_{n}\\ {x}_{n }\\ \\ {x}_{n}\end{array}$$
(3)

where sample \(\begin{array}{c}x\end{array}\) is an \(\begin{array}{c}n\end{array}\)-dimensional random vector \(x=\left({x}_{1},{x}_{2}... {x}_{n}\right)\)' composed of \(n\) indicators. There exists a set of vectors \(\mu\) such that vector \(x\) can be transformed into a new \(k\)-dimensional composite vector \(\left({y}_{1},{y}_{2}... {y}_{k}\right)\)' through a linear transformation. The newly obtained variable combinations \(\left({y}_{1},{y}_{2}... {y}_{k}\right)\)' become the 1st principal component, 2nd principal component, and so forth until the \(k\)-th principal component. The proportion of total variance in \({y}_{1}\) is maximized, and the variance of \({y}_{2}...{y}_{k}\) decreases sequentially.

Affinity Propagation (AP)

AP is a clustering algorithm in which the basic idea is to consider all data points as potential cluster centers (referred to as exemplars). It constructs a network by connecting each pair of data points with lines, forming a similarity matrix. Then, by propagating messages (responsibility and availability) along the network's edges, it computes the cluster centers for each sample (Frey and Dueck 2007). This algorithm exhibits significantly lower clustering errors than other clustering algorithms and requires less time (Cui et al. 2019). The AP analysis in this study was implemented using the AP from the sklearn package in Python (version 3.9.7). The algorithmic procedure is illustrated in Fig. 3.

Fig. 3
figure 3

Affinity Propagation algorithm procedure

Full size image

Analysis of Driving Mechanisms of Wetland Changes

For this study, 15 factors that contribute to wetland change were chosen based on regional characteristics, data availability, and previous research findings (Li et al. 2022; Wang et al. 2022; Zhang et al. 2023; Zhang et al. 2021a, b; Zhang et al. 2021c). Table 3 provides detailed information on each driving factor. ArcGIS Pro (version 3.0.1) was used to divide the research region into a 250 m × 250 m grid. The wetland degradation area and driving factor data for each grid cell were extracted as the foundational data for the OPGD model. In RStudio (version 4.3.0), the GD package was utilized. By comparing the factor detection q-values for every driving factor, the OPGD model determines the optimal parameter discretization scheme for each driving factor. This study employed five different discretization methods, including equal breaks, natural breaks, quantile breaks, geometric breaks, and standard deviation breaks, with the number of breaks ranging from 3 to 9. The process is illustrated in Fig. 4.

Table 3 Main parameters of the driving factors used in this study
Full size table
Fig. 4
figure 4

Optimal data discretization process

Full size image

This study utilized the factor and interactive detectors to investigate the driving mechanisms of wetland evolution under different urban–rural gradients. The \(\begin{array}{c}q\end{array}\)-value was used to gauge the explaining power of different drivers on the dependent variable. The formula below is used to calculate q-values:

$$q=1-\frac{1}{N{\sigma }^{2}}\sum \nolimits_{h=1}^{L}{N}_{h}{\upsigma }_{h}^{2}$$
(4)

where \(\text{q}\in [\text{0,1}]\); \(\text{L}\) represent the number of driver layers; \(\text{h}=1, 2,......\text{L}\); \({\text{N}}_{\text{h}}\) and \(\text{N}\) represent the number of units in layer \(\text{h}\) and the entire region, respectively; \({\upsigma }_{\text{h}}^{2}\) and \({\upsigma }^{2}\) represent the variances of the target values in layer \(\text{h}\) and the entire region. A higher \(\text{q}\)-value signifies a more pronounced driving effect of the factor, while a lower \(\text{q}\)-value suggests a weaker effect (Wang et al. 2016).

The interactive detector is employed to detect interactions between drivers. The interactions between two drivers in the interactive detector can be categorized into five types, as shown in Fig. 5.

Fig. 5
figure 5

Type of interactions between two driving factors

Full size image

Results

Wetland Dynamic Changes

In terms of the spatial arrangement of wetlands within Hefei City, Chaohu Lake is the primary feature. Other wetlands are mostly distributed in the southern Huangpi Lake and the northern reservoir area, with some other small wetlands scattered throughout (Fig. 6). In terms of wetland area, there has been a general decline from 1990 to 2020. It first decreased significantly, then increased slightly from 1990 to 2005, followed by a continuous decrease from 2005 to 2020 (Fig. 7). Hefei's wetland area decreased from 1274.56 km2 in 1990 to 1119.37 km2 in 2020, with a reduction of 155.19 km2 in total area and an overall loss of 12.18%. From 1990 to 1995, the wetland area decreased the fastest, from 1274.56 km2 to 1131.82 km2, corresponding to a dynamic degree of − 2.24%. From 1995 to 2000, the wetland recovered slightly, increasing from 1131.82 km2 to 1145.01 km2, with a wetland dynamic degree of 0.23%. From 2000 to 2005, the wetland recovery growth rate increased from 1145.01 km2 to 1191.99 km2, and the wetland dynamic degree was 0.82%. From 2005 to 2020, taking five years as the node, the wetland area displayed a continuing downward trend across the three stages, with slightly different degrees of change, corresponding to dynamic degrees of − 0.38%, − 0.28%, and − 0.59%, respectively (Fig. 8).

Fig. 6
figure 6

Spatial distribution changes of wetlands from 1990 to 2020

Full size image
Fig. 7
figure 7

Changes in wetland area from 1990 to 2020

Full size image
Fig. 8
figure 8

Wetland dynamic degree from 1990 to 2020

Full size image

In this study, the dynamics of Hefei wetlands were investigated over three time periods (1990–2000, 2000–2010, and 2010–2020). The spatial distribution changes are shown in Fig. 9. The changes in wetlands are mostly due to the decrease caused by the transition of wetlands to farmland and the growth caused by impervious surfaces and the transition of farmland to wetlands. The main bodies of wetland loss are mostly large-area wetlands around Chao Lake and scattered small-area wetlands; the main bodies of wetland increase are mainly Dafangying Reservoir and Huangpi Lake. The wetland restoration and reduction area are shown in Fig. 10, and the source of restoration and the transfer destination of loss are shown in Fig. 11.

Fig. 9
figure 9

Spatial distribution map of wetland changes from 1990 to 2020

Full size image
Fig. 10
figure 10

Wetland restoration and degradation area from 1990 to 2020

Full size image
Fig. 11
figure 11

Sankey diagram of land use transformation from 1990 to 2020

Full size image

From 1990 to 2000, the wetland area displayed a downward trend, experiencing a loss of 174.00 km2 and a restoration of 44.46 km2. The loss area was 129.54 km2 more than the restoration area. It was mainly converted into farmland and impervious surface, accounting for 164.81 km2 and 8.92 km2. From 2000 to 2010, the wetland area exhibited an increasing trend, losing 78.12 km2 and recovering 102.37 km2. The restoration area was 24.25 km2 more than the loss area. This was mainly due to the integration of wetland resources for constructing Dafangying Reservoir and the conversion of farmland and impervious surfaces into wetlands. The respective areas are 94.97 km2 and 7.34 km2. From 2010 to 2020, the wetland area displayed a downward trend, experiencing a loss of 109.99 km2 and a restoration of 60.10 km2. The loss area was 49.89 km2 more than the restoration area, mainly converted into farmland and impervious surface, which were 102.03 km2 and 7.90 km2, respectively.

Identifying the Urban–Rural Gradient

Results of the MSPA analysis in Hefei from 1990 to 2020 are shown in Fig. 12. A landscape index system for dividing urban and rural gradients was constructed based on the method described in 2.4.1 above. The PCA results are shown in Table 4. The first three principal components were chosen as inputs to the AP for this study. In 1990, 2000, 2010, and 2020, the first three principal components collectively explained 94%, 97%, 98%, and 98% of the variation in landscape indicators delineating the urban–rural gradient. According to AP clustering data, this study uses the numerical values of urban–rural gradient indicators to identify urban–rural gradient types, divided into four categories: urban areas, agricultural villages, semi-ecological villages, and ecological villages. The spatial distribution of each gradient type is illustrated in Fig. 13. The urban gradient-type regions have the highest proportion of impervious surface, the agricultural village gradient-type regions have the highest proportion of farmland, the semi-ecological village gradient-type regions have the highest proportion of ecological land and cropland, and the ecological village gradient-type regions have the highest proportion of ecological land.

Fig. 12
figure 12

Results of the MSPA analysis from 1990 to 2020 (a–e are based on using ecological land as the analysis prospect, while e–h are based on using impervious surface as the analysis prospect)

Full size image
Table 4 Principal component analysis results from 1990 to 2020
Full size table
Fig. 13
figure 13

Spatial distribution of urban–rural gradient types from 1990 to 2020

Full size image

Independent Effects of Each Driving Factor on Wetland Changes

Since the wetland area in Hefei generally exhibited a declining trend from 1990 to 2020, this study used the reduced wetland area as the dependent variable and the 15 driving factors as independent variables. It utilized OPGD to discretize the continuous data within the driving factors. The optimal discretization of the independent variables and the number of breaks are shown in Table A.1. From a single-factor perspective, the explanatory power of each driver is expressed as the q-value. All q-values in the three sub-periods (1990–2000, 2000–2010, 2010–2020) passed the statistical significance test (p < 0.01) (Table A.1). The q-values of the driving factors for each period within each urban–rural gradient type region are illustrated in Figs. 14, 15, 16 and 17.

Fig. 14
figure 14

Explanatory power (q) of driving factors in urban gradient-type regions on wetland changes

Full size image
Fig. 15
figure 15

Explanatory power (q) of driving factors in agricultural village gradient-type regions on wetland changes

Full size image
Fig. 16
figure 16

Explanatory power (q) of driving factors in semi-ecological village gradient-type regions on wetland changes

Full size image
Fig. 17
figure 17

Explanatory power (q) of driving factors in ecological village gradient-type regions on wetland changes

Full size image

For wetland changes in urban gradient-type regions (Fig. 14), the impervious surface proportion, slope, and aspect exerted significant explanatory power from 1990 to 2000, corresponding to q-values of 0.177, 0.165, and 0.151. While the explanatory capability of pH, OC, clay, and sand exhibited was relatively low, each being below 3%. From 2000 to 2010, OC and sand showed significant explanatory power, corresponding to q-values of 0.390 and 0.347. Furthermore, precipitation, pH, clay, and silt had relatively low explanatory power, all of which were less than 3%. From 2010 to 2020, impervious surface proportion, potential evapotranspiration, and precipitation had notable explanatory power, corresponding to q-values of 0.065, 0.053, and 0.051, respectively. Conversely, pH, OC, clay, silt, sand, and aspect had relatively low explanatory power, all being less than 2%.

For wetland changes in agricultural village gradient-type regions (Fig. 15), silt and the cropland proportion exhibited remarkable explanatory power from 1990 to 2000, corresponding to q-values of 0.437 and 0.106. On the contrary, the explanatory powers of the impervious surface proportion, potential evapotranspiration, and aspect were all less than 1%. From 2000 to 2010, silt, pH, clay, OC, and sand demonstrated notable explanatory power, and their corresponding q-values were 0.506, 0.328, 0.328, 0.327, and 0.321. In contrast, population density, GDP, and aspect had relatively low explanatory power, all of which were less than 2%. From 2010 to 2020, silt had the highest explanatory capability. The q-value was 0.493. However, both the impervious surface proportion and aspect had less than 1% explanatory power.

For wetland changes in the semi-ecological village gradient-type regions (Fig. 16), silt and the impervious surface proportions exhibited significant explanatory power from 1990 to 2000, corresponding to q-values of 0.423 and 0.125. In contrast, the explanatory power of population density, GDP, clay, PH, slope, and aspect was relatively low, all below 2%. From 2000 to 2010, the silt and clay demonstrated significant explanatory power, corresponding to q-values of 0.484 and 0.213. However, the impervious surface proportion, population density, GDP, potential evapotranspiration, and aspect had relatively low explanatory power, all of which were less than 2%. From 2010 to 2020, the silt, elevation, and the proportion of cropland demonstrated notable explanatory power, corresponding to q-values of 0.186, 0.152, and 0.100, respectively. In contrast, the explanatory capability of the impervious surface proportion and OC was relatively low, both being less than 1%.

For wetland changes in the ecological village gradient-type regions (Fig. 17), silt exhibited significant explanatory power from 1990 to 2000. The q-value was 0.437. While the impervious surface proportion, temperature, potential evapotranspiration, and slope had explanatory power below 2%. From 2000 to 2010, silt, elevation, precipitation, and cropland proportion demonstrated significant explanatory power, corresponding to q-values of 0.280, 0.132, 0.107, and 0.134. However, the explanatory capability of impervious surface proportion, population density, and aspect were all less than 3%. From 2010 to 2020, GDP and elevation had notable explanatory power, corresponding to q-values of 0.335 and 0.200. Meanwhile, population density exhibited the lowest explanatory power, corresponding to q-value of 0.035.

In summary, among the human-induced factors, the proportion of cropland exhibits strong explanatory power for wetland changes across all four gradient-type regions. Additionally, impervious surface proportion demonstrates notable explanatory power for wetland changes, specifically in urban gradient-type regions. Among the natural factors, soil-related factors play a significant role in explaining wetland changes in rural gradient-type, semi-ecological gradient-type, and ecological village gradient-type regions. Additionally, climate factors show an increasing trend in explanatory power for wetland changes across all four gradient-type regions.

Impact of the Interaction of Driving Factors on Wetland Changes

The reduction of wetland area is not the result of a single factor and requires further explanation by interactive testing. The results of interactive detection indicated that the interactions between the drivers of wetland evolution in Hefei are mainly bivariate-enhance, with a few nonlinear-enhance, uni-weaken, and nonlinear-weaken, and no independent factor effects.

The interactive detection results for urban gradient-type regions are shown in Fig. 18. From 1990 to 2000, 9.52% of interactions exhibited uni-weaken, primarily involving interactions between meteorological factors and soil factors. In addition, 10.47% of interactions were characterized by bivariate-enhance, and the remaining interactions demonstrated nonlinear-enhance. The interaction between the aspect and the proportion of cropland had the strongest effect, corresponding to a q-value of 0.69. Interactions such as elevation ∩ aspect, aspect ∩ GDP, and aspect ∩ impervious surface proportion were also relatively strong, with explanatory power exceeding 60%. From 2000 to 2010, interactions between clay and precipitation exhibited nonlinear-weaken, and 27.62% of interactions showed uni-weaken. This primarily involved the interactions of clay and precipitation with other factors. The remaining interactions demonstrated nonlinear-enhance. The elevation ∩ GDP showing the strongest effect, corresponding to a q-value of 0.32. Interactions such as elevation ∩ precipitation and aspect ∩ proportion of cropland were also relatively strong, with explanatory power exceeding 30%. From 2010 to 2020, interactions between OC and potential evapotranspiration exhibited uni-weaken. Additionally, 8.57% of interactions showed bivariate-enhance, while the remaining interactions demonstrated nonlinear-enhance. The interaction between slope and impervious surface proportions had the strongest effect, corresponding to a q-value of 0.37. Interactions such as slope ∩ potential evapotranspiration, slope ∩ temperature, slope ∩ GDP, slope ∩ proportion of cultivated land, potential evapotranspiration ∩ proportion of cropland, potential evapotranspiration ∩ impervious surface proportion, and temperature ∩ impervious surface proportion were also relatively strong, with explanatory power exceeding 25%. In this gradient-type region, interactions involving soil factors and various factor combinations were generally weak across all three time periods, and some interactions exhibited a weakening trend.

Fig. 18
figure 18

Interaction results of driving factors in urban gradient-type regions

Full size image

The interactive detection results for agricultural village gradient-type regions are shown in Fig. 19. From 1990 to 2000, 13.33% of interactions exhibited uni-weaken, and 12.38% showed bivariate-enhance, while the remaining interactions demonstrated nonlinear-enhance. The interaction between precipitation and the proportion of cropland had the strongest effect. The q-value is 0.25. Interactions such as elevation, slope, clay, OC, pH, precipitation, temperature, GDP, and proportion of impervious surface ∩ proportion of cropland were also relatively strong, with explanatory power exceeding 20%. From 2000 to 2010, 11.43% of interactions exhibited uni-weaken, and 12.38% showed bivariate-enhance, while the remaining interactions demonstrated nonlinear-enhance. The interaction between elevation and the proportion of cropland had the strongest effect. The interactions involving the soil factor with other factor combinations were relatively strong, with explanatory power exceeding 30%. From 2010 to 2020, 13.33% of interactions exhibited uni-weaken, and 3.80% showed bivariate-enhance, while the remaining interactions demonstrated nonlinear-enhance. The interaction between elevation and the proportion of cropland had the strongest effect, corresponding to a q-value of 0.20. Interactions such as pH ∩ GDP and slope, aspect, sand, clay, OC, pH, precipitation, temperature, and proportion of impervious surface ∩ proportion of cropland were also relatively strong, with explanatory power exceeding 15%. In this gradient-type region, interactions between the proportion of cropland and various factor combinations were generally strong over the three time periods. However, the interaction between OC and each factor combination was mainly uni-weaken.

Fig. 19
figure 19

Interaction results of driving factors in agricultural village gradient-type regions

Full size image

The interactive detection results for semi-ecological village gradient-type regions are shown in Fig. 20. From 1990 to 2000, the interaction between OC and pH exhibited nonlinear-weaken, 12.38% of interactions exhibited uni-weaken, and 8.57% showed bivariate-enhance, while the remaining interactions demonstrated nonlinear-enhance. The interaction between the proportion of cropland and impervious surface proportion ∩ precipitation had the strongest effect, with the q-values of both being 0.26. Interactions involving elevation, slope, potential evapotranspiration, temperature, and GDP ∩ proportion of cropland were also relatively strong, with explanatory power exceeding 20%. From 2000 to 2010, 13.33% of interactions exhibited uni-weaken, and 1.90% showed bivariate-enhance, while the remaining interactions demonstrated nonlinear-enhance. The interaction between precipitation and the proportion of cropland had the strongest effect, corresponding to a q-value of 0.57. Interactions such as aspect ∩ clay and elevation, slope, sand, clay, OC ∩ proportion of cropland were also relatively strong, with explanatory power exceeding 50%. From 2010 to 2020, 11.43% of interactions exhibited uni-weaken, and 1.90% showed bivariate-enhance, while the remaining interactions demonstrated nonlinear-enhance. The interaction between elevation and the proportion of cropland had the strongest effect, corresponding to a q-value of 0.47. Interactions such as elevation ∩ sand, elevation ∩ pH, and elevation ∩ population density were also relatively strong, with explanatory power exceeding 40%. In this gradient-type region, during 1990–2010, interactions involving the proportion of cropland with various factor combinations were generally strong. From 2010 to 2020, interactions involving both the proportion of cropland and elevation with various factor combinations remained strong. Across all three time periods, interactions involving OC tended to exhibit uni-weaken.

Fig. 20
figure 20

Interaction results of driving factors in semi-ecological village gradient-type regions

Full size image

The interactive detection results for ecological village gradient-type regions are shown in Fig. 21. From 1990 to 2000, 13.33% of interactions exhibited uni-weaken, and 5.71% showed bivariate-enhance, while the remaining interactions demonstrated nonlinear-enhance. The interaction between elevation ∩ proportion of cropland had the strongest effect. The associated q-value was 0.19. The interaction between precipitation ∩ proportion of cropland was also relatively strong. The q-value was 0.18. From 2000 to 2010, 13.33% of interactions exhibited uni-weaken, and the remaining interactions demonstrated nonlinear-enhance. The interaction between population density ∩ proportion of cropland had the strongest effect. The associated q-value was 0.59. Interactions involving the proportion of cropland with other factors (excluding silt) were also relatively strong, with explanatory power exceeding 40%. From 2010 to 2020, 13.33% of interactions exhibited bivariate-enhance, while the remaining interactions demonstrated nonlinear-enhance. The interaction between GDP ∩ impervious surface proportion had the strongest effect, corresponding to a q-value of 0.61. The interaction involving GDP with other factors were also relatively strong, with explanatory power exceeding 50%. From 1990 to 2010, the interaction between organic carbon and each factor combination mainly exhibited uni-weaken in this gradient-type region.

Fig. 21
figure 21

Interaction results of driving factors in ecological village gradient-type regions

Full size image

Discussion

Dynamic Changing Characteristics of Wetland Landscapes

The interannual variation in wetland area characteristics may be a consequence of Hefei establishing a balance between economic development and resource conservation. During the research period, Hefei's wetland area experienced significant changes (Fig. 7). In the initial stage (1990–1995), as urbanization and industrialization rapidly advanced, Hefei experienced rapid economic development and population growth, which caused extensive wetland loss due to land reclamation and urban infrastructure construction. (Mao et al. 2018). In the following decade (1995–2005), to foster harmonized economic, social, and environmental development, China implemented the “China Wetland Conservation Action Plan” in 2000 and released the “National Wetland Conservation Planning Outline (2002–2003)” in 2003. These initiatives elevated wetland protection and restoration to the national level. Against this background, Hefei has proactively aligned with the national strategy, which has, to a certain extent, reversed the trend of a continuous reduction in wetland area. Meanwhile, the Dafangying Reservoir, built from 2001 to 2004, integrated northern wetland resources, farmland, and impervious surface, which also explains the significant increase in wetland area in Hefei from 2000 to 2005. Although Hefei has made great efforts to protect wetlands in the following 15 years (2005–2020), urban expansion, continued economic and population growth, and other reasons have resulted in a further rise in water demand (Deng et al. 2023), and as a result, the wetland area once again shows a decreasing trend year by year.

Influence of Ecological Surroundings and Anthropogenic Intervention Factors on Urban Wetland Changes Across the Urban–Rural Gradient

The research findings indicate that there are differences in the drivers of wetland change in different urban–rural gradient-type regions. Moreover, the interactions between factors often exhibit greater explanatory power, suggesting that wetland changes are jointly affected by ecological surroundings and anthropogenic interventions (Zhang et al. 2021a, b).

In urban gradient-type regions, anthropogenic interventions have consistently been the dominant factor driving wetland degradation (Fig. 14). The greater the population density, GDP, proportion of impervious surface, and proportion of agricultural land, the more severe the wetland destruction will be (Wang et al. 2022). Regarding ecological surroundings, compared with the situation before 2010, when terrain factors and soil factors were dominant, the influence of climate factors on wetland reduction increased after 2010. Climate changes such as reduced rainfall, increased temperatures, or extreme weather, intensify the reduction of wetlands to some extent (Li et al. 2021, 2020b). In urban gradient-type regions, anthropogenic interference is most severe. Land composition reveals a lesser extent of cultivated and ecological land, with impervious surfaces dominating. The large impervious surface generated by urban construction activities reduces natural land, gradually weakening the explanatory power of soil and topographical factors in influencing wetland changes in this gradient-type region. Therefore, minimizing the negative impact of anthropogenic interventions is key to achieving the sustainable development of wetlands in urban gradient regions.

In contrast, ecological surroundings, particularly soil-related factors, emerge as the predominant drivers of wetland degradation in agricultural village gradient-type regions (Fig. 16). At the same time, the proportion of cropland among anthropogenic intervention has a strong explanatory power regarding the evolution of wetlands in agricultural village gradient-type regions, while the explanatory power of GDP on the evolution of wetlands has increased after 2010. Agriculture village gradient-type regions have more human interference. Land use has more impervious surface, less ecological land, and the most cultivated land. Concentrated agricultural activities directly affect soil properties and quality (Wang et al. 2022); thus, the influence of soil factors on wetland reduction is particularly notable in this gradient-type region. Hefei’s economy has developed rapidly since 2010. The encroachment and destruction of wetlands are a result of urban sprawl to meet the demands of urban economic development. Urban expansion has caused the encroachment and disruption of wetlands to accommodate the needs of urban economic development (Mao et al. 2018).

The dominant factors driving wetland degradation are natural in semi-ecological and ecological village gradient-type regions (Figs. 17, 18). Unlike agricultural village gradient-type regions, elevation among natural factors has stronger explanatory power in semi-ecological and ecological village gradient-type regions. The semi-ecological and ecological village gradient-type regions have less human interference and less cultivated land. Among the four urban–rural gradient-type regions, the semi-ecological village gradient-type regions have more ecological land, and the ecological village gradient-type regions have the most ecological land. Most ecological land is natural land, and elevation usually determines the height of the terrain and the direction of water flow. Differences in elevation directly affect the distribution and shape of wetlands (Job and Sieben 2022). Therefore, elevation has a greater explanatory power for wetland changes in these gradient-type regions.

Impact, Limitations, and Prospects

This study explored the differences in how wetland changes respond to ecological surroundings and anthropogenic interventions across different urban–rural gradients based on the OPGD model. This study innovatively introduces urban–rural gradient analysis when analyzing the driving mechanism of wetland evolution. This has significant implications for managing wetlands in urban areas subject to varying levels of anthropogenic interventions and for establishing a balance between economic and social development and wetland conservation.

However, our study still has some limitations to be considered in future research. (1) The dynamic change characteristics of wetlands in Hefei were analyzed in this study based on the CLCD dataset. Limited by the land use classification level of the dataset, this study utilized the first-level classification of land use without further subdivision into the second level for wetlands. This might overlook some information, such as the evolving characteristics and driving mechanisms of different wetland types, potentially affecting the comprehensiveness of the research. (2) This study combines previous research results and data availability, and selects 15 driving factors. However, the drivers may not have been chosen comprehensively enough. For example, anthropogenic interventions are measured from four dimensions: population density, economic growth, changes in the proportion of farmland, and changes in the proportion of impervious surface, without considering the impact of government policies and social culture on the evolution of wetlands. (3) Due to data availability, this study uses soil data from the Second National Land Survey in the World Soil Database. However, soil properties may change during the study period, thus affecting the study’s accuracy, but it still provides valuable reference.

In addition, some issues should be discussed in the future. The research results show that the reduction of small-area wetlands in Hefei City is severe (Fig. 9). In the construction process of urban and rural areas, large-area wetlands have received extensive attention due to their outstanding morphological functions and massive impact on people and nature. In contrast, small-area wetlands that are numerous and widely distributed are often overlooked (Liu and Gu 2022). The ecological structure of small-area wetlands is relatively unstable compared to that of large-area wetlands, and rapid urban development poses a significant threat to small-area wetlands (Yuan and Zhou 2022). Small-area wetlands are crucial providers of ecological services; for example, they serve as stepping stones for biological migration and they regulate hydrology and rainfall flooding. They can alleviate the inadequacy of regional ecological resources and the lack of ecological space resulting from the tightness of land resources (Cui et al. 2021; Zhang et al. 2023). Therefore, future studies could further focus on the evolutionary characteristics of small-area wetlands and their driving mechanisms under the urban–rural gradient.

Conclusions

This study analyzed the dynamic changes in wetlands in Hefei and innovatively constructed an urban–rural gradient analysis framework to investigate the driving mechanisms behind wetland dynamics across different urban–rural gradients. First, a single land use dynamic degree and land use transition matrix were used to analyze the dynamic change characteristics of wetlands from 1990 to 2020. Second, combining MSPA, PCA, and AP, we conducted urban–rural gradient identification and delineation. Finally, OPGD was selected to conduct quantitative statistical analysis on the driving forces of factors affecting wetland changes under different urban–rural gradients.

The results indicate that from 1990 to 2020, there have been significant dynamic changes in wetlands in Hefei. The overall trend shows a decrease in wetland area, with a total reduction of 155.19 km2. This decline primarily manifests in the degradation and disappearance of small wetlands, with wetlands mainly converting to cultivated land and impervious surfaces. The most influential human and natural factors on wetland dynamic changes vary in different urban–rural gradients. In urban gradient-type regions, the primary human and natural factors during 1990–2000 were impervious surface proportion and slope, respectively. In comparison, during 2000–2010, they were the proportion of cropland and potential evapotranspiration. For the period 2010–2020, they were impervious surface proportion and potential evapotranspiration. In agricultural village and semi-ecological village gradient-type regions, the predominant human and natural factors for all three periods were cultivated land ratio and silt. In the ecological village gradient-type regions, the primary human and natural factors for the first two periods were also cultivated land ratio and silt, while for the period 2010–2020, they were GDP and elevation. Additionally, the majority of interactions among driving factors exhibited stronger explanatory power.

This study and its findings provide a basis for understanding and conserving urban wetland resources, balancing urban–rural development and wetland preservation. Additionally, it contributes to advancing the achievement of the United Nations’ Sustainable Development Goals by 2030.